Advances in Geophisycs, H. E. Landsberg, 1974

ADVANCES IN
GEOPHYSICS
VOLUME 17
Contributors to This Volume
FRANS DEMEYER
FRANCIS E. FENDELL
V. B. KOMAROV
A. QUINET
B. V. SHILTN
Advances in
GEOPHYSICS
Edited by
H. E. LANDSBERG
lnstitute for Fluid Dynamics and Applied Mothematics
University of Maryland, College Park, Maryland
J. VAN MIEGHEM
Royal Belgian Meteorological lnstitute
Uccle, Belgium
Editorial Advisory Committee

BERNARD HAURWITZ R. STONELEY
ROGER REVELLE URHO A. UOTILA
VOLUME 17
I974
Academic Press New York San Francisco London

A Subsidiary of Harcourt Brace Jovanovich, Publishers
COPYRIGHT0 1974, BY ACADEMIC PRESS, INC.
ALL RIGHTS RESERVED.
NO PART OF THIS PUBLICATION MAY BE REPRODUCED OR
TRANSMITTED IN A N Y FORM OR BY ANY MEANS, ELECTRONIC
OR MECHANICAL, INCLUDING PHOTOCOPY, RECORDING, OR ANY
INFORMATION STORAGE AND RETRIEVAL SYSTEM, WITHOUT
PERMISSION IN WRITING FROM THE PUBLISHER.
ACADEMIC PRESS, INC.

111 F i Avenue, New York, New York 10003
United Kingdom Edition published by

ACADEMIC PRESS, INC. (LONDON) LTD.
24/28 Oval Road,London NW1
LIBRARY
OF CONGRESS
CATALOG
CARDNUMBER:
52-12266
ISBN 0-12-018817-1
PRINTED IN THE UNITED STATES OF AMERICA

CONTENTS
LIST OF CONTRDUTORS......................................... vii
Tropical Cyclones
FRANCIS
E . FENDELL
1. Introduction ................................................ 2
2 . Aspects of Tropical Meteorology ............................... 17
3 . Models of a Mature Tropical Cyclone ........................... 29
4 . Theory of Tropical Cyclone Intensification ...................... 75
5 . Concluding Remarks ......................................... 86
Appendix A . Estimating the Kinetic Energy and Water Content
of Hurricanes ............................................... 87
Appendix B. The Moist Adiabat ................................. 88
Partial List of Symbols ......................................... 92
References .................................................... 93
Note Added in Proof ........................................... 100
A Numerical Study of Vacillation

A. QUINET
1. Introduction ................................................ 101
2. The Laboratory Simulation of Large-Scale Atmospheric Flow . . . . . 102
3. A Model Atmosphere for the Study of Vacillation ................ 113
4. The Spectral Dynamics and Energetics of the Model ............. 131
5. The Numerical Study of Vacillation ............................ 149
6. Vacillation in the Atmosphere ................................. 178
List of Symbols ................................................ 182
References .................................................... 184
Filter Techniques in Gravity Interpretation

FRANS
DE MEYER
1. Introduction ................................................ 187
2. Convolution Filtering ........................................ 189
3 . Upward and Downward Continuation of the Surface Gravity
Effect ...................................................... 203
4 . Frequency Filtering ......................................... 226
5 . Calculation of Derivatives of Higher Order ..................... 248
Appendices.................................................... 250
List of Symbols ................................................ 254
References .................................................... 256
V
vi CONTENTS
Aerial Methods in Geological-Geographical Explorations

B . V . SHILINAND V. B. KOMAROV
1. Introduction................................................ 263
2 . Radar Aerial Survey ......................................... 265
3. Infrared Aerial Survey ....................................... 282
4. Aerogeochemical Survey: Remote Sensing of Gases and Vapors . . . . 304
6. Conclusions................................................. 321
References .................................................... 322
SUBJECT ................................................
INDEX 323
LIST OF CONTRIBUTORS
Numbers in parentheses indicate the pages on which the authors' contributions begin.
FRANS
DEMEYER,Royal Belgian Meteorological Institute, Uccle, Belgium (187)
FRANCISE. FENDELL, Engineering Sciences Laboratory, T R W S y s t e m Group,
Redondo Beach, California (1)
V. B. KOMAROV,
Laboratory of Aeromethods, Ministry of Geology, Leningrad,
U.S.S.R. (263)
A. QUINET,Institut Royal Me'te'orologiquede Belgique, Bruxelles, Belgium (101)
B. V. SHILI", Laboratory of Aeromethods, Ministry of Geology, Leningrad,
U . S . S . R .(263)
vii
This Page Intentionally Left Blank
TROPICAL CYCLONES
Francis E. Fendell
Engineering Sciences Laboratory
TRW Systems Group. Redondo Beach. California
.
1 Introduction .......................................................... 2
1.1. The Dangers and Benefits of Hurricanes ............................. 2
1.2. Some Observational Facts on Hurricanes ............................. 8
1.3. Tropical Cyclone Generation ....................................... 3
1.4. Properties of Mature Hurricanes .................................... 5
1.5. Path Prediction ................................................... 12
1.6. The Importance of Tropical Cyclones in the Global Circulation . . . . . . . . . . 13
2. Aspects of Tropical Meteorology .......................................... 17
2.1. Stability in the Tropical Atmosphere ................................. 17
2.2. Tropical Cumdonimbi ............................................. 22
2.3. CISK ............................................................ 26
3. Models of a Mature Tropical Cyclone ...................................... 29
3.1. Introduction ..................................................... 29
3.2. The Carrier Model ................................................ 29
3.3. Critique of the Carrier Model ....................................... 33
3.4. Maximum Swirl Speed Estimate According t o the Carrier Model ........ 35
3.5. The Swirl-Divergence Relation for the Frictional Boundary Layer ...... 45
3.6. The Energetics of the Frictional Boundary Layer and Throughput Supply 50
3.7. The Riehl-Malkus Postulate of an Oceanic Heat Source ................ 53
3.8. The Intensity of a Tropical Cyclone and the Underlying Sea Surface
Temperature ..................................................... 55
3.9. Critique of the Riehl-Makus Model ................................. 59
3.10. Numerical Simulation of Hurricanes on Digital Computers ............. 66
.
3.1 1 Implications of Hurricane Models on Seeding ......................... 70
4. Theory of Tropical Cyclone Intensification ................................ 76
4.1. Carrier’s Outline of Intensification ................................... 76
4.2. Critique of the Carrier Model of Intensification ....................... 78
4.3. The Distribution of Cumdonimbi during Intensification ................ 81
4.4. The Time-Dependent Flowfield during Intensification .................. 82
5 . ConcludingRemarks .................................................... 86
Appendix A. Estimating the Kinetic Energy and Water Content of Hurricanes 87
Appendix B . The Moist Adiabat ........................................ 88
Partial List of Symbols ................................................ 92
References ............................................................ 93
Note Addedin Proof .................................................. 100
1
2 FRANCIS E. FENDELL
1. INTRODUCTION
1.1. The Dangers and BeneJits of Hurricanes

I n an average year the Atlantic and Gulf coastal states of America suffer
over $100 million damage and 50-100 fatalities owing to hurricanes; in a
severe year damage will exceed $1 billion (Meyer, 1971a). Hurricane Agnes
from June 19 to 26,1972 caused damage from Florida to Maine that has been
estimated at $3 billion (Anonymous, 1972), and Hurricane Camille in August
1969 cost 258 lives (White, 1972). The threat consists of high winds [over
200 mph is known (Riehl, 1972a)l;in rainfall [up to 27 in. in 24 hr (Schwarz,
1970)-since 1886 hurricanes have caused over 60 floods in the United States
(Alaka, 1968), and Hurricane Agnes is estimated t o have rained in toto 28.1
trilIion gallons (Anonymous, 1972)]; in storm surges, especiallyalong concave-
ly curving coasts [coastal ocean levels can rise 15-20 ft, with Hurricane
Camille setting the American record at 24 f t along the Gulf coast (White,
1972)];1 and in ocean wave heights [wave heights of 45 ft are known (Riehl,
1954, p. 298)]. On the positive side, on a long-term basis the hurricane-
associated rainfall over the Eastern seaboard states is about one-third of the
total annual rainfall (Penner, 1972); without hurricanes, droughts seem inevi-
table on the west coast of Mexico, in southeast Asia, and elsewhere. About
one to three hurricanes cross coastlines of the United States annually.
Globally, there are about eighty tropical storms annually, of which about
fifty intensify into hurricanes; hurricanes occur in all oceans except the South
For the East Coast of the United States, the hurricane-generated storm surge
depends on storm direction, but can be roughly estimated from s = 2.90-9.70 x
+
( v ) ~ 1.33 x , < (v),,,,~ < 149, where ( v ) is~the
~ ~X 1 0 - 3 ( 0 ) ~ a74 ~ maximum
~ wind speed
in miles per hour and s is the surge above normal in feet [based on Saffir (1973)]. How-
ever, the crudity of such simple empirical relations becomes evident when one recognizes
that the storm surge (which may precede, wcompany, or lag the center of the hurricane)
is sensitive to the storm translation speed and size, as well as direction relative to the
coast; the shelf dope and bathymetry; the radius and magnitude of the maximum wind
speed, and the general radial profile of the swirl; and the central pressure deficit. In
addition t o onshore wind-driven waves, with the arrival of a hurricane, and persisting
often for about a day after its passing along a coast with a sloping bottom (myY,of
inclination angle p), arise so-called edge waves. These waves travel parallel t o the coast;
their crests are normal to the coast. I f the component of the hurricane translational
velocity parallel to the coast is denoted U , the period of the waves is (27rU/g sin p) and
the wavelength is (27rU2/gsin p); typically for the East Coast of the United States the
period is 5-7 hr, and the wavelength is 400 km. The amplitude vanishes rapidly from the
shore seaward, and is negligible at the distance of one wavelength; the amplitude is
roughly given by the inverse barometer rule (one centimeter water rise per millibar of
atmospheric pressure drop) so heights of two t o three feet are typical (Munk et al., 1956)
Sometimes waves generated because of the rate of along-shore hurricane movement
relative to shelf depth cause more flooding than onshore waves.
TROPICAL CYCLONES 3
Atlantic (Atkinson, 1971). Under the current state of the art for path pre-
dictions for either military or civilian use, three times the area actually hit
by a hurricane is typically placed under hurricane warnings (Meyer, 1971a;
Malone and Leimer, 1971).Anderson and Burnham (1973)note that a swarth
650 n miles wide will typically be placed under hurricane warnings; destruc-
tion usually covers about a 250 n miles wide path, but 200 n miles to each side
must be placed under warnings owing t o the current 122 n miles mean position
error in 24 hr forecasts of hurricane landfall. If one assumes 10 landfalls a
season, then Anderson and Burnham estimate a $15.2 million savings by the
20 yo of U.S. Gulf Coast residents who take protective measures, the first
year after the mean position error of landfall is reduced to 70 n miles. I n
the greater Miami area, current estimates suggest that a t least $2 million
are spent on preparations whenever hurricane warnings are issued
(Simpson, 1971). Unnecessary preparation by the United States Department
of Defense installations owing to false hurricane and typhoon warnings costs
$8.3 million annually, exclusive of diversion-of-manpower costs. Thus, the
mere threat of hurricanes incurs expenditure of prodigious sums.
1.2. Some Observational Facts on Hurricanes

"A tropical cyclone starts out as a tropical disturbance in which there is a
slight surface circulation and perhaps one closed isobar. When the wind
increases to about 20 knots and there is more than one closed isobar around
the center, it is called a tropical depression. When the wind rises to more than
34 knots, and there are several closed isobars, it becomes known as a tropical
storm. If the winds exceed 64 knots (74 mph), it is classified as a hurricane or
typhoon or cyclone (depending on location) " (Day, 1966, p. 187). Thus any
low-latitude low-pressure circulation is technically a tropical cyclone;
however, probably to emphasize the physical similarity of all very intense
tropical lows despite the plethora of local appellations, in practice tropicaE
cyclone often refers to the hurricane stage only. Following this practice, the
name tropical cyclone will be mainly used t o refer categorically t o the
hurricane stage; hurricane will also be used as a synonym for variation. The
other local names will be used mainly when geographically appropriate.
The local designations for tropical cyclones (Fig. 1) are hurricanes (North
Atlantic Ocean), typhoons (western North Pacific Ocean), papagallos (eastern
North Pacific Ocean), baguios (Philippine Islands), cyclones (North Indian
Ocean), trovadoes (near Madagascar), and willy-willies [near Australia-a
term in use in the early twentieth century, although now it isreservedfordust
devils (Spark, 1971; Vollsprecht, 1972)].
Annually, especially in the late summer and early fall, some disturbances
over tropical oceans warmed by solar radiation (usually a t least 26"-27"C,
often 28°C and higher) intensify into tropical cyclones, which are typically
TROPICAL CYCLONES 5
a t least a thousand miles in diameter and ten miles in height (i.e. they
extend from sea level to tropical tropopause). Not only do most typhoons
form in the autumn, but also the most intense ones occur then (Brand, 1970a).
These vortical storms are cyclonic in the Northern Hemisphere and anti-
cyclonic (North Pole reference) in the Southern Hemisphere, and take many
days to intensify-indicating that the small Coriolis force is the source of
angular momentum and explaining why intensification within fj0 of the geo-
graphic equator is very rare. I n fact, conservation of angular momentum in
itself indicates that a fluid particle in the tropics drawn in about five hundred
miles will swirl a t several hundred miles per hour.
With satellite photography, inspection of broad ocean expanses has im-
proved, and some former estimates about the frequency of tropical cyclones
have had to be revised (Large environmental-data-gathering ocean buoys,
of which two experimental forerunners are currently deployed by the U.S.
Department of Commerce in the Gulf of Mexico, may eventually play a large
role in hurricane detection.) About three-quarters of the annual global total
of fifty tropical cyclones occur in the Northern Hemisphere. I n the North
Atlantic about 60% of the nine tropical storms that typically occur annually
intensify into tropical cyclones; in the eastern North Pacific, 33% of 14;
and in the western North Pacific, 66 % of 30 (Atkinson, 1971).Since there are
daily disturbances in the tropics in autumn, a weak disturbance has a poor
chance ( 10 yo)of becoming a tropical storm, but any disturbance that does
N
manage to become a tropical storm has a good chance t o become a hurricane

(Palm& and Newton, 1969, p. 503). Since globally just over 60 % of tropical
storms become hurricanes, there appears to be no criterion such that once a
developing depression exceeds it, the depression will definitely become a
hurricane. No one today can infallibly predict which tropical disturbances
will intensify into hurricanes.
1.3. Tropical Cyclone Generation

Hurricanes form where there is sustained local convective activity over
warm tropical seas, with surface temperature 27°C or higher (so that low-level
air lifted on a moist adiabat remains convectively unstable relative t o the
undisturbed ambient up to 12 km); where there is enhanced cyclonic shear (as
occurs when the Intertropical Convergence Zone lies a t a considerable dis-
tance from the geographical equator); and where there is weak vertical shear
of the zonal wind. Each of these three criteria warrants elaboration.
The f i s t criterion regarding stability tends t o be fulfilled only in the fall,
and is discussed in detail in the discussion of tropical-atmosphere stability
(Section 2.1). With regard to the need for increased cyclonic shear as well as
warm ocean temperatures, it is noteworthy that, globally, 65% of the
disturbances that later become tropical storms were first detected between
10" and 20" of the equator, with only 13 % poleward of this region and only
22 % equatorward (Atkinson, 1971). The absence-of-a-vertical-wind-shear
criterion may explain the anomalous cyclone season for the northern Indian
Ocean (in the Arabian Sea, the Bay of Bengal, and even the South China
Sea);in these areas, rather than a single autumnal peak, twin peaks in cyclone
frequency occur in fall and spring, with a relatively uneventful summer
season (Palm& and Newton, 1969). (The displacement during the summer of
the low-level trough northward over land so westerlies generally cover most,
if not all, of the North Indian Ocean, is probably another contributing factor
for the lull between the greater peak of activity in the fall and the lesser
peak of activity in the spring.) Gray (1968) suggests that unless cumulo-
nimbi retain vertical integrity, the atmospheric lightening associated with
them is dissipated by being advected in different directions at different
altitudes (" ventilated "). Climatologically, regional differences in vertical
wind shear within the tropics become most evident in the upper troposphere.
Gray (1968) also asserts that the up- and down-drafts in cumulonimbi
themselves help suppress enhancement of vertical wind shear as baroclinicity
increases during intensification of tropical depressions. Vertical wind shear
a t upper tropospheric levels prevents the higher structure of some typhoons
from getting fully organized, although the structure may be well defined in
the lower troposphere; such so-called shallow typhoons are invariably of
minimal hurricane intensity (Varga, 1971).
Particular features of the North Atlantic hurricane season are now enu-
merated. 'As the peak of the hurricane season approaches, the region where
tropical storms reach hurricane intensity moves eastward from the Gulf of
Mexico and the Caribbean to the Cape Verde Islands; as the peak hurricane
season passes, the spawning ground moves westward again t o the Caribbean
(Meyer, 1971a). Coincidentally, there is a latitudinal movement northward in
the first half of the season, then a retreat equatorward (Riehl, 1954, p. 323).
Cyclogenesis poleward of 20"N is a particular characteristic of the North
Atlantic; in fact, for the past four years the tropical North Atlantic has been
mostly free of hurricanes, which have been forming a t 25"N and higher
(Simpson and Frank, 1972). I n 1972, for example, there were only eight
hurricane days in the North Atlantic Ocean (second lowest total since 1930),
while there were 33 hurricane days in the eastern North Pacific Ocean, and
121 typhoon days in the western North Pacific (21 more typhoon days than
average, and the highest number since 1959). Lower than normal sea-surface
temperature and greater than normal vertical wind shear (owing t o strong
high-level westerlies) have been advanced as possible explanations for the
recent decrease in hurricane activity in the North Atlantic.
At one time discussion of tropical cyclogenesis inevitably evoked discussion
of waves in the easterlies (Riehl, 1954). But in recent years attention has
TROPICAL CYCLONES 7
focused on (1)the role of the ICTZ and (2) twice-weekly autumnal disturb-
ances that begin as large cyclonic sandstorms of 1000 n miles extent over
Africa and that drift westward a t up to 10 knots. There may well be two
different sources c'hurricane seedlings, and if so, the relative rolesof latent and
sensible heat in each may differ (Garstang, 1972).
1. Tropical cyclones tend to form on the poleward side of the equatorial
trough; in fact, 80-85 % of synoptic scale tropical disturbances form within
2"-4" of the equatorial trough on the poleward side. Disturbances are rarer,
smaller, and weaker when the trough is closest to the equator. The trough is
maintained by the CISK process (Section 2.3),since low-level meridional moist
inflow sustains the persistent cloudiness of this perennial low-pressure region.
Further, there islow-level cyclonicshear frominteraction of easterlies poleward
of the trough and westerlies equatorward of the trough. On a zonally averaged
basis around the globe, the equatorial trough annually migrates from 15"N
to 5’5, lagging the solar zenith by about two months as it does so (Riehl,
1972a). Byers (1944) attributes the absence of hurricanes intheSouthAtlantic
largely to the failure of the ITCZ to become displaced south of the equator,
even in February, a t longitudes extending from the eastern Pacific to western
Africa. Agee (1972) has presented an interesting sequence of satellite photo-
graphs documenting a case of tropical cyclogenesis in the vicinity of the ITCZ
in late July, 1972.
2. If it is possible t o correlate tropical North Atlantic disturbanceswiththe
ITCZ, it is also possible t o correlate them with disturbances first formed over
the mountainous east African bulge, that migrate westward (Carlson, 1969).
Half the disturbances over the tropical North Atlantic can be so traced;
further, half of these disturbances can then be traced across Central America
to the eastern Pacific. Actually 75 yoof eastern Pacific storms originate east
of Central America. A significant percentage of midseason hurricanes (August
and September) have African origins; analyses of dust samples taken on
Caribbean isles after hurricane passage reportedly confirm the African origin
of the storm systems (Jennings, 1970). The sometimes turbid disturbances
migrating westward from the arid Sahara may not adjust to the maritime
environment until they are over mid Atlantic. An interesting case documented
by Denny (1972) is an African seedling identified on September 7, 1971, which
became a depression on September 11 and Atlantic Hurricane Irene on
September 18. The system crossed southern Nicaragua on September 20 and
regenerated t o eastern Pacific Hurricane Olivia (948 mb central pressure)
before dying on September 28 near Baja, California.
There are many other correlations of tropical cyclone formation that may
be attempted; for example Carpenter et al. (1972) have recently suggested
a major peak in hurricane formation near a new moon, a minor peak a t full
moon, and minima a t last quarter and several days after f i s t quarter.
8 FRANUIS E. FENDELL
However, the correlations with the ITCZ and African seedlings just discussed
have the advantage of inherent plausible physical mechanisms. Discussion of
modeling of tropical cyclogenesis will be taken up in Section 4.
1.4. Properties of Mature Hurricanes

Tropical cyclones have a structure characterized in the mature stage by a
relatively cloud-free calm eye (winds usually well below 15 mph) of about
10-20 miles radius. The eye is characterized by low pressure a t sea level
(often below 960 mb, infrequently below 910 mb) and high temperatures aloft
(often 10°C above ambient a t the same altitude). Such pressure drops are
particularly spectacular in the tropics, where surface pressure usually varies
by little more than 0.3 yo (Riehl, 1954, Chapter 11). The eye is surrounded
by an eyewall, an approximately ten mile wide annulus of intense convection,
torrential rainfall, and deep thick cloudiness (marked by large numbers of
cumulonimbi). Outside the eyewall are convective rainbands that appear like
pinwheels or logarithmic spirals in some satellite photographs and/or radar
displays taken from above the storm (Figs. 2 and 3).
The principal velocity component in much of the storm is azimuthal (or
tangential). Hurricane force winds begin a t the eyewall and extend outward
50 to 70 miles; winds usually fall t o moderate gale level ( -35 mph) a t dis-
tances of 100 to 150 miles from the center. The vertical component of velocity
is largest in the eyewall. I n the Northern Hemisphere, there is low-level
cyclonic in0ow (below 3 km)and high-level outflow (above 8 km, with the
maximum outflow near 12 km); toward the outer regions (between, say, 200
t o 276 miles from the axis) the outflow turns from cyclonic to anticyclonic
(relative t o an observer rotating with the earth). If one subtracts off the
symmetric flow from the total flow, the outflow layer possesses a horizontal
anticyclonic eddy to the right of the path vector and a cyclonic eddy t o the
left; these are seemingly due to ambient streaming around the high-level
outflow (Black and Anthes, 1971). I n the midtroposphere (say, 3 t o 8 km
above 8ea level) there is little radial flow. While there is much spray, thelowest
few hundred feet (at least) of the inflow layer remain cloud-free in as far as
the eyewall (Riehl, 1954).
It is often agreed that there is slow downward motion in the eye and in the
outer regions of the storm t o balance the strong updrafts in the eyewall;
that radiational cooling of about 1.5"C/day attends sinking in the outer,
less cloudy regions; that, either directly or indirectly, the release of the latent
heat or condensation reduces the density in the eyewall t o establish a large
radial pressure deficit relative t o ambient conditions, from hydrostatic
considerations; that a still further pressure deficit from ambient occurs in the
eye owing to roughly dry adiabatic recompression of air that has arisen along
TROPICAL CYCLONES 9
FIG. 2. NASA photograph of Hurricane Gladys west of Naples, Florida taken from
Apollo 7 on October 7, 1968. The maximum wind speed was 65 knots at this time, and
the tropopause was at 54,000 ft.
the near moist adiabat vertical sequence of states that characterize an ap-
preciable portion of the eyewall; and that a cyclostrophic balance (balance of
radial pressure gradient and centrifugal force), together with an algebraic
decay of the swirl with increasing radial distance from the axis, yields on the
average a fairly good estimate of the maximum swirl speed.
Outside the eyewall, the precipitation rate falls off roughly linearly with
distance from the center of the tropical cyclone. Cloud cells in the eyewall
are typically 5 to 20 km thick and a few of the smaller ones can rain over
15 cm/hr. Spiral bands out to 150 km yield 1 cm/hr, and further out 0.25 cm/hr
-though individual convective cells of 1 km horizontal scale can give much
heavier rainfall. I n the low rainfall area the precipitation is probably snow
FIG.3. NOAA photograph of mature hurricane revealing lower-level rainbands

outside the region of overcast (OjC) and upper-level cirrus clouds of the outflow.
that turns t o rain a t the melting level (Meyer, 1971a,b). Incidentally, clouds
in a hurricane are probably strained by the rapid swirl into the spiral band
pattern seen on radar screens or in satellite photographs The spiral bands
give visualization t o parts of the strain pattern, rather than streamline
pattern. A rainband persists typically for one to two hours (Gentry, 1964).
I n addition to rainfall, storm surge, large waves, and high winds, tropical
cyclones can spawn tornadoes and waterspouts of moderate intensity (Orton,
1970). These are usually reported for the region outside the domain of hurri-
cane winds, but this may only mean that twisters are more easily discernible
from the general vortical intensity in the outer portions. One would expect
tornadoes t o be most closely associated with the intense convection and large
swirl of the inner rainbands. However, the fact remains that tornadoes have
been observed mainly in the outer portions (i.e., where the surface pressure
exceeds 1000 mb), soon after landfall of a formerly intense, rapidly decaying
hurricane. Perhaps the intrusion (sometimes associated with the occurrence
of tornadoes) into the storm of midtropospheric ambient dry air is significant.
Most tornadoes are reported for the right forward quadrant, with respect to
an observer looking along the direction of translation. This seems plausible
because for North Atlantic hurricanes that quadrant is also the most severe
TROPICAL CYCLONES 11
with respect t o rainfall, rainbands, and winds (Hawkins, 1971).As previously

noted, only near the center are tropical cyclones axisymmetric to good approxi-
mation; near the outer edges there is asymmetry. The additive translational
contribution t o the aximuthal winds has been cited as one phusible source of
the asymmetry in wind speeds (Riehl, 1954, p. 290). But Riehl also notes that
the largest radial inflow occurs in the right forward quadrant once trans-
lational effects are subtracted out. It may be remarked that the so-called
beta-plane effect (variation of the Coriolis force with latitude for fixed
velocity) is not negligible over a system as large as a hurricane, which is
normally at least of diameter 1000 miles.
The longest recorded lifetime for an Atlantic tropical cyclone was Ginger
(September 5-October 5, 1972); of this 31-day lifespan, 20 were spent as a
hurricane. Previously, Carrie (1957) had lasted 18 days as a hurricane and
Faith (1966) had lasted 26 days as a system (Simpson and Frank, 1972).
An unusually long-lived tropical cyclone in the western North Pacific was
Rita (July 6-27, 1972); of this 22-day lifespan, about 18 days were spent as a
typhoon.
Hurricanes tend to wea.ken moderately rapidly over land. The central
pressure of Camille (1969) rose from 905 to 990 mb in about 13.5 hr after
landfall (Bradbury, 1971)? Agnes (1971). a minimal hurricane with 986 mb
central pressure and maximum winds of 75 mph with gusts t o 95 mph over
the Gulf of Mexico, was downgraded to a tropical storm eight hours after
crossing the Florida panhandle (Anonymous, 1972). In an interesting report
Grossman and Rodenhuis (1972) cite Hurricane Able (1952) and Hurricane
Diane (1955) as examples of hurricanes which only weakly interacted with
their environment and still maintained appreciable energy, circulation, and
precipitation rates after passing inland. For example, 24 hr after Diane passed
inland over the Carolinas on August 17, 1955 the central pressure had risen
from about 985 mb to 1000 mb, and the core rainfall had decreased from 9.1
cm/day to half as much. Convective instability in the core decreased, and
maximum rainfall occurred 75 miles from the center. Kowever, in the next
24 hr the central pressure began to fall a few millibars and the central rainfall
at the center of the storm began to increase; the convective instability in the
core increased again. Hurricane Diane, although inland, encountered no
major orographic changes during this time while moving northeastward.
2According to data collected on 30 typhoons for 1960-1970, tropical storms ap-

proaching the Philippines with maximum winds in excess of 90 knots leave with winds
reduced by 4CL.50yo,while a storm wit,h peak winds less than 90 knots experience only a
10-15 yo reduction recovered within a day after leaving (and storms with peak winds
under 60 knots typically undergo an intensification in crossing the islands). The average
crossing time is 14.5 hr, with the weaker storms crossing more quickly (Brand and
Bellock, 1972).
Grossman and Rodenhuis emphasize the uncertainty in numerical modeling

of the role of latent heat release in complicated situations in which soundings
indicate layers of stable and of convectively unstable air (Section 3.10). This
lack of knowledge renders detailed analysis of how tropical cyclones die inland
dif6cult.
Matano and Sekioka (1971),after examining several typhoons near Japan,
seem to suggest that a tropical cyclone gradually decays without strong
interaction with other atmospheric systems if it moves into extratropical
regions where there is no marked midlatitude baroclinicity. If there is marked
baroclinicity, the tropical cyclone will tend to interact with extratropical
cyclones, either preexisting or else induced on fronts passing close to the
typhoon core.
1.5. Path Prediction

!Fropica1 cyclones move westerly in the trades, usually at 15 to 26 mph,
before turning poleward, often at greatertranslationalspeeds.Except when the
westerlies are furthest north in midsummer, one frequently observes a re-
curvature eastward in the midlatitudes along the westernsideof high-pressure
cells (Riebl, 1972a); usually a decrease in intensity follows a recurvature
(Riehl, 197213). Not only does the latitude of recurvature (if it does occur)
move poleward and then retreat equatorward as the peak typhoon season
arrives and passes, but also the longitude of recurvature tends to move east-
ward and then retreat westward as autumn comes and goes; further, within
two days after recurvature, the storm accelerates typically to two to three
times its translational speed at the point of recurvature (Burroughs and
Brand, 1973).However, there are so many special circumstances that many
exceptions could be cited to virtually every generalization about path. For
example, coexistent tropical cyclones of comparable size and intensity in the
same hemisphere rotate about one another (Fujiwhara effect) (Brand, 1970b),
while binary systems in different hemispheres tend to move parallel (Cox
and Jager, 1969)-as suggested by classical hydrodynamic potential flow
theory for line vortices. Prior p&ssage, within a month, of a previous cyclone
can also have an effect (Brand, 1971). Tropical cyclones can interact with
extratropical cyclones; some typhoons are large enough to alter anticyclonic
highs, the result on path being similar to that just cited for coexistent typhoons
in different hemispheres (Palm& and Newton, 1969). Riehl (1954, p. 289)
notes that hurricanes produce long waves that travel three times as fast as the
storm and provide early warning. When hurricane swell arrives at a coastline,
the normal wave frequency of 10 to 15 per minute, is reduced to 2 to 4; the
wave direction can be interpreted to yield the path.
Tropical cyclone path prediction is in an imperfect state. Since forecast
errors involving the intensity, rainfall, and movement of hurricanes can
have serious consequences, the subject is pursued here briefly. For 1971,
short term (12-36 hr) path forecasts by the National Hurricane Center were
best made by the use of past climatological and analogue data. (What did
previous hurricanes in a similar situation do?) Methods based on historical
data are referred t o as “objective.” For 1971, for long term (48-72 hr), NHC
found methods based on dynamical principles superior (R. Simpson, private
communication). Today the average error for 24-hr predictions of hurricane
movement is 129 n miles; the average landfall error for a 24-hr prediction is
about 100 n miles. The reduction is due to the closer monitoring of tropical
storms as they approach the Atlantic and Gulf coasts (Simpson, 1971). The
less accurate forecasts often entail faster moving storms, and storms a t
latitudes poleward of the trades, where recurvature may occur.
Many dynamical techniques treat the tropical cyclone as a point vortex
steered in a current, which has been smoothed t o remove the influence of the
circulation of the storm itself. The steering current may be the flowat a specific
level, usually in the mid- or upper troposphere [Byers (1944, p. 447) cited
10,000 ft]. More recent work emphasizes the use of a steering layer; Riehl
(1954, p. 345) advocated the use of a pressure weighted mean flow from the
surface to 300 mb, over a band 8” latitude in width centered on the storm, to
predict hurricane direction and speed. Today prognostic flow is sometimes
used t o predict path. Further, the barotropic model of Sanders and Burpee
(1968) averages over the depth of the troposphere fron 100 to 1000 mb, and
does not involve reduction of the tropical cyclone to a point vortex. Despite
such improvements, the outlook for reducing path forecast errors by more
sophisticated dynamical prediction models is not favorable:
Numerical models for predicting the movement and development of hurricanes
remain a frail source of guidance to the hurricane forecaster for three reasons.
First, an error in direction of movement as small as 8 t o 10 degrees-nominally
a n acceptable one in a 24-hour forecast for extratropical storm centers-can yield
disastrous results in hurricane warnings if followed literally. Second, the perform-
ance of most hurricane prediction models depends significantly upon the initial
direction of movement of the center, which in turn depends upn an exact know-
ledge of the current center position and the position 6 and 12 hours earlier. The
average positioning error is more than 20 nm, and often leads to initial direction
errors of 15-20 degrees. Finally the forecaster remains hard put t o identify and
diagnose the frailties of numerical prediction models for individual forecasts. All
too often this has led to near abandonment of the guidance materials and the
application of empirical and individual experience factors in decision making.
(Simpson, 1971, p. 1.)
1.6. The Importance of Tropical Cyclones in the Global Circulation

Lorenz (1966, p. 409, 418) suggested that. hurricanes were of secondary,
rather than primary, importance in the global circulation of the atmosphere
(Fig..4). However, he noted that it was unclear how great a role hurricanes
DIRECT FERREL CELL DIRECT HADLN

CELL INDIRKT
DIRECT 1 LATITIBE
lo G R ONTAL ZONE
CIRCULATION
FIG.4. Schematic features of the atmosphere i n winter. Latent heat evaported from
tropical oceans to the lower-level trades is carried toward the ITCZ by the sketched
meridional flow; the air rises as the ITCZ is approached, and the latent heat ultimately is
released as sensible heat and gravitational potential energy. Height of the local tropo-
pause is marked. Portions of the general circulation are less well d e h e d in summer. TA
denotes tropical air; MLA, niidlatitude air; PA, polar air; STJ, subtropical jet stream;
and PFJ, polar front jet stream. (By permission from PalmBn and Newton (1969,
p. 669), Academic Press, New York).
play in maintaining the currents of larger scale. He also noted that hurricanes
often do not appear in numerical simulations of the general circulation on
advanced digital computers. Of course, appearance is hardly to be expected
when restraints on computing time severely limit resolution; for example, the
Mintz-Arakawa model is currently treated for a 5-deg longitudinal and a
4-deg latitudinal spatial grid with a 6-min time step, such that about 20 min
of computer time is needed to simulate a day of climate (Rapp, 1970; Gates
et al., 1971). The resolution for disturbances of dimension less than lo00 n
miles is poor, but two weeks of computer time per day of simulated climate
would be required on most computers if satisfactory resolution on a 100 n
mile scale were sought. Meteorologists a t the Rand Corporation in Santrt
Monica, California planned to program the two-level Mintz-Arakawa model
of the general atmospheric circulation for the advanced ILLIAC IV com-
puter. Even if such plans are realized, a grid no finer than 1" x 1" seems
practical; thus, if and when hurricanes are added t o the global circulation
model, their role may have to be introduced by parameterization guided by
more fundamental studies (such as a subroutine with finer grid activated on
appropriate occasions). After all, cumulus convection, radiational cooling,
and turbulent mixing must currently be parameterized into computer simu-

lations of tropical cyclones themselves.
Manabe et al. (1970), in their computer simulation of the global circulation,
do observe the two-week intensification of tropical disturbances of scale 2000-
3000 km near the ITCZ in regions where tropical cyclones are observed;
further, there is low-level convergence and upper-level divergence. However,
the resolution is not great, the pressure fell only to about 980 mb, and the
disturbance appears more like subtropical cyclones (warm core above 800 mb,
especially above 400 mb, and cold core below 800 mb). Appreciable doubt
remains whether these disturbances are to be identified as hurricanes; Bates
(1972) discusses possible inadequacies in the parameterization of cumulus
convection adopted.
Some observational evidence has been presented that in certain regions
during certain seasons hurricanes do play a role in the global circulation. From
interpretation of satellite photographs of tropical storms and typhoons for
1967-1969, Erickson and Winston (1972) have suggested that substantial
heat and moisture transfer from tropical cyclones to midlatitude westerlies
plays a role in the autumnal build-up of the planetary scale circulation. A
tropical storm is believed to transport energy from the lower to upper tropical
atmosphere; the energy is then carried to fronts in the midlatitudes in broad
extensive cloud bands extending northeastward over 20" latitude. The warrn-
ing so conveyed to the extratropics intensifies the midlatitude zonal westerlies
around the 300 mb level. Satellite photography suggests that unless upper-
tropospheric westerlies favor a heat-exporting outflow from the cyclone
toward the subtropics, as evidenced by a n upper-level cloud plume emanating
from the storm, intensification can be suppressed and the pressure deficit
abruptly reduced. When the Hadley cell circulation is weak in late summer,
perhaps hurricanes do serve locally as a substitute mechanism for conveying
energy from the tropics to the extratropics.
That hurricanes could locally and seasonally play such a role seems a t least
possible. First, some models of hurricane structure require for self-consistency
an export of heat from the storm to the rest of the atmosphere. For example,
various models require an export of roughly loz6ergslday (Palm& and Riehl,
1957; Erickson and Winston, 1972);unless this energy (in the form of sensible
heat), produced by transformations within the hurricane, is removed by
other members of the general circulation via high-level advection on the peri-
phery of the storm, the models break down (Riehl, 1954, p. 338; Palm& and
Riehl, 1957, pp. 158-159). Anthes and Johnson (1968), by applying the theory
of available potential energy, estimate that about 4 x lo2*erg/day are con-
tributed by a hurricane to the global scale. Second, a hurricane has been
estimated to possess a kinetic energy comparable t o that of a hydrogen bomb,
or 4 x erg (Battan, 1961; see also Appendix A), and one has rained
over 95 in. on one location in four days (Silver Hill, Jamaica in November,
1909) (Alaka, 1968). The daily production of condensed water for precipita-
tion in a mature hurricane has been estimated at 1.6 x l0ls gm (Ooyama,
1969, p. 29). A weather system with such energy and water content, that
extends radially hundreds of miles and vertically from sea level to the
tropical tropopause, and that persists for weeks, would seem no local accident.
Third, hurricanes occur annually, mainly in the autumn after the long heating
of the tropical oceans by solar radiation, and hurricanes do turn poleward
after drifting westward in the trades. These facts invite the previously stated
speculation that hurricanes &reseasonally part of a substitute mechanism for
relaxing energy poleward when the Hadley cell mechanism is not sdlicient.
If so, then the paradox discussed by Bates (1972, pp. 2, 14) is perhaps res-
olved:‘‘ TheHadley cell of the summer hemisphere is weak or non-existent. ...
There seems to be a paradox in the fact that while the Hrcdley cell is most
intense in winter, the frequency of oceanic tropical disturbances, which one
would expect to be an important contributor to its rising branch, is greatest
in summer.’’ Even if temporally and longitudinally varying eddy transfer
associated with pressure troughs, rather than the zonally symmetric mean
meridional (Hadley) cell, is the primary mechanism by which moisture and
angular momentum are transported from the tropics to the midlatitudes
(Riehl, 1954, Chapter 12; Riehl, 1969a), still, a significant portion of the eddy
transfer occurs in the upper troposphere. The mode by which quantities to be
transferred are convected to appreciable height in the tropics is cumulonimbus
clouds, and, as discussed later, these occur in abundance in hurricanes. The
basic suggestion remains: hurricanes may play some role in the fall in the
export of energy from the tropics to the midlatitudes by what they export
themselves, and by what they convect to the upper tropospheric levels in the
tropics for other mechanisms to export.
However, on an annual and global basis, vertical transfer in the tropics is
carried out by smaller scale, but more numerous convective systems (PalmBn
and Newton, 1969, p. 572). In fact, it is now suggested that only about one
percent of the total energy annually exported from the tropics to the mid-
latitudes is conveyed by hurricanes. For convenience, suppose in a year there
are 100 hurricanes, each lasting 18.25 days and covering lo6 miles2. The
surface area of the tropics (with 30” serving as the limit) is about lo8miles2.
The fraction of the total annual total static enthalpy H transferred by
hurricanes is
R = ( 100)( 18.25)lo6(aH/az)o , h/( 365)10*( aH/az)o, ~
where the numerically unassigned quantity in the denominator is the ambient

sea level transfer rate, and that in the numerator is the rate above ambient
associated with hurricanes (the transfer coefficients have been cancelled in
numerator and denominator). According to Carrier et al. (1971), the ratio of
the two unassigned quantities is about 0.2, so R 21 0.01. The value of R would
be larger if the model of Malkus and Riehl (1960) were used. These authors
assert that a t the sea surface in a moderate intensity hurricane (966 mb central
pressure) the Bowen ratio (ratio of sensible t o latent heat transfer) is about
0.2, and that the latent heat transfer is augmented by a factor of 10-12 over
the ambient level in the trades.
2. ASPECTSOF TROPICAL
METEOROLOGY
2.1. Stability in the Tropical Atmosphere

Stability in the tropical ambient atmosphere introduces consideration of
circumstances rare in the extratropics. For convenience these properties of the
tropical atmosphere crucial t o tropical cyclogenesis are introduced a t this
point.
Stability is conveniently described in terms of the total static enthalpy
H where
Here cPis the heat capacity of the atmosphere (effectively that of dry air, and
independent of temperature for present purposes) ; T is the static temperature;
g the magnitude of the gravitational acceleration; z the height above sea level;
L the relevant latent heat of phase transition for water substance; and Y
the mass fraction of water vapor. Throughout this paper, whenever H is
(loosely) described as a temperature, reference is to (Hlc,). Actually the total
enthalpy of a fluid particle is the total stagnation enthalpy
where q is the magnitude of the velocity vector (the wind speed).

The magnitude of the various contributions in the tropics to H and H ,
are worth mentioning. As shown in Fig. 5 , the static enthalpy decreases
from a sea level maximum of about 350°K ( L= heat of condensation plus
fusion) to a midtropospheric minimum of about 330"a t about 650 mb (around
13,000 ft), before recovering the sea level value of 350"just below the tropical
troposphere (that is, at almost 100 mb or 50,000 ft). This midtropospheric
minimum is observed with monotonous regularity throughout the tropics in all
seasons, although the local sea level and near-tropopause maxima are reduced
from the 350"level in the equatorial trough to 340"near the subtropics. Over
thelowestfew tens of millibars there may be a layer of well-mixed air in which
H is virtually constant, but this is of no major consequence here. The dy-
namic contribution of (q2/2)is always small, and usuallynegligible; for example,
even in a 200 mph hurricane, (q2/2cp)= 4"K, or just over a 1 yo contribution
to (H/cp)near sea level. For stability one can virtually always consider H ,
EQUIVALENT POTENTIAL TEMPERATURE, ee
200
rather than H , . At sea level, the static temperature contribution T is about

300°K near the trough, the latent heat contribution ( LY/cp)is about 50"K,
and, of course, the gravitational potential energy contribution (gzlc,) is zero.
The moisture content of the tropical ambient falls off roughly exponentially
with increasing height; the e-folding distance is crudely 650 mb. Whatever
modest variation in H normally occurs from day to day over a locale in the
tropics is more likely due to changes in water vapor content than tochangesin
temperature (Johnson, 1969, pp. 122-123); the lapse rate alters very little as
one moves from cloudy areas t o clear areas, but there is a drop in the dew
point (Gray, 1972b, p. 14). I n any case, certainly by 400 mb (or 25,000 ft)
for current purposes H A 8, where the potential temperature-like quantity
(actually an enthalpy) 8 is defined by
(4) O=c,T +gz
A quantity describing the stagnation potential-like enthalpy may also be
defined :
At 50,000 ft the gravitational contribution (gzlc,) is about 150"K, or over

40 yoof 8 (or H ) , and hence no longer negligible.
Clearly H plays a role much like the equivalent potential temperature
used by most meteorologists, but H is preferred here. For one thing, what a
dry-bulb and a wet-bulb thermometer measure (of relevance below) is
readily understood in terms of H , and 8 , . However, unless L is constant, H is
not rigorously a thermodynamic state variable. However, L varies slowly
with temperature and H remains a very useful concept.
The condition that convection is absent (mechanical equilibrium) is that
the entropy increases with height. If air is approximated as a perfect gas, the
unsaturated atmosphere is stable if (aH/az)> 0, convectively unstable if
(aH/az)< 0. If the atmosphere is saturated, then the atmosphere is stable
if (aH/az) > 0, and statically unstable if ( a H / a z )< 0. Convectively unstable
means a perturbed particle will resume its former position unless lifted enough
Fro. 5. (lop) The equivalent potential temperature 8. as a function of height z and

pressure p , from measurements near Barbados in the Lesser Antilles in July-August,
1968. Four characteristic weather types are represented: average (solid), suppressed
convection (dotted), moderately enhanced convection (long dashes), and strongly
enhanced convection (short dashes). (From Warsh e l a2. 1971, p. 127; with permission of
the American Royal MeteorologicalSociety.) (bottom)Soundings taken at Gan Island in the
Indian Ocean (OO"41'S; 73'09'E). The heavier curves (solid for static temperature,
dashed for dew point) are the average for 42 wet days; the lighter curves are the average
for 113 dry days. Measurements were made in Julys 1960-1964. (From Johnson, 1969,
p. 122; with permission of the Royal Meteorological Society.)
FIG.6. The average total static temperature (Hlc,) for the Northern Hemisphere in
winter. The curve marked A denotes the locus of the minimum of ( H / c p ) free
; convection
can most readily occur below level A, and only undilute ascent can continue much above
A; undilute ascent continues t o level B, where sea level values are recovered. (By per-
mission from Palmen and Newton, 1969, p. 574. Academic Press, New York.)
for the onset of condensation, in which case it will rise until its density dis-
crepancy relative t o ambient is reduced to zero (i.e. rise to a new equilibrium
position). Statically unstable means unstable without the requirement of a
suEciently large displacement. If the atmosphere is dry, or if for some reason
one wants to consider stability excluding the role played by condensation
(dry ascent as opposed t o moist ascent), then the atmosphere is stable if
(aO/az) > 0, and unstable if (ae/az) <0.
I n the extratropics, H and 8 both increase monotonically with altitude
usually, so the atmosphere is stable to both dry and moist ascent (Fig. 6).
Only in exceptional circumstances, as in thunderstorms, does penetrative
convection, with its vertical (as opposed t o slantwise) ascent, occur. I n the
tropics 8 increases monotonically with height, but H (as previously noted) has
a midtropospheric minimum (Fig. 7);thus air is stable to dry ascent, but air
in. the lower atmosphere is unstable to moist ascent (convectively unstable).
How turbulent mixing, cumulus convection, radiational cooling, cumu-
lonimbi (see below), and large-scale circulation maintain this condition of
convective instability deals with tropical meteorology in general, and lies
largely outside the scope of this review of tropical cyclones (although such
distinction will probably soon prove unsound with incipient progress in
tropical cyclogenesis). Actually a full understanding of how the tropical
ambient is maintained does not exist; the complexity is indicated by the fact
that, as will now be explained, the H profile generates different types of
clouds, which in turn sustain the H profile.
I n the convectively unstable tropical atmosphere, the larger the lapse rate,
the more suppressed is the cumulus activity. Only as the ambient lapse rate
- 50
-40
-30
--
N
0
X
4
90
-20
- 10
1 (OK)
FIG. 7. Static temperature T, potential temperature-like measure (6/cp), total static temperature (H/cp),and the total static
temperature for a hypothetical atmosphere saturated at the actual local temperature and pressure (I?/cp)vs. height and pressure for
the West Indies ambient for September. Based on data given by Jordan (1957). A parcel at height z1 will become buoyant a t z o
where (Hlc,) at z1 exceeds ( a / c , ) at zo( > z , ) ; the inequality is satisfied for air in the planetary boundary layer, and little else.
approaches the moist adiabat is there penetrative convection. For example,

Malkus (1960) notes that -(aT/az) 6.S0C/km from 900 t o 200 mb in the
N
tropics normally, but -(aTlaz) 6.0°C/km in the inner rain area of Hurricane
N
Daisy (1958). There have been many confirmations since, that the less pro-
nounced the midtropospheric minimum of the total static enthalpy in the
tropics, the more convective activity is likely t o be present (Garstang et al.,
1970; Aspliden, 1971).
If a glob of very low-level tropical air (buoyant element) is displaced
vertically enough for condensation to occur, ascent will continue many
kilometers virtually to the tropical tropopause, under the stability criteria
just presented, if the ascent is rapid enough for no mixing or radiational
cooling t o occur during ascent. Near sea level air would rise dry adiabatically
until saturated (roughly, 50 mb) ; thereafter, enough precipitation would fall
out to leave the air just saturated a t the local pressure and temperature, the
air retaining the latent heat of phase transition. Such a locus of thermodyn-
amic states is conventionally referred t o as the moist adiabat; for rapid
ascent in which ambient processes are too slow to act, H s ~ o n s t(Air
. ~ that
has risen to its level of neutral buoyancy will not be unstable t o descent,
because any compressional heat from work done on the fluid by gravitational
forces cannot be absorbed by the condensed water substance, which has
precipitated out. However, descending flow will accompany ascent, and this
will be addressed in Section 2.3.)
The core of a towering cumulonimbus is described by a moist adiabat.
The tropical ambient must be close to moist adiabatic for cumulonimbi t o
be significant; this is in fact the case (Riehl, 1969a, b). If the ambient were
moist adiabatic, cumulonimbi would be unnecessary because the ambient
would be too unstable (there would be gross convective lifting). If the
ambient were far removed from moist adiabatic, cumulonimbi would be
nonexistent because the atmosphere would be too stable (there would be no
lifting, just heating up of the air).
2.2. Tropical Curnulonimbi

About one percent of the area within the tropics is receiving precipitation
on average; this area is likely to be in the equatorial trough, but is not uni-
formly distributed (Riehl, 196913).The precipitation is concentrated in large-
scale disturbances consisting of a few cumulonimbi immersed in a larger
number of cumuli (Figs. 8 and 9); Gray (197213) estimates that 10-20 yo of
The convective instability of the tropical ambient explains how cumulonimbi can
occur, but cumulonimbi explain how high static enthalpy levels of the upper tropical
troposphere can arise. This interdependence exemplifies the earlier remark on the link
between the presence of clouds and the circumstances regarding stability.
cb
FIG.S. The total static temperature (Hlc,) over the ocean in the equatorial trough and
in the tropics near the subtropics, as a function of pressure. Both typical cumulus loci
(cu) and also cumulonimbus loci (cb) are noted. (Based on, with permission, Palmen
and Newton, 1969, p. 575. Academic Press, New York.)
the rain areas are covered by active cumulonimbi. Agglomerations of cumulo-

nimbi are sometimes described as cloud clusters. Cumulonimbi tend to align
themselves in easGwest rows, or bands; the bands can be tens to hundreds of
kilometers long (Kuettner, 1971), and are typically 20-50 km wide (Charney,
1971). The easGwest alignment is consistent with the low-level wind field of
large-scale disturbances in the tropics. The ITCZ, itself composed of one or
more narrow bands of vigorous cumulonimbus convection, is typically of
width 300 km (Charney, 1971).It is emphasized that the ITCZ, while a region
of variable winds, appreciable cloudiness, large precipitation, and low pressure
(Godshall, 1968), does not consist of a long unbroken band of heavy cloudin-
ess. Rather, it consists of intermittent cloud clusters (with strong low-level
convergence, appreciable vertical movement with precipitation, cyclonic
vorticity, and upper-level outflow with possibly anticyclonic relative vor-
ticity) with interstitial clear areas (with divergence, subsidence, and anti-
cyclonic vorticity) (Holton et al., 1971).&Gray (1972b) also emphasizes this
patchwork nature of the entire tropical belt, which consists of 20 yo cloud
area (which contains the cumulonimbi, than can rain a t the rate of 2.5
These authors emphasize that the ITCZ consists of westward propagating cloud
clusters that pass at four- to five-day intervals.
24 TRANCIS E. FENDELL
FIQ.9. NASA photograph taken in June, 1966 from Gemini IV of cumulus, cumulo-
nimbus, and cirrus clouds over the Pacific Ocean off the western coast of Central America;
the view is northeast toward Mexico.
gm/cma-day), and 40 yovariable cloud area and 40 yoclear area (this 80 yoof
the area yields negligible precipitation). There tends to be moist ascent in
the cloudy areas, and dry descent in the less cloudy areas, though by evapor-
ation of liquid water and by diffusion of water vapor, water substance is
transferred from the cloudy to less cloudy regions. If the tropics do consist of
cloud clusters of 500 km scale separated by clear (or a t least mostly cloud
free) areas of 200-10,000 km scale, then one anticipates that there is a greater
incidence of cumulonimbi relative t o cumuli in those clusters near the ITCZ
as opposed to those clusters in the trades (see below). Hence, there would be
appreciable upper (lower)tropospheric detrainment from those clusters in the
ITCZ (trades). If the clusters propagate westward a t intervals of several days,
then the cIusters would be associated with thin wave troughs (i.e., low pressure
regions) and the clear areas, with ridges.
These cloud syst,:ms are described here because in some still incompletely
understood manner, they play a role in tropical cyclogeneisis (Palm& and
Newton, 1969, p. 585).Although tropical cyclogenesis is marked by the exist-
ence of relatively little variation of the zonal wind with height, cloud
clusters in which the easterly wind speed aloft exceeds that near the surface
tend t o spawn storms. Cumulonimbi are given particular attention because
although they cover one one-thousandth (PalmBn and Newton, 1969, p. 440)
to possibly four-thousandths (Gray, 1972b, p. 33) of the area of a n equatorial
belt extending 10"latitude on each side of the equatorial trough, they cover
one-tenth of the area of the trough itself (Bates, 1972, p. 2). Tropical cyclo-
genesis often occurs near the equatorial trough, and cumulonimbi occur in
concentration in the hurricane eyewall (Section 4.3). I n fact, the absence of
cumulonimbi in subtropical cyclones occurring off western India, over the
Pacific, and over the Atlantic serves as one means of distinguishing these
storms (in which maximum intensity is reached in midtroposphere, and in
which a warm core occurs at high levels but a cold core a t low levels) from
tropical cyclones (Walker, 1972). [Simpson and Pelisser (1971)discuss various
so-called hybrid storms in and near the tropics, that may have hurricane force
winds a t least transiently, but do not have conventional hurricane s t r ~ c t u r e . 1 ~
A cumulonimbus (cb) is typified by a radius of 2 km, a lifespan of 30 to
40 min, a height of 14 km, a peak water substance content of 4 gm/cm3,
maximum updraft speed of 25 m/sec, a thermal anomaly of 6"C, and a total
rainfall of 2.4 cm. I n contrast, a cumulus (cu) is typified by a radius of 1 km
or less, a lifetime of 10 t o 25 min, a height of 6 km, a peak water content of
1 gm/cm3, maximum updraft speed of 6 m/sec, a thermal anomaly of 4"C,
and a total rainfall of 0.3 cm (Lopez, 1972a). Cumulonimbi are observed to
overshoot their neutral stability level and then oscillate about it; further,
only the core of a cloud rises undilute because only the larger clouds reach
upper tropospheric levels (Malkus, 1960). Riehl and Malkus (1961) estimate
that a cumulonimbus conveys about 6 x lo1O gm/sec and that a fluid particle
rises the full vertical extent in about 30 min.
In the summer tropics there is far more local ascent and descent than might
be suggested by the mean synoptic scale motion (Johnson, 1969; Gray,
1972b); in fact, if air in the equatorial trough rose only with the mean vertical
flow, the ascent would be so slow that radiational cooling alone would stop it
in midtroposphere (PalmBn and Newton, 1969) and cumulonimbi would not
exist. Gray (1972b) suggests that there is ten to twenty times the mean s p -
optic convergence in the lower summertime tropical atmosphere, and this is
the source of the initial forced vertical displacement thatleadsto condensation
and permits free convection to continue undiluted ascent in the unsaturated
conditionally unstable ambient conditions. Without one or two orders of
magnitude more up and down recirculatory local motion than suggested by
Further discussion of hybrid (or semitropical) storms, which rely for energy source
not only on latent heat release but also on baroclinicity associated with positioning of
warm and cold air masses, is given by Spiegler (1972).
the mean synoptic scale motion, the water substance distribution in the sum-
mertime tropics (in which individual cumulonimbi can rain a t 2.5 cm/day
even though the evaporation rate at the sea surface is a uniform 0.5 cm/day)
is inexplicable. This large local recirculation may have implications for the
momentum balance because it would seem to inhibit vertical wind shear
development.
Malkus and Riehl (1960) described the tropical cumulonimbus as a hot
tower because sensible heat from condensation was supposedly available t o
warm the ambient surrounding air. Lopez (1972b), however, asserts that
cooling tower would be more apropos because the condensational heat is
primarily used t o raise the air in the column t o its level of neutral stability
(i.e., the heat goes into potential energy). The air that is detrained from the
cloud serves to cool the environment because the heat extracted from the
ambient t o reevaporated condensed moisture in the detrained air exceeds any
sensible heat transferred to the surrounding environment. Measurement23
around cumuIus activity indicate that the static temperature often falls (e.g.,
Riehl, 1969b, p. 588). Of course, the ascent in a cloud cluster does indirectly
engender drying and warming through the subsidence of surrounding air
that inevitably accompanies ascent in the cloud; this heating counteracts
heat loss to detrainment and radiation. Thus on a large scale cumulonimbus
activity generates a net warming, but the effect on the locally surrounding air
is cooling (Gray, 1972b). Most models parameterizing the role of cumulus
convection in large-scale disturbances have the direct sensible heat transfer
model, rather than the indirect warming-by-subsidence mechanism, in mind
(see Section 3.10). The Riehl-Malkus hot-tower model is inconsistent with the
existence of large local recirculatory motion (Gray, 1972b).
2.3. CISK
Among the most important and fruitful concepts introduced in tropical
meteorology in several decades is an idea due to Charney (Charney and
Eliassen, 1964; Charney, 1971); the concept of CISK (conditional instability
of the second kind), in addition to whatever contribution it may make to
extratropical meteorology, seems of great utility in describing tropical phe-
nomena of widely different scales, ranging from asingle cumulonimbus to
hurricane rainbands t o cloud clusters to tropical cyclones to the ITCZ (the
intertropical convergence zone is, for present purposes, taken as identical
with the equatorial trough, the doldrums, and the trade confiuence). I n all
these phenomena there is low-level convergence with large moisture content
in the presence of appreciable Coriolis force. Phillips (1970) notes that while
the physical procssses involved in CISK can be qualitatively described, much
remains to be done with respect t o detailed quantitative description (Geissler,
1972).
The CISK process is the feeding of convective activity in a swirling flow

by frictionally-induced inflow in a surface boundary layer. The idea is that
rotating flows like that in a local low over the ocean tend to suppress converg-
ence except in frictional boundary layers (about a kilometer or two thick and
contiguous to the ocean). I n this frictional layer there is radial inflow because
the Coriolis and centrifugal forces (which balance the radial pressure gradient
above the boundary layer) are diminished by friction; in the frictional layer
radial advective transport responds to the '' partially unbalanced " pressure
gradient. The radial inflow in the boundary layer is furnished by slight sink-
ing in the swirling flow above the boundary layer, in those regions where the
circulation relative to the earth decreases with radial distance from the center
of the low. I n the tropics the converging boundary layer flow is warm and
moist, because of transfer from the warm ocean during inflow and because the
boundary layer air is warm and moist a t the outset. The boundary layer flow
erupts near the center of the disturbance in the region where the circulation
relative to the earth of the fluid above the boundary layer increases with
radial distance. I n the conditionally unstable tropical atmosphere the core
of the erupting, rotating column can rise vertically without dilution t o the
tropical tropopause. [Buoyant elements from higher in the boundary layer
may have somewhat lower total static enthalpy, and may rise to a level of
neutral buoyancy somewhat lower than that of sea level air. I n fact, a buoyant
element may contain air from a wide range of the boundary layer thickness;
the core of the element, which does not mix with outside air, may not mix
internally either, so there may well be shedding off of fractional pieces at
various upper tropospheric levels. This refinement need not be pursued here.
But i t should be noted that there is good laboratory evidence that entrain-
ment into a buoyant plume is reduced appreciably with increasing rotation
rate of the environment, for all but very small rotation rates (Emmons and
Ying, 1967).] The hydrostatically computed weight of the convective column,
which thermodynamically is described by a moist adiabatic locus of states, is
taken to be less than the weight of a column of ambient air. The difference in
weight of the two columns implies a radial pressure gradient, which (by the
conservation of radial momentum) in turn implies swirling. The swirling leads
to further downflux into the surface frictional layer, further inflow, further
moist adiabatic ascent, and hence maintenance (or even augmentation) of the
radial pressure gradient. A crucial additional point about the CISK process
is that it furnishes the basis of explaining how, instead of competing, cumulus
scale activity can cooperatively interact with cyclone scale activity to their
mutual enhancement.
The emphasis in CISK on both Coriolis force and conditional instability
suggests that both the ITCZ and tropical cyclogenesis occur near, but not at,
the equator. The critical point of the CISK process is that the weight of the
convective column is lighter than that of the ambient gas. From a facile view,
28 FRANUIS 1.FENDELL
the release of the appreciable heat of condensation during moist adiabatic

ascent can only lighten the column, and the existence of a radial pressure
gradient is obvious. However, Gray (197213)has recently viewed this typical
summary of the CISK mechanism as an incomplete description of certain
tropical phenomena, because it fails to emphasize the actual indirect heating
mechanism associated with cumulus activity and consequently fails to em-
phasize the large local up and down recirculation above the mean synoptic
motion (Section 2.2).
Another problem with quantitative analysis of the CISK process is that the
surface frictional layer over the oceans may not be adequately described
even in time average by a linear quasi-Ekman layer theory, and particularly
not at low latitudes. For instance, the zonal translational speed, latitude,
and scale of some tropical disturbances are such that the frictional layer
does not have time to equilibrate, and steady linear Ekman theory is in-
appropriate. Gray (1972a) notes that dissipation of kinetic energy in the
tropical layer so exceeds production that indeed there must be mesoscale
and synoptic scale sinking of new momentum-carrying fluid to sustain the
boundary layer. However, Gray claims that, contrary to Ekman’s theory,
measured planetary boundary layer thickness does not increase with de-
creasing latitude; he suggests vertical stability and turbulence scales, rather
than rotation, determine the boundary layer thickness. Gray’s general
remarks, it may be noted, are not addressed to the exceptional circumstances
of a tropical cyclone. Mahrt (1972a,b)has attempted theoretically to demon-
strate the sensitivity of tropical disturbance development to the details of the
subcloud level moisture convergence; especially equatorward of the ITCZ
where the. Coriolis parameter is small and its latitudinal variation is
large, nonlinear advective acceleration may so alter the boundary layer
flow that spiraling may be in the opposite sense to that predicted by Ekman’s
theory. The thickness, latitudinal variation, and strength of cross-isobarflow
may deviate from what would be predicted by linear analysis. Also, whether
the amount of fluid emerging from a quasi-linearEkman layer isfully adequate
to describe the growth of finite-sized tropical disturbances is a point of current
contention; perhaps at least some entrainment in excess of the Ekman
pumping into the cloud base will be found necessary for the early-stage
growth of tropical disturbances. For example, alternatives to CISK (baro-
tropic instability owing to north-south wind shear) and modifications to
CISK (low-level convergence by a frictionless, internal-wave-type mechan-
ism) have been proposed to describe early-stage growth. Finally, Petrosyants
(1972) comments that the ITCZ in the Atlantic Ocean is most clearly evident
in the eastern part, where a trade inversion prevents the development of
deep layer tropical convection, and mainly trade cumulus clouds are found;
Petrosyants questions compatibility of this observation with the deep con-
vective penetration concept of CISK as an explanation of the ITCZ.
3. MODELSOF A MATURETROPICAL
CYCLONE
3.1. Introduction
I n sorting out the thermohydrodynamics of a tropical cyclone, one is
faced with understanding the interaction of two scales of phenomena: the
larger cyclone scale (-500 to 1000 miles), and the smaller cumulus scale
( 1 mile). The cyclone must feed the cumulus scale, which in turn sustains
N
the cyclone scale, in a cooperative interdependence. Such ideas suggest that

the CISK process is not only relevant t o the maintenance of a n individual
cumulonimbus cloud, but also (properly scaled) to the maintenance of the
tropical cyclone in the large (Section 2.3).
The cyclone scale thermohydrodynamics of the mature stage, taken to be
axisymmetric and quasi-steady for most purposes to a lowest order of ap-
proximation, will be set forth in terms of a model introduced in the last three
years by Carrier and his co-workers (Carrier, 1970, 1971a,b; Dergarabedian
and Fendell, 1970, 197213; Carrier et al., 1971; McWilliams, 1971). These
articles delineate overall dynamics and thermodynamics, and imply (in con-
trast t o other models t o be introduced later) no major augmentation of
ambient sensible and latent heat transfer is needed t o explain hurricanes.
These articles emphasize that many details and refinements in structure still
remain to be quantitatively treated.
3.2. The Carrier Model

Subsequently (Section 4) transient analysis will be devoted to describing
how a severe vortical storm may be generated in the tropics. Here, the quasi-
steady mature hurricane is studied.
The Carrier model, on the basis of subdividing the tropical cyclone into
segments where different processes and scales predominate, is a four-part
analysis. The four regions, indicated in Fig. 10, are the throughput supply I,
the frictional boundary layer 11, the eyewall and efflux region 111, and the
eye IV. Some of this subdivision is conventional, some not. Besides clarifying
locally dominant physical processes, subdividing permits retention of the
minimal number of terms in locally valid quantitative formulation; this
procedure simplifies the mathematical solution in a manner unavailable to any
direct finite-differencing of uniformly valid equations.
The Carrier model is closed for convenience-there is no very significant
amount of mass convected across any boundary, although massmay bediffused
across any boundary. The cylindrical-like volume encompassing the entire
storm has the sea surface for its bottom; its sides lie far enough from the
center (about 500 to 1000 miles) so that the winds are virtually reduced to
ambient, and the swirl relative to the earth is taken as zero for compatibility
with the Ekman condition for no radial inflow across the outer boundary. The
30 FRANCIS E . FENDELL
top of the storm is taken to be that height (-150 mb) a t which sea level air
in the outer part of the storm, if lifted rapidly so that the total enthalpy of a
fluid particle remained constant because relatively slow ambient-maintaining
processes would not have time t o act, would no longer be unstable relative
t o the local ambient air. Such rapid lifting is, of course, the moist adiabatic
ascent discussed in Section 2.1. This “ instability lid ” lies a t so great a height
that there is virtually negligible swirl, as explained below; the ambient
pressure and temperature a t this height are taken to describe all radial
positions a t this height, from the center to the outer edge. Thus, the top of the
storm is an isothermal, isobaric, constant altitude lid with no water vapor
content for current purposes.
Discussion now turns to describing each of the four regions comprising the
tropical cyclone in some detail.
I n region I there is warm moist air typical in stratification of the ambient
atmosphere in which the hurricane was generated. This air spun up under
conservation of angular momentum as it moved in toward the axis of sym-
metry during the formative stage. As the mature stage was approached,
a gradient wind balance of pressure, Coriolis and centrifugal forces choked
off any further inflow; the inflow is only enough to prevent the eyewall
I11 from diffusing outward, and that requires only an exceedingly small
radial flow. The air in I, then, is rapidly swirling, the azimuthal velocity
component greatly increasing and the pressure greatly decreasing from the
edge to the center. Under such a radial profile for the swirl, there is a small
downflux from the throughput supply I into the frictional boundary layer IS.
The small downflux leads to a large net mass flux into the frictional layer
because of the large area involved. Furthermore, the downdraft is only a gross
temporal and azimuthal average because locally and transiently there is
intense convective activity by which clouds form and rain falls.
I n the frictional boundary layer 11, the only region in which angular
momentum is not conserved but is partially lost to thesea, thereisappreciable
influx. I n fact, the azimuthal and radial velocity components are of compar-
able magnitude; typically, for fixed radial position, the maximum inflow speed
a t any axial position in the boundary layer is about one-third the maximum
azimuthal speed. [This fraction is close t o the one reported by Hughes
(1952) from flight penetration of hurricanes a t altitudes of 1000 f t or less.]
The vertical velocity component is much smaller. The reason for the inflow is,
as previously discussed, that the no-slip boundary condition reduces the
centrifugal acceleration, and a relatively uncompensated pressure gradient
drives the fluid toward the axis of symmetry (so-called “tea cup effect ”).
Far from the axis in I1 the classical balance of the linear Ekman layer (friction,
pressure, and Coriolis forces) suffices; since the downdraft from I to I1 is
probably fairly independent of radial position (especially far from the
eyewall) for swirl distributions of practical interest, three-quarters of the

flux inward through I1 comes from downflux across the interface between
I1 and I where ( r o / 2 )< r <ro ,where r o is the radial extent of thestorm. Closer
in to the axis the nonlinear accelerations, especially the radial acceleration
of radial and tangential momentum and also centrifugal acceleration, must
enter.
As the pressure gradient in I lets more air sink into 11, the influx into I1
drives boundary layer air moderately rapidly up a cloudy eyewall 111.I n the
eyewall hydrostatic and cyclostrophic approximations hold; the locus of
thermodynamic states is the moist adiabat based on sea level conditions in the
eyewall (this particular point of the Carrier model will be examined in the
critique to follow). The swirl is reduced in I11 owing to boundary layerfriction.
Near the top of I11 the flow moves further away from the axis of symmetry
such that, in the outflow, the air seems to an observer on earth to be rotating
opposite in sense t o the rotation in I and I1 (which is cyclonic). The air in I11
slips over the air in I with no interaction; there is no large radial pressure
gradient in much of 111, unlike I.
At low altitudes the eyewall flushes moist air out of IV, and at high
altitude rained-out air is entrained into IV. I n time the eye becomes drier and
better defined; it is the central core in which relatively dry air sinks, is
warmed by compression, and is entrained out into the eyewall or recirculated
within the eye (stagnation a t the base of the eye would permit transport
processes to cool the eye). The relatively light eye permits much greater
pressure deficits from ambient and hence supports much higher swirling
speeds. At a fixed altitude, the density in the eye is less than that in the eye-
wall, which in turn is less than that of the ambient gas at the storm edge,
Since (except in cumulonimbi) the hydrostatic approximation is uniformly
valid, spatially and temporally, in a hurricane, the sea surface eye pressure is
less than the sea surface eyewall pressure, which is less than the sea surface
ambient pressure. While Carrier has not explicitly written so, i t would seem
that as a general trend, the smaller the eye radius, the more spin-up under
conservation of angular momentum. Hence the more intense storm might
generally have a relatively small eye, the eye broadeningas thestorm weakens.6
Data on this matter appears contradictory. Using all recorded data for Atlantic
hurricanes for 1961-1968, Sheets (1972) found no correlation between the size of the
radar eye and storm intensity indicators (such as maximum wind speed or minimum
sea level pressure); there was reportedly a slight tendency toward smaller eye diameters
at lower latitudes. Questions concerning the use of radar by itself to determine the eye
radius will be discussed in Section 3.4. In contrast, the eye diameter increased typically
from 20 n miles t o 33 n miles, while the intensity typically decreased by 33 yo,according
to data for typhoons crossing the Philippines during 1960-1970; furthermore, for typhoons
near the Philippines the eye diameters for intense typhoons (maximum wind speed over
90 knots) are typically 13% smaller than for weaker typhoons (maximum wind speeds
Carrier’s model thus pictures the tropical cyclone as a once-throughprocess

in which a “fuel supply”-the warm moist air in I, which is part of the
tropical cyclone at its inception and is convected with the storm-is slowly
exhausted. The storm weakens because the air drifting downintothe frictional
layer toward the end is typical of the higher tropical environment and hence
of lower total static enthalpy. Eventually the fuel supply is exhausted, and
the boundary between I11 and I sinks toward 11. Some models (e.g., Eliassen
and Kleinschmidt, 1957) picture a recirculation through the storm of outflow
air; the storm does not survive long enough for this, nor could such recycled
air maintain the storm.
There is one important omission to the foregoing description that has been
intentionally deferred: the energetics of the surface frictional layer and asso-
ciated questions of airlsea transfer of latent and sensible heat. The relevant
quantity to consider is the total stagnation enthalpy (the sum of static en-
thalpy the heat associated with condensible moisture, gravitational potential
energy, and kinetic energy contributions). This quantity is conserved at
roughly its ambient stratification throughout regions I and 11; therefore, it is
described by a profile that decreases with height fmm 10oO mb to 650 mb,
and then increases with height, as mentioned above. The implication is that
the heat and mass transfer from the ocean to the atmosphere is about the
same within the hurricane as in the ambient. This transfer helps compensate
for the rain-out in the spiral bands and helps maintain the warm, moist
nature of the air in I. [Occasionally the Carrier model is still grossly mis-
represented as proposing adiabatic conditions (constant total stagnation
enthalpy in I1 so the net heat and mass transfer from sea to air is zem); such a
solution cannot possibly satisfy the parabolicboundaryvalueproblemdescrib-
ing the energetics of the frictional boundary layer because it obviously
violates the initial condition at r = T o , the outer edge. In fact, if the supple-
mental 0ux from the ocean is entirely eliminated, as from passage over land,
the spin-down time is O(a/v1 aSZ1 where2) the eddy viscosity v e miles2
hr-l, the normal component of the rotation of the earth SZ i 2 x 10-1 hr-l,
and the height of the throughput supply a G 1 mile-so the spin-down time
is half a day to two days.] The model of total stagnation enthalpy fixed at its
ambient stratification breaks down in the eyewall 111; there the vertical
velocity component is at least one, probably two orders of magnitude larger
than the relatively small downdraft into the boundary layer; the result is
under 90 knots). Incidentally, if the size is determined by the average diameter of the
closed surface isobar, the size of a storm decreases by 1 7 % in area in crossing the
Philippines, and near the Philippines the intense typhoon has a mean outer circulation
diameter 60-150 n miles greater than that of a weaker typhoon (Brand and Blellock,
1972).
that convection dominates the slow ambient-sustaining processes, so for a t

least portions of the eyewall the total stagnation enthalpy is virtually
constant a t its local sea level value, which (as just explained) is roughly its
ambient sea level value.
3.3. Critique of the Carrier Model

A closed system seems naturally definable since a single hurricane probably
does not interact significantly with the entire planetary atmosphere, especially
when structure rather than path is under consideration. However, the closed
system is considered optional by Carrier because the same conclusions would
also hold for a n open system, although the arguments would be far more
difficult to construct. Clearly treatment of asymmetric effects should be a
future goal of the Carrier model since complete elucidation of the spiral-band
phenomena and other properties are probably unattainable otherwise. Initial
emphasis on axisymmetry has been the traditional path of development: for
tropical cyclone models. Nevertheless, examination of the modification of
the axisymmetric theoretical solution by accounting for the beta-plane effect
(variation of the Coriolis parameter with latitude) may well prove interesting.
The disruption of axisymmetry might prove appreciable, and further, par-
ticular intensity might be ascribed by theory to the outer western edge of the
storm with respect to an observer looking along the direction of translation;
if the height of the tropopause over the storm increased with latitude, this
effect would not be expected, and it is in fact not observed. As noted earlier,
what is observed is that the storm remains basically axisymmetric, with
particular intensity occurring in the inner portions of the northeast quadrant.
The sketch in Fig. 10 is schematic, but it will appear below that in lieu of a
greatly augmented oceanic sea-to-air latent and sensible heat transfer, the
Carrier model relies upon the pressure deficit achievable by dry adiabatic
recompression in the eye to be available to maintain dynamic equilibrium in a
system with wind speeds known to reach 200 mph. Thus, some small but finite
outward sloping of the eyewall a t least down to midtropospheric levels, if not
lower, is required for internal consistency in the Carrier model. Such ideas were
present in the workof Haurwitz (1935)and in the early papers of Palm& (1948).
The particular emphasis placed on careful scrutiny of the dynamics and
energetics of the frictional inflow layer is a major contribution of the Carrier
model. [Incidentally, the thickness of this layer is too great to let increased
frictional drag explain the accelerated decay after landfall, unless the coast-
line is mountainous. Even then, the orographic effects are more likely t o be
felt in deleterious premature rain-out owing to lifting of I, which would result
in relatively dry air descending into the boundary layer and running inward
toward the eyewall, so that moist adiabatic ascent would no longer be possible.
The rate of decay of intensity owing to sea-to-air enthalpy transfer reduction
t t
ALTITUDE,
I I -
t-l MI)-
*
FIG.10. This conjectured configuration of a mature hurricane with rough order of
magnitude dimensions is not drawn to scale. The subdomains are: I, throughput supply, a
region of rapid swirl and very slow downdrift; 11, frictional boundary layer; 111, eyewall;
and IV, eye. Across the boundary layer there is about a 100 mb drop, and across I, a
further drop of about 200 mb; the pressure a t the top of the hurricane is about 150 mb,
i.e. the top is near the tropopause. The eye-eyewall interface is taken to slope outward,
though the effect may not be so pronounced as sketched.
after landfall, according to the Carrier model, was discussed earlier. The fact
that friction effects the low-level moisture convergence suggests small
increases in friction may even cause intensification.] However, the most
novel contribution of the Carrier model is identification of the “fuel supply”
in I, which requires only ambient level sea-to-air total stagnation enthalpy
transfer for maintenance. That the fluid in I does sink down into I1 is suggested
by the observation by Gentry (1964, p. 64) that a t lower altitudes the tem-
peratures are lower in the outer rainbands than in the surrounding air. There
is little undiluted ascent from the inflow to the outflow layer. Rather, even the
air which ascends in the outer rainbands also descends again in the storm
area, and is not immediately carried away through the outflow layer in the
upper troposphere.
The grossest feature of the Carrier model is the absence of any refinement
t o the eyewall structure. While for decades modelers have presented mean
soundings in the eyewall that suggest moist adiabatic ascent from the surface
layers (Palm& and Newton, 1969, pp. 477482), Shea (1972) emphasizes that
the ascent is limited to cumulonimbi that cover only 10 to 20 yoof theeyewall
area. Subsidence occurs in the regions of the eyewall outside cumulonimbi.
Using flight data that include intensifying and decaying as well as mature
tropical cyclones, Shea asserts that the relative humidityprobablyisnot 100 %
throughout the eyewall, and there is a small midtropospheric minimum in
mean vertical profiles of the equivalent potential temperature (far less pro-
nounced than the ambient minimum). I n fact, wet-bulb effects may have led
to spuriously low temperature measurements a t midtropospheric heights
(Gentry, 1964). I n any case, the failure to delineate this structure does not
have significant repercussions for Carrier’s work on the mature hurricane, but
does have important implications on his work on intensification, to be dis-
cussed later (Section 4).Shea also suggests that the gradient wind balance
does not hold to good approximation in the eyewall; if so, this is probably
due to a contribution from the transient partial derivative of the radial
velocity component, and is not taken t o modify analyses of a mature hurricane
appreciably.
I n the following sections three specific analyses among the many carried
out by Carrier and his co-workers t o help substantiate the model will be
briefly reviewed. The three analyses involve maximum swirl speed estimation,
the dynamics of a nonlinear Ekman layer, and the energetics of the surface
frictional layer. The quantitative results achieved t o date concerning the
Carrier model have been attained without large-scale digital computation.
The emphasis on subdivisional investigation of the four regions of the storm,
with interfacial compatibility, permits a substantially analytic approach.
It should be noted that although certain linearizations are sometimes adopted
for tractability by Carrier and his co-workers in the course of their analysis,
Carrier’s model for the mature hurricane is definitely nonlinear, and solutions
derived by such linearizations are acceptable only if, a posteriori, they can be
demonstrated t o satisfy the original nonlinear boundary value problem with
acceptably small error. Without prior proof of internal self-consistency by
subdivisional analyses, a full numerical treatment of the basic boundary value
problem would seem premature.
3.4. Maximum Swirl Speed Estimate According to the Carrier Model

An upper and lower bound on the central pressure deficit achievable in a
known spawning atmosphere will now be set forth using the Carrier hurricane
model, hydrostatics, and the thermodynamics of moist and dry air. Specific-
ally, the weights of various columns of air in the storm will be determined
in the light of different moisture content and thermodynamic processes
nvolved. The bounds on the central pressure deficit can then be translated
into an estimate of bounds on the maximum swirl speed through dynamics
(the radial momentum equation). Fletcher (1955)had suggested the useof the
cyclostrophic balance once pressure deficits were known, and Malkus (1958)
had suggested that pressure deficits could be calculated from moist adiabatic
considerations for the eyewall and dry adiabatic considerations for the eye.
Here the concepts are combined to achieve quantitative bounds, but just as
important, to demonstrate that hurricane speeds could be achieved without
36 FRANCIS 1.FENDELL
requiring any enthalpy transfer from the ocean greatly i n excus of the ambient
transfer, provided the eyewall i s not perfectly vertical. Miller (1958) performed
somewhat similar calculations to those to be described and also noted
that the Riehl-Malkus postulate of large oceanic latent and sensible heat
transfer to air flowing in through the frictional layer was not required to
explain low central pressure.
The f k t step is to neglect the frictional boundary layer 11, which is rel-
atively thin and across which, except for hydrostatic variations, the pressure
does not change according to lowest order boundary layer theory.
The variation of pressure p , density p, and the temperature T with height
above the ocean z, for any ambient tropical atmosphere in which a hurricane
forms, may be computed from
(6) Pa = Pa R a T (a = dry air)
(7) p v = pv RET / u (v = water vapor; u = 0.622)
(8) P =Pa +Pv 9 p =P a +P v , P" = P(T)(RH)
(9) dpldz = -pg
(10) T=f(P),RH=g(P)
where the temperature profile f ( p ) and the relative humidity (RH) profile
g ( p ) are taken as known from measurement. The saturation pressure P ( T )is
well tabulated for vapor and liquid phases above freezing, and vapor and solid
below freezing (Keenan and Keyes, 1936); a convenient and accurate expres-
sion for P(T) in millibars was given by Tetens (Murray, 1967):
(1W P(T) = 6.1078 exp[a(T - 273.16)/(T - b)]

a = 21.8745584
b = 7.66 I
over ice
b = 35.86 I
a = 17.2693882
over water
where T is in OK. The integration proceeds from the sea level upward in alti-
tude z; data typically extend from about 1000 mb to 150 mb (Fig. 11).
The top of the storm is normally taken as the height at which the ambient
total stagnation enthalpy (for which the kinetic energy contribution is negli-
gible) recovers its sea level value, as noted earlier; here, however, a slightly
different procedure explained below will be used. The sea level ambient state
is henceforth denoted by subscript s.
In a fully developed storm the air rising up the eyewall ascends in cumulo-
nimbus clouds, and thus follows moist adiabats. The initial states to be used
for the moist adiabats are not necessarily known a priori. If one believes
the total stagnation enthalpy is constant along a streamtube in the surface
frictional layer from the ambient to the eyewall, then the sea level tropical
, ;J
t , I
I
I I
38 FRANCIS E. FENDEIL
ambient state may characterize the total stagnation enthalpy of the stream-
tube rising in the eyewall closest to the storm center (procedure A). [Actually,
mixing inevitably occurs so perhaps a lower total stagnation enthalpy
characteristic of some height above sea level should be used, but the small
distinction is not worth the effort (Carrier, 1971b, p. 158).]While such a choice
for the initial state of an eyewall moist adiabat is sometimes made, and while
such results will be presented here, another preferable procedure B will also be
developed. I n this alternative procedure, the temperature and relative
humidity of the sea level state of the moist adiabat will be taken as known,
but the initial pressure will be taken as unknown (to be determined by itera-
tion for self-consistency to be explained below).
There is also a comment worth noting concerning the equationthatdescribes
the moist adiabat. If one simply takes dH, = 0 with L held constant (Charney,
1971, pp. 357-358), where again H , is the total stagnation enthalpy, then after
manipulation with (6)-(9):
The [d(q2/2)/dp] contribution is negligible and is henceforth dropped. A

more careful derivation from basic principles (Appendix B) yields a slightly
different expression
However, if one takes p e pa [so x = 1 in (12)] and p e pa, which incurs a n

error of only about three percent a t most a t any point intheflodeld, thenthe
two expressions are equivalent.
The procedure for computing the moist adiabat is to use the dry adiabatic
relation T p Y l ( y - - l ) , y = 1.4, from the sea level eyewall conditions t o that
N
pressure at which R H = 1, a t which point one switches to the moist adiabat

and continues the calculation. The integration is terminated a t that pressure
p1 for which the temperature calculated from the moist adiabat and from the
ambient are equal; this temperature is denoted TI, and the height above sea
level a t which T, occurs is denoted z1 (the “lid” on the cyclone). One then
integrates (6)-(9) and the differential equation for the moist adiabat from
z=zl (where T = T, and p =pl) to z = O ; RH =1 initially during this
integration, but one switches to the dry adiabat where appropriate. Under
procedure A the sea level pressure for the eyewall p(z = 0) = p e is that
computed. Under the superior procedure B, an estimate of the sea level
eyewall pressure p e had to be adopted to compute the moist adiabat, and this
value must be recovered for convergence. If no eye existed in the vortex-as
seems to be the case for some tornadoes and waterspouts-then the just
calculated p ( z = 0 ) =p e , p(z = 0 ) = pe would characterize conditions a t
the center of the vortex.
The lid on the storm as calculated here will be higher in procedure B than
in procedure A, for a given ambient. The taller the storm, themoreintenseitis
according to calculations developed here. The same relation holds observation-
ally (Riehl, 1972a, p. 248).
I n a mature hurricane a pressure deficit in excess of ( p s - p e ) is achieved
by having rained-out air entrained from the eyewall sink in a relatively dry
eye under adiabatic recompression. Thus in a hurricane ( p s - p e ) is a lower
bound on the central pressure deficit. For a n upper bound on the deficit
that may be achieved, one may adopt the idealized model that the eye is
completely dry (so no compressional heat is lost to reevaporation) and that the
air entrained into the eye is drawn from the top of the eyewall (or, inany case,
has T = TI, p = p l , Y = 0 at z = zl). The relevant equations are (6)-(9),
RH = 0, and T P ( Y - ~ ) / Y ; integration in the direction of decreasing z yields
N
p ( z = 0) = p c ( < p e < p s ) and p(z = 0) = p,( < pe <ps)-the density discre-

pancies so calculated are a t most 25 yo and Fletcher (1955) estimates the
density does not vary by even 15 yo,so the density may be held constant a t its
ambient value throughout the dynamical calculations now discussed.
If one adopts the cyclostrophic balance, holds p constant a t (say) ps , and
(since the core is observed not to rotate) lets
(14) w(r) = 0 O<r<R
= (v)msx(R/r)" RIpI
then
(15) (Vlrnax sPA/P~I"~

[ 2 n ( ~-
First, for a one-cell vortex (when there is no eye so the moist adiabat calcula-
tion is appropriate all the way to the axis), a rigid body-like rotation lies near
the core so then
(16) ~ ( r=) (v)rnax(r/R) 0 2r I R
=(~)max(R/r)" RI rI 00
and from the cyclostrophic balance

(17) ( ~ I n m x= [(2n/(n+ 1 ) ) ( ~ s- ~e)/psIl/~
Clearly, (15) holds for the mature hurricane; the maximum swirl is larger in
+
(15) than in (17) by the factor ( n 1)1/2because none of the pressure deficit
from ambient needs t o be expended to maintain rotation of the central
column. Next, although power law decays of swirl with radial distance are
frequently adopted and suffice for current purposes, i t will become evident
that other forms are at least as plausible, and more convenient, for r 2 R.
I n any case, Miller (1967) suggested from limited data that 0.5 < n <0.65
usually suEces, and Riehl (19634 had chosen n = 0.5; earlier, Byers (1944,
p. 436) had also recommended n A 0.5 and Hughes (1952) considered n 5
0.6-0.7 suitable for an average hurricane. The upshot is that estimates will
be made on the limits n = 0.5 and n = 1.0, although n A 0.5 seems the more
realistic. Finally, the gradient wind equation would probably be more ap-
propriate than the cyclostrophic equation, but the more complicated formula
would give maximum speeds reduced by only five percent from those obtained
from the simple forms (15) and (17).
First, results using procedure A (i.e. taking the total static enthalpy H
constant throughout the radial inflow as well as eyewall region) will be given.
If H is held constant a t sea level ambient conditions b, = 1014 mb, Ts=
299.4"K,(RH), = 0.841, then use of (12) for the moist adiabat gives conden-
sation a t 972 mb (Te 295"K, z A 180 ft). One finds z1= 48,150 ft, p , =
138 mb, T,= 203°K. The eyewall pressure a t sea level p , = 978 mb [with
T,i296"K, (RH)eI 0.9751.The eye pressure a t sea level p , = 894 mb (with
T, = 346°K);this upper bound reflects, it is reiterated, the idealization of a
completely moisture-free eye. As a variant, if sea level ambient conditions were
revised slightly for computing the moist adiabat only [ps= 1014 mb, T,=
299.4"K, (RH), = l.O)], then use of (12) gives z1 = 52,500 ft, p , = 110 mb,
T,= 196.6"K. also,^, 5 945 mb, T,= 297°K;p , = 850 mb, T,= 352.7"K.
As an illustration of the change attendant upon using (13) in place of (12),
one finds for H referenced to the unsaturated sea level ambient [ p s= 1014
mb, TS=299.4"K, (RH),=0.84], saturation again occurs at 972 mb, z e
450 ft; however, x1 = 46,500 ft, p 1 = 150 mb, T, = 205.4"K; further, p , =
988 mb, T,= 297.2"K,(RH), = 0.94 (Figs. 12and 13).Ifthesealevelambient
state is taken as saturated for computing the moist adiabat only, z1 = 50,530
ft, p1 = 122 mb, T1 = 199.6"K;whence, p , = 958 mb, Te = 2973°K.
These results, t o reiterate, have followed procedure A [moist adiabat
based on sea level ambient, as given by Jordan (1957) or modified t o be satu-
rated a t the same temperature and pressure]. The results have been presented
here because such calculations have been performed in the past, and because
the sue of the variances resulting from the use of (13) in place of (12) seemed
worth investigating.
For the superior procedure B (iteration to determine the consistent pressure
a t the base of the eyewall moist adiabat), only (13) will be used. If one adopts
T,= 299.4"K, (RH), = 0.84, one finds convergence for p e = 958 mb (Figs.
14 and 15); incidentally, z, = 50,100 ft, p , e 125 mb, T,s 200°K; further,
saturation occurs a t 918 mb, z A 1300 ft. If L is given the value of the heat of
condensation only, (H/cJs = 343.4"K (sea level ambient), and (H/cP),5
346.4"K (sea level under the moist adiabat). [Under (13), (Hlc,) is not pre-
400-
3
-5
Q.
600-
m-
I
200 220 240 260 280 300 320 340 360
T CK)
FIG.12. The ambient pressure-temperature curve is based on data for the Caribbean
in September given by Jordan (1957). The moist adiabat is based on having sea level
ambient air undergo dry adiabatic expansion until saturation, then moist adiabatic
ascent; the resulting sea level pressure deficit from ambient gives a lower bound on the
central pressure deficit under the adopted model. An upper bound on the deficit is fur-
nished by having the air that rose on the so-called moist adiabat, recompressed dry adia-
batically back down to sea level. Altitudes ar0 associated with the thermodynamic
states by use of hydrostatics and the equations of state for dry air and water vapor.
-Ambient, --- moist adiabat, dry adiabat.
cisely constant on a moist adiabat.] Also, p c = 875 mb, T,= 349°K. Thus it
would seem that the sea level total static temperature needs only to increase
about 3"K, with relative humidity and sea surface temperature held fixed,
for a n appreciably reduced pressure under eyewall cumulonimbi cores.
According to (15), for the deficit ( p s - p e ) = 56 mb, = 155 mph for
n = 0.5 and A 219 mph for n = 1.0. Any eyewall bending wouldpermit
appreciably higher speeds; in fact, since ( p , - p c ) = 139 mb, theoretically
sufficient bending could explain any speed up to 244 mph for n = 0.5 and
345 mph for n = 1.0. If one again takes T,= 299.4"K,but now lets RH = 1.0,
one finds convergence for p , = 890 mb. Here zI = 57,500 ft, p , = 84 mb,
T,= 191.8"K. In addition, ( H / c , )=
~ 359.4"K; so an increase of about 16°K
in the total static temperature could explain any recorded hurricane wind
intensity without any eyewall bending if the sea level eyewall air were satu-
rated, and if a moist adiabat were taken to be an adequate approximation
t o the eyewall sounding. I n fact, increases above 16°K appear impossible
under these conditions as long as the sea surface is taken as an isothermal
I 1 I I 1
0 10 20 30 40 50
6)
FIG.13. The density versm altitude relation for each of the three thermodynamic
loci of Fig. 12 are presented. The density variations from ambient within the storm are
less than 30 %, and speeds are far below sonic everywhere.~ Ambient, ---moist
adiabat, dry adiabat.
surface. Of course, use of the moist adiabat for the entire eyewall and the
dry adiabat for the eye involves idealizations yielding upper estimates, but the
large magnitude of the winds so calculated suggests the ideas behind the
estimates are correct.
At this point a brief summary of the calculated results and their implication
seems apropos. Even though cumulonimbi occupy only 10 to 20 % of the
eyewall (Charney, 1971, pp. 358-359), results have been presented for moist
adiabatic ascent and hydrostatics in the eyewall which suggest that significant
pressure deficits relative to ambient may be achieved by moist adiabatic
ascent of air whose total static temperature (Hlc,) is increased by only 1 yo
over ambient, and very intense storms may be achieved by ascent of air with
a 5 yoincrease over ambient. The 5 yoincrease involves arbitrarily increasing
the eyewall relative humidity a t shipboard height to 100 yo from an ambient
level of 84 Yo, and it would seem more likely that the means by which the
winds of very intense hurricanes are achieved is that the eye-eyewall inter-
face, as well as it can be defined, slopes a t least slightly outward from the top
1014 I I I I I I I
200 2M 240 260 280 330 320 340 360
T PK)
FIG.14. The three curves are analogous to those of Fig. 12, except that the moist
adiabat, while still taken to be based on ambient sea level values for relative humldity
and temperature, IS now computed from a n iteratively determined sea level pressure,
consistent with the definition of the top of the storm as an isobaric, isothermal surface
of constant altitude.~ Ambient, --- moist adiabat, --- dry adiabat.
of the inflow layer to the top of the storm. The calculation of wind speed for a
storm with an eye implicitly assumes suchsloping. Theadditionaldryadiabatic
compression-associated pressure deficit available to sustain swirling is reduced
if the eye-eyewall interface is vertical to a certain height, and only then slopes
outward; the additional pressure deficit is unavailable if the eye-eyewall
interface is perfectly vertical virtually to the top of the storm. (Whether or
not the eye-eyewall interface slopes outward does not alter the conceptual
thermodynamic stratification of the eye nor the sea level eye pressure deficit
from ambient.) Dissenting publications will now be discussed.
Malkus and Riehl(l960)find that moist adiabatic ascent of ambient sea level
air to the level of neutral buoyancy produces a sea level pressure of 1000 mb;
insufficient detail is presented to ascertain what precise calculation was con-
ducted. I n any case, such a pressure drop is insufficient t o sustain even a mod-
erate hurricane, and since these authors believe the eyeeyewall interface to
be virtually perfectly vertical, another means of achieving and sustaining winds
known to arise in hurricanes was required. The means ultimately postulated is
latent and sensible heat transfer from sea to air greatly in excess of ambient
transfer rates; the air ascending in the eyewall would then possess large
enough equivalent potential temperature to achieve the pressure deficits nec-
essary t o sustain even the highest known hurricane winds. The Riehl-Malkus
1010
900
700
h
n
9
n
500
100
0 10 20 30 40 50
(e)
FIG.15. In this replotting of Fig. 14, explicit pressure versus altitude curves for the
ambient, the moist adiabat taken t o characterize the eyewall, and the d r y adiabat
recompression taken to characterize the eye are presented. The moist adiabat is com-
puted by the iterative procedure described in the caption to Fig. 14. -Ambient,
-_- moist adiabat, dry adiabat.
extra oceanic heat source postulate will be examined below (Section 3.7-3.9).
Here it will be noted that even extrapolation of low-speed transfer coefficients
to extreme hurricane conditions could not justify the equivalent potential
temperature increases required by Riehl and Malkus; thus for very intense
hurricanes (only) an outward sloping of the eyewall was conceded by these
modelers. Such a procedure directly contradicts the conclusions of Shea
(1972), based on consideration of flight radar data; Shea finds that the eye-
eyewall interface slopes modestly outward for moderate hurricanes, but
questions whether it does so for intense hurricanes. However, radar data are
based on returns from precipitation and err on the side of verticality because
the strongest returns will come from the torrential rains of cumulonimbi,
rather than from weaker or decaying cumuli (Palm& and Newton, 1969, pp.
487491). Direct observational distinction between an interfacial slope of a
few degrees and no slope a t all, especially when some cloudiness may occur
in the eye, is difficult, and observational evidence can be cited t o support
either position. For instance, Palmen (1948) sketches an eyewall with ap-
preciable outward slope down t o nearly inflow-layer heights, based on obser-
vations of a September 1947 hurricaneoff Tampa, Florida. PalmenandNewton

(1969) construct a schematic diagram of Hurricane Helene, based on obser-
vational data of September 26, 1958, t h a t indicates an outward eyewall slope
commencing a few kilometers above the sea surface. On the other hand,
Riehl (1854, pp. 312-313) cites reports published in 1945 that described an
observed eye-eyewall interface t,hat did not have appreciable outward slope
until 30,000 ft. Definitive proof concerning whether or not dry adiabatic
recompression contributes t o the pressure difference available to sustain
hurricane winds appears nonexistent.
This section is concluded by noting that maximum wind speeds in hurricanes
are estimated, in the absence of aircraft reconnaissance, by means of previously
correlated formulas involving the area of the overcast " circle " on satellite
photographs (Hubert and Timchalk, 1969) (Fig. 3).
3.5. The Swirl-Divergence Relation for the Frictional Boundary Layer
Because the maximum speed achieved in a tropical cyclone is rarely much

over 200 mph, the Mach number rarely reaches even 0.3.Hence, when examin-
ing the dynamics (as opposed t.0 the energehs), an incompressible const>ant-
property model suffices.
A steady axisymmetric flow of an incompressible fluid is now studied t o
confirm the crucial point that), under rapid swirling, there is downflux from
region I to region 11, and sufficient downflux enters the surface frictional
layer to account for the mass flux up the eyewall. The analysis will be carried
out, in a noninertial coordinat,c system rotating at the constant speed of that
component of the rotation of the earth which is normal t o the local tangent
plane. Because the boundary layer divergence under an impressed swirl
(the major constraint furnished by the boundary layer on the inviscid flow
above i t ) is relatively small in magnitude, careful formulation and solut>ion
of the coupled quasilinear parabolic partial differential equations and bound-
ary conditions describing the layer are required.
The relevant equations are
(18) v-q=o
(19) V(q"/2)+(Vxq) x q + 2 8 , x q = - ~ ~ - v V x ( V x q )
+
whcrc @ = ( p i p ) (ae +
x r)"/2 gz. the gravitational acceleration g = -g2,
+
the vrlocit'y in nonincrt,ial coordinates v =Re x r q, the component of the
rotat.ion of the earth normal t o the local tangent plane !& = fi2, and the Irine-
matic viscosity (later given eddy-diffusivity values) is v.
Nondimensionalizat,ion is effected by letting q' = q / ( Y o Q ) l ' z , 1)' =
@/(Y0Q), r' = r/(Y?o/Q)1’2, and E = v/Yo where the Ekman number E < 1
and Yocharacterizes the circulation away from the boundary (such as the
maximum swirl speed times the radius at which it occurs). Dropping primes,
one has
(20) v.q
(21) V(q72) + (V x q) x q +2k x q = -vp --EV x (V x q)
These equations are studied in axisymmetric cylindrical polar coordinates
(22) q=UP +vd + wk, r = rB + z &
Away from the boundary (i.e. in region I) the following expansions are
adopted:
(23) p=?r(r,z)+*.-, v= V(r,z)+-*-,
w = El'' W(r,z ) +- - ,
* u =O ( W / ~ )
Substitution of (23) in (20) and (21) gives the gradient wind equation:
(24) rrz=O, W,=O, rrr=2V + V2/r
Subscripts r and z here (and x and 5 below) denote partial differentiation.
The axially invariant solution is complete when ~ ( ror ) V ( r ) is specified
[here V ( r ) will be given]; W ( r ) is found by matching the solution to (24)
t o the solution for the frictional layer 11, and in this sense W ( r )is determined
by the boundary layer dynamics.
If 5 = zE-'la [which implies that the frictional layer is O(E1I2)in thickness]
and if near the boundary
(251 u = Ub(r, 5 ) f .'
* 2 2) = vb(r, 5) + ' '. ,
w = E'''Wb(r, 5 ) +. * ' , p =p b ( r , 5) + * ' ' ,
then the axial component of the momentum conservation equation degener-
ates to (ap,,/a()= 0 in conventional fashion, so the pressure field in the bound-
ary layer is known from (24). If
(26)
$=rub, Y=rV, $=rub, iZ=2-1/2wb , x = r2, I==21/25,
then in ternis of dimensional quantities $ = r u / Y o i,h = rv/Y,, tij = )
w / ( 2 Q ~ ) ~ [" =
, z/(v/2Q)'/', x = Qr'/Y,,; the boundary layer thickness is
O(U/Q)~/'.In terms of quantities introduced in (26), one has from (20) and
(21), upon dropping the tildes,
(271 +I + wy =0
(28) $41 +wdr + w2 - - $"/2x - (* - Y) - +n = 0

(29) $*I +W& ++-h = O
Matching of expansions gives

(30) (+a:+-to, t,h+Y(x) given
and a t 5 = 0 no-slip conditions are adopted:

(311 5 = 0: += w = $h=o
Initial conditions are conveniently given bynotingthat a t x = xo ,for xolarge
enough, the solution is given by discarding all the nonlinear terms and retain-
ing the linear equations treated by Ekman, in which x enters parameterically
only. The solution to the balance of Coriolis, pressure, and friction forces is
well known:
(32) 4 = -[Y”(~)]sin(2-”~5)exp(
-2-1/25)
(33) t,h = [Y(z)][l- c o ~ ( 2 - l / ~ ~ ) e x p ( - 2 - ~ ~ ~ ~ )
(34) +
w = 2-1/z[Yz(x)]{1- [ ~ i n ( 2 - ~ / ~ <cos(2-1~2~)][exp(-2-1~25)]}
)
Specifically what is sought is w(r, C-tco) = W ( r ) for Y ( r ) of interest.
For r large, from (34)
(35) w(r, 5 -+ 00) = W ( r )= 2-1’2Y,(x)
Numerical integration by finite-difference methods is formidable because

the flow component in the timelike direction, u, is, in successively thinner
strips lying parallel t o the boundary, alternately in the direction of inte-
gration (numerically stable) and opposite to it (unstable). Though the radial
flow is, on net, in the direction of integration, the integration is marginally
stable. The only finite-difference results applicable to the hurricane problem
known t o the author were given by Anthes (1971) and are discussed later
(Section 3.10).
George (Canier et al., 1971) and Dergarabedian and Fendell (197213)
independently but simultaneously applied the method of weighted residuals
to the boundary value problem, and found that, except near the eyewall
where the method was inadequate, the linear result given in (35) sufficed
for the nonlinear problem as well (Fig. 16).Thus, for a form like Y = A(1-
z/xo) [for which +(ro, 5 ) = 0 according t o (32)-as required by the closed
system model] w ( r , 5 --f co)-+ -(A/xo), a small negative constant quantity
(equivalent to about 0.005 mph downdraft for physically interesting values).
Incidentally, the form just mentioned for the rapid swirl in the inviscid
flow above the frictional inflow layer is stable for parametric values of prac-
tical interest, according to Rayleigh’s criterion (Greenspan, 1968, pp. 271-
272).
0.12
La
9
0.4
0.2
-0.06 I I I I I I I I
0 2 4 6 8 10 12 14 16 18
X
FIG.16. Nondimensionalresults for the frictional boundary layer obtained by the method of weighted residuals, from Dergarabedian
a d Fendell (1972b). The divergence w(z, 5 --f a)) and the volumetric flux 6 = -j." +(z,5) d[ are presented for the impressed swirl
'Y = 1 - x / z o , zo = 20, believed pertinent to a hurricane outside the eyewall. Except near the axis where nonlinear inertial effects
dominate, the linear Ekman layer result, w(z, 5 rn) =YZ/2lIa,is an excellent approximation to the numerical results. The volumetric
flux 8(z)is thus linearly proportional to (zo - z)to good approximation. Normalized residuals indicate large errors for a < 3, and
discount the premature eruption as an artifact of the method. The solution by Carrier (1971a)for small z indicates the adequacy of
the linear Ekman result for the divergence to within a factor of two. Since v t (1/76)milesa/hr and R = (1/16)rad/hr, dimensionally
the results imply the boundary layer is of thickness O ( V / R ) ~=/ 0 ~ ( 1 mile), the d o d u x into the boundary layer is ( 2 ~ R ) l ' ~ w ( z ,
5 + 00) = 0(6 x miles/hr), and the volumetric flux erupting up the eyewall is (2n-T,av/L2)1'a6(z) = 0 ( 7 x lo3 miles3/hr)where
'Yo characterizes the eyewall relative angular momentum per unit mass. The implication is that the fluid initially in the boundary
layer sustains the hurricane for about a week, and the fluid lying above the boundary layer (with supplementary replenishment of
moisture from the ocean) can readily sustain the hurricane for more than another week.
The reason weighted-residual calculations fail near the axis, as discovered

by Carrier (1971a) and McWilliams (1971) by modified Oseen linearization
and by Burggraf et al. (1971) by seminumerical analysis, is that the structure
of the boundary layer so varies with radial distance that adequate represen-
tationin terms of one set of orthonormal polynomials is difficult. Par from the
axis of symmetry, friction is important across the entire layer of thickness
0 ( v / ! 2 ) l i z ;near the axis, fiiction is significant only in a small sublayer near
the “wall” of thickness O(r2v/Y)1i2,and the remainder of the inflow layer of
O ( V / ~thickness ) ~ / ~ is inviscidly controlled. However, for conditions of interest
in hurricanes, (35) is everywhere correct to within a factor of two, and often
far better.
Still further confirmation of the adequacy of the classical linear Ekman
swirl-divergence relation for the nonlinear frictional inflow layer under a
hurricane vortex has recently been given, in as yet unpublished work, by
K. K. Tam of McGill University. Professor Tam has applied the theory of
differential inequalities to the solution proposed by the Carrier group to the
boundary value problem (27)-(31) and indicated the satisfactory accuracy
of the result.
But while the solution may satisfy the boundary value problem, one may
ask how well the boundary value problem reflects the actual situation,
since a constant eddy viscosity has been adopted to model the turbulent
transfer. It would seem premature to adopt a second-order (or field-closure)
model that reduces the Reynolds time-averaged formulation to a determinate
set by means of a differential relationship between the rate of change of the
Reynolds stress and diffusion, production, and dissipation of turbulence;
still, an eddy viscosity invariant with distance from the air-sea interface
seems outdated, and further, some small slip might better model the boundary
condition a t the air-sea interface, taken planar.
It should be noted, however, that only one quantitative result from the non-
linear Ekman layer plays a role in the Carrier model, and that is the swirl-
divergence relation appropriate a t the extreme outer edge of the frictional
inflow layer. Furthermore, the eddy viscosity is almost always taken as
independent of distance from the “wall” over a t least the outer ninety
percent of the turbulent boundary layer thickness, aside from intermittency
effects, which play very little role in altering the profiles of time-averaged
macroscopic variables and are very often entirely neglected in calculations. I n
fact, one does not really know how to model slip and effective roughness
effects in the hurricane viscous sublayer, so restoration of eddy-viscosity
dependence on the normal coordinate would be a very speculative undertaking
a t best.
Other workers introduce a constant or wind-dependent drag coefficient to
model frictional effects near the air-sea interface for computer solutions.
There seems to be no inherent reason why either of these empirical, phenom-

enological devices (drag coefficients, eddy viscosity coefficients) is superior
to the other; either could be made to succeed by proper adjustment (curve-
fitting). Both face the problem that empiricism developed for relatively small
wind speeds and relatively smooth interfaces must be extrapolated t o higher
wind speeds and very rough seas (Ooyama, 1969, p. 19). There exist attempts
to study the frictional inflow layer entailing both simpler and more detailed
modeling than that just presented; for example, Riehl (1963a) invokes con-
servation of potential vorticity for the frictional inflow layer, and Riehl
and Malkus (1961),who assert that dissipation of kinetic energy in the surface
layer is independent of distance from the center, find that they must adopt
exceptionally large eddy transfer coefficients to resolve problems that arise
in computing a refined mechanical energy budget.
3.6. The Energetics of the Frictional Boundary Layer and Throughput Supply
For the frictional boundary layer I1 and throughput I, Carrier et al. (1971)
take the following approximations as adequate for the quasi-steady mature
phase: (1)the Prandtl and Schmidt numbers are unity; (2) the hydrostatic
approximation holds; (3)the boundary layer approximation holds (derivatives
normal to the boundary exceed those tangential to the boundary, but velocity
components parallel to the boundary exceed those normal t o the boundary);
(4) the eddy transfer is adequately modeled by the laminar flux-gradient
relations for diffusion of mass, momentum, and heat (as given by Fick,
Newton, and Fourier, respectively), except that the augmented kinematic
(eddy) viscosity may vary with radial position (but not with axial position);
and (5) the mixture of dry air and water vapor may be taken as a perfect gas
with constant heat capacity over the range of temperatures of interest here.
I n view of the limited understanding of quantitative formulation of cumulus
convection and turbulent transfer, on the cyclone scale of interest here these
five approximations seem reasonable. Clearly the familiar Reynolds analogy,
together with the constancy of the eddy viscosity over most of the boundary
layer, is being adopted. It can then be shown that for L held constant, the
following equation, a generalization of those given by Crocco in fluid dynamics
and Shvab and Zel’dovich in combustion, holds in the meteorological context
of interest here :
(36) u(aH/&) +w ( ~ H / ~=zv(a2H/az2)

) +D
where D denotes radiational loss, taken by Carrier et al. (1971) to be repre-
sentable as
with f (z) (to be discussed below) known. The H used throughout this section
is H , as defined in (3); the subscript t is dropped t o avoid double subscripts
below. The boundary-initial conditions are taken to be
(37) z = 0 : H = H,; z =zl: H = H ,
(38) r = r o : H = Ha@)
where, again, ro is the outer edge of the storm and z1 is the top of the storm.
The ambient profile H,(z) is given; H , = HI = N,(z = zl)= Ha@= 0).
Taking H, = H,(z = 0) implies that for r > R (where R is the radius of the
eyewall), the ocean surface temperature is a constant and the water vapormass
fraction (which takes on its saturation value at the nominally plane sea
surface z = 0 ) is taken independent of pressure (see below). Since the thermal
conductivity of water greatly exceeds that of air, uniform sea surface tern-
perature appears to be a good approximation. That N,= H,(z = 0) follows
from the definition of the lid on the storm.
The terms, from left t o right, in (36) represent radial advection; axial
convection; turbulent diffusion and cumulus convection (both parameterized
in v, which will henceforth be treated as constant, though this is not nec-
essary); and radiation loss. Throughout I, and in I1 a t r = r,, , (36)may be
approximated as
(39) w ( a H / a ~-) V ( ~ ~ H /=
~ -?f (Zz ~) H)
this follows because +(z,,)= 0 [cf. (32)] and u(aH/ar) is negligible in I
[cf. (23)]. In view of the boundary conditionsf(z) must be chosen to permit
H ( z ) = Ha(z),where H,(z) is given (Jordan, 1957); Carrier et al., (1971) take
w(r, , z ) E wa(z) = 0, but there appears to be no need t o require this. Thus,
(40) w,(z)(~H,/~z)
- V(a2H,/ik2)= - f ( z ) H ,
In fact, taking H = Ha@)throughout I and I1satisfies the initial and boundary

conditions, and renders u ( a H / & )= 0. Further, the function w(r, z ) is available
from the nonlinear boundary layer dynamics just discussed; the dynamics
~ 5 + 00) = W ( r ) ,the value a t the outer edgeof thebound-
reveals ( w ) ~=, w(r,
ary layer-and W is independent of r over a wide expanse. OnIy if W is
appreciably different from w, can H depart from H a , and such a variance
occurs only in the eyewall, where vertical convection dominates all other
terms in (36) because (by simple continuity considerations) w is increased
two orders of magnitude over its maximum magnitude in I or 11.Hence, for
a t least portions of the eyewall,
(41) w aHpZ = 0
and the relevant boundary data for this hyperbolic suboperator is that given
a t z = 0 by (37). All this discussion, more carefully argued in Carrier et al.
(1971), leads to two significant results:
1. Throughout the frictional boundary layer and the throughput supply,
outside the eyewall the total stagnation enthalpy is approximately fixed
a t its ambient stratification. Furthermore, the small correction owing to
inertial effects in EL hurricane is readily seen to be a decrease of H with z
such that the enthalpy gradient a t z = 0 is increased slightly (about ten per-
cent or so).
2. In portions of the eyewall, the total stagnation enthalpy is, to good
approximation, constant at its sea level value, i.e. columns of air are very
nearly rising on a moist adiabat.
Numerical values of interest are the ambient net sealair enthalpy transfer
and the eddy viscosity (Carrier et al., 1971):
-pv[aH(r,z =o)/az] & --p@H,(r, 2 =O)/az]

(42) = 1.8 x lo5 ergs cm-2 sec-l
v = 2.7 x lo5 cm2 sec-l
Because this result is not compatible with many existing hurricane models,
it becomes necessary to identify the reasons for disagreement. Hence in the
following sections the Riehl-Malkus theory is scrutinized.
It may be worth emphasizing that (36) yields a steady solution for the
frictional inflow layer with all inertial terms considered. It is not based on
obviously inadequate models of the f i o w layer, such as frictionless or purely
horizontal flow.
However, this result is really not quite so accurate as suggested by Carrier,
Hammond, and George, owing t o the omission of compressibility effects. A
decrease in gas density attends the decrease in gas pressure (of up to 12 %)
as one moves radially inward in the frictional inflow layer from the outer edge
toward the eyewall This decrease in air density is negligible for the dynamics,
but not the energetics. If the gas density decreases as one goes radially inward,
but the total stagnation enthalpy sea to air transfer remains approximately
a t the ambient rate (Carrier, 1971b, p. 158 footnote), then the vapor mass
fraction (ratio of vapor density to gas density) increases because there is less
mass to accept the same transfer. Carrier et al. (1971, p. 162) acknowledge the
effect, but fail to account for it in their calculations. It seems more appro-
priate to study the vapor density, which a t sea level depends only on sea
surface temperature, rather than the vapor mass fraction. A rough &st
estimate is that perhaps a 20 yoincrease over tropical ambient levels in latent
and sensible heat transfer from the ocean t o the atmosphere occurs near the
eyewall of hurricanes owing to this density reduction effect.
3.7. The Riehl-Malkus Postulate of an Oceanic Heat Source
While the exposition of the mature tropical cyclone has been given here in
terms of the Carrier model, most published works adopt the framework of the
Riehl-Malkus theory alluded to earlier (Riehl, 1954, 1963a; PalmBnandRiehl,
1957; Malkus, 1958, 1962; Malkus and Riehl, 1960; Riehl and Malkus, 1961).
The cornerstones of this theory have been summarized by Malkus (1962,
p. 232):
The new model relates core maintenance to mechanism, namely, cumulonimbus

convection and Bea-air exchange. To sustain the required pressure gradients, two
coupled processes are necessary: j h t , a greatly naagnijied oceanic irnput of senaible
and latent heat, and secondly, the undilute release of the latter in concentrated hot tower
ascent, 80 that air of high heat content is pumped rapidZy into the upper troposphere.
The quantitative establishment of these crucial relationships was carried out in a
joint analytic and observational framework (Malkus and Riehl, 1960).
The important role of large numbers of cumulonimbi in the eyewall has

been confirmed, although the Riehl-Malkus concept of a direct hot-tower
heating mechanism has since been challenged by Lopez (1972b), as discussed
in Section 2.2. Here the other major contention [“. . . it is postulated that
lowering of surface pressures in hurricanes arises mainly through an ‘ extra ’
oceanic heat source in the storm’s interior ” (Malkus and Riehl, 1960, p. 12)]
is questioned. There is undoubtedly some augmentation of the ambient level
sensible and latent heat transfer from sea to atmosphere over a wide ocean
expanse under a hurricane, and the warm autumnal tropical seas are in a
major way responsible for the atmospheric total static enthalpy profile to
begin with; however, an augmentation of over an order of magnitude in
oceanic heat transfer concentrated, according t o Malkus and Riehl, largely in
the region 30 to 90 km from the cyclone center seems unnecessary and
unlikely (Section 3.8).
The Riehl-Malkus theory, as noted earlier in Section 3.4, takes the eye-
eyewall interface t o be effectively perfectly vertical so some mechanism
other than dry adiabatic recompression must be found to sustain hurricane
force winds. Why this mechanism is stated to be an “extra” oceanic heat
source is understood by reconstructing their logic. It may be worth noting
a t this point that Riehl and Malkus permit the relative vorticity (rather than
the relative velocity) to vanish a t the outer edge of the storm; workers using
their concepts tend to permit convective, but not diffusive, transport across
the outer cylindrical boundary, while workers on the closed Carrier model
permit diffusive, but not convective, transport across boundaries.
Malkus and Riehl (1960, pp. 3, 7) acknowledge the existence of a low-level
frictional inflow layer, although in a sample ca€culation there is .flux from
the boundary layer from the outer edge of the eye out to 500 km a t an average
ascent rate of 0.7 mph. Thus air flows in through the boundary layer from
the outer edge of an open system, rather than sinking down into the frictional
layer as in Carrier's model. The boundary layer air, according to Malkus and
Riehl, undergoes adiabatic expansion as it spirals inward toward the center,
yet it remains isothermal. This requires a vast, rather localized increase in
sea-to-air transfer between the ocean and the contiguous atmosphere. That
the gradient normal to the air/sea interface of temperature and of water vapor
mass fraction is large enough to be consistent with vastly increased air/sea
transfer is accepted as possible:
I n the outskirts of a hurricane the temperature of the inflowing air drops slowly
due to adiabatic expansion during (horizontal) motion toward lower pressure. It is
one of the remarkable observations in hurricanes that this drop ceases at pressures
of 990-1000 mb and that thereafter isothermal expansion takes place. Presumably,
the temperature difference between sea and air attains a value large enough for the
oceanic heat supply to take place at a sufficient rate to keep the temperature
difference constant (Malkus and Riehl, 1960, p. 9).
The actual transports [between sea and air], of course, are very large in the
hurricane compared to the trades. Sensible heat pickup is 720 cal/cma/day, an
increase by a factor of 50 over the trades . . .; latent heat pickup is 2420 cal/cm2/day,
higher by a factor of 12-13 (Malkus and Riehl, 1960, p. 12).
Similar arguments are made elsewhere by Riehl (1954, pp. 28&287):

.
Many published records, notably those by Deppermann . ., have proved that
the surface temperature outside the eye is constant or decreases very slightly
toward the center. The implications of this remarkable fact passed without notice
.
until Byers .. drew attention to it. The temperature of the surface air spiraling
toward a center should decrease if adiabatic expansion occurred during pressure
reduction. For instance, air entering the circulation with the average properties of
the mean tropical atmosphere should reach the 930 mb isobar with a temperature
of 20.5"C and specific humidity of 17 gm/kg. Because of condensation, a dense fog
should prevail at the ground inward from the 970 mb isobar. But this is never
observed. It follows that the potential temperature of the surface air increases along
the inward trajectories. We also know that the specific humidity increases and that
the cloud bases remain between a few hundred and 1000 feet
The surface air thus acquires both latent and sensible heat during its travel toward
..
lower pressure. .
A source for the heat and moisture increment is obvious. The ocean is greatly
agitated, and large amounts of water are thrown into the air in the form of spray.
It is hard to say where the ocean ends and where the atmosphere begins! As the air
moves toward lower pressure and begins to expand adiabatically, the temperature
difference between ocean and air suddenly increases. Since the surface of contact
between air and water increases to many times the horizontal area of the storm,
rapid transfer of sensible and latent heat from ocean to air is made possible. I n the
outskirts, say beyond the 990 mb isobar, the turmoil is less and the process of heat
transfer is not operative.'
'
From " Tropical Meteorology " by Herbert Riehl. Copyright 1954 by McGraw-Hill
Book Company. Used with permission.
The Riehl-MaIkus theory that greatly augmented heat and mass transfer
sustains the tropical cyclone has, in fact, been parameterized into all
existing computer simulations (see Section 3.10). For example, Rosenthal
(1971b) closely reflects the Riehl-Malkus theory and notes similar logic in
the work of another computer modeler of tropical cyclones (Ooyama, 1969):
Air-sea exchanges of sensible and latent heat have long been considered impor-
tant ingredients in the development and maintenance of tropical storms. Palmbn
(1948) showed, on a climatological basis, that tropical storms form primarily over
warm ocean waters (TSes > 26°C). Malkus and Riehl (1960) showed that the deep
central pressures associated with hurricanes could not be explained hydrostatically
unless the equivalent potential temperature, 8., in the boundary layer was 10”t o
15°K greater than t h a t of the mean tropical atmosphere. Byers (1944) pointed
out that the observed near-isothermal conditions for inward spiraling air in the
hurricane boundary layer required a source of sensible heat t o compensate for the
cooling due to adiabatic expansion. . . .
Ooyama (1969) fround drastic reductions in the strength of his model storm when
the air-sea exchanges of sensible and latent heat were suppressed. He pointed out
that at sufficiently large radii, the boundary layer is divergent (the so-called Ekman
.
layer “sucking”). .. This subsidence tends t o decrease the boundary layer 8.
since 88,/8z < 0 in the lower troposphere. Ooyama argued that unless the energy
supply from the oaean can again raise the 8. of the boundary layer air t o sufficiently
large values before the inflowing air reaches the inner region, the convective activity
will diminish in those regions and, hence, the storm will begin to weaken.
Ooyama’s line of reasoning can be extended to show that evaporation is far more
important than sensible heat flux. The air sucked into the boundary layer has a
higher potential temperature than the original boundary layer air. The subsiding
air has a smaller 8, only because it is relatively dry. (Rosenthal, 1971b, p. 772).
3.8. The Intensity of a Tropical Cycbne and the Underlying Sea Surface
Temperature
Palmen and Riehl (1957, p. 156) state: “Evidently, a cyclone will decay
rapidly if it encounters thermally unfavorable conditions. The generation
term in [the equation for conservation of kinetic energy] w ill then decrease
rapidly; it will become negative if the air ascending in the core is cooler than
the surroundings in spite of release of latent heat. This would happen when a
storm moves over a relatively cold ocean surface, over a continent, or when
colder air masses invade the cyclone near the surface.”
I n discussing the Riehl-Malkus theory, one should distinguish between what
these authors themselves assert, and what others citing their work have added.
For example, Shuleykin (1970,1972)goes beyond Riehl andMalkus andasserts
that the “. . . power of the hurricane . . . increases sharply with the tem-
perature of the underlying water surface ” (Shuleykin, 1972, p. 1).However,
attempts to correlate central pressure deficit and sea surface temperature,
without sufficient account of other factors, seems simplistic. This particular
point will be now discussed.
56 FltLLNCIS E. FENDELL
First, ocean temperatures are difficult t o determine accurately from cur-

rently available records and by currently available methods (Perlroth, 1967 ;
Hidy, 1972, p. 1091). Next, Gentry (1969a, p. 406) presents data relating
". .. the maximum intensity of several tropical cyclones to the temperatures
of the sea beneath them and shows that both severe and weak tropical
cyclones occur when ocean temperatures are relatively high. This suggests
that variations in parameters other than the transfer of heat from the ocean
t o the atmosphere also influence the storm's intensity . . . ." Actually Gentry's
data (Fig. 17) indicate that some intense tropical cyclones lie over relatively
cold water ( <28.0°C). While Brand (1971) cites a supposed correlation of
central pressure deficit with sea surface temperature throughout the lifetime
of Hurricane Ester of September 9-26,1961, it may be noted that during one
twelve-hour period of constant sea temperature, the central pressure rose 15
mb; for three days while the sea temperature hovered about 84"F, the central
pressure nonmonotonically rose from 930 mb to 955 mb; and at various
times when the sea temperature was a t 86"F, the central pressure was a t
values as low as 927 mb and as high as 953 mb. Perlroth (1967) shows that for
Hurricane Ginny of October 21-28, 1963, for five days while the sea tem-
perature hovered near 80"F, the central pressure fell from 995 mb t o 970 mb
(Fig. 18). Perlroth (1969) cites the importance in tropical storm generation of
considering not only high surface temperature (above 26°C) but also small
oceanic temperature variation with depth (less than 7°C within 200 f t of the
surface). Perlroth emphasizes that the oceanic environment is only one factor
in tropical storm intensity, and the sea surface temperature is only one factor
in the oceanic environment.
3
-s
10107-
1000-. .' I I
..
a
1 . I 1. a
1.0 I -
-50
n
$2
y
2
Ln
990-
980-
a
a
. -668
-81
5
4
. . .
y
n
970- -93
a
960- -102 3
2
Y .*
2 950- -111 g
2
9 940 - a - 120 5
930- -128 5
3 920- -136 3
z
2 910- -143 ,D
$ 900- -149 ;
;
a
SEA SURFACE TEMPERATURE (OC)

-
FIG. 17. Measurements of minimum sea level pressure versus local sea surface
temperature for several tropical cyclones (from Gentry, 1969a, p. 406; reproduced with
permission of the American Meteorological Society). Each dot represents relation for an
individual storm selected from 1965-67 seasons.
I I I I I 1 I I I I I 1 1 1 I I I 1 -
PRESSURE
-
50 r 1005- &----
Q\
\
'1
1WOt \
\
995 - -
990- -
dLI m 985 -
3
I-
80 975 -
970 -
21
211 I 22 23 24 25 26 27 28
965
102 202' 06, lkz 'i2z 1z
: 2k d8z 1Lz I d4z lk, dlz lbz 2bz' d6z Ikz Id22 1!2z 222
FIG.18. The central sea level pressure is plotted as a function of the local sea surface temperature for Hurricane Ginny, 21-28
October 1963, which passed over the Gulf Stream on 24 October (from Perlroth, 1967, p. 266). Solid line indicates extrapolated
temperature reading besed on composite sea surface temperature chart, 11-19 October, 1963.
Finally, the fact that hurricanes leave cold wakes in the ocean might suggest
that they do drain the upper sea layers of heat. However, the reason for the
sea surface temperature dropping about two degrees Centigrade after hur-
ricane passage is the upwelling of lower, colder water owing to convection
currents induced in the ocean by the pressure deficits of the hurricane
(Perlroth, 1967). Slow-movingintense tropical cyclones can cause upwelling,
from depths of 40 to 65 m, within the radius of hurricane force winds;
compensatory downwelling of as much as 80 to 100 m occurs 45 to 110 n miles
from the path of the storm (Revesz, 1971; Leipper, 1967). The warm ocean
water near the center of the hurricane flows radially outward, and is replaced
by colder water from lower depths; outside 65 n miles from the center, below
the outflow, there is compensatory inflow toward the center of the hurricane
t o complete the convection cell (Leipper, 1967; Wright, 1969).
Leipper finds the temperatue readings a t depths in the outer portion of the
hurricane path t o be higher than the temperatures before the storm. Thus a
significant portion of the heat of the upper layers would seem to be trans-
ported radially outward, vertically downward, and then radially inward.
Leipper and Wright, however, postulate that all the heat of the relatively
warm upper oceanic level in excess of that of the colder upwelling is trans-
ferred to the atmosphere as an augmentation of the ambient rate8; Leipper
suggests a transfer rate of 4500 cal/cm2-day under Hurricane Hilda (1964),
I n this regard deductions from data taken off Barbados in August 1968 may be of
interest (Warsh, 1973). Warsh characterizes the convective activity of the tropical
atmosphere on a given day in terms of graduated modes, ranging from mode one (severely
depressed convection with almost no cumulus) to mode six (severely enhanced convection
with mostly cumdonimbi and heavy showers). Warsh finds the sensible heat flux from
the ocean t o the atmosphere increases only extremely slightly from mode one t o mode six.
In contrast, the latent heat flux decreases t o one-half in magnitude from mode one to
mode four, then increases from mode four t o mode six to nearly its mode one magnitude.
Since the equivalent potential temperature in a mode six atmosphere is almost every-
where 335°K or higher, and since the latent heat flux from the sea in a mode six atmos-
phere is still less than the equivalent of one-half centimeter of evaporated water per
day, the results would not seem to augur well for a sometimes postulated order of magni-
tude increase in latent heat transfer over ambient levels under even a severe storm.
Ostapoff et al. (1973) also deduce, from oceanographic observations in the ITCZ during
a broad light rainfall from cumuli, that the latent heat transfer was appreciably reduced
owing to high humidities, but the sensible heat transfer was somewhat increased, with
the result that the Bowen ratio had an average value of about 0.2 (over twice the tropical
norm), and a maximum value of 0.247. Finally, even though the measurements were
taken over free water surfaces in the laboratory or over lakes a t low wind speeds, the
results of Easterbrook (1969) are interesting as further evidence that intuitive expecta-
tions about latent heat transfer can err. Easterbrook finds that the vortical wind patterns
created by waves on free water surfaces can suppress vertical turbulent transport of
water vapor and reduce evaporation, such that larger height-to-period ratios for surface
waves can lead to smaller mass transfer across the two-phase interface.
and Wright suggested 7200 cal/cm2-dayunder Typhoon Shirley (1965). The

previously cited figure of Malkus and Riehl (1960) for a moderate hurricane
was 2420 cal/cm2-day. Leipper acknowledges that the transfer so predicted
is so much larger t o the right of the path that even observed asymmetries
in intensity are inadequate for his estimation. Further reasons for doubt
concerning Leipper's calculation for Hurricane Hilda will be given in
Section 3.9.1. [Most discussions emphasize the cold wake left in the upper
oceanic layers by intense, slowly moving hurricanes; however, the reduction
of surface layer temperature by upwelling and mixing may be occurring
directly under the hurricane. Black and Mallinger (1972, p. 74) find evidence
of radial outflow and downwelling of warm upper layer water for Hurricane
Ginger (1971); the upwelling of colder water was found to reduce the sea
surface temperature by 4°C on one day, and by 2.5"Con another day when the
hurricane was translating a t a higher speed. Reduced latent and sensible
enthalpy transfer from sea to atmosphere ensued, with a reduction in inten-
sity. Thus, the interaction of a hurricane with the ocean may be such that
reduced transfer from the sea, not augmentation of transfer over the ambient
level, may occur (cf. Section 3.9.1).] The hurricane-induced cooling in the
ocean may be anticipated to be reduced to 1°C or less for particularly stable
ocean stratification, as would be the case for very cold water below the
well-mixed layer.
Leipper and Volgenau (1972) later introduced the concept of a hurricane
heat potential, which they applied to oceanographic data for the Gulf of
Mexico. The idea is that all the heat content of the upper oceanic layer (top
500 cm) implied by water temperature in excess of 26°C is available to sustain
a hurricane. The authors acknowledge that their theory implies that in a
hurricane, the air controls the sea temperature, although in almost every
other physical situation, the sea is regarded as characterizing the air tem-
perature. If all this heat were really available for transfer, why would a
hurricane be required (rather than just the normally operative turbulent
transfer mechanisms) to effect the transfer? One wonders how such a large
potential could be developed, and why hurricanes are not always formed over
waters with surface layer temperatures above 26°C.
3.9. Critique of the RiehkMatkus Model

A critique of the Riehl-Malkus reasoning that an extra oceanic source of
latent and sensible heat maintains the tropical cyclone might begin by noting
that use of the adiabatic expansion relation p N TYi(Y-l)is inappropriate for
a (turbulent) frictional layer. Such a relation implies that the entropy is
constant for each fluid element. Riehl (1954, p. 26) is correct when he assumes
that neglect of diffusion gives almost a 6°C drop (with condensation) in passing
from 1014 mb (autumnal sea level ambient) to 930 mb. I n fact, the inflow layer
60 FRANCIS 1. FENDELL
is largely cloud-free, and the same adiabatic expansion taken as dry produces
a 7.3”Cdrop. Such a large temperature drop does not occur (though whether
the static temperature is really constant as Riehl believes is uncertain), and
should never have been expected. But Riehl (1954, p. 286) was surprised at
measurements he interpreted to imply constant static temperature during
inflow in the frictional layer, and passed immediately from the (clearly in-
appropriate) isentropic expansion to the postulate of latent and sensible heat
transfer from the ocean greatly in excess of ambient transfer 1evels.O
3.9.1. Evaluating the Arguments for Greatly Augmented Enthalpy Transfer.
There are no indisputable semiempirical theories or direct measurements of
the heat and mass transfer from the ocean to the atmosphere in a hurricane:
“unfortunately there is little information on C , (‘ the nondimensional co-
efficientfor the air-sea energy interchange ’) under hurricane conditions, other
than the semispeculative guess that the exchange coefficientsof latent heat,
sensible heat, and momentum are probably of the same magnitude ” (Ooyama,
1969, p. 15). Hidy (1972, pp. 1086 and 1097) notes the lack of quantitative
progress on air-sea transfer, cites the paucity of data on the drag coefficient
at sea level for winds over 20 mph, and mentions the difficulties in extra-
polating data accurately; in fact, one of few known results casts doubt on the
validity of the Reynolds analogy relating turbulent transfer of heat and
momentum at the air-sea interface. There have been attempts to extrapolate
to hurricane conditions empircal laws relating ocean evaporation rates to
wind speed. The actual measurements refer to low wind speed and involve
relatively dry air, far from saturation; extrapolation to hurricane winds and
nearly saturated low-level tropical air is unjustified and misleading. Garstang
(1967)emphasizesthat a critical assumption in the theory of Riehl and Malkus
is that the sea-to-air transfer coefficients increase linearly with wind speed
indefinitely into the hurricane regime. Zipser (1969, p. 813) found that using
such linearly extrapolated bulk turbulent transfer relations, which imply
an order of mapitude increase in latent heat transfer under moderately
strong tropical disturbances over the ambient rate of transfer, would have
produced clouds in an observed tropical disturbance with unsaturated con-
vective downdrafts on a time scale an order of magnitude too short. Zipser
also notes that Riehl and Malkus expected downdrafts around cumulonimbi
Kraus emphasizes that the Riehl-Malkus theory depends on an isothermal inflow
layer and discusses whether frictional effects are fully accounted for in that theory:
“The importance of [an] additional energy supply to air in the hurricane, before it
ascends ill the warm core, was stressed particularly by Malkus and Riehl (1960). The
argument would be weakened somewhat if a (verbal) suggestion by Carrier was found
to be true, Carrier maintains that surface friction must cause a deceleration and compres-
sion of a surface air parcel along its trajectory; this compression compensates to some
extent for the expansion that would be associated otherwise with the externally imposed
pressure reduction” (&am, 1972, p. 208).
t o be saturated, but the observed downdrafts were unsaturated and so

details of the Riehl-Malkus predictions on sea-to-air transfers near cumulo-
nimbi must be modified. I n the absence of more quantitative evidence Riehl
(1954, p. 287) asserts that much sea spray is tossed into the air in a hurricane.
This does not in itself assure augmented net enthalpy transfer from sea to
air, since heat must be drawn from the air to evaporate the drops for later
condensation of the water vapor in the eyewall and inner rainbands. Evapor-
ation of spray leads to a temperature decrease as well as a dew point increase,
such that the total stagnation enthalpy remains unchanged. Heat loss from a
spray particle (which acts like a wet-bulb thermometer) to surrounding air
of relative humidity of 0.85 or higher will be small.
Possibly the closest approximation t o a direct measurement of the Riehl-
Malkus postulate of an augmented oceanic heat source in a hurricane are the
tritium-tracing measurements by Ostlund (1968, 1970) in Hurricane Faith
(1966). The tritium deposited initially in the stratosphere following American
and Russian fusion bomb tests in 1961 and 1962 temporarily created trace
amounts in the troposphere that exceeded the amounts in the ocean; deter-
mining tritium concentrations could theoretically determine the source of
water substance in a hurricane. While the experiment cannot be repeated
without the resumption of atmospheric nuclear testing, and while the present
author does not regard the experiment as highly reliable, Hidy (1972) re-
ported the result without reservations. &tlund found that the ambient rate
of latent heat transfer from sea to atmosphere in the tropics (which is equiva-
lent to evaporation of about 0.5 cm/day of water) was increased by a factor
of about three within the inner 100 km radius of the hurricane, as opposed to
the order of magnitude increase predicted by the theory of Riehl and Malkus.
There seems t o be no indisputable direct evidence for the internal ocean heat
source postulate of the Riehl-Malkus theory; the question does remain whether
or not Carrier and his co-workers have obviated the indirect justification,
specifically, the assertion by Riehl and Malkus that the oceanic heat source is
necessary to satisfy the laws describing conservation of energy and momentum
in the storm.
I n this respect the earlier quotation from Rosenthal (1971b, p. 772),
citing Ooyama (1969, pp. 1&15), should be discussed. Rosenthal asserts that
higher level air sinking down into the surface frictional layer will be drier
and of lower equivalent potential temperature than the original boundary
layer air. If the quasi-steady approximation of the Carrier model is justified,
then in the mature stage the total stagnation enthalpy is a function mainly of
position. According to Carrier and his co-workers, the plume and radiative
transfer mechanisms that carry heat and moisture up into the ambient tropical
atmosphere persist, with neither augmentation nor diminution to any ap-
preciable degree, in the presence of the storm. These transfer mechanisms help
compensate for rain-out in the outer spiral bands such that by the time a
particle has slowly sunk into the boundary layer a t 0.005 mph, i t has been
enriched in total stagnation enthalpy to the ambient value a t its current height.
I n other words, much of the sea-to-air transfer of latent and sensible heat
continues to pass across the boundary layer with little diminution, just as
in the ambient, such that the enthalpy of air originally in the 700900 mb
strata increases t o that of air originally in the boundary layer owing to en-
richment of its total static enthalpy as it slowly descends. Only as the hurri-
cane leaves the tropical oceans is this ambient transfer reduced; the air
entering the boundary layer eventually is of lower total static enthalpy,
since it comes from air which is originally higher in the tropical ambient
(hence colder and drier), and since that air is not appreciably enriched as it
descends. I n this way traverse over ocean patches of varying temperature can
help cause the well known nonmonotonic perturbations in hurricane intensity
within the general level of strength computed from the spawning ambient as
discussed earlier (see also Section 4.4).
I n this critique of the Riehl-Malkus model, one should note a later modifi-
cation of earlier statements. Malkus and Riehl (1960, p. 17) write: “The
total added heat energy from the ocean is on the order (for the moderate
storm) of 2.5 cal/gm, while the average normal heat content (latent plus
sensible) of tropical air is about 80 cal/cm.” Thus the enthalpy contribution
added by the ocean to ambient air enthalpy is about three percent for a
moderate storm. “ The heat . . . gained by the air is only a minute fraction
of that carried inward through the cylinder a t the distance of the 1OOO-mb
isobar . . . Nevertheless, it is these small increments that produce the strong
inward warming. . . . They are thus of utmost importance for generation and
maintenance of tropical storms, even though they may be wholly neglected in
a general energy balance. . . .” (Palm& and Riehl, 1957, p. 156). That such a
marginal amount of heating is critical to both generation and maintenance
seems intuitively surprising. Later Malkus (1962, pp. 247-249) limited the
role of greatly augmented enthalpy transfer strictly to the maintenance of
the hurricane, since during development the high winds that supposedly
permit the increased turbulent transfer do not exist.
Hawkins and Rubsam (1968) have computed the structure and budgets
for Hurricane Hilda for October 1, 1964, when it was an intense storm over
the Gulf of Mexico. An unprecedented five-level collection of data was avail-
able for exhaustive analysis, and the total static enthalpy budget was cal-
culated very similarly t o the approach used by Malkus and Riehl (1960).
Hawkins and Rubsam found, however, that (upon inclusion of the postulated
latent and sensible enthalpy flux from the sea) the amount of energy estimated
t o be radiated away from the top of the storm was twenty times what seems
physically possible. The kinetic energy budget computed for Hilda was also
a t significant variance from the one computed by Riehl and Malkus (1961)
for Hurricane Daisy (1958). I n view of this, Leipper's (1967) suggestion that
the oceanic heat transfer to Hilda would be underestimated, using Malkus's
formulas, by 50-90 yoseems incomprehensible.
3.9.2. Temperature Measurements in Hurricanes. If numerous accurate

temperature measurements in rapidly swirling, heavily raining portions of
the hurricane could be readily carried out, resolution of doubt about the
existence of a large ocean heat source within hurricanes could be quickly
accomplished. However, such measurements are still subject to controversy
and inaccuracy. Aircraft measurements tend t o be taken above the i d o w
layer for safety reasons: ". , .the preferred altitude for reconnaissance traverses
of a hurricane is from 10,000 to 14,000feet " (Meyer, 1971a, p. 58). Currently
used instrumentation may suffer from slow response time, from wet-bulb
effects, and from the need to deduce the static temperature from a measure-
ment that includes the dynamic contribution (Meyer, 1971b, pp. 19-37).
Riehl and Malkus (1961, p. 188) discuss possible unresolved errors in mid-
tropospheric temperature measurements of a few degrees Centigrade in
magnitude.
First, to a reasonable degree of accuracy a dry-bulb thermometer measures
+
[ T (q2/2c,)], and a wet-bulb thermometer (with knowledge of the pressure)
+ +
yields [ T (q2/2cP) (LY/cp)](Hess, 1959, p. 61). At very low altitudes, the
dry-bulb temperature is the total potential-like temperature ( O,/cp), and the
wet-bulb temperature is then related to the total stagnation temperature
(H,/cp), because the missing gravitational contribution (gz/c,) is negligible.
Here the recovery factor (usually denoted r ) for the thermometers is taken as
unity; for the relevant Mach and Reynolds numbers this seems satisfactory,
since for a well-designed instrument r >0.9 for turbulent flow in air. The
belaboring of the dynamic contribution would normally be unjustified in
meteorology, but a 200 mph wind is equivalent to a contribution of about
-
4°C; this amount is in excess of one percent of both the static temperature
( 300°K) and also the total stagnation temperature ( 350°K) for the sea
N
level autumnal tropical ambient. Whether temperature measurements made

several decades ago, when hurricanes generally were not believed t o be cap-
able of such speeds, are properly corrected for this dynamic contribution is
not always clear. It should also be recalled that, while the theory needs some
improvement, the still quite useful result of Carrier et al. (1971) implies that
the wet-bulb temperature would remain approximately radially invariant in a
traverse of a hurricane along a ray and at constant height up to about 700 mb
altitude, from the outer eyewall to the outer edge.
Deppermann (1937, p. 5), whose work is frequently cited by Riehl, writes:
". . . remarkably uniform diurnal temperature oscillations are maintained
both before and after a typhoon; but when the station is under typhoon in-
fluence, a reduction of the maximum temperature by about 3°C occurs, while
the minimum temperature remains the same." If by maximum and minimum
temperatures Deppermann means dry-bulb and wet-bulb temperatures,
respectively, and if one assumes Deppermann correctly accounted for the
dynamic contribution (qa/2cp),then Deppermann finds that the static tem-
perature T decreases by about 3°K in a typical typhoon, while the total
stagnation temperature increases by a few degrees Centigrade (5'4°C a t the
very most). I n any case, there is no evidence of large increases in (Hlc,), as
an oceanic heat source would allegedly cause. Also, if (Hlc,) is roughly
radially constant in the inflow layer, then a decrease in T of about 3.7"C
from sea level autumnal tropical ambient is necessary for condensation, and
this is reportedly not achieved in a typical typhoon, so a cloud-free frictional
inflow layer is no surprise in Carrier's model.
Arakawa (1954, p. 119) reported that the air temperature dropped 1°C
from 28°C to 27"C, and the wet-bulb temperature remained about constant
radially at approximately 26.5"C, during the passage of a n 898 mb typhoon
over a Japanese weather ship in October 1944; only in the eye did the two
temperatures rise sharply. Winds rose t o about 120 mph in the eyewall.
Again, this seems compatible with the Carrier-Hammond-George predictions,
which include no oceanic heat transfer within hurricanes much above ambient
level. Palm& and Newton (1969, p. 478) assert with no elaboration that the
measurement implies an increase of the equivalent potential temperature
from 360" to 385"K, and thus is a confirmation of the internal heat source
postulate; the author does not understand how this result was achieved.
To avoid difficulties in boundary layer measurements, Riehl (1963a,
p. 277) studied equivalent potential temperatures deduced from aircraft
data taken a t the 245-250 mb level a t the inner edge of the eyewall. Little
complication from moisture is expected a t such heights, and the variation of
equivalent potential temperature with height is not expected t o be large.
Riehl reports equivalent potential temperature of 370°K for a hurricane of
central pressure of 960 mb, but provides no further details on the instrument-
ation or data reduction whatever. If the result is correct, very significant
evidence exists for an oceanic heat source. However, Dr. R. H. Simpson of the
National Hurricane Center in Miami, Florida has informed the author that a
vortex thermometer was used (private correspondence). I f one recalls that
250 mb corresponds very roughly t o approximately 35,000 ft, then the remarks
of Gentry (1964, p. 12) are illuminating: " The wind tunnel tests of the vortex
probe under icing conditions (Ruskin and Schecter . . .) indicated that the
recorded temperature might be too high due to the change of state on impact
a t probe entry. . . . No evidence is readily available as t o whether much ice
accumulated on the vortex probe of the aircraft flying a t about 35,000 f t or
higher. Ordinarily, little liquid water would be encountered a t the higher

elevations and the ice accumulation should be minor. Some data collected by
theresearch aircraft in 1961 and 1962 indicated, however, that a t times the ice
accumulation might be significant. There is the possibility, therefore, that the
temperatures recorded for the upper troposphere are too high by an unknown
amount. There will probably always remain a question as to the accuracy of
the temperature measurements until some absolute standard of comparison
is developed."
3.9.3. On the Magnitude of Possible Increases in the Wet-Bulb Temperature

at Sea Level. Some insight into possible increases in the total stagnation
+ + +
temperature (HJc,) = T (gzlc,) (y2/2c,) (LY/c,) at sea level within a
hurricane will now be sought. Very near z = 0, q I 0. Also, the air tempera-
ture assumes the temperature of the sea; since the sea temperature is held
spatially constant in all hurricane analyses known to the author [with one
experimental exception (Anthes, 1972, pp. 473474)], and since the sea tem-
perature for an intense slowly translating hurricane decreases toward the
center of the storm if i t changes a t all, the conservative approximation for
current purposes is to hold T everywhere constant a t the ambient sea level
value for the tropical autumn. Then the only changes in (H/cp)are due to
changes in (LYlc,,) or more specifically, to changes in Y since L and cp are
taken constant for current purposes. Further, with only one percent error.
y = PvlP
(43) +
= dRH)P(T)/Cp, a(RH)P(T)I
G o(RH)P(T ) / p
so the vapor mass fraction Y increases as the relative humidity R H increases
or as the total pressure p decreases, for temperature T fixed. I n the autumnal
tropical ambient, if L represents latent heat of condensation plus glaciation,
at sea level (LYIc,) G 50°K. The ambient sea level relative humidity is 0.84
and the pressure is 1014 mb. [The relative humidity is unity a t the air-sea
interface but sea level measurement generally means readings a t shipboard
height, or more phenomenologically, it means measurement in the postulated,
roughly thirty foot thick constant-stress layer near the air-sea interface
(Hidy, 1972, p. 1084).] For a moderate hurricane with central pressure of
966 mb, the increase in (Hlc,) is 2.5"K, and for an intense hurricane with
central pressure of 910 mb, the increase in ( H / c p )is 5.7"K-provided R H is
held a t 0.84. If R H increases from 0.84 t o unity under the eyewall, then ( H / c p )
increases by 12.5"K for the moderate hurricane and 16.3"K for an intense
hurricane-although the justification for adopting such an increase in R H
appears unestablished, and Anthes and Johnson (1968, p. 297) cite evidence
that the relative humidity of the boundary layer tends t o remain between
80 and 90 yo.Malkus and Riehl (1960, p. 16) claim the equivalent potential
temperature, which is clearly closely related to (Hiep),increases by 12'5°K
from ambient to eyewall in the moderate hurricane [although Riehl (1963a,
p. 277) seems to augment this t o 16"K], and by 35°K in the intense
hurricane. By the reasoning adopted here the physical basis of the 12.5"K
increase becomes evident, but the 35°K increase for an intense hurricane
represents a physically impossible relative humidity of over 100 yo. Suc-
cinctly, while Carrier and his co-workers discuss increases in the first normal
derivative of H at sea level in hurricanes over ambient on the order of 10 yo,
Riehl and Malkus discuss occasional increases in H itself on the order of 10 %.
Carrier and his co-workers neglected the full effect of decreased air density as
the eyewall is approached, and found H t o be radially constant; the conject-
ure here is that the error in H so incurred is about one percent or so.
3.10. Numerical Simulation of Hurricanes on Digital Computers
I n recent years there has been a proliferation of computer models of the

tropical cyclones (Yamasaki, 1968; Ooyama, 1969;Rosenthal, 1970; Sundqvist,
1970; Kurihara, 1971; Anthes, 1972).These models warrant a review in them-
selves, and only a limited discussion is undertaken here.
First, it must be understood that these models are by no means exact
solutions of the full conservation equations, subject to appropriate boundary
and initial conditions. I n fact, to paraphrase Lorenz (1967), one might remark
that detailed reproduction of the hurricane lifespan by numerical methods
might not appreciably increase physical insight, because the total behavior
is so complex that the relative importance of various features might be no
more evident from examination of the computer solutions than from direct
observations of the real atmosphere. In any case, computer solutions are today
far from exact solutions; in fact, today no computer solution even attempts
t o treat observed initial data (Ooyama, 1969, pp. 35-37; Rosenthal, 1971b,
p. 35))but rather describes how a general hurricane grows and dies-in fact,
works in computer simulation emphasize results for the later stages of intensi-
fication and the peak intensity portion of the lifetime. The adopted initial condi-
tions almost invariably have pronounced axisymmetric circulations, andonly a
major alteration of parameterizations prevents inevitable hurricane formation.
Parameterization also entails compromise with exact solution. Because
many phenomena on the cumulus scale and smaller are crucial to the
hurricane scale flow, computer solution today is feasible only if the roles of
turbulent diffusion, radiative transfer, and (especially) cumulus convection
are simulated by parameterization. No one knows with certainty how to
describe the role of these often shorter-time-scale phenomena in terms of
variables occurring in the hurricane scale solution, or even if adequate descrip-

tion of this type is possible. Reviewing the parameterizations of cumulus
convection currently used in numerical simulations of hurricanes, Ooyama
(1971, p. 744) acknowledges that “. . . it is generally agreed that these
methods are essentially stopgap measures which are used for want of better
alternatives.” Gray (1972b, p. 57, 60) writes: “. . . the authors (Gray and
Lopez) feel that it is very difficult (or impossible) at the present time for any
numerical model of tropical motion to come t o grips with the real problems
of incorporating the cumulus convection in terms of the broader scale flow. . . .
I n that the cumulus cloud is such a distinctive physical unit, it would appear
that it should be independently treated. . . . The physics of the cumulus-
broadscale interaction may not necessarily be overcome by applying the
primitive equations to even smaller grids and time steps without additional
insight into the character of the individual convective elements.” While there
are some physical concepts that guide currently employed parameterizations
(such as weak integral constraints on energy and water substance), the param-
eterizations remain rather arbitrary and experimental (for a summary of
them, see Haltiner, 1971; Bates, 1972).Often the parameterizations are freely
adjusted until the cyclone scale results are consistent with measurements
and/or intuitwi. The computer results are very sensitive to the assumptions
made concerning the formulation of cumulus convection and eddy transfer
(Ramage, 1971, p. 40; Haltiner, 1971, p. 262; Garstang, 1972, p. 619). Without
any disparagement. such adjustment of parameterizations may be termed
curve-fitting. What is learned from current computer experiments is moot. Of
course, no model can today entirely escape such curve-fitting. The question to
be pondered is how refined a solution is currently warranted in view of the
limited physical understanding of certain critical phenomena? The validity
of a parameterization of cumulus convection might be established by fixing
all constants and function-! forms, and by showing that such a parameteri-
zation successfully describes a wide variety of atmospheric phenomena; such
success w ould furnish reason to scrutinize the incorporated physics carefully.
A further complication in evaluating numerical models is that sometimes
the finite differencing, intentionally or inadvertently, adds t o the stated
formulation in the sense that effects not in the differential equations are
present in the difference equations. The most accurate differencing of the
differential equations is not necessarily preferred, but rather numerical
techniques are rated “. . . on intuitive meteorological inspection of results ”
(Anthes et al., 1971, p. 747). For instance, errors introduced by less accurate
upstream-difference schemes were retained to simulate lateral mixing when
more accurate central-difference schemes gave less realistic results (Rosenthal,
1970, pp. 657-658). Later the differential formulation was modified so that
an accurate differencing of the modified expression for lateral diffusion gave
68 FRANCIS E. BENDELL
the numerical contribution desired. However, the modified expression ". . . is

not very satisfying from a physical point of view" (Anthes et al., 1971,
p. 747). What can be learned from such procedures again seems moot.
Even if the parameterizations for cumulus convection, turbulent mixing,
and radiative transfer were adequately known, and the finite-differencing
were adequately accurate, the question of feasibility in seeking a uniformly
valid solution for the entire cyclone would remain. Large gradients occur over
relatively small scales in important subregions of the storm (e.g., the frictional
bouiidary layer and the eyewall), while small gradients occur over relatively
large scales in the bulk of the storm (the rapidly swirling regions and the out-
flow layer aloft). On current computers usually a more or less fixed grid of 10
km.or 20 km radial resolution and a t most thirteen layers of vertical resolu-
tion is adopted by practical considerations; the overwhelming bulk of the grid
points then lie outside the eye, the eyewall, and the frictional boundary
layer-where important processes are occurring . The domain sizes are quite
limited (typically 440 km) so relatively weak boundary conditions (requiring
only that purely advective influx occur a t the side boundaries) are employed
and much of the storm lies outside the domain of computation. Garstang
(1972, p. 619) states numerical results are often very sensitive to the treat-
ment of the outer limits. An alternative to direct numerical solution of the
uniformly valid equations is t o subdivide the storm into natural portions
where different gradients and phenomena are operative. This is the approach,
with provision for interfacial compatibility, used by Carrier and his co-
workers.
Some evidence for uncertainty concerning current numerical models for the
hurricane is provided by results for the frictional inflow layer. First, with his
model, Rosenthal (1971b) performs numerical experiments in which the
latent heat transfer from the sea is greatly augmented over ambient but the
sensible heat transfer is set to zero. The goal is to show that the latent heat
transfer is the important portion of the total enthalpy transfer. Of coume,
Rosenthal is well aware that the sensible heat transfer is required to permit
the latent heat transfer (Riehl, 1954, p. 336). The computation is cited as
evidence of how incompletely the full conservation equations are represented in
current models. Thus the current computer programs cannot prove the
physical validity of the Riehl-Malkus oceanic heat source postulate; the
solutions which the current computer programs generate merely reflect the
incorporation of the Riehl-Malkus postulate in their formulation.1° Next,
1°Black and Mallinger (1972, p. 64) characterize most models as giving a 60%
reduction in maximum wind speed for a 2°C drop in sea surface temperature. Doubt
concerning such sensitive dependence of hurricane intensity on sea surface temperature
was discussed in Section 3.8. Rocent,lythe existence of a closed oceanic convection cell
induced by hurricane winds, already discussed,has been further confirmed by additional
the study by Anthes (1971) of the incompressible axisymmetric flow in a

frictional layer under an intense Rankine-like vortex furnishes an opportunity
to test how well the finite-difference techniques used in computer simulations
of hurricanes succeed in a problem of reasonable complication t o which the
answer is apparently known. Interestingly, Anthes introduces an initial value
problem as a convenient device for achieving the desired steady solution,
without claiming validity for the transient phase necessarily computed. There
are t h e e points about the solution that will be noted. (1) The multilayer
solution reveals that the one-layer treatments of Ooyama, Rosenthal, and
Yamasaki underestimate the tangential and (particularly) the radial velocity
components of the frictional layer, and give a premature eruption from the
Ekman layer at incorrectly large radial distances from the center. (2) While
both diffusion coefficients are held constant, Anthes takes the radial diffusion
coefficientfour orders of magnitude larger than the axial. Either by this means,
or by large truncation errors through forward-differencing of advective terms,
for plausible solution radial diffusion plays a role in regions of the flow
where boundary layer theory suggests radial diffusion should be negligible.
(3) The Ekman layer divergence changes from a weak downdraft to a large
updraft about 125 km from the axis of symmetry. The spuriousness of this
result, and the actual two-part structure of the frictional layer a t distances
from the axis a t which nonlinear terms are significant [as revealed by analytic
and more sophisticated numerical techniques (Carrier et al., 1971; Carrier,
1971a; Burggraf et al., 1971; Dergarabedian and Fendell, 1972b)], remain
obscured (Section 3.5).
Nevertheless, more elaborate computer simulations are continually devel-
oped. Anthes (1972)has attempted to account for departures from azimuthal
symmetry. The axis of the simulated hurricane vortex undergoes a curious
anticyclonic rotation a t approximately 8 mph about the center of the storm.
measurements on Typhoons Trix (which was situated over the central South China Sea
on October 23-24, 1952) and Wilma (which was similarly situated on October 28-29,
1952), and on other storms (Ramage 1972). Ramage (1972, p. 491) notes: “When they
struck the Philippines Trix and Wilma were ... equally intense typhoons. Trix
reintensified over the South China Sea and Wilma probably did. Paired aircraft recon-
naissances on the 23rd and 28th and the 24th and 29th . . . show little intensity difference,
in contrast to Ooyama’s numerical hurricane model . . . in which a smaller surface
temperature difference is associated with a significantly large intensity difference.”
Furthermore, the path for Wilma was much less influenced by oceanic cooling owing to
the prior passage of Trix than the work of Brand (1971) (previously cited in Section 1.5)
would suggest, even though Brand considered typhoons generally more separated in
time. This is one more item of evidence suggesting that ocean temperatures are not
solely determining influences on hurricane behavior. Ramage (1972, p. 493) concludes:
“The history of typhoon Wilma suggests that surface temperature may not always
control storm intensity nor significantly affect storm movement, for Wilma moved
along the cold water wake of Triz and remained intense until reaching Indochina.”
70 FRdNCIS E. FENDELL
Anthes suggests the path of a cyclone is not influenced by horizontal sea

surface thermal gradients, although Brand (1971) claims observational
evidence t o the contrary. Newer computer models are evolving that (unlike
most of the models previously discussed) do not isolate the storm from its
surroundings, which may largely determine the hurricane path. However,
these newer models often must employ a grosser grid in order to describe a
larger region.
3.11. Implications of Hurricane Nodels on Seeding

The current mode of attempting artificial modification of tropical cyclones
is to introduce silver iodide crystals in the supercooled water believed t o
exist in appreciable, naturally unfrozen quantities high in the eyewall, and
perhaps also in rainbands and cumulus clouds both close t o and far from the
eyewall. I n a typical field experiment, over two hundred rocket cannisters,
each containing about 150 gm of silver iodide, are released a t about 33,000 f t
over a 30 km distance (entirely outside the eye), along a ray 90" clockwise
t o the direction of storm translation; such an experiment takes two to three
minutes to execute.ll I n laboratory studies, a gram of silver iodide produces
1Ol2 to 1014crystals, which serve as ice nuclei and which grow to 3 p diameter.
The goal is that there will be lo3 nuclei/liter of space over a depth of several
kilometers a few minutes later; this concentration is reduced to negligible
levels in a couple of hours (Gentry, 1971a; Penner, 1972).The silver iodide will
hopefully cause the supercooled water droplets to freeze, and thus to release
the heat of fusion (Battan, 1969).
The proposed mechanism by which this heat release (and supposed attend-
ant temperature rise of a couple of degrees Celsius, and density decrease) is
eficacious has been altered on several occasions (Simpson and Malkus, 1964;
Rosenthal, 1971a; Gentry and Hawkins, 1971), and a t times several alter-
native possible mechanisms have been set forth (Gentry, 1971a,b). A concise
history of the various suggestions has been given by Riehl (1972a, pp. 267-
*l Typically five runs of the kind described, at two-hour intervals, starting from the
radius of maximum winds and proceeding radially away from the center, characterize the
so-called eyemod experiment. Most of the discussion is oriented toward this type of
experiment. However, it may be noted that the seeding of Hurricane Ginger (1971) dis-
cussed below was a so-called rainsector experiment: all the water-containing clouds in a
45" sector from 50 to 100 n miles from the center are seeded a t about 22,000 f t to divert
low-level inflow to premature eruption and high-level outflow outside the eyewall, with
resultant dispersal of latent heat energy over a wider region of the storm (Hawkins et al.,
1972). I n concept, a rainsector experiment consists of four fifty-minute seedings, at fifty-
minute intervals. The reason a reinsector, rather than eyemod, experiment was attempted
on Ginger was that it was a large diffuse storm (moderate gale winds out to 260 n miles,
circulation out to 400 n miles) with only a n ill-defined eyewall (maximum winds of about
60-70 knots at 40 t o 60 n miles from the center).
270). One accepted point is that seeding in the nascent eye of a developing
tropical storm should be avoided since this procedure would probably abet
intensification (Rosenthal, 1971a). The most recent proposal has been t o seed
cumuli outside the eyewall and above the 0°C isotherm (roughly, 600 mb or
15,000 ft) with silver iodide. (Of course, it is more difficult to find seedable
clouds, and there is a greater expanse to cover, outside the eyewall.) The
aim is not to capitalize directly upon the heat of fusion released by supercooled
water droplet freezing, but rather to cause these cumuli outside the eyewall
to grow several thousand feet in height from midtropospheric to upper tropo-
spheric levels. This would hopefully initiate a chain of events from which
an order of magnitude more heat than that associated with freezing will
indirectly be realized. This added heat would come from some low-level in-
flowing warm moist air erupting not up the eyewall, but prematurely up the
growing cumulus, with associated condensation and additional glaciation. If
entirely successful, proponents have suggested that perhaps a substitute
eyewall further from the center, and hence (from conservation of angular
momentum) with reduced swirling speeds, could be realized from this pre-
maXure eruption of boundary layer air (Gentry and Hawkins, 1971; Rosenthal,
1971~).
Fukuta (1972) suggests that the field experiments conducted to date have
employed overseeding of supercooled water droplets to form small ice parti-
cles; the energy release would be small and the effect over in less than half an
hour. Instead, Fukuta suggests reducing the seeding material to 1 yo or even
0.1 yo of that used in past field experiments, t o form large precipitable ice
particles. The latent heat of fusion these falling particles could extract
from the inflow-updraft stream would supposedly raise the central pressure
for hours, reduce swirling, and induce spreading out of the storm. A detailed
analysis fully substantiating either the concepts of Gentry and Hawkins or
the concepts of Fukuta has not appeared.
Clearly, like numerical simulation of hurricanes, the seeding of hurricanes
warrants an entire review unto itself. I n fact, with the proposed extension of
the U S . seeding effort [Project Stormfury, a joint effort of the Department
of Commerce (NOAA) and the Department of Defense (Navy)], from the
western North Atlantic to the western North Pacific Oceans (Mallinger, 1971),
this might be an apt time for such a review.
Attention will be confined here to a few observations, especially to what the
hurricane models suggest about the seeding procedure.
Even proponents of seeding anticipate only a 10 t o 15% decrease in
maximum winds. I n the one case in which larger decreases were noted, the
anomaly is now attributed to synoptic peculiarities relating t o upper level
outflow (Hawkins, 1971). This is not to disparage such decreases; statistical
treatment of a model suggests multimillion dollar annual savings in damage
from hurricane seedings that would reduce peak winds by Gfteen percent
(Boyd et al., 1971). The problem is that the storm is naturally oscillating in
intensity by the same order of magnitude, so it is very difficult to distinguish
natural and artificially induced changes. I n fact, the gradual increase of the
maximum winds of Hurricane Ginger by 15 yo during the day after initial
seeding on September 26, 1971 was ascribed to natural forces by Project
Stormfury personnel, as was the 11% decrease after the second seeding on
September 28 (Lieb, 1972). Fujita (1972) could find no evidence that seeding
altered Ginger; since the central pressure of this weak hurricane was only
980 mb at %he times of seeding, he suggests that perhaps the cloud-top
heights were too low for the silver iodide to be effective.The threat of litigation
has constrained the number of seeding experiments severely.
The National Hurricane Research Laboratory (Gentry, 1969a,b, 1970)
has discussed a six t o twelve-hour cycle of amelioration after seeding; a
physical basis for this time scale has yet t o come forth (see below). Further, if
the central pressure deficit is reduced as reported, eyewall seeding must alter
the eye in a still unidentified manner, although Black et al. (1972) suggest
that one hour after seeding outside the eye, the eye may expand more than
is consistent with natural variability. More complete post-seeding probing
of the tropical cyclone would be helpful in evaluating these claims.
Rather similar computer models have produced different guidance with
regard to current seeding practice. The reasons for the discrepancy are not
fully available because some details remain unpublished. Rosenthal (1971~)
notes that both the magnitude and the duration of the heating taken t o
simulate seeding in his computer experiments seem excessively large. Never-
theless, in those numerical experiments believed t o most closely simulate the
field experiments performed on Hurricane Debby, August 18 and 20, 1969,
Rosenthal (19714 predicts an increase after seeding in the maximum wind,
and over wide radial extents an augmentation in wind level, a t 700 mb; the
comparable measurements a t 12,000 f t indicated a reduction in winds (Haw-
kins, 1971). Rosenthal predicts a decrease in the maximum wind a t sea level
after seeding (though in substantial portions of the hurricane, surface winds
are predicted t o increase); no post-seeding measurements were made a t sea
level. Quantitative detail is omitted here because Rosenthal (1971a, p. 415)
notes: ". . .at best, the results should be considered qualitative guidance material."
Sundqvist (1972) finds his numerical model predicts an increase in maximum
winds from silver iodide seeding, which serves to release more effectively the
latent heat in clouds below 0°C; the increase is about 10 % if seeding is carried
out in the region of intense convection. Within 20 hr of cessation of seeding,
the storm is predicted t o return to its preseeding state; seeding does not
initiate a cascade of effects that permanently alters the storm. Estoque (1971,
p. 4), noting the tentative nature of his conclusions, remarks:
We have [studied] the properties of a hurricane model which is suitable for

simulating the effects of artificial seeding. The main improvement of this model
over previous ones is the explicit inclusion of cloud micro-physical processes, in-
cluding the prediction of the liquid water distribution. Thus, the model is able to
simulate more realistically the release of the latent heat of fusion due to artificial
seeding. The model has been used to simulate artificial seeding over three different
radial locations in the vicinity of the maximum in the surface wind. The expected
reduction and outward displacement of the maximum radial surface pressure
gradient did not materialize. Instead, the pressure gradient and also the wind
increased in intensity. However, the seeded storm intensity decreased after the
seeding stopped. The greatest effect occurs when seeding is done over a region just
inside the surface wind maximum location.
A n interesting feature of the unseeded hurricane is the occurrence of periodicity
in time of the hurricane intensity. The magnitude and the period of this periodicity
is about 10 m sec-' and 8 hours, respectively. This magnitude is about the same
order as that of changes induced artificially by seeding. If this periodicity is real,
it should be taken into account in the interpretation of seeding experiments in the
actual atmosphere.
The effectiveness of current seeding practices remains unresolved. If the

Carrier model is valid, silver iodide seeding can only transiently upset the
stable hurricane configuration. Under this model, warm-fog dispersal methods
would have t o be applied to the entire " throughput supply " layer of warm
moist air t o achieve the significant goal of premature rainout in outer spiral
bands. The layer of warm moist air is so spatially extensive that such at-
tempts seem somewhat impractical. Dergarabedian and Fendell (1971) have
discussed the use of warm-fog dispersal methods for lower levels in a hurricane
for premature rainout of the throughput flux; almost simultaneously, so did
Matthews (1971, p. H-8), albeit more optimistically:
The feasibility of warm cloud modification prior to cold cloud modification

should be examined because warm cloud modification may permit growth of small
warm clouds t o temperatures at which cold cloud modification will be effective.
The combined use of warm cloud and cold cloud modification techniques would
permit selective seeding in all regions without cloud top temperature restrictions.
Seeding to divert a path seemingly holds little better promise than seeding
to alleviate intensity, since there appears to be no way to discern what path
alterations were due t o human intervention under current understanding.12
l 2 Dr. R. Cecil Gentry, director of the National Hurricane Research Laboratory and
of Project Stormfury, stated recently that budgetary reductions had grounded the four
aircraft used in cloud-seeding experiments, and probably no full-scale attempts a t
hurricane modification would be undertaken in the next three years (Anonymous, 1973).
If and when seeding of hurricanes is resumed around 1976 perhaps activities will be ex-
tended to the North Pacific where more storms suitable for seeding occur annually.
Gray (1973)is also pessimistic about the effectivenessof silver iodide seeding
to modify hurricanes artificially, especially since such seeding can be directly
effective only as long as the lifetime of the cumulus cloud. Others have pro-
posed the alternative of spreading evaporation retardants on tropical oceans
to weaken incipient tropical cyclones (Gentry, 1969a), or even to preclude
hurricanes altogether; casting aside questions about the survivability of such
thin layers in the ocean and about pollution, one must wonder whether or not
fewer, but more violent, hurricanes would result, since the storms may well
play an essential (often, humanly beneficial) role in global energy and water
substance transport. Since the damage from a hurricane in the United States
tends to increase faster than the fourth power of the maximum sustained
surface wind speed (Howard et al., 1972), and since (by extrapolation from
data on typhoons inthenorthwesternPacific) twenty percent of hurricaneshave
wind speedsin excess of 120mph, a thirty percent reduction of maximum hurri-
canewindspeed remainsa very desirable goal. Gray ( 1973)believes a more safely
executed and more effective (if more expensive) means of attaining such a goal
than by silver iodide seeding is by dispersal of two to three million pounds of
0.1 p sized carbon particles in the surface fiictional layer at radial distances
outside the outflow level cirrus shield (which extends out generally about 200
miles from the center for moderately severe storms) but inside the surround-
ing anticyclone circulations (which typically border the outer edge of the
hurricane on the poleward side while in the trades). The carbon black dust
(Downie, 1960; Frank, 1973)is envisioned to cause O.5"-l0C heating per hour
for ten hours over an area of 40,000-80,000 kmawhen dispersed to a 10 yo
areal coverage, because it absorbs 15 yo of the solar radiation daily incident
on the tropical surface and then rapidly transmits the heat by conduction
to adjacent air flowing at low altitudes into the hurricane. Ideally, the result-
ing 5"-1OoC rise in air temperature increases buoyancy in the frictional bound-
ary layer, and the consequent convection leads to a dry downdraft which
inducesan increase in moisture flux from the ocean to the surface inflow layer.
The result is hopefully that about ten percent of the mass inflow (whichwould
reach the base of the eyewall from the edge of the cirrus shield in a day and a
half or so) prematurely erupts, and the maximum wind speeds in the eyewall
are significantlyreduced on the time scale of about one day. Thus, the ultimate
effect sought is that envisioned by Gentry and Hawkins (1971),but the means
employed differs. Rosenthal's numerical model (Rosenthal, 1970, 1971a,b,c)
predicts such a carbon dust seeding strategy will succeed in lowering the
maximum wind speed, although rainfall and wind levels in the outer portion
of the storm will increase. Three possible difficulties, besides expense of the
operation, are dispersal problems owing to clumping of the carbon particles
(whichmight be difficult to counter cheaply and without performance degmd-
ation); anticipating the storm path (so the particles remain outside the cirrus
shield for ten hours but then flow into the storm, especially into the particularly
severe right-hand semicircle with respect to a n observor looking along the
path) ; and being certain midtropospheric moisture levels a t surface pressures
of 995-1005 mb (where the cirrus shield often ends) can sustain tall aumulo-
nimbi towers (otherwise, according to the Carrier model, the energy-enriched
erupting air eventually returns to the surface layer and proceeds inward to
the eyewall, perhaps t o intensify the storm).
CYCLONEINTENSIFICATION
4. THEORYOF TROPICAL
There exists no satisfactory theory explaining how, annually, a small

number of tropical disturbances intensify to become hurricanes, and why in
contrast most disturbances do not. Charney (1971) notes that, unlike in the
midlatitudes, tropical temperature gradients are weak, and tropical cyclones
most often form over the tropical oceans where the surface temperatures are
particularly uniform. No theory of how pronounced baroclinicity develops
in the barotropic tropics is likely to emerge until the detailed local balances
and large-scale circulations of the tropical ambient are better understood
quantitatively, because early stages of cyclogenesis will probably involve
perturbation about that ambient.
Carrier (1971b) has presented an intensification process by which his quasi-
steady mature model evolves in time from a tropical depression. Tracking
back t o even earlier evolution seems premature a t the current state of under-
standing. Although Carrier’s theory has been reproduced by Penner (1972)
without objection, Carrier’s theory of intensification is not currently entirely
satisfactory. Thus, it is used here only as an interesting vehicle t o raise some
significant points about the briefer, later stage of intensification, as opposed
t o the longer, earlier stage. Ooyama (1969, pp. 35-37), among many others,
has written about an initial very modest rate of pressure fall in intensifying
tropical cyclones, followed by an appreciably accelerated rate of fall-it is
this latter stage which is principally addressed here. [A particularly dramatic
example of such rapid central sea level pressure fall after a slower decline in an
earlier stage is furnished by Typhoon Irma, which moved northwestward in
the central Philippine Sea in early November 1971 (Holliday, 1973). On
November 8 Irma passed from a tropical depression t o a storm; during the
next two and one half days the central pressure fell about 20 mb. However, in
the succeeding 24.5 hr the pressure fell from 981 to 884 mb, a minimum pres-
sure exceeded by only two storms on record. During its peak intensity stage,
the typhoon exhibited the occasionally observed double-eyewall structure,
one ring of dense tall cloudiness occurring a t a radius of only 3.5 n miles and
the second concentric ring a t a radius of 20 n miles.]
4.1. Carrier’s Outline of Iderm&atwn
Carrier begins with a schematic view of an axisymmetric tropical depression

in which there is a weak Rankine-vortex-like swirling, the maximum azi-
muthal speed lying relatively far from the axis of rotation. There is radial
inflow in region I during the transient phaae (Fig. 19). The swirling would
quickly establish a shear layer beneath it; there is weak upflux out of (down-
fluxinto) the surface layer where the relative circulation increases (decreases)
with radial distance. Siice linear theory correctly predicts the down&x in the
mature stage with rapid swirl, linear theory probably s d c e s to predict the
boundary layer divergence during intensification. However, since equilib-
ration of the bounday layer requires times of O(Q-l), or roughly sixteen hours
in the tropics, and since this is possibly an appreciable fraction of the later-
stage, rapid-intensiiication time scale, a transient linear theory will be re-
quired. It must yield a radial influx through the boundary layer I1 in excess
of the radial inflow speed in I. (Incidentally,the times characterizingsignficant
change in the thermodynamicstate of the lower tropical ambient are probably
on the order of 100 hr, a time span partly related to the typical magnitude of
eddy viscosity appropriate for the tropical atmosphere. Hence, unless the
intensification time from depression to cyclone greatly exceeds 100 hr-
and this seems dubious-the thermodynamicstate of the ambient may possibly
be justisably taken as fixed throughout the later stages of intensification.
This statement by no means implies that the ambient can be taken as fixed
throughout the entire intensificationprocess. The very fact that descent speeds
in the outer portions of even a mature hurricane are roughly comparable to
ambient tropical descent speeds suggests that the tropical ambient probably
cannot be held fixed throughout intensification.)
The air erupting from the boundary layer begins to displace the air initially
in the central core of the developing storm. The air initially present in the core
is of slightly lower pressure than the ambient air at the edge of the storm,
but not vastly different in vertical stratification. The air erupting from the
boundary layer under the Rankine-vortex-likeswirl displaces the air initially
present in the core verticalIy upward; since there is a “lid” on top of the
storm, the vertically displaced initial air is, near the top of the core, squeezed
radially outward.
The following competitbn develops. The new air rising out of the boundary
layer is drawn entirely from relatively warm moist air near the bottom of the
atmosphere. Thus, displacing the air initially present in the core is relatively
light air. On the other hand, the convective motion of new air is small, espec-
ially at early times, and the ambient processes (turbulent diflwion, radiational
cooling, cumulus convection) try to maintain the original, near-ambient strati-
fication in the core. If the convective displacement wins out, then the core
becomes lighter and lighter, relative to a column of air a t the outer edge of
the storm. This paragraph contains those points of Carrier's intensification
model that seem most unsettling, and will be subject t o close scrutiny in the
critique below.
However, Carrier's basic proposition is that the swirling in T has led to a
downflux into 11, a spiraling inward in the boundary layer and an upflux into
the core, and a lightenng of the core by hydrostatic considerations. For
dynamic consistency, the centrifugal force (anticipated to be the dominant
inertial effect) must increase t o balance the augmented radial pressure grad-
ient. Since angular momentum is conserved in I, where friction is negligible,
the fluid particles must necessarily move in closer t o the axis of symmetry
(axis of rotation). The result is that in time, in the Rankine-vortex-like swirl
distribution, the maximum azimuthal speed increases in magnitude and the
position of the maximum lies closer to the axis (Figs. 19 and 20). Hence, the
more the pressure falls in the core, the more fluid sinks into the boundary
layer t o spiral inward, erupt upward, and cause further pressure reduction
in the core. If the crucial early competition is resolved in favor of the organ-
ized convection, ultimately the particles erupting out of the boundary layer
rise so quickly that they lie on a moist adiabat, and the greatly lightened core
is entirely flushed of its original fluid.
While most descriptions (Palmbn and Newton, 1969) tend to picture the eye
as formed gradually as the pressure deficit develops, in the Carrier model the
central core may well be competely flushed of ambient-like air, so that the air
in the core lies on a moist adiabat, before much trace of an eye is to be found.
At Grst, a Rankine-vortex-like swirl holds everywhere. From this fully devel-
oped one-cell structure with a t most only the rudiments of an eye, a well-
defined two-cell structure with a calm center region emerges rapidly, probably
in much less than an hour, owing to inertial oscillation, in the following way.
As the pressure falls in the core relative to the ambient, the particles in I
necessarily move in closer to the axis t o permit a compensating centrifugal
force t o develop. Once the core is flushed and moist adiabatic ascent character-
izes the full height of the core, no further pressure deficit can be generated. By
inertia, the spinning particles continue t o move in, a dynamic imbalance is
created, and a radial acceleration develops t o force the particles away
from the axis of symmetry. This reverse motion creates a rarefaction a t the
center, and relatively dry warm motionless air sinks down the axial column
to form an eye. This air may be air from above the storm or rained-out, slowly
swirling air entrained out of the top of the moist adiabatic column (Fig. 21).
Because there is no appreciable swirl (hence no associated pressure gradient)
in the eye, there is no frictional boundary layer under the eye. The moist
adiabatic column becomes an annulus displaced from the axis,i.e., the eyewall;
the inertial oscillations of the eyewall eventually damp in time.
"TfrI
ill"
I
Cl--------- c
0 R r r
*
0 R
FIG. 19. In this schematic view by Carrier of the flow configuration and circum-
ferential velocity distribution in an intensifying tropical depression at some early
time t = tl (say), the interface C-C between the new and initial air in the core is idealized
as horizontal for convenience. However, it is easy t o show that for all but very modest
circulations (i.e. except for angular velocities less than three times the ambient), a rigidly
rotating core (which implies uniform updraft vdocity a t the top of the nonlinear frictional
layer) could not accept all the fluid pumped through the Ekman boundary layer. Thus a
uniform ascent in the core is dynamically impossible, and in fact ascent in the eyewall is
concentrated in cumulonimbi. For the gross balances being discussed, many points can
be made without accounting for such refinements in eyewall structure, though ultimately
details of the structure are crucial. (From Carrier, 1971b, p. 146.)
4.2. Critique of the Carrier Model of Intensijhztion

Carrier has offered more than the conventional linearized stability analysis
t o determine what scales of motion grow in what time interval (Yanai,
1964; Ogura, 1964; Yamrtsaki, 1969). [Bates (1970) did study the growth t o
finite amplitude in the ITCZ of nonzonally symmetric equatorial wave
disturbances of 200 km longitudinal scale, which are observed to move
westward a t 13 knots. The time scale for early growth owing t o barotropic
c
0 R r r
Fro. 20. Intensification from tropical depression to hurricane has progressed to a more
advanced stage in this schematic diagram, holding at t = t2 > t l , according to Carrier's
model. The magnitude of the maximum swirl is increased and its position lies closer to the
axis of symmetry. (From Carrier, 1971b, p. 148.)
instability was about two days; frictional dissipation eventually curtails

further growth effected by conversion of condensational heat t o eddy avail-
able potential energy. Because Charney's concept that condensational heating
in the tropics is dependent on convergence of moisture in the planetary
boundary layer was incorporated, the disturbance had a warm core with
maximum intensity in the lower atmosphere. Holton (1972) furnishes further
lucid insight into equatorial wave disturbances, and in fact concisely discus-
ses many other topics arising in tropical meteorology.] Carrier has also
emphasized the role of lightening of the core rather than (say) large-scale
horizontal turbulent diffusion. However, there are problems concerning the
nature of the convective lifting in the core, as now discussed.
80 FRANCIS 1.BENDELL
I I
0 R r a
FIG.21. Schematic picture by Carrier of the flow configurationwhich prevails when the
radius of maximum swirl R is increasing and an incompletely formed eye is being filled
with relatively dry and motionless air, which sinks down from the top of the storm under
dry adiabatic compression. With the formation of an eye lighter in weight than the
eyewall, the terminal stages of intensification and the beginnings of quasi-steady
mature stage structure are realized. (From Carrier, 1971b, p. 150.)
The ambient distribution of total stagnation enthalpy is such that when a

column stratified like the ambient is lifted as a whole adiabatically, warm air
is squeezed radially outward a t the top, even if warm air is added a t the
bottom (Fig. 7). The net effect for small convective motion, according to a
method of characteristics numerical integration of the resulting hyperbolic
boundary value problem, is not a lightening of the developing core (Dergar-
abedian and Fendell, 1972a, pp. 62-67).13
A closely related observation is to note that Carrier (1971b), in his discus-
sion of intensification, seemed more concerned with how a hurricane is formed
than whether it is formed. Hence, he adopted effectively the following pro-
portionality:
pressure fall from ambient new air into core
-
total pressure fall for a moist adiabatic core total air required to flush core
This statement is true a t the outset and, if a hurricane forms, a t the end, of
flushing the core. I n fact, however, such a simplification lets a hurricane
always be formed. I n truth, the pressure fall for small times is less than that
predicted by the fractional-volume proportionality. As the quasi-steady
l3 A caloulation illustrating that undiluted lifting of the autumnal Caribbean ambient

produces reduced temperatures at all tropospheric levels was also carried out by Riehl
(1963b).
TROPICAL. CYCLONES 81
mature state approaches, the initially slow pressure fall probably accelerates
t o a rapid decrease, and the thermodynamic stratification of the emerging
eyewall asymptotically approaches a near moist adiabat state.
The problem cannot be satisfactorily buried in parameterizations of
turbulent transfer, radiative cooling, and cumulus convection in the core.
What appears needed is explicit recognition that a broad uniform ascent in
the core does not account for the actual mechanism of convection, which
would seem t o entail ascent in a concentrated number of intensified cumulo-
nimbi. The somewhat tentative nature of this remark stems from statements
like that by Palmkn and Newton (1969, p. 490): “It seems clear from . . .
observations and from descriptions of turbulent encounters that, in some
cyclones, ascent in the inner region is dominated by cumulonimbus, . . . while
in others a’broad-scale and more uniform ascent takes place.”
If cumulonimbi do play an essential role in the core ascent during intensi-
fication, some semblance of an eye would probably form a t an earlier stage
of hurricane development than Carrier suggests.l4 The eye would probably
not be entirely created on a time scale of 20 min a t the end of intensification,
as Carrier suggests, although inertial oscillations a t the end of intensification
might rapidly delineate a previously ill-defined eye.
The tritium-tracing analyses by Ostlund (1968) of Hurricane Hilda (1964)
and Hurricane Betsy (1965) suggest that while the eye may have partially
consisted of air entrained from the eyewall, there had been substantial sub-
sidence of stratospheric air down into the eye. Thus, Carrier seems correct
in citing two possible sources for the air in the eye.
4.3. The Distribution of Curnulonimbi during Intensification

Riehl(1972a, p. 249) notes: “ Very fine weather, with hardly a cloud in the
sky, often precedes the arrival of a hurricane by 1 day, when the storm is
moving westward in the tropics. I n the past this deceptively beautiful weather
has been the cause of many disasters with heavy loss of human life in the
Caribbean and elsewhere.”15I n a possibly related observation, Lopez (197210)
calls attention to the work of Oliver and Anderson (1969), in which satellite
photographs are used t o indicate that the majority of hurricanes form around
a circulation center in the clear area ahead of (rather than directly under) a
l4 Consistent with his conception that i t is the compressional heating of the compen-
satory downdraft that warms the air, because heat directly released by cumulus con-
vection is almost entirely expended in raising the buoyant column, Gray (Shea, 1972,
pp. 118-122) suggests that turbulent diffusion and mixing convey heat from the eye to
warm the eyewall.
l6 From “Introduction to the Atmosphere” by Herbert Riehl. Copyright 1972 by
McGraw-Hill Book Company. Used with permission.
82 FBANCIS E. FENDELL
convective region. The concentration of cloudy convective regions amid clear

calm regions in tropical cyclogenesis may be related to characteristics of
subsidence that, by continuity, must accompany convection.
However, perhaps also there is a concentrating of preexisting dispersed
cumulonimbi that is a concomitant of large-scale tropical cyclogenesis. This
admittedly speculative subject is included because it may be worth pursuing.
Furthermore, Malkus et d.(1961) observed that for the inner rain area of
radius 200 n miles in Hurricane Daisy (1958), in the prehurricane stage there
were 60 cumulonimbi that covered one percent of the area; during intensi-
fication two and one-half percent of this area was covered by cumulonimbi.
I n the mature stage there was a persistence of cumulonimbus patterns rel-
ative to the center, and there were 200 cumulonimbi that covered four percent
of the area. By studying measurements in Hurricanes Carrie (1957) and Cleo
(1957),as well as Daisy, Gentry (1964) estimated five percent coverage of the
inner eighty nautical mile radius area by cumulonimbi (total cloud coverage
in this area was under 25%). Gentry adds that both aircraft penetration and
radar returns suggest that the area of strong convective activity is larger in
more intense hurricanes and is located closer to the center.
It might also be speculated [in contrast to conjectures by Gray (1972b,
p. 64) on conditions for tropical cyclogenesis]that for a synoptic scale pressure
difference to develop not all the air ascending in the cumulonimbi should
locally subside, but some of the descent should be deferred to a larger hori-
zontal scale.
4.4. The Time-Dependent Fbwjeld during Intensification

Although the energetics of the core may need revision, Carrier's modeling
of the dynamics in the outer portions during the transient phase should
remain substantially unchanged. Thus, this portion is developed here in some
detail.
For axisymmetric intensification, a set of uniformly valid equations for the
dynamics is now given in cylindrical polar coordinates. If subscripts r, 2, and t
denote partial differentiation, for a noninertial coordinate system rotating
a t the speed of the normal component of the rotation of the earth (i.e. normal
to a plane locally tangent to the earth at about 15" latitude so Q G 0.06 rad
hr - l) :
(44) ut -+ uu, -+ wu,- 2Qv - (w2/r)-k p - Ipr = vu,
(45) +
(W* u(1"u), + Mrv)z + 2Qru = v(r%
(46)
and
(47)
These equations hold in 0 < r 5 T o , 0 < z < z1 where, again, z1 is the top of
the storm (a slippery lid) and ro is the outer edge. The boundary-initial
conditions of relevance here are
(48) r = r,: u = y = 0; z = O : u =v = w = O
(49) t = t o : u, v small
There is angular momentum in the initial flowfield in an inertial frame of

reference, and this is conserved with the anticipated radial inflow toward the
axis of symmetry; the picture of the flow is that given in Fig. 19.
For convenience the following notation is adopted. The radial, azimuthal,
and axial velocity components will be denoted by capitals ( U , V , W ) in I
and by lower case letters (u,v, w ) in the frictional boundary layer 11.
From results given earlier for the mature stage boundary layer dynamics,
and from other considerations developed below for the transient boundary
layer, it is anticipated that Y = rV is a linear function of r2, and independent
of z; then there is a downflux into the frictional boundary layer I1 from the
supply region I which is independent of r to lowest order. It follows from (45)
and (47),together with (48)and (49):
(50) '4' = r V ( r ,t ) = R[RO2- R2(t)](roZ

- r2)/(roz- R2)
(51) @ = rU(r,t ) = RR(rO2- ra)/(ro2 - R2)

and
(52) W ( r ,z, t ) = (zRRz/[r,a- R2(t)]}

where R(t,) = R , .The associatedpressurefieldfrom (44) and (46)is discussed
later. This solution holds in I, i.e. in R < r 2 r o . At t = t o , in I, i.e. in R, 5
r 5 ro , Y = 0. The particles at r = R , at t = t o , at a later time t > t o lie
at r = R < R , . Typically, ro =O(lOOO miles), R , =0(200 miles), and (R,/12)
I R(t)IRo.
The next step is to develop the boundary layer in I1 that lies under the
swirling flow in I given by (50)-(52). The equations for the viscous layer are
taken to be the linear transient parabolic equations appropriate for low
+
Rossby number flow. While the inequality (02 Ya) Q2R4may not form-
ally hold everywhere in the flowfield, it is widely valid and, in addition, from
experience with the mature quasi-steady storm, the key output, the boundary
layer divergence, is adequately given by linearized theory. Hence,
(53) 4 3 - 4t = -2Q*
(54) v*ez - *t =2 q
where
(56)
(57)
subject to
(58) ~ ( rz ,= 0, t ) + -(@ + iY), X(T, z -+ 00, t ) + 0 , ~ ( rZ,,0)= 0
If (roa- Ra)+roa, an excellent approximation,then it is convenient to define
(59) P(r, t ) E Pl(t)[l- (P/ro")]
= [Boa - RZ(t)][l - (r2/r0Z)].
Carrier (1971a) studied the special form P l ( t )= Roaexp(.t). For the more
general form (€is), by Laplace transformation (Dergarabedian and Fendell,
1972a, pp. 56-57), the relevant boundary layer divergence is
(60) w(r, z + co,t )
+2!22i)1/a][exp(st)]ds
C+im
= ( ~ ~ / ~ / r ~ ~ ) R e ( 1 / 2 xP1(s)[(s
i) - 2sZi)/(s
6 /Q](ROa
G ( v / ~ ) 1 ~ 2 ( 1 / r 0 2 ) [ (- ~ Ra)
7 ) ~ / ~-
for the R(t) of practical interest. The divergence, a function of time only, is
largest in magnitude in the steady state; if R +( R0/12)typically,
w(r, z-+ co) = -(Ro/ro)a(vC!)1~2
= 0(10-3 mph)
The net volumetric flux down into 11,and hence (in an incompressible model)
up into the core, Q(t), is given by
(611 Q = Q0[(6/7Q)a/at - 11P - (x/R0)21
where the quasi-steady discharge into the base of the core I11 is given by
(62) &o = .rr(r,Z - R2)(Ro/ro)"vC!)"~

The previously discussed requirement for compatibility between (1)the
nonlinear spin-up of the inviscid flow in I under conservation of angular
momentum and (2) the pressure deficit generated between a column in the
core and a column in the ambient owing to partial flushing of the core with
boundary layer air driven radially inward by the outer inviscid flow, is now
made quantitative. It is necessary to model the radial and azimuthal velocity
components in the core; kinematic and continuity considerations suggest that,
a t least in the lower regions of 111, for this purpose it is adequate to take
(63) ru = qi = r2R/R
(64) rv = y5 = Q(RO2- R2)r2/R2
Accordingly (44) is integrated from r = 0 to r = ro a t some fixed z above the
boundary layer; the profiles for u and v are given by (63) and (64)for 0 <r < R
and by (50) and (51) for R < r 5 r , . The term uu, in (44) is a perfect differen-
tial and yields precisely zero; wu, yields zero because u is independent of z
and w is small anyway. It follows that the time-dependent, Coriolis, centri-
fugal, and pressure gradient terms yield, for p = const. and upon neglecting
terms of O(R2/rO2) relative to those of order unity,
(65) p ( r o ,z = 0 , t ) -p(r = 0, z = 0, t )
= [pQ2ROZ(Ro2 - R2)/R2]{1+[2 ln(ro/R)- 1]R2/Ro2}
+ P(d/dt)ERAln(Rb.o)I
The last term on the right-hand side is characterized by the inverse square
of the time for intensification; this time is probably much larger than the time
for equilibration of the boundary layer. Thus normally the time-derivative
term can be neglected.
However, when inertial oscillations associated with the continued radial
influx and rebound of air in I occur, a shorter time scale enters and the time
derivative term must be retained. Carrier (1971b, p. 157) suggests that the
time scale of the oscillation is 0(4R,2/l2RO2)where Rf is the final equilibrium
position of the eyewall R(t). For typical values, with R, G 250 miles, this
time scale is 20 min. I n fact, holding the convection-core pressure deficit
from the ambient fixed at about 35 mb, holding p a t its ambient sea level value,
letting R , G 250 miles and r o G 1000 miles, one finds from (65)-by dropping
the temporal derivative-that R = Rf = O(35 miles). Restoring the temporal
term, adopting as initial conditions R = O(35miles) and R s -a( RO2- R2)12R
(Carrier, 1971b, pp. 152 and 155), and holding the core pressure deficit
relative t o the ambient fixed at 35 mb (near quasi-steady conditions), one
finds from (65) that R decreases to about 16 miles in about a half-hour before
R = 0. Thus, numerical values confirm the inertial oscillation period suggested
by Carrier. Such oscillations probably reinforce the eye, rather than create it,
as noted earlier (Section 4.2). The oscillation of the eyewall is dwelled upon
because it seems to be the first proposal of a mechanism internal to the
hurricane that might explain a crudely periodic variability in intensity of 10
to 20% in peak winds, on the time scale of a few hours; such variability
appears to be observed sometimes, and attribution to changes in the ambient
86 FRANCIS 1 . FENDELL
atmospheric and underlying oceanographic states may not be always satis-

factory.
The increased low-level radial inflow toward the center of the developing
cyclone seems consistent with the previously discussed reorganization of the
statistically steady cumulonimbus activity which helps maintain the ambient
enthalpy transport process. Such radial inflow implies a stronger circulation
(or vorticity), and this implies a decreased mixing of updraft air with sur-
rounding air; the result would be a decreased density in the updraft region
and hence a decreased pressure. As cumulonimbus activity increases in the
central region of the incipient cyclone, there is probably diminution of such
activity in the surrounding region. However, the downdraft in the surrounding
region should differ little from the ambient since the total low-level inflow
into the whole region will not have changed appreciably. Thus, the antici-
pation is that the flowfield presented in this section probably remains sub-
stantially correct, within the axisymmetric approximation.
6. CONCLUDING REMARKS
Several viewpoints not universally accepted have been adopted in this
review. The existence of, and necessity for, a greatly augmented oceanic heat
source in the inner regions of a hurricane as the basic sustaining mechanism
has been questioned. The validity of the hot-tower thesis of heating from the
tropical cumulonimbus has been scrutinized. The incomplete understanding
of the processes and time scales of tropical cyclogenesis and intensification
has been emphasized. The conviction that improved quantitative under-
standing of the tropical ambient is prerequisite t o further progress on the
deepening of tropical disturbances has been stated. Reasons for reservations
concerning the probable success of current hurricane seeding techniques have
been offered.
This unconventionality hopefully conveys two major points to the reader.
First, any serious worker in hurricanes would be very remiss not t o consult
the referenced works directly and decide for himself whether or not the
viewpoints adopted here are in fact valid. Second, this review on tropical
cyclones describes a vital subject in rapid flux to which many basic physical
and mathematical contributions are still to be made. This is not a field in
which only exquisite refinements t o elaborate computer programs remain. I n
fact, many valuable theoretical contributions in recent years have been
achieved by Carrier and Gray via an approximate solution of simplified
models with only modest computing, as opposed t o attempting direct numer-
ical simulation of the full conservation laws. The meteorologist with insight
will hopefully agree that the tropical cyclone is neither a closed subject, nor
a subject to which only those with access to very sophisticated computing
facilities can aspire to contribute.
THE KINETIC
APPENDIXA. ESTIMATING ENERGY
AND WATER
CONTENTOF
HURRICANES
To estimate very quickly and roughly the total kinetic energy content of a
mature hurricane of moderate intensity, one must evaluate I where
I =-
1
2
1" pw2 d V
= 7T 1z=.?1/2
t=O
t=ro
P[~ir)J2rdr
The speed is virtually all swirl, and about half the nine mile high, axisym-
metric storm is rotating like (say) a potential vortex. The swirl is (suppose)
100 mph a t the edge of the (nonrotating) eye re = 30 miles, and the storm
extends out to ro = 500 miles. The density may be set t o a constant average
value of 5 x gm ~ m - Then
~ .
(-4.2) I A +pz1(w,,,)2re2 1.dr/r

70
= 1024 ergs
Battan (1961, p. 21) characterizes the kinetic energy of a hurricane as 1 O 1 O

kW-hr, or 3.6 x loz3erg.
An upper bound on the rainfall per day for a hurricane hovering over a
spot is now calculated according to the Carrier model. The vertical velocity
down into the boundary layer in the mature stage is
(A.3) w(r, z-+ co,t -+a)

G -(Ro2/ro2)(vQ)1'2
The volumetric flux down into the boundary layer is, for ro29 Ra,
(A.4) Q0= m-02w(r,z -+ co,t -+00) A -TR~~(YQ)~/~
this is also the flux up into the eyewall. The mass of air per unit time is given
by pQo where p is fairly constant a t a fixed height. The water content of this
mass of air, which falls out as precipitation, is given in mass per time by
(A.5) J = 7rpRoZ(vQ)"2Y(Z = z*)

where z t is a typical ambient-profile height for characterizing the mass fraction
of water vapor for the flux into the eyewall. For p A 1.1 x gm cm-3,
v = 2 x lo6 cm2 sec-l, Q s 6.25 x hr-l, and [from the data of Jordan
(1957)l Y ( z ,= 1000 ft) G 1.65 x lopz, J A 1.8 x 10'l gm 6ec-I. If this
falls entirely in an annulus from 20 t o 30 miles, one gets 4.3 x gm cmA2
sec-l, or since 1 gm of liquid water occupies about 1cm3, about 370 cm day-'.
The record rainfall over a spot in one day is associated with a hurricane:
117 cm (Baguio, Philippines in July 1911) (Paulhaus, 1965). Of course, for a
88 FRANUIS E. FENDELL
hurricane translating at 20 mph, no point is likely to lie under the eyewall

for more than an hour, and (370/24) & 15 cm hr-l. (Typically, a hurricane
is observed to drop a total of 15-30 cm of rain on coastal areas which pass
under the central portions of the storm.)
Because of the flood-spawningtorrential rains over Virginia associated with
Hurricane Camille, it is interesting to consider whether the water vapor
content of this hurricane at landfall is consistent with the large total rainfall
over the Southeast. From isohyetal maps published by Schwarz (1970) it
may be crudely estimated that in eastern Kentucky, Virginia, and West
Virginia an area of 150 x 180 miles received an average of five inches of rain
and an area of 100 x 360 miles received three inches. In addition 5 x lo4
miles2of Mississippi, Alabama, Louisiana, and Tennessee received an average
of three inches. This totals 2.58 x l O I 5 gm,which is probably a conservative
estimate. From Jordan’s data one can estimate that the Caribbean hurricane
contained
(A4 Q = nr02 1; PY dz = nro21,
where p Y = p, = density of water vapor. For an AugustSeptember ambient,

I , = 4.71 gm Equating Q to the total rainfall, one finds ro 260 miles.
This is an interesting value in that Camille was a smdl intense hurricane
(De Angelis, 1969) with hurricane winds extending outward from the center
about 60 miles and gale winds, 110 miles (Meyer, 1970, p.5). However, the
results here indicate merely that ro = O(260 miles) satisfies a conservative
estimate of the total precipitation requirement; it is not intended to imply r0
could necessarily be taken so small, though if one fits Carrier’s proposed
swirl profile [given below (35)] to the data, ro = O(225 miles).
APPENDIXB. THEMOISTADIABAT
Because of the significant role it plays in understanding tropical cyclones,
a derivation of the relation describing the moist adiabatic process seems
worth including. The following is based on unpublished notes of G.F. Carrier
of Harvard University.
When moist air expands adiabatically,the temperature eventually decreases
to a value at which the water vapor density reaches its saturation level. For
all further decreases in temperature, condensation occurs. Here it is assumed
that the condensedwater has zero volume (a good approximation),and that the
liquid precipitates out as it is formed.
An air-vapor mixture is at temperature 117, the mass of air in volume V
being m a , and the saturated vapor mass in V being m, . The corresponding
densities are pa = (m,/V) and p, = (mv/V ) ; the pressure p = p a +pV. A
relationship between T and p (say) as T changes is obtained by studying
the moist adiabatic process in several steps (Fig. 22). The mixtureis separated
into two fluids (dry air and vapor), to be later recombined; this step is taken
to require no expenditure of work.
Initially, the mass m, of vapor occupies the volume V as in the top box of
Pig. 22. That volume is decreased by a process in which dm, of the vapor is
condensed while it and the remaining vapor are a t temperature T and pres-
sure pv. Thus, the new volume (t,hatof the second box) is V[l - (dm,/m,)].
The work done by external forces during this process is ( p , V dm,/m,), and is
indicated by the arrow to the left of the boxes. The heat gained Ldm, is
shown at the right. The vapor now expands adiabatically and isentropically
+
to volume ( V d V ) as shown in box 3; the new temperature T is given by
P.1) T ' / T = { V [ l -(dm,/m,)]/(V +dV)}yv-l
= 1 - (yv - l)(dmv/mv +
+ d V / V ) ..-
In the third step, the gas in box 3 is heated to temperature ( T +dT), as
indicated in box 4. The heat that must be added during this process is
P.2) +
(mv- dmv)(c,),[dT ( r v - l)(dm,/mv dV1 W I +
= m,[(c,), dT +
R,T dVIV] RvTdm, + + * * *
PVV dmJmv - - Ldmv
-
pvV (dmJmV + dV/V)-
1 ( 1 4 ~ 1 (dmJm,
) + dVM)
m, (cy,dT*RvTdV/V)+R,T dmv
m -dm
T + dT
P, dV - B-
T (1 - (V-l)dV/V)
ma (cv dT + ROT d V M
FIG.22. Schematic diagrams of tho sequence of thermodynamic steps by which the

equation describing the moist adiabatic process is derived. The upper diagram sketches
operations on the water vapor component, and the lower diagram sketches operations on
the dry air component; the diagrams read from top to bottom. Arrows on the left of thc
diagrams denote work done on tho system (arrow in) and work done by the system
(arrow out). Arrows on the right of the diagrams indicate heat removed from the system
(arrow out) and heat added to the system (arrow in).
where (c,,), is the heat capacity a t constant volume of the vapor, (c,,), the
heat capacity a t constant pressure of the vapor, R, = (c,,), - (cV),, and
y v = (c,)v/(cu)v '
The dry air, on the other hand, starts off in the bottom box of volume V
with mass ma a t temperature T . It expands isentropically to volume ( V d V ) +
so that its temperature becomes T [ l - (ya- l ) ( d V / V ) ] Henceforth
. yv =
y a = y . The dry air is then heated a t constant volume to temperature
+
(T d T ) by the addition of heat in the amount rna[(c,,),dT RaT(dVl V ) ] .+
Thus a t the end of the process, a vapor mass (mv - dm,) and a dry air mass
ma occupy a volume ( V + +
d V ) a t temperature (T dT). By definition, the
process undergone will have been the moist adiabatic one if the net heat added
+
is zero, and if the final (p, dp,) is the saturation value for the final temper-
ature (IT +dT):
(B.3) ma[(cu)adT + Ra T dV/VI+
~ , [ ( G , ) ~ ~ T + R , T ~+VR ,/ TVd]m , = -Lam,
(B.4) Pv = mv/V = Pv(T)
where p,( T ) is the function that describes saturation density.
By the equation of state for dry air,
(B.5) d V /V = -dpa/pa = d TIT -dpa/pa
But (rlp,/pv) is comprised of two contributions: that due to change of volume
and that due t o the loss of mass via condensation. Thus
pV
--
+
dp, - mv - dm, V
Pv m, V+dV
=1 - (dm,/m,) - (d V /V ) +- -
*
Hence, after division by V and using the equations of state,

(B.7) {pa[(cu)a + Ral dT -dPa) + {pv[(cu)v + RvldT - d P ) = -(L/ V )d(pv V )
or
where P(T ) is the saturation vapor pressure.

Alternatively, the moist adiabat may be derived as a special case of the
equation for conservation of energy, written as the sum of internal plus kinetic
energy for a multicomponent mixture. Here Cartesian tensor notation will be
used instead of vector notation, and a comma denotes partial differentiation.
Furthermore, the symbolism previously adopted in the main text for describ-
ing the dynamics of an effectively one-component gas will not be used here.
The air velocity is denoted u,, with utu,=qa and ut S,, = w, where the
gravitational acceleration g, = - g Si, . The water vapor velocity is (u, vt). +
The vapor quantities pv, (c,,)~,(c,),., Rv and the air quantities pa, (c~),,
( c ~ ) , ,Ra have the same designations as above. The specific internal energy
of the air is e, ,and of the vapor ev . The specific heat of the liquid water sub-
stance is c l . The heat lost by radiation and all other processes is &, with units
(massllength-time3).The conservation of energy for current purposes may be
written (upon neglect of the kinetic energy contribution relative to internal
energy contributions)
(B.9) baea + ~ v e v l t, +[Paee,ui + p v e v ( ~ i+vi)l.<
= -(pa + + +
p v ) P ( u i j ~ t ) , j (kT,t),t-ctTw -Q
where ui, is the total stress tensor, k is the thermal conductivity, and
(B.lO) Pv.t + +
pv@* vt).t= --w
In general, w (the mass of water vapor condensed per unit time) must be
specified explicitly or through a prescribed mechanism; when the fluid is
saturated and (dT/dt) <0, w is known.
For steady flow with densities, pressures, and temperature dependent on the
vertical coordinate z only, with negligible diffusion so v, = 0 and utf= -par,,
in the hydrostatic approximation p s +
pg = 0, it follows from continuity that
(B.ll) (PaW),z=O, (pvw),z=--w
hence,
(B.12) + + +
pa Wea, 2 pvwev, 2 - wev PW. z ct T o = 0
With the aid of the continuity equations and the equations of state,
(B.13) +
~ a ~ [ ( ~ p ) a2T l .~vw[(cp)vTl,z - WP, z = w(ev + Rv T - CIT )
Since
(B.14) ev = cl T + L(T) - Rv T
and
(B.15) = -wpa(pv/pa). 2 = -(1/V d(pv v)
upon substitution and cancellation of the common factor w, one recovers
(B.7).
ACKNOWLEDQMEJYTS
The author is very deeply indebted to Prof. George F. Carrier of Harvard University
for generous and indispensable guidance and assistance on every aspect of the work
described here. He also wishes to thank Dr. Paul Dergarabedian of The Aerospace Corpor-
ation for many stimulating discussions. However, full responsibility for all errors lies en-
tirely with the author. All computationswere programmed by Mr. Phillip s. Feldman, of
TRW Systems. This study was initiated owing to the interest of Capt. Hugh Albers of
the Interdepartmental Committee on Atmospheric Sciences (Department of Commerce);

the encouragement and helpful suggestions of Drs. John Perry and Lawrence G.
Roberts of ARPA Information Processing Techniques (Department of Defense), Drs.
Robert Rapp and W. L. Gates of Rand Corporation, and Mr. James Murray of the Army
Research Office (Durham) are gratefully acknowledged. Finally, the author wishes to
thank Dr. H. E. Landsberg for the opportunity to present this paper. This work was
carried out under contrmts DAHC04-71-C-0025and DAHCO4-67-C-0016 with the U.S.
Army Research Office (Durham, North Carolina). Although this work was sponsored
in part by the Advanced Research Projects Agency of the Department of Defense under
ARPA Order 1786, the views and conclusions contained in this paper are those of the
author and should not be interpreted as necessarily representing the official policies,
either expressed or implied, of ARPA or the U.S. Government.
LISTOF SYMBOLS
PARTIAL
c1 Heat capacity of liquid watec W Azimuthal velocity component of q1
cp Heat capacity at constant pressure Wi Velocity vector for water vapor
cV Heat capacity at constant volume relative to dry air; velocity vector
e Internal energy for a fluid in a noninertial system
E Ekman number, v / Y o V Azimuthal velocity component of q1
g Magnitude of gravitational accelera- in the inviscid flow exterior to the
tion, IS11 frictional boundary layer; volume
g1 Acceleration of gravity W Vertical velocity component of pI
+
H Total static enthalpy, c,T gz+ LY W Vertical velocity component of qr in
+
H , Total stagnation enthalpy, H q2/2 the inviscid flow exterior to the
L Specific latent heat of phase transition frictional boundary layer
mr Molecular weight of species i X fir2/y$,;a factor in the moist adiabat
n Power of algebraic decay of swirl with based on d H = 0
radial distance Y Water vapor mass fraction, pV/p
p Pressure z Altitude above sea level
P Saturation vapor pressure
q Wind speed, lqtl
pi Velocity vector for a fluid in non- Greek Symbob
inertial frame rotating with earth y C&"
Q Net volumetric flux per time; heat p Density
loss per voume-time 4 ru [sometimes, r ( u - U);sometimes,
Q o Steady-state net volumetric flux per nondimensional]
time @ Tu
r Radial coordinate in cylindrical polar h
t,
rw [sometimes, T(V -'v);sometimeq
coordinates nondimensional]
R Radial position of maximum azimuth- Y rV
al velocity w Ratio of molecular weight of water
R6 Gas constant for species i , R/m, vapor to dry air
R Universal gas constant ul, Total stress tensor
RH Relative humidity y Kinematic viscosity (usually given
t Time turbulent transfer values)
’2 Temperature w Mass of water vapor condensed per
u Radial velocity component of 9, volume-time
ut Velocity vector for dry air Coriolis parameter (component of the
U Radial velocity component of pt in the angular velocity of the earth perpen-
inviscid flow exterior to the dicular to a plane locally tangential
frictional boundary layer to the sea surface in the tropics)
8 azimuthal coordinate in cylindrical Subscripts

polar coordinates; potential-like a Dry air; ambient
+
enthalpy, cpT gz c At the axis of symmetry at sea levcl
0, Stagnation potential-like enthalpy, 0 Eyewall a t sea level
8 + 9”P i Vector (Cartesian tensor notation)
[ z/E1Ia o Outer edge of hurricane; initial value;
typical value
s Ambient at sea level
Superscript t Total
Total derivative with respect t o time v Vapor
REFERENCES*
Agee, E. M. (1972). Note on ITCZ wave disturbances and formation of Tropical Storm
Anna. Mon. Weather Rev. 100, 733-737.
Alaka, M. A. (1968). “Climatology of Atlantic Tropical Storms and Hurricanes,” ESSA
Tech. Rep. WB-6. US Govt. Printing OEce, Washington, D.C.
Anderson, L. G., and Burnham, J. M. (1973). Application of economic analyses t o hur-
ricane warnings t o residential and retail activities in the U.S. Gulf of Mexico coastal
region. Mon. Weather Rev. 101, 126-131.
Anonymous (1972). Hurricane Agnes: The most costly storm. Wealherwise 25, 174-184.
Anonymous. (1973). U.S. hurricane control test program cut. 5 0 s AngeZes Times 92, 4
February, Part 8, p. 6.
Anthes, R. A. (1971). Iterative solutions t o the steady-state axisymmetric boundary-
layer equations under an intense pressure gradient. Mon. Weather Rev. 100, 261-268.
Anthes, R. A. (1972). Development of asymmetries in a three-dimensional numerical
model of a typical cyclone. Mon. Weather Rev. 100, 461-476.
Anthes, R. A., and Johnson, D. R. (1968). Generation of available potential energy in
Hurricane Hilda (1964). Mon. Weather Rev. 96, 291-302.
Anthes, R. A., Rosenthal, S. L., and Trout, J. W. (1971). Preliminary results from a n
asymmetric model of the tropical cyclone. Mon. Weather Rev. 99, 744-758.
Arakawa, H. (1954). On the pyramidal, mountainous, and confused sea in the right or
dangerous semi-circle of typhoons. Pap. MeteoroZ. Qeophys. 5, 114-123.
Aspliden, C. I. (1971). “ On Energy Distribution in the Tropical Troposphere,” Rep.,
Dept. Meteorol., Florida State University, Tallahassce.
Atkinson, G. D. (1971). “Forecasters Guide to Tropical Meteorology,” Tech. Rep. No.
240. USAF Air Weather Serv., Scott AFB, Illinois.
Bates, J. R. (1970). Dynamics of disturbances on the Intertropical Convergence Zone.
Quart. J . Roy. Metecrrol. Soc. 96, 677-701.
Bates, J. R. (1972). Tropical disturbances and the general circulation. Quart. J . Roy.
Meteorol. SOC.98, 1-16.
Battan, L. J. (1961). T h e Nature of Violent Storms.” Doubleday, Garden City, New
York.
Battan, L. J. (1969). “Harvesting the Clouds-Advances in Weather Modification.”
Doubleday, Garden City, New York.
Black, P. G., and Anthes, R. A. (1971). On the asymmetric structure of the tropical
cyclone outflow layer. J . Atmos. Sci. 28, 1348-1366.
Black, P. G., and Mallingcr, W. D. (1972). “The Mutual Interaction of Hurricane Ginger
* This article was completed in February 1973.

94 KRANCIS E. FENDELL
and the Upper-Mixed Layer of the Ocean,” Appendix D, Project Stormfury Annual
Report 1971. Nat. Hurricane Res. Lab., Coral Gables, Florida.
Black, P. G., Senn, H. V., and Courtright, C. L. (1972). Airbourne radar observations of
eye configuration changes, bright band distribution, and precipitation tilt during
the 1969 multiple seeding experiments in Hurricane Debbie. Mon. Weather Rev.
100,208-217.
Boyd, D. W., Howard, R. A., Matheson, J. E., and North, D. W. (1971). “Decision
Analysis of Hurricane Modification,” Proj. 8503 Final Rept. Stanford Res. Inst.,
Menlo Park, California.
Bradbury, D. L. (1971). “The Filling Over Land of Hurricane Camille, August 17-18,
1969,” Satellite and Mesometeorol. Res. Proj. Res. Pap. 96. Dept. Geophys. Sci.,
University of Chicago, Chicago, Illinois.
Brand, S. (1970a). “ Geographic and Monthly Variation of Very Large and Very Small
Typhoons of the Western North Pacific Ocean,” Tech. Pap. No. 13-70. Nav. Weather
Res. Facility, Norfolk, Virginia.
Brand, S. (1970b). Interaction of binary tropical cyclones of the western North Pacific.
J. Appl. Meteorol. 9, 433-441.
Brand, S. (1971). The effects on a tropical cyclone of cooler surface waters due t o upwelling
and mixing produced by a prior tropical cyclone, J. Appl. Metwrol. 10, 866-874.
Brand, S., and Bellock, J. W. (1972). “Changes in the Characteristics of Typhoons
Crossing the Philippines ” Teoh. Pap. No. 6-72. Environ. Prediction Res. Facility,
Nav. Postgrad. School, Monterey, California.
Burggraf, 0. R., Stewartson, K., and Belcher, R. (1971). Boundary layer induced by a
potential vortex. Phys. Fluid8 14, 1821-1833.
Burroughs, L. D., and Brand, S. (1973). Speed of tropical storms and typhoons after
recurvature in the western North Pacific Ocean. J. Appl. Meteorol. 12, 452-458.
Byers, H. R. (1944). “General Meteorology,” 2nd ed. McGraw-Hill, New York.
Carlson, T. N. (1969). Some remarks on African disturbances and their progress over the
tropical Atlantic. Mm. Weather Rev. 97, 716-726.
Carpenter, T. H., Holle, R. L., and Fernandez-Partaggas, J. J. (1972). Observed relation-
ships between lunar tidal cycles and formation of hurricanes and tropical storms.
Mon. Weather Rev. 100, 451-460.
Carrier, G. F. (1970). Singular perturbation theory and geophysics. S I A M (Soc. Ind.
App .Math.) Rev. 12, 175-193.
Carrier, G. F. (1971a). Swirling flow boundary layers. J. Fluid Mech. 49, 133-144.
Carrier, G. F. (1971b). The intensification of hurricanes. J. Fluid Mech. 49, 145-158.
Carrier, G. F., Hamrnond, A. L., and George, 0. D. (1971). A model of the mature hurri-
cane. J . Fluid Mech. 47, 145-170.
Charney, J. (1971). Tropical cyclogenesis and the formation of the Intertropical Con-
vergence Zone. I n ‘‘ Mathematical Problems in the Geophysical Sciences. I. Geo-
physical FIuid Dynamics (Lectures in Applied Mathematice, Vol. 13)” (W. H. Reid,
sd.), Vol. 1, pp. 336-368. Amer. Math. SOC.,Providence, Rhode Island
Charney, J. G., and Eliassen, A. (1964). On the growth of the hurricane depression.
J . AtWaO3. SCi. 21, 68-75.
Cox, J. L., and Jager, G. (1969). “A Satellite Analysis of Twin Tropical Cyclones in the
Western Pacific,” Tech. Memo. WBTMSOS 5. ESSA Space Operations Support Div.,
Silver Spring, Maryland.
Day, J. A. (1966). “The Science of Weather.” Addison-Wesley, Reading, Massachusetts.
DeAngelis, R. M. (1969). Enter Camille. Weatherwdse 22, 173-179.
Denny, W. J. (1972). Eastern Pacific hurricane season of 1971. Mon. Weather Rev. 100,
276-293.
Deppermann, C. E. (1937). “Wind and Rainfall in Selected Philippine Typhoons.”
TROPIUAL CYCLONES 95
Manila Weather Bureau Central Observatory Report, Bureau of Printing, Manila,

Philippines.
Dergarabedian, P., and Fendell, F. (1970). On estimation of maximum wind speeds in
tornadoes and hurricanes. J . Astronuut. Sci. 17, 218-236.
Dergarabedian, P., and Fendell, F. (1971). “Tornado and Hurricane Thermodydro-
dynamics,” Tech. Rep. TRW Systems, Redondo Beach, California.
Dergarabedian, P., and Fendell, F. (19724. “Tropical Cyclones,” Rep. 18524-6000-R0-01.
TRW Systems, Redondo Beach, California.
Dergarabedian, P., and Fendell, F. (1972b). The surface frictional layer under a hurricane
vortex. J . Astronaut. Sci. 20, 9-34.
Downie, C. S. (1960). Cloud modification with carbon black. In “Cumulu~Dynamics”
(C. E. Anderson, ed.), pp. 191-209. Pergamon, Oxford.
Easterbrook, C. C. (1969). “A Study of the Effects of Waves on Evaporation from Free
Water Surfaces,” Res. Rep. No. 18. Bureau of Reclamation, U.S. Department of the
Interior, US Govt. Printing Office, Washington, D.C.
Eliassen, A., and Kleinschmidt, E. (1957). Dynamic meteorology, In “Handbuch der
Physik” (S. Fliigge, ed.), Vol. 48, pp. 1-64. Springer-Verlag, Berlin and New York.
Emmons, H. W., and Ying, S. J. (1967). The f i e whirl. In “Eleventh Symposium
(International) on Combustion.” pp. 476-488. Combustion Institute, Pittsburgh,
Pennsylvania.
Erickson, C. O., and Winston, J. S. (1972). Tropical storm, midlatitude cloud-band
connections and the autumnal buildup of the planetary circulation. J . AppZ.
Metewol. 11, 23-36.
Estoque, M. A. (1971). “Hurricane Modification by Cloud Seeding,” Rep., Rosenstiel
School of Marine and Atmospheric Science, University of Miami, Coral Cables,
Florida.
Fletcher, R. D. (1956). Computation of maximum surface winds in hurricanes. Bull.
Amer. MeteoroZ. SOC.36, 247-260.
Frank, W. M. (1973). “Characteristics of Carbon Black Dust as a Large-scale Tropo-
spheric Heat Source,” Pap. No. 196. Dept. Atmos. Sci., Colorado State University,
Fort Collins.
Fujita, T. T. (1972). “Use of ATS Pictures in Hurricane Modification,” Satellite and
Mesometeorol. Res. Proj. Res. Pap. No. 106. Dept. Geophys. Sci., University of
Chicago, Chicago, Illinois.
Fukuta, N. (1972). “Modification of Hurricanes by Cloud Seeding,” Rep., Denver
Res. Inst., Denver, Colorado.
Garstang, M. (1967). Sensible and latent heat exchange in low latitude synoptic scale
systems. Tellw 19, 492-508.
Garstang, M. (1972). A review of hurricane and tropical meteorology. Bull. Amer.
Meteorol. Soc. 53, 612-630.
Garstang, M., La Sew, N. E., Warsh, K. L., Hadlock, R., and Peterson, J. R. (1970).
Atmospheric-oceanic observations in the tropics. A m p . Sci. 8, 482-495.
Gates, W. L., Batten, E. S., Kahle, A. B., and Nelson, A. B. (1971). “A Documentation
of the Mintz-Arakawa Two-level Atmospheric General Circulation Model,” Rep.
R-877-ARPA. Rand Corp., Santa Monica, California.
Geissler, J. E. (1972). On the verticaI distribution of latent heat release and the mechanics
of CISK. J . Atmos. Sci. 29, 240-243.
Gentry, R. C. (1964). “A Study of Hurricane Rainbands,” Nat. Hurricane Res. Proj.
Rep. No. 69. US Weather Bureau, Washington, D.C.
Gentry, R. C. (1969a). Project Stormfury. Bull. Amer. MeteoroZ. SOC.50, 404-409.
Gentry, R. C. (1969b). Project Stormfwy, 1969. Pap., Tech. Conf. Hurricanes. 6th, 1969.
Gentry, R. C. (1970). Hurricane Debbie modification experiments. Science 168, 473-475.
Gentry, R. C. (1971a). To tame a hurricane. Sci. J . 7,49-55.

Gentry, R. C. (1971b). Project Stormfury, 1971. Bull. Amer. Meteorol. Soc. 52, 775.
Gentry, R. C., and Hawkins, H. F. (1971). “A Hypothesis for Modification of Hurricanes,”
Appendix B, Project Stormfury Annual Report 1970. Nat. Hurricane Res. Lab.,
Coral Gables, Florida.
Godshall, F. A. (1968). Intertropical Convergence Zone and mean cloud amount in the
tropical Pacific Ocean. Mon. Weather Rev. 96, 172-175.
Gray, W. M. (1968). Global view of the origin of tropical disturbances and storms.
Mow. Weather Rev. 96, 669-700.
Gray, W. M. (l972a). “A Dia,gnostic Study of the Planetary Boundary Layer Over the
Oceans,” Pap. No. 179. Dept. Atmos. Sci., Colorado State University, Fort Collins.
Gray, W. M. (197213). “Cumulus Convection and Larger-scale Circulations. Part 3.
Broadscale and Mesoscale Circulations,” Pap. No. 190. Dept. Atmos. Sci., Colorado
State University, Fort Collins.
Gray, W. M. (1973). Feasibility of Beneficial Hurricane Modification by Carbon Dust
Seeding,” Pap. No. 196. Dept. Atmos. Sci., Colorado State University, Fort Collins.
Greenspan, H. P. (1968). “The Theory of Rotating Fluids.” Cambridge Univ. Press,
London and New York.
Grossman, G., and Rodenhuis, D. (1972). “The Effcct of Release of Latent Heat on the
Vorticity of a Tropical Storm Over Land,” Tech. Note BN-722. Inst. Fluid Dyn.
Appl. Math., University of Maryland, College Park.
Haltiner, G . J. (1971). “Numerical Weather Prediction.” Wiley, New York.
Haurwitz, B. (1935). The height of tropical cyclones and the “eye” of the storm. Mon.
Weather Rev. 63, 45-49.
Hawkins, H. F. (1971). Comparison of results of the Hurricane Debbie (1969) modifi-
cation experiments with those from Rosenthal’s numerical model simulation
experiments. Mon. Weather Rev. 99, 427-434.
Hawkins, H. F., and Rubsam, D. T. (1968). Hurricane Hilda, 1964. 11. Structure and
budgets of the hurricane on October 1, 1964. Mon. Weather Rev. 96, 617-636.
Hawkins, H. F., Bergman, K. H., and Gentry, R. C. (1972). “Report on Seeding of
Hurricane Ginger,” Appendix B, Project Stormfury Annual Report 1971. Nat.
Hurricane Res. Lab., Coral Gables, Florida.
Hess, S. L. (1959). “Introduction to Theoretical Meteorology.” Holt, New York.
Hidy, G. M. (1972). A view of recent air-sea interaction research. Bull. Amer.
Meteorol. SOC.53, 1083-1 102.
Holliday, C. R. (1973). Record 12- and 24-hOur deepening rates in a tropical cyclone.
Holton, J. R. (1972). “An Introduction to Dynamic Meteorology.” Academic Press,
New York.
Holton, J. R., Wallace, J. M., and Young, J. A. (1971). On boundary layer dynamics
and the ITCZ. J. Atrnos. Sci. 28, 275-280.
Howard, R. A., Matheson, J. E., andNorth, D. W. (1972). The decision to seed hurricanes.
Science 176, 1191-1202.
Hubert, L., and Timchalk, A. (1969). Estimating hurricane wind speeds from satellite
pictures. Mon. Weather Rev. 97, 382-383.
Hughes, L. A. (1952). On the low-level wind structure of tropical storms. J . Meteorol. 9,
422-428.
Jennings, G. (1970). ‘‘ The Killer Storms-Hurricanes, Typhoons, and Tornadoes.”
Lippincott, Philadelphia, Pennsylvania.
Johnson, D. H. (1969). The role of the tropics in the global circulation. I n “ The Global
Circulation of the Atmosphere ” (G. A. Corby, ed.), pp. 113-136. Roy. Meteorol.
SOC.,London.
Jordan, C. L. (1957). “A Mean Atmosl,here for the West Indies Area,” Nat. Hurricane
Res. Proj. Rep. No. 6. US. Dept. of Commerce, Washington, D.C.
Keenau, J. H., and Kcyes, F. G. (1936). “Thermodynamic Properties of Steam Including
Data for Liquid and Solid Phases.” \Viley, New York.
Kraus, E. B. (1972). “Atmosphere-Ocean Tnteraction.” Oxford Univ. Press (Clarendon),
London and New York.
Kuettner, J. P. (1971). Cloud bands in the earth’s atmosphere. Observations and theory.
Tellus 23, 404-426.
Kurihara, Y. (1971). An eleven-layer, asisymmetric, primitive equation model of a
tropical cyclone. Bull. Amer. Meteorol. Soc. 52, 769.
Leipper, D. F. (1967). Observed ocean conditions and Hurricane Hilda, 1964. J . Almos.
Sci. 24, 182-196.
Leipper, D. F., and Volgenau, D. (1972). Hurricane heat potcntial of the Gulf of Mexico.
J . Plbys. Oceanogr. 2, 2 18-224.
Lieb, H . S. (1972). Project Stormfury Director Reveals Results of Hurricane Ginger
Seeding Experiments,” News Release NOAA 72-4. U.S. Dept. of Commerce,
Washington, D.C.
Lopez, R. E. (1972a). Cuinulus Convection and Larger-scale Circulations. Part 1. A
Parametric Model of Cumulus Convection,” Pap, KO. 188. Dept. Atmos. Sci.,
Colorado State University, Fort Collins.
Lopez, R. E. (197213). “ Cumulus Convection and Larger-scale Circulations. Part 2.
Cumulus and Meso-scale Circulations,” Pap. 189. Colorado State Univ. Fort Collins.
Lorenz, E. N. (1966). The circulation of the atmosphere. Amer. Sci. 54, 402-420.
Lorenz, E. N. (1967). ‘i The Nature and Theory of the General Circulation of the Atmo-
sphere.” World Meteorol. Organ., Geneva.
Mc\Villiams, J. C. (1971). The boundary layer dynamics of symmetric vortices. Ph.D.
Thesis, Harvard University, Camhridge, Massachusetts.
Ma,hrt, L. J. (1972a). A numerical study of the influence of advective accelerations in an
icicalizetl, low-latitude, planetary boundary layer. J . Atmos. Sci. 29, 1477-1484.
Rlahrt, L. J . (197213). A numerical study of coupling between the boundary layer and
free atmosphere in an accelerated low-latitude flow. J . Ataaos. Sci. 29, 1485-1495.
Malkus, J. S. (1958). On the structure ant1 maintenance of the mature hurricane eye.
J . Meteorol. 15,337-349.
Malkus, J. S. (1960). Recent developrnonts in studies of penetrative convection and an
applicatiori to hurricane cumnlonimhun tuuvrs. I n ‘‘ Cumulus Dynamics ” (C. E.
Anderson, ed.), pp. 66-84. l’ergamori, Osforcl.
nlalkus, J. S. (1962). Largc-scale interactions. In “ T h e Sea” (M. N. Hill, ed.), Vol. 1,
pp. 88-294. Wiley (Intersciencc), New York.
Blalkus,J . S.,and Riehl, H. (1960). On the dynamics and energy transformation in steedy-
stat.e hurricanes. Tellus 12, 1-20.
Rlalkus, J. S., Ronne, C., and Chafee, 11. (1961). Cloud patterns in Hurricane Daisy, 1968.
Tellzts 13, 8-30.
Mallinger, W. D. (1971). “Project Stormfury Experimental Eligibility in the Western
North Pacific,” Appendis I,, Project Stormfury Annual Report 1970. Nat. Hurricane
Rex. Lab., Coral Gablos, Florida.
Malone, M. J. and Leimer, D. R. (1971).“Estimation of the Economic Benefits to DOD
from Improved Tropical Cyclonr Forecart.ing,” Rep. Headquarters Military
Airlift Command, Scott AFB, Illinois.
Manabe, S., Holloway, J . L., Jr., and Stone, H. &I. (1970). Tropical circulation in a time-
integration of a global model of t h r atmosphere. J . Atmos. Sci. 27, 580-613.
Matano, H., and Sekioka, 31. (1971). Some aspectn of extratropical transformation of a
tropical cyclone. J . Meteorol. Soc. J a p . 49, 736-743.
98 FRANCIS 1. FENDELL
Matthews, D. A. (1971). “Ice-phase Modification Potential of Cumulus Clouds in

Hurricanes,” Appendix H, Project Stormfury Annual Report 1970. Nat. Hurricane
Res. Lab., Coral Gables, Florida.
Meyer, J. W., ed. (1970). “Airborne Severe Storm Surveillance,” Vol. 1. Tech. Note
1970-43. Lincoln Lab., Lexington, Massachusetts.
Meyer, J. W. (1971s). Toward hurricane surveillance and control. TechnoZ. Rev. 74,53-60.
Meyer, J. W. (1971b). “Airborne Severe Storm Surveillance,” Vol. 2, Tech. Note 1970-43.
Lincoln Lab., Lexington, Massachusetts.
Miller, B. I. (1958). On the maximum intensity of hurricanes. J. Meteorol. 15, 184-195.
Miller, B. I. (1967). Characteristics of hurricanes. Science. 157, 1389-1399.
Munk, W., Snodgrass, F., and Carrier, G. (1956). Edge waves on a continental shelf.
Science 123, 127-132.
Murray, F. W. (1967). On the computation of saturation vapor pressure. J. Appl.
Meterol. 6, 203-204.
Ogura, Y. (1964). Frictionally controlled, thermally driven circulations in a circular
vortex with application to tropical cyclones. J . Atmos. Sci. 21, 610-621.
Oliver, V. J.,and Anderson, R. K. (1969). Circulation in the tropics as revealedby satellite
data. Bull. Amer. Meteorol. SOC.50, 702-707.
Ooyama, K. (1969). Numerical simulation of the life-cycle of tropical cyclones. J .
Atmos. Sci. 26, 3-40.
Ooyama, K. (1971). A theory on parameterization of cumulus convection. J. Meteorol.
SOC. Jap. 49, 744-756.
Orton, R. (1970). Tornadoes associated with Hurricane Beulah on September 19-23,
1967. Mon. Weather Rev. 98, 541-547.
Ostapoff, F., Tarbeyev, Y., and Worthem, S. (1973). Heat flux and precipitation estimates
from oceanographic observations. Science 100, 960-962.
C)stlund, H. G. (1968). Hurricane tritium. 11. Air-sea exchange of water in Betsy 1965.
Tellua 20,577-594.
C)stlund, H. 0. (1970). Hurricane tritium. 111. Evaporation of sea water in Hurricane
Faith 1966. J. Geophys. Rea. 75, 2303-2309.
Palmbn, E. (1948). On the formation and structure of tropical cyclones. Geophyaica 3,
26-38.
Palmbn, E., and Newton, C. W. (1969). “Atmospheric Circulation Systems.” Academic
Press, New York
Palmbn, E ,and Riehl, H. (1957). Budget of angular momentum and energy in tropical
storms. J. Meteorol. 14, 150-159.
Paulhaus, J. L. H. (1966). Indian Ocean and Taiwan rainfalls set new records. Mon.
Penner, S. S. (1972). Elementary considerations of the fluid mechanics of tornadoes and
hurricanes. Astronaut. Acta 17, 351-362.
Perlroth, I. ( 1967). Hurricane behavior as related to oceanic environmental conditions.
Tellus 19, 258-268.
Perlroth, I. (1969). Effects of oceanographic media on equatorial Atlantic hurricanes.
TeZZus 21, 2 3 6 2 4 4 .
Petrosyants, M. A. (1972). The national tropical experiment TROPEX-72. Izv. Acad. Sci.,
USSR, Atmos. Oceanic Phya. 8, 519-522.
Phillips, N. A. (1970). Models for weather prediction. Annu. Rev. Fluid Mech. 2,251-292.
Ramage, C. S. (1971). “Monsoon Meteorology.” Academic Press, New York.
Ramage, C. S. (1972). Interaction between tropical cyclones and the China Seas.
Weather 27,484-494.
Rapp, R. (1970). Climate modification and national security. Pap., Metm-ol. Tech.
Exchange Conf., 1970.
Revesz, W., Jr. (1971). Comparison of effects of various tropical storms on the vertical
temperature of the ocean using pictorial representation. Master’s Thesis, Naval
Postgraduate School, Monterey, California.
Riehl, H. (1954). “Tropical Meteorology.” McGraw-Hill, New York.
Riehl, H. (1963a). Some relations between wind and thermal structure of steady state
hurricanes. J. Atmos. Sci. 20, 276-287.
Riehl, H. (1963b). On the origin and possible modification of hurricanes. Science 141,
1001-1010.
Riehl, H. (1969a). On the role of the tropics in the general circulation of the atmo-
sphere. Weather 24, 288-303.
Riehl, H. (1969b). Some aspects of cumulonimbus convection in relation to tropical
weather disturbances. Bull. Amer. Meteorol. SOC.50, 587-595.
Riehl, H. (1972a). “Introduction t o the Atmosphere,” 2nd ed. McGraw-Hill, New York.
Riehl, H. (1972b). Intensity of recurved typhoons. J. AppZ. Meteorol. 11, 613-615.
Riehl, H., and Malkus, J. (1961). Some aspects of Hurricane Daisy, 1958. Tellw 13,
181-213.
Rosenthal, S. L. (1970). A circularly symmetric primitive equation model of tropical
cyclone development containing an explicit water vapor cycle. Mon. Weather Rev.
98,643-663.
Rosenthal, S. L. (1971a). A circularly symmetric primitive equation model of tropical
cyclones and its reponse to artificial enhancement of the convective heating func-
tions. Man. Weather Rev. 99, 414-426.
Rosenthal, S. L. (1971b). The response of a tropical cyclone model t o vsriations in
boundary layer parameters, initial conditions, lateral boundary conditions, and
domain size. Mon. Weather Rev. 99, 767-777.
Rosenthal, S. L. (1971~).“Hurricane Modeling at the National Hurricane Research
Laboratory (1970),” Appendix C, Project Stormfury Annual Report 1970. Nat.
Hurricane Res. Lab., Coral Gables, Florida.
Saffir, H. S. (1973). Hurricane wind and storm surge. Mil. Eng. 65, 4-5.
Sanders, F., and Burpee, R. W. (1968). Experiments in barotropic hurricane track
forecasting. J . AppZ. Meteorol. 7 , 313-323.
Schwarz, K. (1970). The unprecedented rains in Virginia associated with the remnants
of Hurricane Camille. Man. Wedher Rev. 98, 851-859.
Shea, D. J. (1972). ‘‘ The Structure and Dynamics of the Hurricane’s Inner Core Region,”
Pap. No. 182. Dept. Atmos. Sci., Colorado State university, Fort Collins.
Sheets, R. C. (1972). “Some Statistical Characteristics of the Hurricane Eye and Mini-
mum Sea-level Pressure,” Appendix I, Project Stormfury Annual Report 1971.
Nat. Hurricane Ree. Lab., Coral Gables, Florida.
Shuleykin, V. V. (1970). The power of a tropical hurricane as a function of the underlying
sea surface temperature. Izv., Acad. Sci., USSR, Atmos. Oceanic Phya. 6,729-739.
Shuleykin, V. V. (1972). Development and decay of a tropical hurricane under various
thermal conditions. fzv., Acud. Sci., USSR, Atmos. Oceanic Phys. 8, 1-17.
Simpson, R. H. (1971). “The Decision Process in Hurricane Forecasting,” NOAA Tech.
Memo. NWS SR-53. Nat. Weather Sew. Southern Region Headquarters, Fort Worth,
Texas.
Simpson, R. H., and Frank, J. R. (1972). Atlantic hurricane season of 1971. M a n .
Weather Rev. 100, 256-267.
Simpson, R. H., and Malkus, J. S. (1964). Experiments in hurricane modification.
Sei. Amer. 211, 27-37.
Simpson, R. H., and Pelisser, J. M. (1971). Atlantic hurricane season of 1970. Man.
Spark, E. H. (1971). The willy-willy. Bull. Amer. Meteorol. SOC.52, 574-575.
Spiegler, D. B. (1972). Cyclone categories and definitions: Some proposed revisions.

Bull. Amer. Meteorol. SOC. 53, 1174-1178.
Sundqvist, H. (1970). Numerical simulation of the development of tropical cyclones
with ten-layer model. Part I. Tellua 22, 359-390.
Sundqvist, H. (1972). Model tropical cyclone behavior in experiments related to modi-
fication attempts. Tellua 24, 6-12.
Varga, G. (1971). Some observations on the structure of shallow typhoons. Master's
Thesis, Naval Postgraduate School, Monterey, California.
Vollsprecht, R. (1972). Willy-willy, hurricane or cockeyed bob? Bull. Amer. Meteorol.
SOC.53, 888.
Walker, J. M. (1972). The monsoon of southern Asia: A review. Weather 27, 178-189.
Warsh, K. L. (1973). Relation of sea-air interface energy fluxes t o convective activity in
the tropical Atlantic Ocean. J . Geophys. Res. 78, 504-519.
Warsh, K. L., Echternacht, K. L., and Garstang, M. (1971). Structure of near-surface
currents east of Barbados. J. Phys. Oceanogr. 1, 123-129.
White, R. M. (1972). The national hurricane warning ;rogram. Bull. Amer. Meteorol.
SOC. 53, 631-633.
Wright, R. (1969). Temperature structure across the Kuroshio before and after Typhoon
Shirley. Tellua 21, 409-413.
Yamasaki, M. (1968). Detailed analysis of a tropical cyclone simulated with a 13-layer
model. Pap. Meteorol. Geophys. 19, 559-586.
Yamasaki, M. (1969). Large-scale disturbances in the conditionally unstable atmosphere
in the low latitudes. Pap. Meteorol. Qeophys. 20, 289-336.
Yanai, M. (1964). Formation of tropical cyclones. Rev. Geophys. 2, 367-414.
Zipper, E. J. (1969). The role of organized unsaturated convective downdrafts in the struo-
ture and rapid decay in a n equatorial disturbance. J . Appl. Meteorol. 8, 799-814.
Note Added in Proof. Several recent measurements in hurricanes seem to cast doubt on
the Riehl-Malkus oceanic heat source postulate. First, properly adjusted measurements
i n the frictional boundary layer of Hurricane Ginger (September, 1971), a minimal
hurricane in which winds did not much exceed 90 mph, indicate that at the height of
500 f t above sea level the static temperature decreased 1"-1.6"C from edge t o eyewall.
The need for oceanic heat transfer to maintain a supposedly isothermal boundary layer
appears nonexistent, if this measurement proved t.ypical. Also, both radar measurements
and satellite photography of the eyewalls of Eastern Pacific Hurricane Irah on 23 Sept-
ember 1973 (with maximum wind speed of 130 kt) and of North Atlantic Hurricane
Ellen on 21 September 1973 (with maximum wind speed of 120 k t andminimum central
pressure of 962 mb) indicated significant outward sloping of the eyewall [see, e.g.,
NASA Johnson Space Center Skylab Program photograph SL3-118-2189]. In fact, radar
suggests that the eyewall of the moderately intense cyclone Ellen sloped outward at 46"
above 15,000 ft; past dismissal of radar data reporting significant outward eyewall
sloping may have been unjustified. As a hurricane matures, the data suggest the intense
convective updrafts in the core tilt more from the vertical (perhaps permitting pmipi-
tation, re-evaporated during descent, t o enrich the low-level inflow), and persist longer
(50 min, as opposed to 10-min lifespan further from the core) [Black, P. C. (1974).
Preliminary assessment of handheld photographic observation of tropical storms from
space. In "Skylab Visual Observations," Chap. 16. NASA Johnson Space Center,
Houston, Texas]. The fact that compressional heating in the eye can directly contribute
to the lightening of a column of air whose base lies in the highly swirling eyewall, obviates
the need for an oceanic heat source to achieve that pressure deficit from ambient oom-
patible with observed hurricane wind speeds. On the other hand, if the low-level inflow
layer were to prove saturated near the eyewall, augmentation within the hurricane of
ambient-level heat transfer from the ocean to the atmosphere would still be required.
A NUMERICAL STUDY OF VACILLATION
A. Quinet
lnstitut Royal MCtCorologique de Belgique, Bruxelles, Belgium
1. Introduction ....................................................... . . . . . . . . . . . 101

2. The Laboratory Simulation of Large-Scale Atmospheric Flow.................
............ 102
2.1. The Various Simulated Geophysical Flows . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 102
2.2. The Essential Distinction between Hadley and Rossby Flows . . . . . . .
2.3. The
2.3. The Characteristic
Characteristic Features
Features of Vacillation .........................
of Vacillation
3. A Model Atmosphero for the Study of Vacillation . ......
3.1. General Comments on thP Thcwretical Approach to the Study of
............. .........
Vacillation ...................................................
Vacillation
3.2. The Tensor Formulation of Quasi-Hydrostatic Flows . . .
4. The Spectral Dynamics and Energetics of tho Model

4.1. The Spectral Representation . . . . . ..........................
....................... 131
4.2. The Spectral Equations . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . 141
4.5. Comments on the Various Physical PI,
5. The Numerical Study of Vacillation . . . . . . . . . . . . . . . . . 149
5.2. The Potential Energy Vacillation ...........................

5.3. The
5.3. The Kinetic
Kinetic Energy Vacillation . . . . .
Energy Vacillation .......................
5.4. ......................................
5.4. Tho Various Vacillation Cyc1c.s ........................ 175
6. Vacillation in the Atmosphere . . . . . . . . . . . . . 178
.......................
. . . . . . . 180
List of Symbols . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
1. TNTRODUCTION
It is common knowledge that the motion of the atmosphere is extremely
complex. However, the upper level \reather maps exhibit flow patterns with
more or less permanent and characteristic properties. Perhaps the most
surprising observation is the extremely restricted number of scales of motion
as compared with the a priori infinite number of degrees of freedom involved
in the general thermohydrodynamical equations. The existence of moving
waves whose progression is associated with a change in the tilt of their
through and ridge lines, as well as the alternation between quasi-zonal
and essentially meridional (lobed) circulations, are also typical features of
the large-scale atmospheric flow.
101
102 A. QUMET
It may therefore be believed that the apparent complexity of the weather

maps should result from a few, highly selective processes linked in a more or
less simple way. To identify these basic mechanisms and to state clearly
how they interplay is the fundamental problem of large-scale atmospheric
dynamics. To achieve this aim i t is extremely useful to study simple geo-
physical flows that can be generated in the laboratory. In this respect,
vacillation constitutes one of the most interesting phenomena.
The following text is devoted to the numerical study of vacillation. Section
2 gives a description of this kind of flow as it is known from laboratory
experiments. Section 3 considers the minimum set of conditions to be fulfilled
by the simplest possible numerical model to be able to simulate the phe-
nomenon. Since vacillation appears in flows with various geometrical con-
straints, the tensor formulation of the equations has been adopted. The
reader who is not, and does not want to become familiar with this technique
can go directly to Section 3.3.2 where he will find the classical two-layer
quasi-geostrophic model equations. Section 4 gives the spectral form of these
equations when the fluid moves in a channel. Section 4 describes also the
various processes capable of taking place in the model and interprets them
in terms of the flow pattern. I n order to allow an interpretation based on
energy processes too, the spectral energetics of the model is formulated.
Section 5 is a detailed numerical study, from both the dynamics and the
energetics points of view, of two types of vacillation of the kind emphasized
from laboratory experiments. Finally, some cases of ahmospheric situations
presenting striking similarities with vacillation are pointed out in Section 6.
2. THELABORATORY
SIMULATION
OF LARGE-SCALE FLOW
ATMOSPHERIC
2.1. The Various Ximulated Geophysical Flows
Laboratory experiments are one of the primary sources of knowledge in

physics. Having carefully collected sufficient experimental observations, a
theoretical explanation may be elaborated, leading to new results which in
turn must be checked against new experiments. I n this way, theory and
experiment usually interplay in the study of physical phenomena.
Direct experimentation however is not possible in the case of some geo-
physical phenomena such as the dynamical processes associated with the
large-scale atmospheric motion systems having linear horizontal dimensions
of several thousands of kilometers. Nevertheless, making use of the concept
of dynamic similarity (Rossby, 1926), one may try to build up laboratory
devices which are capable of simulating some features of the actual atmo-
A NUMERICAL STUDY OF VACILLATION 103
spheric flow. There have been many attempts in this direction in the past
(for a n historical review, see Fultz et al., 1959).
Almost twenty-five years ago, such experiments, first suggested by
Rossby (1947) and Starr were carried out by D. Fultz a t the University of
Chicago. I n his initial experiments, Fultz (1949) generated thermally driven
circulation in a rotating hemispherical shell of liquid. For suitable conditions
of rotation rate and thermal forcing, he observed flow patterns for which the
speeds of flow and changes with latitude compare reasonably well with the
wind speeds observed in the atmosphere, provided the speeds are expressed
in an appropriate unit, i.e. the equatorial rim speed.
This quantitative agreement stimulated further work in the laboratory
modeling of the atmosphere. It should be recalled that, in analogy with the
earth, where the resultant gravity field is nearly normal to the earth's mean
sea level surface, Rossby (1926) suggested using as a bottom surface for the
tank the paraboloid whose shape is the equipotential surface of the resultant
of the local gravity field and of the centrifugal force due to the rotation of
the vessel. For the rotation rate imposed in laboratory arrangements to
simulate the atmospheric conditions, the slope of this equipotential surface
is extremely small.
If one now considers a nearly radial gravity field operating on a fluid in
motion on a sphere, i t is clear that this configuration of the force field is
impossible to reproduce with laboratory devices. It should however be noted
that the effect of the curvature of the earth can be taken into account by
other factors. For instance, in the case of a barotropic flow in arotatingtank,
the theorem of conservation of absolute vorticity shows that an inward
radial depth decrease of the fluid is dynamically analogous t o an inward
radial increase of entrainment vorticity (usually denoted by f in meteorology).
It should be noted that the use of a bottom surface normal to the resultant
force field operating on the fluid in the tank suppresses the radial depth
gradient. On the other hand, the radial increase of depth in an hemispherical
shell operates in an opposite direction t o the one imposed to simulate the
poleward increase of the earth's vorticity in barotropic flows.
This is perhaps one of the reasons why Fultz quickly dispensed with the
hemispherical arrangement and used a cylindrical vessel, the so-called dish-
pan, which is, for a given experiment, rotating a t constant angular velocity Sl
(simulating the west to east rotation of the earth) around the vertically
oriented axis of the cylinder. Differential heating is provided, a t the bottom
disk, by steady symmetrically distributed heating a t the periphery (simulating
equatorial conditions) and cooling a t the center of the disk (simulating polar
conditions).
Experiments with a cylindrical annulus rotating around its vertical axis
were initiated in 1950 by R. Hide a t Cambridge University. I n these experi-
104 A. QUINET
ments, the heat source is at the outside wall which is a t constant temperature
T, (simulating low latitude heating) and the cold source is a t the inside
wall a t constant temperature T, (simulating high latitude cooling, T,< Tb).
When experiments are performed a t different rotation rates and different
intensities of the thermal forcing (i.e. for different south-north temperature
contrasts) different types of flow patterns are generated. Roughly speaking,
the dishpan experiments and those with a cylindrical annulus give the same
qualitative observations. However, the reproducibility of the results is
much better in the annulus experiments, presumably because the flow is
much more constrained by the geometry of the system and consequently
less sensitive to random perturbation in the initial conditions of operation.
For some intensities of thermal forcing and some rotation rates, the flow
is symmetric with respect to the rotation axis and proceeds as in a trade wind
cell, with upward motion at the warm rim and downward motion a t the cold
source. While moving northward in the upper layer (from the heat source
t o the cold source) the fluid is deflected by the rotation so that southerly
winds are gradually changed into winds with a westerly component. At the
“ pole,” the fluid particles subside and give rise, when moving to the south
(from the cold source to the warm source), to northeasterly trade winds in
the bottom layer. From these observations presenting obvious similarities
with what is known about the tropical atmosphere, Fultz et al. (1959) have
called this flow a Hadley flow.
For other operating conditions, unsymmetric flows with traveling waves
arise. Some of these have a striking resemblance t o the Rossby waves which
are observed on upper level synoptic maps a t middle and high latitudes.
This similarity has led to the term: Rossbypows.
Careful experiments have shown that it was possible to distinguish several
types of Rossby flows, each of them constituting a Rossby regime. Thus, Hide
(1968) was the first (1950) to observe a steady Rossby regime in which the
waves progress a t constant speed and without changing their shape. The ex-
istence of another Rossby regime was also discovered by Hide (1953, 1958),
namely a periodic regime in which the Rossby waves undergo periodic changes
in their amplitude, shape, and progression (wave speed). Hide called this
phenomenon vacillation. Finally, nonperiodic flows have also been observed,
first (1950) by Long and Owens working with the dishpan arrangement at
Chicago (Fultz et al., 1964) and are classified into the irregular Rossby regime.
To sum up, the Hadley 00w involves a single type ofregime, called the Hadley
regime, while Rossby flows can fall into one of the three types: steady,
vacillating, or irregular Rossby regimes.
The critical conditions of thermal forcing and rotation rate for the transi-
tion between the Hadley and the steady Rossby regimes as well as between
steady Rossby regimes with different wavenumbers have been experimentally
FIG.1. Spectrum diagram of transition curves obtained by Fultz et al. (1959) in an

annulus of water; abscissa is proportional t o the square of the rotation rate fi and the
ordinate is proportional, for a given rotation rate, to the imposed thermal contrast A T .
The “knee ” curve separates the symmetric Hadley flow to the left of the curve and the
Rossby flow. Figures indicate the wavenumber prevailing in the Rossby flow. Each
experiment was performed in such a way that the radial temperature difference was
raised slowly from zero after reaching the desired constant rotation rate.
determined by Fultz et al. (1959). Figure 1 gives the curves obtained by

these authors working with a cylindrical annulus. Moreover, Fowlis and Hide
(1965) have also given the critical transition conditions between the steady
wave regime and the irregular regime. Vacillation arising between these two
types of regimes is considered by thew last authors as a transitional phe-
nomenon whose domain of appearance in the plane of external parameters
(Q and AT = T,- T,) is rather restricted as compared with the domain
of existence of the other regimes.
It is interesting to recall that the experiments of Fultz et al. were performed
in such a way that the fluid was first brought into solid rotation with the
cylindrical annulus whose walls were initially kept a t the same temperature.
Having reached the desired rotation rate, the walls were then very slowly
heated and cooled to the desired temperature values T , and T,, respectively.
This procedure allows one to state unequivocally the critical conditions
of rotation rate and thermal forcing for the establishment of the different
regimes. It is however completely artificial. Indeed, as noted by Fultz et al.
(1959), different regimes can be reached for the same values of Q and AT
depending upon the way the steady values of these parameters are attained.
This “hysteresis,” making different states of motion correspond to the same
values of Q and AT, is a consequence of the nonlinear character of the equa-
tions governing the flow (Lorenz, 1962; Quinet, 1973b).
106 A. QUINET
2.2. The Essential Distinction between Hadley and Rossby Flows

Before dealing with vacillating flows in more detail, one may recall that the
experiments have shown that there are fundamental differences between the
two classes of Hadley and Rossby flows, the most characteristic one being
the process of heat transport. I n the Hadley flow, this transport is accom-
plished by meridional circulation (in planes containing the axis of rotation)
as a t low latitudes in the atmosphere, while in all Rossby regimes the quasi-
horizontal disturbances ensure the major part of this transport, as in the
atmospheric westerlies. In laboratory experiments, Hadley circulations in
particular are associated with slow rotation rates Q. On the other hand, the
vertical component ofthe earth’s rotation is zero a t the equator and increases
poleward. Thus, in laboratory experiments as well as in the atmosphere,
the flow tends to be three-dimensional if the vorticity of the absolute rotation
is notably less than the vorticity of the relative motion. The comparison
between the relative and entrainment vorticities is expressed by the Rossby
number
(2.1) RO= .rZ/LQ= iJ2Q
where S is the horizontal relative velocity scale, L the length scale, and [ the
mean vertical component of the relative vorticity of the flow. This number
also expresses the ratio between the relative acceleration and the Coriolis
acceleration. Quasi-geostrophic motions, for which the Coriolis acceleration
dominates, are consequently associated with the low values of Ro, that is to
say with high rotation rates R. On the contrary, when these rotation rates
are small, there exists a n unbalanced part of the Coriolis force (the ageo-
strophic part) and the zonal flow is systematically deflected in a direction
normal to the flow, forcing the fluid particles to travel along horizontal
spirals. Vertical (ageostrophic) motions, imposed by continuity, then take
place a t the cold (descending motion) and warm (ascending motion) sources
and allow the spiraling fluid particles to travel between the heat sources,
ensuring in this way the closure of the heat transfer cycle.
At higher rotation rates, quasi-geostrophic flows become possible. I n this
case, the flow remains essentially two-dimensional (quasi-horizontal), the
fluid particles approaching the cold source being able to progress later south-
ward on the same horizontal surface as a consequence of the balance between
the Coriolis forces and the horizontal pressure gradient forces. Consequently,
the heat transfer is now accomplished by horizontal eddies, the fluid particles
successively traveling quasi-horizontally from south to north and then from
north to south. Since the thermal wind law holds for quasi-geostrophic flow,
there is a strong correlation between the flows in the different layers. This
is not so for Hadley flows where trade or antitrade winds prevail in the lower
or upper layer respectively.
A NUMERICAL STUDY O F VACILLATION 107
The sudden transition from the Hadley to the Rossby regime when the
rotation rate is progressively increased suggests the intervention of some
mechanism of dynamic instability. According t o Bolin (1952) and following
the theory of baroclinic (Charney, 1947; Eady, 1949) and barotropic (Kuo,
1949) instability, the dynamic instability condition would result from the
increase of the wind shear (both horizontal and vertical) that might be
associated with the increase of the rotation rate. But the observed restabiliza-
tion of the symmetric Hadley regime at high thermal forcing (upper part of
Fig. 1) then appears as paradoxical. Indeed, it can be expected that intense
thermal forcing generates high horizontal temperature gradients within the
fluid and consequently, according to the thermal wind relation, high vertical
wind shear. The paradox was solved by Lorenz (1962) who pointed out that
intense thermal forcing gives rise to intense upward motions of warm air
near the outer rim and downward motion of cold air near the center (or the
inner rim). Consequently, the static stability of the fluid increases with the
thermal forcing so that finally the flow becomes baroclinically stable again
and the symmetric Hadley regime is reestablished. I n this context it has to
be emphasized that baroclinic instability is considered as the single mechanism
responsible for the transition from the Hadley to the steady Rossby regime.
This has been indirectly confirmed by Lorenz (1962) who generated steady
Rossby regimes in a model in which the process of barotropic instability was
suppressed.
Let us also point out that, if vertical motion plays an essential role in
realizing the critical conditions of static stability allowing the reappearance
of Hadley flows, some differences can be anticipated according to the way
in which the increase of static stability can be accomplished. I n the Hadley
regime, the cellular structure of the flow ensures a self-stabilization. This is
not the case in Rossby flows where the motion is essentially quasi-horizontal.
Accordingly, the reestablishment of a Hadley regime from a Rossby regime
will presumably need a higher thermal forcing than the one compatible with
a directly established Hadley flow. This hysteresis in the observation of
Hadley flows according to the sequence of generated flows has indeed been
observed by Fultz et al. (1959) and has been explained by Lorenz (1962).
The laboratory simulation of both Hadley and Rossby flows strongly
suggests that, in spite of the obvious geometrical distortion, laboratory
experiments are able t o capture some of the very essential properties of the
large-scale atmospheric motions. Interest will be more particularly focused
here on Rossby vacillating flows which appear sufficiently simple as to
remain " understandable " and, consequently, capable of improving our
knowledge of the large-scale atmospheric motions. Indeed, the distinction
between vacillation and nonperiodic flows seems to be more a question of
degree than one of fundamental character, as already denoted by the fact
108 A . QUWET
that they both belong to the same class of flow. The observation of several
sequences of upper air synoptic maps suggests that the evolution of the
atmospheric 00w pattern could be interpreted as some " perturbation " of a
vacillation cycle. If this could actually be confirmed by a detailed analysis,
much would be gained about the understanding of the behavior of the
atmosphere.
2.3. The Characteristic Features of Vacillation
2.3.1. Vacillation in the Laboratory Flows. It has already been mentioned

that Hide has been the first to recognize a new phenomenon, he called
vacillation, in which the waves of the flow pattern are submitted t o periodic
fluctuations in amplitude, shape, and progression rate. This type of flow
has also been extensively studied by Fultz and his collaborators. Figure 2
gives one example of the upper flow observed in a cylindrical annulus during
a vacillation cycle whose period is 16.25 revolutions (one revolution simulating
one day). The time interval between two successive photographs is four
revolutions. The most peculiar feature is the periodic change in the horizontal
plane of the orientation of the trough and ridge lines of the five-wave, while
the zonal flow is mainly concentrated in a meandering jet stream. Moreover,
the amplitude of the five-wave also exhibits fluctuations, increasing from the
tilted five-wave pattern of Fig. 2a to the strong upper cyclonic circulation
of Fig. 2c.
One of the essential results of laboratory experiments has been to show
that vacillation is accompanied by a periodic variation in the heat trans-
ported from the warm to the cold source. A representation of these fluctua-
tions can be given by measuring the temperature difference between the
inlet and outlet water circulating a t the cold bath of the annulus as repre-
sented in Fig. 3. Fultz et al. (1959) have also shown that the meridional
profile of the zonal wind undergoes substantial fluctuations during the cycle.
On the one hand, the position of the maximum zonal wind fluctuates in
latitude, being successively near the outside and the inside wall. On the
other hand, the intensity of the zonal wind varies simultaneously with the
same period. These fluctuations in wind speed are such that the maximum
zonal westerlies are attained near the time of the minimum heat transport
a t the cold source.
Obviously, many properties of the flow exhibit the same kind of periodic
fluctuations. Thus large positive values of the zonally averaged relative
northward eddy angular momentum transport r2[u'v1],due to the fluctuations
u', v’ of the velocity components u (u > 0, westwind) and v (v > 0,south-
wind) from their zonal average [u],[v] along the latitude circle of radius r,
are followed by smaller values. As a rule, the maximum transport is located
(b) (d)
FIG. 2. Sequence of photographs showing a kinetic energy vacillation cycle. Notice

the changes in the orientation of the trough axis from NW-SE in Fig. 2b to NE-SW
in Fig. 2d and the change of amplitude of the wave between Fig. 2a and Fig. 2c. The
period of the cycle is l e t revolutions; the time interval between two successive photo-
graphs is 4 revolution8 (courtesy of 1).Fultz).
FIQ.3. Trace obtained by Fultz el al. (1959) of the temperature difference between
inlet and outlet water a t the cold source during vacillation. The period of the cycle is
about 12 revolutions.
110 A. QUINET
a t middle latitude (approximately midway between the warm and cold

sources) while the subsequent smaller values are slightly shifted to the north
(in the direction of the cold source).
According to Fultz et al. negative values of the momentum transport
occur when the waves a m tilting from N W toward SE, a situation which
arises generally when the troughs are deepening t o generate upper cyclones,
somewhat after the time of occurrence of the minimum value of the heat
transport a t the cold sourm.
The relative northward eddy angular momentum trensport P[u’u’] in
10-*rbZRaunits, estimated after Fultz et d.(1959) data, across the latitude
circles r = 0.8fb and r = 0.6rb, is shown in Fig. 4. The figures on the abscissa
N v’u‘l
r=O.B rb
-2,
-2 I
FIQ. 4. Relative northward eddy momentum trenaport rl[u’v’] in lO-‘r,*R* units

amom latitude circle r = 0.8rb and r = 0.6rb during the vacillation cycle, being the
angular velocity of rotation and rb the radius of the outer rim; the inner rim is at radius
0.4rb. T h e small cimlea denote the time of occurrence of maximum (Mx)and minimum
(Mi) heat transport to the cold source; numbers in absoiaaa are revolution numbers
(After Fultz et al., 1869).
denote revolution numbers so that the vacillation period is about 12 days.

The outer rim is a t a radius rb and the inner rim lies a t 0 . 4 ~The ~ . mean
transport during one vacillation cycle a t 0.8rb is definitely toward the north.
The small circles denote respectively the time of occurrence of maximum
(Mx)and minimum (Mi) heat transport at the cold source. Moreover, Pfeffer
and Chiang (1967) noted that Fultz’s data can also be used to demonstrate
that the barotropic transfer rate between eddy kinetic energy and zonal
kinetic energy undergoes substantial fluctuations in magnitude and also
perhaps in sign during the cycle. Consequently, while baroclinic processes
are necessarily associated with the thermal forcing, nevertheless vacillation
A NUMERICAL STUDY O B VACILLATION 111
involves also nonnegligible barotropic energy exchanges. Moreover, the

cyclic increase and decrease of the transfer between the two types of kinetic
energy proves that the zonal flow interacts with the eddies so that nonlinear
mechanisms play an essential role in vacillating flows.
More recently, Pfeffer and his collaborators a t Florida State University
(1965), proceeded to very careful experiments, especially in the domain of
vacillating flows. While recognizing that the difference is presumably rather
one of degree than one of intrinsic nature, they were led to distinguish two
kinds of vacillation.
A first one, of the kind presented in Fig. 2, is called a “ tilted trough vacilla-
tion.” Moreover, in order to draw attention t o the very fact observed by
Fultz et al. (1959) that the fluctuations of the tilt of trough and ridge lines
are associated with fluctuations of the zonal and eddy kinetic energies,
Pfeffer and Chiang (1967) suggest also calling this type of flow a “kinetic
energy vacillation.” But Pfeffer et al. (1965) and Pfeffer and Chiang (1967)
also reported the existence of a different kind of vacillation, a n example of
which is given in Fig. 5. The essential difference from the case reported in
(a1 (b) ( C )
FIG.5. Sequence of photographs showing extreme stages of a potential energy vacilla-

tion cycle. Notice the change in the amplitude of the wave without noticeable change
of its tilt. The period of the cycle is 92 revolutions; the time interval between two
successive photographs is 46 revolutions (courtesy of R. Pfeffer and W. Fowlis).
Fig. 2 lies in the fact that the trough and ridge lines no longer exhibit fluctua-
tions in orientation during the cycle but that the wave amplitude undergoes
major oscillations. The wave may even nearly vanish during the cycle so
that the flow appears then as quasi-zonal (Figs. 5a and 5c). Therefore, these
authors have proposed calling this type of vacillation “ amplitude vacillation.”
As a matter of fact, amplitude vacillation is most common with high viscous
fluids and appears at a lower rotation rate and higher thermal forcing than
the tilted trough vacillation (Fowlis and Hide, 1965).
112 A. QUINET
2.3.2. Meteorological Interest of Vacillation. The similarity of all the

characteristics of vacillation with what is known about the atmosphere
makes vacillating flows of comiderable meteorological interest. The tilting
of the planetary waves has indeed been recognized by Starr (1948) to play a
fundamental role in the necessary northward transport of angular momen-
tum of the atmosphere by the large-scale atmospheric eddies. The positive
correlation between the fluctuations of northward and eastward velocity
components required for a northward transport of angular momentum across
latitude circles is realized when the troughs exhibit a SW t o NE tilt as, for
example, in Fig. 2a and Fig. 2d. Thus, in this respect, laboratory vacillating
flows do simulate one of the most fundamental processes of the atmospheric
large-scale turbulence. Figure 5 and, less obviously, Fig. 2 also suggest a n
analogy between vacillation and the recognized quasi-periodic redistribution
of the angular momentum within the atmosphere giving rise to the so-called
index cycles (Namias, 1954). I n this respect, Fig. 5a or Fig. 2a would be
typical of a high index, quasi-zonal circulation and Fig. 5b or Fig. 2c of a low
index one. The estimates of the heat transport show that, in the atmosphere
as well as in the laboratory flows, the low index configuration is associated
with a maximum value of the heat transport to the cold source while high
index circulation corresponds t o a minimum value of this transport. Finally,
the period of vacillation has been observed to vary between large ranges,
depending on the rotation rate 2 ! and the thermal forcing AT. Nevertheless,
these periods are most often of the order (in the corresponding units, recalling
that one day i s equal t o one revolution) of several weeks which is of the order
of magnitude of the duration of the atmospheric index cycles (Namias, 1954).
Moreover, the evolution of the meridional profile of the zonal wind during
a vacillation cycle, as recalled in Section 2.3.1, is consistent with what is
observed during a n atmospheric index cycle.
Pfeffer and Chiang (1967) notice that in the extreme case of low index
during an amplitude vacillation (Fig. 5b) the flow pattern closely looks like
what is known as a blocking situation in synoptic meteorology. However,
blocking situations do not generally repeat themselves around a whole
hemisphere. Nevertheless configurations of three to five blocking highs are
not uncommon, and have been reported by Pfeffer and Chiang (1967). The
transition from a high index (quasi-zonal) circulation of the atmosphere t o
a configuration of a few blocking highs, regularly distributed around the
hemisphere, is generally rather abrupt and does not reappear with a definite
periodicity in time. According t o the cases reported by Pfeffer et al. (19651,
a particular feature of this major change of the whole atmospheric circulation
of the Northern hemisphere would happen usually in winter, either in late
December or in January, and be followed by an extreme maximum amount
of zonal available potential energy. The energy study of this type of process
A NUMERICAL STUDY O F VACTLLATION 113
(Krueger et al. (1965)) has shown that it is characterized by a rapid adiabatic

conversion from potential energy into kinetic energy. On this base, Pfeffer
and Chiang (1967) have suggested also calling the amplitude vacillation a
“potential energy vacillation.”
All these considerations lead to the idea that potential energy vacillation
should essentially result from cyclic variation of the conversion rate between
available potential energy and kinetic energy only (baroclinic process) while
the appearance of kinetic energy vacillation necessarily implies the existence
of substantial transfers between eddy and zonal kinetic energy (barotropic
process). However, Pfeffer and Chiang (1967) recall that vacillation un-
doubtedly involves nonlinear effects, which make it impossible to isolate a
single energy process, either barotropic or baroclinic, in the energy cycle of
the flow. A detailed analysis of the energy cycle of potential and kinetic
energy vacillation would therefore considerably improve knowledge of the
phenomenon. This problem will be dealt with in particular in this review
with the aid of numerical simulation.
3. A MODELATMOSPHERE
FOR THE STUDYOF VACILLATION
3.1. General Comments on the Theoretical Approach to the Study of Vacillation

Theoretical studies (Davies, 1953, 1959; Kuo, 1954, 1957; Lorenz, 1953)
have been devoted to the different t>ypesof flow observed in laboratory
experiments in order to derive, in particular, the critical operational con-
ditions of thermal forcing and rotation rate for the appearance of the different
flow regimes. The Hadley regime and the conditions of transition from this
regime t o the Rossby regimes have especially been extensively discussed by
the authors mentioned above.
More recently, Lorenz (1962) put forward the fruitful idea of using in this
approach the spectral form of a set of thermohydrodynamical equations
capable of describing the fluid motion. To simulate the laboratory flow,
Lorenz uses a quasi-geostrophic, two-layer model of an inviscid and non-
heat-conducting fluid, the friction and heating effects being taken into
account in a roughly parameterized form. The spectral equations are obtained
by expanding the dependent variables into their spectral components in a
functional space and retaining in the equations a few scales of motion only.
This procedure seems specially suited here since, in most cases, the actual
laboratory flow, and particularly vacillation, involves only very few waves.
As the infinite number of degrees of freedom of the initial formulation is,
in this way, reduced to only a few, the theoretical study turns out t o be
greatly simplified. This technique, however, does not suppress the essential
nonlinear character of the equations.
114 A. QUIN'ET
The Lorenz (1962) determination of the critical conditions of the instability

of the Hadley regime in a cylindrical model is in good qualitative agreement
with the laboratory measurements in an annulus region, as reported by
Fultz et al. (1959) (Fig. 1). The quantitative agreement, however, is rather
poor. I n order t o obtain a better concordance between the theory and the
experiments, Merilees (1968) used the correct cylindrical annulus geometry
and added to Lorenz' model the effects of the viscosity and of the heat
conductivity of the fluid as well as those of the lateral boundary. In this way,
he demonstrated that these physical factors play an important role in the
dynamics of the laboratory flows.
Nevertheless, and even surprisingly, Lorenz (1963) has simulated all types
of laboratory regimes in his physically highly simplified model where,
moreover, the channel geometry was used. This seems to support the hypo-
thesis that the different regimes observed in laboratory experiments con-
stitute a general characteristic of thermally driven rotating fluids. If the
geometry of the experiment and the physical properties of the fluid influence
the critical conditions of transition between different regimes or between
different wavenumbers within the Rossby regimes, they do not critically
alter the global mechanical behavior of the system. Consequently, and
although we are aware that our simplification could appear as rather drastic,
we shall use the channel type Lorenz' model for the study of vacillation.
Needless to say, the aim is to capture only some of the essential, nonlinear
aspects of the phenomenon. First the dynamics will be considered and then
its energetics. Finally, the existence of vacillation in the atmosphere will
be briefly presented.
3.2. The Tensor Formulation of Quasi-Hydrostatic Flows
3.2.1. General Characteristics of Tensor Calculus. Vacillation is possible

with systems of various geometrical shapes. Therefore, it will be useful to
express the equations of flow in the most general form, valid in any coordinate
system. For this purpose, use will be made of tensor calculus.
Let us first specify a distinction between (i) a reference body R, which will
be rigid and (ii) a coordinate system, rigidly attached to R (imbedded in R)
or not. The following convention will prove to be convenient: an index repre-
sented by a Greek letter takes the values 1, 2, and 3 while one represented by a
Roman letter will take the values I and 2 only.
The velocity of a material point P of coordinates z'(t), x 2 ( t ) ,z3(t),where t
is time, with respect to a reference body R is given by
(3.1) Va=va -ra (a= 1,2,3)

where va is the velocity of P with respect to the coordinate system (x’, x2,x 3 ) ,
(3.2) v a = dX,(t)/&t = 2,
while ra is the velocity of R with respect to this coordinate system. Obviously,
the velocity of all the fluid particles form a velocity field
(3.3) 7"= V,(d, 5 2 , 23, t )
It should be noted that V a being defined with respect to R is a vector,

independent of the choice of the coordinate system, while both va and ra
depend upon this choice.
The kinetic energy per unit mass with respect t o R then assumes the form
(3.4) K = 4ga, V " V , = *v, V"

I n (3.4), as everywhere else in this section, the usual summation convention
of tensor calculus must be applied: if an index (dummy index) appears twice
in an expression, once as a subscript and once as a superscript, it has t o be
summed for all allowed values of the index. The symmetric tensor g,, = gsa
is known as the metric tensor of the three-dimensional space, usually
introduced from the expression of the square of the length 6s of the vector
82,:
(3.5) (SS)2 = g,,(xY, t ) SX" 6x6
the unit length being chosen. The components V , and V" are respectively
the covariant and contravariant components of the vector V and are associ-
ated by the metric tensor according to
(3.6) V a = gad Va
a property which has already been used in (3.4).The covariant components
V a , v, , r, associated respectively with the contravariant components V", va,
r a are such that
(3.1') V,=V,-Tr,
For the sake of completeness, let us also recall that the so-called "physical
components " of V are given by
These are the components of the vector V referred to a Cartesian system

coinciding with the axis of the basis vectors of the arbitrary coordinate
system xa. These basis vectors are defined a t each point and are tangent to
the coordinate curve along which xa alone varies. It will also prove necessary
to use the twice contravariant tensor gaS defined by
116 A. QUINET
where (cof g,,) denotes the cofactor of gas in the determinant

(3.9)
so that
(3.10) gavgYB = 6,s
where S,b is the Kronecker tensor
(3.11) 8,,=1 if a=P, SaB=O if a+/3
With the usual notation of a partial derivative
(3.12) a,(. * -) = ( a p x q . * .)
and considering K as a function of t , xa, and xa ( a = 1, 2, 3), the equation
of the fluid motion with respect t o R may be written (Defrise, 1964)
(3.13) ( q d t ) / ( a K / a q- aK/axa = -zwaP V B- a,+ - p - 1 a,p
where dldt * .) is the derivative along the flow and 2wasV , is the Coriolis
(a
acceleration. When E is the velocity o f R with respect to an absolute reference

body (entrainment velocity) the vortieity tensor 2waPof the absolute motion
of R is given by
(3.14) 2w,B = 8, E , - a, E , = -20Ba
+
I n (3.13) is the potential of the body forces, p the density of the fluid, and
p the pressure. Actually, the right side of (3.13) should also include the
viscous and turbulent stresses; these however will be disregarded here in
accordance with the previous assumption concerning the viscosity of the
fluid.
For the study of large-scale atmospheric motion it is convenient to use the
pressure as the vertical coordinate so that the coordinate system ( X I , xz,
x3 =p ) is not imbedded in R. Van Isacker (1963) adopts arbitrary " horizon-
tal " coordinates xi (i= 1, 2) a t the mean sea level of the earth's surface and
x3 = p . Moreover he takes for the length of the vector (ax', 6x2, 6x3 = 8p)
the approximate form
(3.15) (6s)Z = gr,(xl, XZ) 6x1 62f + p-2g-2(6p)2
where g is the (constant) acceleration of gravity and p = p(xa,t ) . The choice
of the metric (3.15) implies three geometrical approximations: (1) the isobaric
surfaces are considered as being horizontal for measuring distances; ( 2 ) the
distance between two points ( x l , x2, p ) , (x'l, xr2,p ) of the same isobaric
surface is supposed to be independent of the pressure p ; and (3) for measuring
the vertical distance Sz, use is made of the hydrostatic equation
(3.16) 6z= - ( p g ) - l s p
A BUMERICAL STUDY O F VACILLATION 117
along the ascending vertical (2' = a, x2 = h ) of each mean bea level point
(a, b ) . These approximat<ionsare acceptable for the description of the large-
scale flow of a shallow atmosphere in quasi-static motion. From a geometrical
point of view, the splitting between horizontal and vertical coordinates in
(3.15) implies that the three-dimensional space (x', x 2 , p ) is layered into
isobaric sheets by the pressure parametcr p .
Let us immediately note that, according to (3.15),
(3.17) g31= gr3= 0 for i = 1, 2 and g33= (pg)-2
(3.18) Q=g33y
where y is the determinant I g,, I so that
g3' = g i 3 =0 for i = 1, 2 and g33= l/g33
(3.19)
gl1= g22/y, gZ2= S l h , g12 = g2I = -912ly
and
(3.20) g t k p= a,', k = 1, 2
Finally, the velocity of the carth (R) with respect to the coordinate system
(xl, x2, p ) is given by
(3.21) yl = r2 0
the coordinates xl, x2 being imbedded in R, while
(3.22) r3 = apiat
(xl, x2, z being kept constant) so that
(3.23) V 1 = 211, 8 2 = 112, v3 =w - y3
with the usual notation

(3.24) u3 = dpidt =w
Tensor calculus has been used by several authors in meteorology (see, for
instance, Van Mieghem and Vandenplas, 1950; Langlois and Kwok, 1969).
A systematic treatment of tensor calculus specially applied t o atmospheric
mechanics has been given by Defrise (1964).
3.2.2. The Equations of Large-Scale Quasi-Static Flows. According to (3.4)

and (3.1)
(3.25) aK/axa= g,,(x, - r B )= g,, VB = V ,
aK/ax' = - r B ) ( x u- rv) a, gav- gov(xV,
- r Y )a, r4
(3.26)
= &V4Vva,gpv - V , a,r@
118 A. QUINET
so that
The equation of motion (3.13) then assumes the form

(3.28) dV,/dt - &VBVYa.gsy + V , a,@ V 6= -aa+ - p - l aap
or the equivalent, more usual form
(3.28') dIr,/dt - V B V ,riB V , a,rB + + 2Wu8V B= -aa+ -p-1 aap
with the Christoffel symbol
(3.29) r:6 =&gvE(aagc6 + a6ga6 -a a,g aB ) -
Equation (3.28) can also be put in the more convenient form (Van Isacker,
1963; Phillips, 1965)
(3.29') aV/at + V(*V6 V B )-V x curl(V + E) = -V+ -p-lVp
r
free of symbols and where x is the vector product and V the usual " del "
operator.
When (3.15) is taken into account, the equation of the horizontal motion
deduced from (3.28) can be written
(3.30) d ' l / ' , / d t - - V ' ~ k a , g , ~ - S V 3 v 3 a , q ~ ~v+3 a , r 3 t 2 w i , v a = -az+
for i = 1, 2.
The third equation of (3.28) is always replaced, in the study of large-scale
atmospheric flow, by the hydrostatic approximation
(3.31) a=+= -,,-I
which filters out the meteorologically nonsignificant compressibility waves

(see, for instance, Holton, 1972). The usefulness of this approximation results
from the fact that the period T of the compressibility waves in the atmo-
sphere can be much shorter ( ~ ( 1 sec) than the period of the large-scale
Rossby waves (T> lo5 sec). Accordingly, dealing explicitly with the com-
pressibility waves in atmospheric models would necessitate using a n integra-
tion time step which is about lo5 times less than when the flow is considered
t o be hydrostatic. One may nevertheless recall that this approximation is
justified only by the empirical fact that the large-scale motions of the atmo-
sphere are quasi-static.
This simplification, although offering the greatest practical advantages,
violates the conservation laws expressing the fundamental physical principles
which are intrinsically contained in the original equation (3.28). Energetic
consistency of approximate sets of hydrodynamic equations specially suited
for studying the atmosphere has been discussed by Lorenz (1960), Van
Isacker (1963), and Phillips (1965). As the diagnostic equation (3.31) replaces
the equation of motion along the vertical, a consistent approximate kinetic
energy is defined as the kinetic energy of the horizontal motion only. I n this
case, the theorem of the (approximate) kinetic energy holds provided (3.30)
reduces to
(3.32) d V , / d t - - 3 V j V ' " a , g l ' " + 2 W , ~ ~ 3-a,4
=
Note that the terms which have been suppressed in (3.30) may be considered
as negligibly small from scale analysis considerations.
I n analogy with (3.29'), (3.32) can also be put in the form
(3.33) av,/at + a , w j 7') - etj vx +f) + a3 v,= -a, 4
where 5 is the " vertical " component of the relative vorticity
(3.34) 5 =el' a, V j = ( 1 / d P ) ( a l ~-
, a, v,)
and f the vertical component of the vorticity of the earth's rotation
(3.35) f = 2%12/3/
called the Coriolis parameter. The erjand e,, tensors are defined by
ell = eZ2 = e l l = e22 = 0
-
(3.36) el, = - eZ1= 47; el2 = -e21= 1/47
and satisfy the relation
(3.37) etfg,, ekl= -9"
With (3.15) the continuity equation (Defrise, 1964) assumes the form
(3.38) (1/4;) a d ; V,) + 83 w =0
where the first term on the left-hand side is the isobaric divergence of the
velocity field ( V1,V 2 ) ,
(3.39) 8 f Wd;)add; V , )
Finally, the thermodynamic equation assumes the form
(3.40) c,, dTldt - p - l dpldt = H
where T is the absolute temperature, cp the specific heat of the fluid at
constant pressure, and H the rate of heating per unit mass.
With the ideal gas state equation
(3.41) P = pRT,
120 A. QULNET
where R is the specific gas constant for dry air, Eqs. (3.31))(3.33),(3.38),and
(3.40)form a closed differential set in the six unknown functions V1,V2,W,
4,
p, and T and determine, at least in principle, the evolution of a n inviseid
dry atmosphere once suitable initial and boundary conditions are prescribed.
The fluid used in laboratory experiments (generally water) can be con-
sidered as being incompressible so that p is a function of T only and is
expressed by the thermal expansion law
(3.41’) p =POP - E(T - To)]
where E is the coefficient of thermal expansion of the liquid and po its density
at temperature T o .Consequently, except for using (3.41’)instead of (3.41),
the large-scale atmospheric flow and the geophysical laboratory flows are
governed by analogous sets of equations. I n the remaining part of this
paper, we shall limit ourself to the atmospheric case.
3.3. The Equations of the Two-Layer Balanced Model

3.3.1. The Equations of Quasi-Qeostrophic Flow. We shall now establish
the invariant (tensorial) form of the equations of a quasi-geostrophic two-
layer model. The third equation of motion having been replaced by the diag-
nostic hydrostatic equation (3.31), we are concerned here only with the two
equations (3.33) of the horizontal motion.
Let us f i s t recall the invariant form of the Laplacian
(3.42) V2A = (lid;)a,(y1’2gi’a, A )
and of the Jacobian
(3.43) J ( A , B ) = eU a, A . a, B
in the two-dimensional space (xl, x 2 ) , A and B being two scalars. Van
Isacker (1963) considers also the invariant differential operator
(3.44) &(A, B ) = g” a, A * a, B
equivalent to V A * V B in the usual “del ’) operator notation, and finds it
convenient to introduce the operator S defined by
(3.45) S ( A , B ) = (l/d;)
al(y1‘2gi’A8, B ) = AV2B +&(A, B )
which is also the divergence of gifAa, B in (xl,x2)-space.
Following Helmholtz’ theorem, the isobaric velocity field can be repre-
sented with the aid of two functions i+h and x,namely
(3.46)l 17 = -ei’ *
a, +9” a,x
The reversed convention regarding the sign of x is also used.
so that the covariant components of V are
(3.47)
+ is the &earn function and x the velocity potential of the isobaric wind
v = ( V l , V2).
Recalling the definitions (3.34) and (3.39) of the isobaric vorticity 5 and
divergence 6 respectively, we have
(3.48) 5 = VZ*
(3.49) 6 = VZX
so that, according to (3.46), $,I expresses the nondivergent, rotational part
of the wind and x its divergent, irrotational part.
Operating with ejt a,(...) and (y1l2)-la l ( y 1 / 2 g J-..)
i on the equations of
the horizontal motion (3.33) we obtain a prognostic equation for 5 and 6
respectively, the so-called isobaric vorticity and divergence equations.
Formally, the unknown functions V , and V 2 are now replaced by 5 and x,
a substitution which proves useful in dynamic meteorology.
At this stage, it is essential to mention two attitudes in the mathematical
modeling of the atmosphere. One is based on the empirical fact that, in a
shallow atmosphere, the large-scale wind field ie nearly determined by the
geopotential field 4. Assuming that this condition strictly holds leads to
replacing the equation of the isobaric divergence of the wind by a diagnostic
equation, the so-called balance equation, between the geopotential field 4
and the stream functionJ! , I only. From a physical point of view, the balance
equation ensures a permanent adjustment between the 4- and the #-fields
and, consequently, radically suppresses the gravity inertial waves. Such
models are usually called balanced modele. It then becomes possible to use
numerical integration time steps of the order of lo4 sec which are sufficient
to adequately describe the evolution of the remaining Rossby waves (periods
of the order of lo5 sec).
The other attitude consists of maintaining the isobaric divergence equation.
But, for the atmosphere, the integration time step is then necessarily of the
order of lo2 sec, which seems extremely painstaking as compared with the
period of the Rossby waves. However, because gravity waves are supposed
to be important in some circumstances, in particular when the air flows over
mountains, this procedure is now widely used in forecasting models which
are then referred to as primitive equation models. Finally, it should be noted
that J. Van hacker (personal communication, 1971) recently developed a
new type of atmospheric model in which the gravity inertial waves are
Btered out without imposing a strict condition of balance between the 4- and
122 A. QUINET
+-fields. The main advantage of Van Isacker’s procedure is the avoidance of

the often difIicult problem of the resolution of a nonlinear balance equation,
though allowing integration time steps of the same order or magnitude as in
balanced models.
For our purpose, a balanced model seems quite appropriate since vacillation
appears as a pure Rossby type flow. As the balance equation results from
various simplifications introduced in the original equation of the divergence
of the wind (see, for example, Van Mieghem, 1973), some restrictions are also
imposed on the other equations of the model in order to maintain energetic
consistency. Moreover, different types of balanced models can be formulated,
corresponding to different balance equations and energetically consistent
models (Lorenz, 1960; Van Isacker, 1963; Phillips, 1965). The simplest form
of the balance equation shall be used here, namely the quasi-geostrophic
balance equation
(3.50) va4= S(f,$4

the differential operator S having been defined in (3.45). This choice is com-
patible with the quasi-geostrophic nature of Rossby flows, as discussed in
Section 2.2, and offers the substantial advantage of providing a balance
equation easily tractable from the mathematical point of view.
Since the equation of the isobaric divergence of the wind is replaced by a
diagnostic balance equation, so that a, 6 = 0, the kinetic energy budget has
to deal with the kinetic energy of the nondivergent flow only. This restriction
is analogous to the exclusion of the kinetic energy of the vertical motion,
when the equation of the vertical wind is replaced by the diagnostic hydro-
static equation. With this new definition of the kinetic energy, the total
energy of a nondissipative insulated quasi-geostrophic model will be con-
served provided the relative vorticity equation assumes the form (Van
Isacker, 1963)
(3.51) (a/w= - 4 h T ) -X(f, x)

where the absolute vorticity r] is given by
(3.62) 7=5+f
Writing
(3.53)
the continuity equation (3.38) transforms into
(3.54) 0 = v=x
when the upper boundary condition

(3.55) w=O for p=O
is introduced.
Taking (3.46) into account, the thermodynamic equation (3.40) may be
given the form
(3.56) e) -s(e,x) - ap(evzx)+ ( w , ) ( p o / ~ ) ~
a w t = -w,
where 8 is the potential temperature
(3.57) 8 = T(PO/P)k= -(l/wPo/P)K a w 1nP
with p o = 100 cbar and K = R/c, . Using (3.57), the balance equation (3.50)
transforms into
(3.58) cppo V28= - ( a / a p K ) w ,$4
so that the system of the two prognostic equations (3.51) and (3.56) and the
diagnostic balance equation (3.58)forms a closed differential set in the three
unknown functions t,h, x(or X ) , and 8 . The diagnostic equations (3.57),
(3.411, and (3.54) then allow one to compute 4, p, and w , and thus complete
the description of the flow.
3.3.2. The Equations of the Two-Layer Model. For the numerical integration
of the system (3.511, (3.56), (3.581,the atmosphere is subdivided into isobaric
layers. Obviously, it is necessary that this partitioning not violate the law
of conservation of total energy. Here the procedure followed by Lorenz
(1960) in the case of a two-layer model is recalled. Its vertical structure is
schematically represented in Fig. 6. The three isobaric surfaces of the model
are denoted by p o , pz , and y4 ; p o refers to the surface pressure and will be
put equal to 100 cbar, p z = y0/2, while a t the uppermost level p s = 0. Each
x,
of the functions lfi, and 8 is replaced by two functions Ifig; xl, x3; I3 el,
P4=0
p = Po 0.0.9
2 2
FIG.6. Vertical cross section of the two-layer model. Functions 8 < , xl, and #, are
x,
defined at odd levels p , with i = 1, 3; their expression in terms of the variables 8, u,
#, and T are also reported.
124 A. QUINET
defined a t the isobaric levels p 1 = 3p0/4 and p z =p0/4 respectively and

depending on the two space coordinates x1 and x 2 only (not on p ) . For
convenience, the subscript of any variable will refer to the isobaric level
where this variable is defined. Note that all the dependent variables are now
scalars SO that no confusion is possible concerning the meaning of this
subscript with respect to those used in tensor calculus.
With the usual boundary condition for a flat earth
(3.59) w=O for p = p o
and the upper limit condition (3.55),we have, by virtue of (3.54),
(3.60) V2X=0 a t p = p o and p = p 4 = 0
By virtue of the continuity equation (3.53),the function X , defined a t the
surface p , separating the layers is given by
v2x, - VZX, = v2
iopoiz apt
x 3 ( x 1 , x2)
so that taking (3.60) into account,

(3.61) VZX, = -&p0V2x3
I n the same way,

(3.62) v2x2- v2xo= v2xz= 8 p 0 V 2 x 1
Consequently, the continuity of w in p 2 implies
(3.63) x 3 x1 - - x 3 -
The finite difference form of (3.51) and (3.56) is obtained by replacing the
derivatives with respect t o p by finite differences, thus
(3.64) (W)V2$,= -4#, 9 rl,) -&f, Xj)
and
(3.65) ( W e , = -J($j, 0,) - - X ( O j , xj)
-
e,-1v2xj-1- e j + 1 v 2 x , + 1
Pj- 1 -PI + 1
for j = 1 and 3, H , being the heating rate in the layer j, while some rule is
CP
+- (A
H , Po IC
necessary to define e2 in terms of the two thermal variables el and e3. I n

the same way the finite difference form of (3.58) can be written
(3.66) c,p0"V2~,= 4p1" - p 3 R ) - 1 m $1 -$3)
which is the balance equation for the two-layer model.

A NUMERICAL S T r n Y O F VACILLATION 125
3.3.3. The Energetics of the Two-Layer Model. The aim of this sect'ion is
t>oensure the energet,ic consistcncy of the formulation of the t,wo-layermodel.
Hence, it will deal with t'he conservative, insulated model only.
In an atmosphere in strict hydrost'atic equilibrium, the gravitation potential
energy P and the internal energy 1 are proportional (Margules, 1903).
Therefore, in large-scale atmospheric energet,ics, there is no reason to deal
separately wit,li these two quantities which are always associated in the
definition of the total potential energy
(3.67)
where dm is the mass element of the fluid and dC an element of horizontal

area, the integration being extended t o the system under consideration.
In the two-layer model, (3.67) is approximated by
According t o (3.4), the kinetic energy of the nondivergent horizontal

motion is given by
(3.69)
where the subscript I,!I indicates that only the rotational part of the wind is
taken into account. Using (3.46),we have
(3.70)
or, with (3.37) and (3.44),
(3.71)
I n the two-layer model (3.71) asclumcs the form
Before expressing the energy budgets, it is perhaps worth recalling that

when the system is closed (no net exchange of mass with the surroundings)
and mechanically and thermally insulated (no net work and heat flux
through the boundary) the horizontal average of the (two-dimensional)
divergence of any vector of the flow field vanishes. The property is easily
obtained in mathematical form from Stokes' theorem. Physically it expresses
that the net flux of any flow quantity across the lateral boundary is zero.
On t,he other hand, the horizontal average of a Jacobian is easily seen t o be
126 A. QUINET
zero too. Indeed, the Jacobian always involves a stream function + and,
recalling (3.46), assumes the form
(3.73) JW 8 = vlki a, t
where f is a flow variable. Such a term expresses the transport of by the
nondivergent wind and simply redistributes 5 on the given isobaric surface.
Obviously such a process cannot contribute to modify the average value
o f t on the surface.
According t o (3.67) and (3.65),the budget of P + I for an insulated model
then assumes the form
(3.74) ( a / a t ) (+
~ I ) = cpg-lp;yplx -p3x) J x2we, d x
I n the same way, by virtue of (3.72) and (3.64),
(3.75)
(3.76)
Thus, from (3.74) and (3.76), the balance equation (3.66) ensures the
conservation of the total energy of the insulated nondissipative two-layer
model. Note that this condition leaves O2 arbitrary.
Of considerable interest in atmospheric energetics also is the concept of
available potential energy, originally introduced by Margules (1903) and
which represents the (small) portion of the total potential energy capable of
being transformed into kinetic energy. Conventionally, the available potential
energy is defined as the excess of the total potential energy above its value
obtained by an adiabatic rearrangement of the mass leading t o a stably
stratified atmosphere with horizontal isobaric (or isentropic) surfaces. Van
Mieghem (1957) has shown that this reference state is the one of minimum
total potential energy. An approximate form of the available potential
energy, presenting the main advantage of being expressible only in terms of
the current state of the fluid, has been derived by Lorenz (1955).
I n the two-layer model considered, the potential temperature is defined at
a few levels only, so that a continuous redistribution of the mass cannot be
envisaged. Nevertheless, it is still possible to define a state of minimum
potential energy.
Introducing, with Lorenz (1960), the following new variables 8 and a:
(3.77) e = *(el + e3), = +(e3 - 8,)
so that 6 is the vertically averaged potential temperature and u a measure
of the static stability, the total potential energy (3.68) can be written
(3.78) P +I =c p g - l p , ,
s (ae - bu)
+
with a = [(3/4)" (1/4)K]/2 and b = [(3/4)"- (1/4)n]/2. Now, Eq. (3.65)
with H , = 0 implies the conservation of the mean square potential tempera-
ture
(3.79) +(e12 + e32) =82 + u2
where a bar denotes a horizontal average, provided (Lorenz, 1960)
(3.80) 02 = +(4 + 03) = f3
Recalling that the mean square potential temperature O2 u2 is conserved +

and that 0 is also an invariant for adiabatic flow, the quantity
- -
(3.81) 62 +1,2 +el2 =urn~
+ +
with 8 = 0 8' and u = 0 a’, is also conserved. Accordingly, 15 has a
maximum value urn when u' = 8' = 0, i.e. when the isobaric surfaces (where
the temperature field is defined) are also isentropic. Clearly, in this case, the
total potential energy (3.78) reaches its minimum value attainable by adia-
batic process. The available potential energy in the two-layer model is then
given by
A =cp 1
g- ip0 [(ae - au) - ( a -~bum)] a = c, g-lpo b - u) a
(3.82)
or, taking (3.81) into account,
(3.83) A = ~ , 9 - ~ p ~ b S ( 8 ' 2 + ~ ) / ( udZ~ - - )
Obviously, from (3.82) and (3.81), the budget of A assumes the same form
as the budget of P + I so that the adiabatic two-layer model conserves also
the sum of the (so defined) available potential energy and of the kinetic
energy. Note that this condition imposes the definition of B2 according to
(3.80).
For the sake of completeness, let us also mention that the energetic con-
sistency of less simplified models allows one t o define (Lorenz, 1960) #2 as
follows :
(3.84) *2 = +($I+ $3)
It is then convenient to introduce, together with (3.77) defining f3 and u,

the new variables
128 A. QUINET
so that is the stream function a t the mean level p2 and T the stream function
of the vertical wind shear.
3.3.4. The Basic Physical Processes of the Model. With the new variables
$, T , 8, and u, the balance equation (3.66) assumes the form
(3.86) bc, V28 = X( f , T)
When the vorticity equations (3.64) are added for j= 1 and 3, we obtain,
taking (3.63) into account,
(3.87) (a/at)v2*= -*{J($19 71)+J(h7311 9
(3.88) = -J(i,b, v2$ f f ) -J(T, v2T)
Recalling (3.73),the local rate of change, (3.87),of the vertically averaged

relative vorticity V2$ is given by the mean absolute vorticity advected
by the nondivergent wind in the two layers. Thus, Eq. (3.87) governs the local
barotropic change of relative vorticity. Considering (3.88),a , V y results from
+
an advection of mean absolute vorticity V2# f by the nondivergent mean
flow and an advection of shear vorticity V27 by the shear flow. Obviously, the
horizontal mean of V2+ is left unaltered by these processes.
On the other hand, subtracting (3.64) for j = 1 from (3.64) for j= 3, the
vertical shear vorticity equation assumes the forms
(3.89) (a/wv2T= i[J(*l,
71)- 4 * 3 Y 73)l +Kf,x)
(3.90) = -[J(T, v2* + f ) +J(*, V27)1+4f7 x)
The two terms in square brackets in (3.89) represent the difference between
the advection of absolute vorticity in the upper ($3) and the lower (lCI1)
layers. The third term in (3.89) represents the generation rate
(3.91) S(f9x) =f v2x+ &(f>x )
of shear vorticity by the divergent wind (x).
By virtue of the balance equation (3.86), the vertical wind shear ( T ) is
related t o the isobaric temperature gradients a t the mean level p 2 . For
instance, in the particularly simple case of constant f, the balance equation
(3.86) leads to the thermal wind equation
(3.92) ~ V S i= -(bc,/f)e" a,6
-etj 8 , =
where V S idenotes the vertical shear wind. Thus Eq. (3.89) expresses the
local baroclinic change of relative vorticity. From (3.90) the differential
advection term (the square bracket term) consists of an advection of mean
absolute vorticity by the shear wind and a n advection of shear vorticity by
the mean wind. These processes, as well as those appearing in (3.88),simply
transport vorticity from place to place, and do not alter the horizontally
averaged values of V z $ and 8%.The actual source of vorticity in the model
is represented by the S term in (3.90),involving the divergent wind potential
X. This term is associated with the (baroclinic) conversion processes between
potential and kinetic energy [see Eq. (3.75)]. Thus, the change of shear
vorticity is made up of two contributions: (i) an initial one, represented by the
square bracket of (3.89), which locally modifies, but does not change in the
mean, the shear vorticity of the flow; and (ii) a source term, the S term in
(3.89),which represents the iilfluence of the conversions between kinetic and
potential energy and which alters, both locally and in the mean, the shear
vorticity of the flow.
With the new variables (3.77),the adiabatic form of the thermal equations
(3.65) becomes
(3.93) 0 ) - J ( 7 , u) +S(u,
aejat == A(*, x)
and
(3.94) aa/at= --J(T, 0) - J ( $ , 0) +Q(e, X )
I n both Eqs. (3 93) and (3.94)the first two terms on the right side represent
the local rate of temperature change due to heat advection by the non-
divergent wind. These terms are similar to the vorticity advection terms in
+
(3.88) and (3.90), with 0 substituted by V2$ f and u by V27. The third
terms S and Q express through the continuity equation (3.54), the adiabatic
temperature changes resulting from vertical motions.
3.3.5. Friction and Heating Ejfects.
3.3.5.1. Frictioii , It has been rccogiiizcd (Lorenz, 1962) that friction plays
an essential role in the dynamics of models intended to simulate laboratory
flows. I n order to take frictional cffccts into account, Lorenz introduces a
drag a t the surface p o proportional to the wind in the lower layer and also a
drag a t the surface p z proportional to the difference between the winds in the
upper and lower layers. If thc coefficients of proportionality, i.e. the co-
efficients of friction, are 2F and F‘ at p o and p z respectively, the additional
terms on the right side of (3.88) and (3.90)assume, respectively, the forms
(3.95) - FV2$ + FV’T
and
(3.96) FV2$ - ( F +2F’)V2-r
3.3.5.2. Heating. The parameterization of the heating will also be per-
formed following Lorenz’ procedure (1962). The heat exchange between the
130 A. QUINET
underlying surface and the lower layer is assumed proportional to the differ-
ence between the preassigned temperature of the surface O* and the temper-
ature of the lower layer. Similarly, the heat exchange between the lower and
the upper layers is taken proportional to the difference between the temper-
ature of these layers. If the heating coefficients are denoted by 2H and H'
for the lower and upper layer respectively, the additional terms on the right
side of (3.93) and (3.94) are, respectively
N 1.085, a2 = (3"
a1 = (4/3)" + 3-")/2 N 1.29,
and
= (3" - 3-")/2 N 0.81
3.3.5.3. Evaluation of the friction and heating coeflicients in the atmosphere.

It may be of interest to evaluate the order of magnitude of F and H in the
atmosphere. The atmospheric coefficient of friction F* is usually introduced
by the following expression
(3.99) F*g(a/aP)(PI v I V)
for the frictional drag. There is a large dispersion of the values of F* depend-
ing upon whether it is directly measured or computed by an indirect way or
adjusted in order to improve the results in atmospheric models. Obviously,
there is also a large divergence within each set of estimations based on a given
method. Roughly speaking, the values of F* reported in various papers
range from to (Stessel, 1969). For comparison, with the formulation
adopted here, we write
(3.100) 2F I Vl I = F*9f 2 I v2 I "POP
where the subscripts 1 and 2 refer to levels p 1 and p,, respectively. Mid-
latitude typical values (Statistiques Quinquennales, 1971) for the wind speed
and the air density
I V, I N 20 m sec-l, I V, I 2: 30 m sec - ', pz N 0.7 kg m-3, g N-10 m S ~ C - ~
lead to
(3.101) 3x < F 23 x sec-l
The coefficient of heating is considered to be of the order of magnitude

(see, for instance, Doos, 1969)
(3.102) Hlc, N sec-l
Consequently, in order to reduce the number of parameters in the model,

the approximation
(3.103) F = H/G,
does not seem unreasonable. Furthermore, we shall assume, with Lorenz

(1962), that
(3.104) F = 2F’ = H/c, = 2H‘/c,
which corresponds to a vertical decrease of friction and heating as ( p / ~ ~ ) ~ .

Hence, the friction and heating effects are represented with the aid of a
single coefficient F . As a consequence of the commodity hypothesis (3.103)
and (3.104)it seems unnecessary, and even somewhat misleading, to maintain
the more or less sophisticated refinement introduced in (3.97) and (3.98)
and therefore just the numerical values
(3.105) a, = 1, a2 = a3 = 0
will be used that still warrant that the direction of the heat transfers be
correctly represented.
4. THESPECTRAL
DYNAMICS
AND ENERGETICS
OF THE MODEL
4.1. The Spectral Representation

The spectral form of the equations of the model is obtained by repre-
senting each dependent variable by a series using a complete set of ortho-
normal functions F , . The particular choice of these functions depends on
the geometry of the system. We consider here an infinite channel consisting
of periodically repeating “ units ” of width 7rL and of fundamental length
2rL. Obviously it is then sufficient to study the dynamics and energetics of
a single “unit.” If x1 = x and x 2 = y are rectangular coordinates respectively
oriented along (from W to E) and across (from S t o N) the channel, the
fundamental plane domain is defined by
(4.1) 0 I x l L <2n, 0 IyIL < 7r
and the components of the metric tensor are simply
(4.2) 911 = Yzz = 1, 912 = g21= 0

132 A. QUINET
so that
(4.3) y=l
and
(4.4) g l l = g22 = 1, g’2 == g 2 1 z0
A suitable enumerable set of functions F , orthogonal on the domain (4.1)is
vo.o=1
vm, = 4%
cos(rny/L)
(4.5)
vm,, = 2 sin(rny/L)cos(nx/L)
q$,,, ,,= 2 sin(rny/L)sin(nx/L)
m and n being integers (rn, n = 1, 2 , . . .). These functions are such that
jOznLd x IOnLdyF, F j = 2 2 L 2 8$
where 8,’ is the Kronecker symbol (3.11).Thus each scale of the fields of
motion is characterized by a pair of integers (rn, n), where rn refers to the
meridional wave shape or mode and n to the zonal wavenumber. The de-
composition of the fields into their spectral components not only allows a
clear and easy identification of each scale of motion but also provides the
possibility of further simplifying the model equations by discarding, in the
development of the dependent variables, all but a small number of scales.
Vacillation has been identified as consisting of a cyclic modification of a
pattern including a single longitudinal wave n. Consequently, we shall use a
set of functions F, including only one single zonal wavenumber n (n # 0).
Moreover, it was shown that the wave motion cycle is accompanied by a
concomitant variation of the zonal flow. Consequently vacillation is definitely
a nonlinear phenomenon where the interaction of the wave with the zonal
flow plays an essential role. The minimum set of basic functions t o be used
in a channel for describing this kind of interaction consists of the five func-
tions y z ,o , n , 91;. ,,, v2,
n, and I&,,, with n # 0. Obviously we also need
the function q ~ associated
~ , ~ with the horizontal mean of the dependent
variables ($, 7,8, . . .). On the other hand, the function vl, is indispensable
to model the laboratory thermal forcing condition. Thus, our ordered mini-
mum set of basic functions F , is
(4.6) {Yo =yo. 0 7 F1= v1,0, F2 = vz.0 >

F , ~1~ n , F , 3 ~ ) F.
n ,; ,~ v zn, ,P, &, n}
Consequently, the dependent variables involve (at most) seven spectral
components: (i) a component (0, 0) equal to the horizontal average of the
A NUMERICAL YTVDY OF VACILLATION 133
dependent variable considered; (ii) two zonal components, (1, 0) and (2, 0),
of the first and second mode respectively; and (iii) a wave component n of
the first two modes (1, n) and ( 2 , n ) . Then the #, T, and 8 fields assume the
forms
(4.7)
2
#(x,Y, t ) = L2f C {#m, o( t) vm. O(Y)
m= 1
+ $m, n ( t ) v m , n(x, Y) + #h, n ( t ) d n . n(5, Y))
(4-8)
(4.7')
i= 1
The spectral Coefficients $ 1 , T ,, 8 , , i = 1, 2, . . . , 6, and B0 are functions of

time t only and become the new variables. The factors L2f and BLZf with
B =f /6 . c, where b = [(3/4)"- (1/4)"]/2 allow one to deal with dimension-
less spectral coefficients by using L and f as units of length and time
respectively.
It is possible to restrict the static stability u so that it varies only with
time without breaking down the recognized fundament)al structure necessary
to simulate vacillating flows. Accordingly, and for the sake of simplicity, u
is specified only by a horizontal nondimensional mean component a0 :
(4.10) a(t)= BL2j'uo(t)
Obviously, the equation of the statir stability (3.94)must remain compatible

with the hypothesis (4.10).Therefore (3.94) should be averaged horizontally,
~ - -
(4.11') at?/at = aU/at = -J(T, 8 ) - J ( # , U) + &(8, X) +
134 A. QUINET
a horizontal bar denoting a horizontal average over the domain (4.1). I n

(4.11’),S2 is given by (3.98)with the values (3.105)for a,, a2,and u 3 .Recall-
ing that the system is closed and mechanically insulated, the horizontal
average of the Jacobians and of the divergence V(6Vx) coming from the Q
term vanish in (4.11’) which reduces to
-
(4.11) aulat = -evzX + S,
Then the velocity potential x appears in the equation of the model under its
Laplacian form only, so that it suffices t o use
where, again, the spectral coefficients u,(t)are dimensionless.

I n order t o simulate both the laboratory conditions and the major com-
ponent of the earth’s thermal forcing on the atmosphere, we use for e*:
(4.13) 6*(Z, Y) = BL”f@, 0 To. 0 6 2 0 Tl,o(Y)l+
We consider this forcing t o be steady so that f3Zu and 0:. are simply con-
stants.
The amplitude A,,,, and the phase pm, of the wave (m, n) of mode m
and wavenumber n of any field C will be associated with the spectral coefici-
ents Cm, and (26, in the following way:
(4.14) sin(my/L)[C,, .(t)cos(nx/L) +C6, n( t)sin(nz/l)]
= sin(my/L)A,, n(t)cos[(nZ/L) - P m . n(t)l
so that
(4.15) Am, n = 1 (C;, n +C6? n)1’2)
where the vertical bars denote the absolute value and
(4.16) p m , n =tan-’(C6, nlcm, n)
Consequently, pm, varies counterclockwise (dp,,,, ./dt >0 ) in the (A, p )

plane when the wave progresses eastward (dxldt > O ) . It is then clear that
the sign of the expression
(4.17) A,, n- 4 2 , n sin(pz. n - pi,n) 1 GI,n GIn -G, n (72. n
or, using (4.6),

(4.17‘) Al.nA2, nsin(p2.n--l,n)=C3C,-CC,C5
allows one to determine the relative position of the two modes of the wave n
in terms of their spectral components C,, n , Ci, n , CZs, and Ca, n . Indeed,
for eastward moving waves, the wave (2, n ) lags behind the wave (1, n ) if
pa. - pl,, > rr so that sin(p2, - p l , ) < 0 or, equivalently, if Cl, ,,Cb. ,
- Ci, ,,C2, is negative.
4.2. The Spectral Equations

With a constant Coriolis parameter f and the static stability restricted
according to (4.10),there remain only two differential operators, namely V a
and J , in the equations of the model. The basic functions F , [Eq. (4.5)]
being the eigenfunctions of the two-dimensional Laplacian in Cartesian
coordinates, we simply have
(4.18) LaV2Fi= -atapt
where the eigenvalue -at2 stands for - ( m a + n 2 ) when F , is either q m a n
or c$,,, n . The Jacobian of two basic functions
J ( F j , Fk) = a, Fj a, F k - a, F k a, F , = - J ( F k , Fj)
is represented by a limited series using the basic functions (4.6) only,
6
(4.19) LzJ(F j , F k )= C c i j kP i
1=0
where, owing t o the orthogonality of the Fi’s on the domain (4.1),
The interaction coefficients cij k are completely antisymmetric:

(4.21) cjjk = cjk, = ckfj = -cjik = - C i k j = -c k j i
For the basic set (4.6) the only nonzero coefficients c t f kcan be deduced from
-
(4.22) CI3+/5= c256/4 = C,,,/8 = ~ 2 , , / 8= -(81/2/15~) ’ ‘TI.
Figure 7 schematically describes the associated scale interactions allowed in

the model. The upper connections represent the self-interaction of each
i 2.0 i ii , n I
FIG. 7. Schematic representation of the scale interactions included in the model.

The upper connections represent the self-interaction of wave (1, n) or (2, n) with the
zonal component (1, 0); the lower connections represent the triple interaction between
scales (2, 0 ) , (1, n), and (2, n). In each parenthesis the first integer refers to the mode
and, the second one to the zonal wavenumber.
136 A. QUINET
component ( l , n) or (2, n) with the zonal component of the first mode (1,O).
The lower connections represent the interactions between the three scales
(2, O ) , (1, n),and (2, n).
When the expressions (4.7)-(4.10)are introduced into the equations for
$, T , 8, and u, we obtain (Lorenz, 1963) the spectral form of the equations of
the model
6
(4.24) i i =
$<k=l
a;2(a,Z- a k 2 ) C t j k ( T 1 #k + 7k$j) --aC2wr + k$l - 2kr1
+
6
(4.25) 8%=
f<k=l
c z j k ( d j #k -8k # j ) + OO@t - k(ei - ui) k@t*
6
(4.26) d.0 =- C O t w , -2kuo + k(80 - B0*)
l= 1
where a dot denotes a time derivative with respect to the nondimensional

time to=ft and where k =Ff -I. Taking into account (4.8) and (4.9) the
balance equation (3.86) reduces simply to
(4.27) O * = T ~ if i # O
The system (4.23)-(4.27) involves the 26 unknowns #, , 71,Oi, w , , a = 1, 2,
,. . , 6 , do, and u0. The balance equation (4.27) allows one to eliminate six
unknowns 8, or 7, (i = 1, 2, . . . , 6). Moreover, using (4.27) it is still possible

to eliminate the six unknowns w t , i = 1, 2, . . . , 6 from (4.24) and (4.26),so
that there finally remains a set of 14 prognostic equations in T , , (or 0,)
i = 1, 2, . . . , 6, B o , and u,, . If
(4.28)
u = - 8 d 2 n / ( 1 5 ~ ) ,!? = n2/(n2 1) +
,!?I = (n2+ 3)/(n2+ 4), B” = (n2 - 3)/(n2 + 4)
these equations can be written
(4.41) $0 =
6
(4.42) 6,= - 1 eiWi
i=l
where the following notations are used:

(4.43)
(4.44)
138 A. QUINET
(4.50) w 5 =[ i + (4 +n % , ] - l ( - 4 t ~ [ 7 ~-$ ~(2.n' +7)71$61

- gar774 - (2n2+ 1b2 + ( 4 + m2)k($, -
$2 $41 75))
(4.51) w6 = [I + ( 4 + n 2 ) ) o o l - 1 { 4 a [t,hl~ 5- (2nZ+ 7b1yU

$.
8a[773 $2 - (2n2+ ll7Z
$31 + +
nz)k($6- T6)1
Note that according to (4.29)the waves ( 1 , n ) and ( 2 , n ) do not interact with

the mean zonal flow of the first mode as a consequence of the vanishing
coefficients a12-aka in this equation. Note also (Lorenz, 1963) that the
equations of the model are invariant for a change of sign of the variables of
the second mode (subscripts 2, 5, and 6).
Before dealing with the energetics of the model and in order to avoid any
ambiguity, some dynamic4 properties of the field patterns will be presented
in the next section.
4.3. Some Dynamical Properties of the Field of Motion
4.3.1. The Tilt of Wave n. Let us consider a field G represented by

6
(4.52) C(S, y , t ) = 1SC,F ,
i=o
where S is a scale factor making the Fourier coefficients ct dimensionless.

The equation W / a x = 0 of the trough and ridge lines of this field assumes the
form
(4.53)
nx c4
t g -=
+ 2c6 cos y/L
L c3 + 2c5 cos y / L
The slope with respect to the W-E direction of the trough and ridge lines is
then given by
Consequently, the change with time of the ct coeEcients generally not only
gives rise to the progression of the wave but also to a modification of its
shape. Note that the sign of the slope of the wave pattern does not change
over the interval 0 < y 5 XL and is determined, according to (4.54) and
(4.17'),by the relative position of modes 1 and 2 of the wave n. For instance,
if the wave (2, n ) lags behind the wave ( 1 , n ) , then c4c5 - c3 C g > 0 so that
6y/6x > 0 and the flow pattern exhibits a SW-NE tilt.
4.3.2. The 8-N Transport of W-E Momentum. Let us denote a zonal

average by square brackets and the deviation along a latitude circle with
respect to the zonal average by primes. I n particular, we have for the zonal u
and meridional v wind components, respectively,
(4.55) v1= u = [u] + u‘, v2 = 2, = [v] + 0’
where obviously
(4.56) [u‘] = [v’] = 0
The mean S-N transport rate of zonal momentum (W-E momentum)
across a latitude circle in a layer of pressure depth of one unit is then given by
(4.57) [vul = [vl[ul+ [V’U’I
The first term [v][u] on the right-hand side of (4.57) is associated with the
existence of a mean meridional circulation [w] a t the considered latitude. The
eddy momentum transport [v‘u‘] is due to the covariance between the
fluctuations of v and u along a latitude circle and results from the eddies.
Recalling the definition (3.46) of V and taking (4.2)-(4.4) into account,
the nondivergent wind a t the mean level p 2 is simply given by
6
(4.58) u$= -a,+= -L2fc+f ayp,
1=1
and
(4.59) v* = a,*= L2f c
6
1=1
*f
With the basic functions (4.6)we have
(4.60)
and
(4.61) [%I = 0
the zonal average of terms linear in sin(nx/L) or cos(nx/L) being equal to
zero. This yields for the fluctuations:
(4.62)
(4.63) V = L2f c $+a, F ,

6
l =l
Taking (4.58)-(4.63) into account, we obtain for the mean S-N transport
rate of zonal momentum, (4.57),across a latitude circle by the nondivergent
wind a t the mean level:
140 A. QUINET
Consequently, according to (4.17), the transport of W-E momentum is

northward if the ( 2 , n) wave of the +-field lags behind its (1, n) wave. As a
consequence of (4.54), this transport is accomplished when the wave tilts
from S-W toward N-E (6y/6x>O), a classical result of the theory of
momentum transport (Starr, 1948). From (4.64), [w u is ] symmetric with
respect to the central latitude, where it reaches its extremum value.
The convergence of the mean meridional momentum transport a t a given
latitude is given by
Let us consider the +field a t a particular time so that +:, i,b2#

-
$L2,
is fixed and supposed to be positive. From (4.64) i t can be seen that, in this
case, the transport is toward the north along the whole width of the channel
and maximum a t its center. The convergence of momentum being positive
t o the north and negative t o the south of the center of the channel [Eq. (4.65)],
the zonal wind tends to increase northward and t o be concentrated into a jet
a t the maximum convergence latitude, yo, easily obtained from (4.65)
(4.66) yo= L cos-1 ( - l / d 3 ) , T / 2 <yo <7r
When the time flows, I&, i,h2, - t&,, varies and, consequently, so does
the latitude of the jet. If 4;. fi2, - 4;. keeps its sign, the maximum
zonal wind remains in the northern region while it will travel across the
center of the channel when +;, 9h2, - t,h1. 4;. changes its sign.
Obviously, the x-dependence of the momentum transport, wu, related to
the presence of the disturbances, creates a longitudinally meandering jet.
Let us also note that, according to (4.64),the transport increases with n, i.e.
with decreasing wavelength.
4.3.3. The Northward Sensible Heat Transport. The northward sensible

heat transport per unit mass across a latitude circle a t the mean level is
proportional to the covariance
In a manner similar to that followed in the preceding subsection, we find

The correlations between the waves (1, ti) of tlhe $- and 0-fields and between
the waves (2, 71) of the same fields the [terms in sin2 yIL and sin2 2ylL in
(4.68)] determine a systematic nort.hward or southward heat transport at all
latitudes. According t o (4.68)and (4.17) the transport is northward when the
wave ( m , n ) , with m = 1 or 2 , of tjhe temperature field lags behind the same
wave of the stream field. I n cont,rast,,the heat transport associated with the
correlations between two different modes of the wave n of the 4- and &fields
[the term in sin y/L sin 2y/L in (4.68)] changes sign a t the central latitude
y = 7rL/2.
4.4. The Spectral Energetics of the Model
4.4.?. The Globul Forms. The tot(a1potentiaI energy (3.75) being a linear
function of 0 and u, we immediately obtain from (4.9) and (4.10)
(4.69) P + I = cb-'faOo - bu,)
where the scale factor c is given by
(4.70) c = 2x2 L4p0 9 - y

The kinetic energy (3.72)can be given the form
(4.71)
which can be partitioned into the zonal kinetic energy
(4.72)
and the eddy kinetic energy
We shall also consider each term of (4.71) as the spectral component i of R

The available potential energy (3.83) is given by
(4.74)
where, according to (3.81),
+c
6
(4.75) u& = (702 012
I=1
142 A. QUINET
Each term of (4.74) can be considered as the spectral component i of A

while the classical zonal A , and eddy A , available potential energy are
given by
a
(4.76) A, = [c/(uo + U0rn)l c 812
f=l
and
+
6
(4.77) = [ci(uo u0m)I C
t=3
@t2
When use is made of the equations of the model (4.29)-(4.42),the budget

of the energy forms (4.69), (4.72), (4.73), (4.76), and (4.77) can be written
(4.78) a ( p + ryat = -(cz +c,) +G
(4.79) aAz/at = -c, -cA-CAR +GZ+GZR
(4.80) aAE/at= -CE+CA+CAR +GE+%,
(4.81) aK,/at = c, -C K - Dz
(4.82) aKE/at= CE + CK - DE
The terms appearing on the right sides of (4.78)-(4.82)can be given the follow-
ing interpretation:
(4.83)
is the conversion rate of A , into K , .

6
(4.84) c,= -cfc eiwf
t=3
is the conversion rate of A , into K p .
(4.85‘)
is the transfer rate of A , into A , resulting from sensible heat advection.
(4.86’)
is the transfer rate of A , to A, due to the evolution of a. .
(4.87)
is the generation rate of A , by external heating.

8
(4.88) + C
GE = [ W k / ( u o aom)I i = 3 e t ( @ t * - 0,)
is the generation rate of A , by external heating.

(4.89)
is the generation rate of A, due to nonadiabatic effects on a. and the reference

state (uom),
(4.90)
is the generation rate of A , due to nonadiabatic effects on cro and the refer-
ence state (aom).
(4.91)
is the transfer rate of Kz into K ,
(4.92)
is the dissipation rate of K z by friction.

R
(4.93)
is the dissipation rate of KE by friction, and finally,

(4.94) G = cflcb-y(a + b)(e,* - e,) + (a + 2 ~ 4 ~ ~ )
is the generation rate of total potential energy by external heating.
144 A. QUINET
The energy structure of large-scale quasi-static flows as expressed by

(4.79)-(4.82)is well known in meteorology (Lorenz, 1955). One can simply
point out that the spectral equations allow one to assess easily the role of the
time changes of u,,and uOmon the conversion and generation rates of avail-
abIe potential energy, while this effect is generally overlooked in most studies
of atmospheric energetics. Basically there are two possible sources of eddy
kinetic energy. The first one consists of a baroclinic conversion C, of eddy
available potential energy into eddy kinetic energy. Note however that this
process necessarily implies a previous transfer C , from zonal into eddy
available potential energy or an external generation G, of eddy available
potential energy which establish along the latitude circles the temperature
fluctuations needed for the conversion of potential into kinetic energy (Van
Mieghem, 1952). The second possible source is a direct barotropic transfer
C, of binetic energy from the zonal flow toward the eddy flow.
4.4.2. The Individual F o r m . For convenience, we shall use throughout

this section the following notation 1
-
(4.95) E = -(8%‘%/37r)n; E‘ = G~~~ = -(322/2/15~)n;
-
E” = cZs6= c254 = -(642/2/15n)n
We shall consider here the contributions t o the energy budgets (4.83)-(4.94)
of each of the interactions involved in the model and which are schematically
shown in Fig. 7. In order to avoid any canfusion, let us nevertheless recall
that there is no self-interaction of the components (1, n)or (2, n) of the @-field
with its component ( 1 , O ) .
4.4.2.1. The barotropic transfers of kinetic energy. The single barotropic

transfers of kinetic energy allowed in the model concern the scales i = (2,0),
j= (1, n), and k = (2, n). According to (4.91) and (4.91’))they assume the
form
(4.96) (aK/at)?’ = Cf&“(a?- Uk2)T
(4.97) (aK/at)7T= c@’(ak2 - atz)T

(4.98) (aK/at);T = Cj& (Ui2- u,2)T
where the superscript BT recalls that we are considering only the barotropic
changes of kinetic energy. The coeficients
(4.99) u12=4, aj2=1+n2, ak2=4+n2
occur twice with opposite sign so that the total kinetic energy is conserved.
The expression for T is
(4.100) =$Z,O($l, n$k. n - $;, n$z, n) + $2.0(71.n7a, n -Ti, n 7 ~ n).
+ 72. 0(71. n K . n -7;. n$z. n ) +72.0($1, n.6. n -$;, n 7 2 , n)
For given values of c, E", a t , a;, and ak2,the sign of T governs the direction
of the barotropic kinetic energy transfers between the components i, j,and
k of the kinetic energy as in a purely barotropic flow (Fjortoft, 1953). Taking
into account the values of a,2, uT2,and uk2 from (4.99) and recalling the
negative value (4.95) of E", the kinetic energy is transferred from scale (1, n)
toward the scales ( 2 , O ) and (2, n) when T > 0. From (4.100), the sign of T is
determined by the zonal flow of the second mode ( $ z . o and T ~ , and ~ ) the
relative position of waves (1,n) and (2, n) of the $- and fields. For instance,
with t+hZ, > 0, the first group of terms in T contributes positively provided
the wave (1, n) of the $-field lags behind its wave (2, n ) with, according to
(4.17), a maximum contribution when the phase difference between these
waves reaches the value n / 2 . Obviously each group of terms in T may receive
a similar interpretation. Nevertheless, the sign of the whole expression T,
resulting from the contributions of the four groups of terms, cannot be a
priori stated in a simple way.
4.4.2.2. The baroclinic transfers of available potential energy. Let us first

consider (see Fig. 7) the baroclinic transfers between the zonal available
potential energy of the first mode A , . and the eddy available potential energy
of the first A l , or second A z , mode. According to (4.85) and (4.85'), these
are given by
or
where the asterisk indicates that only the contribution of the transfer between
A l , o and A l , n or A 2 , n to the value of is considered. Recalling that
E and E' have negative values, (4.95), available potential energy is transferred
from scale (1, 0) toward scale (1, n) [or (2, n ) ] if the wave (1, n) [or (2, n)]of
the temperature field lags behind the wave (1, n) [or (2, n ) ]of the stream field.
Here, as everywhere else in this discussion, it is assumed that f'l,o >0, i.e.
the zonal temperature field of the f i s t mode is such that the temperature
decreases northwards. It is well known from linearized perturbation theory
146 A. QUINET
(see, for instance, Thompson, 1961) that the growth of baroclinic disturb-
ances imposes such a configuration of the stream and temperature fields.
Thus, in a purely zonal flow, the transfer of available potential energy from
scale (1, 0) toward scale (1, n ) [or ( 2 , n)] appears as a necessary step in the
conversion of the initially exclusively zonal svailitb1e potential energy into
kinetic energy of the disturbance. According to (4.68) this transfer implies
meridional heat kransport ,
For the additional interactions between scales (2, 0 ) , (1, n), and (2, n ) we
get, from (4.85) and (4.85 ),
Obviously, these transfers conserve the total available potential energy and
each T ican easily be given an interpretation in terms of the flow pattern.
Let us emphasize the completely different structure of the transfers (4.103)-
(4.105)of available potential energy as compared with the barotropic trans-
fers (4.96)-(4.99)of kinetic energy between different scales of motion. I n the
present case, there are three separate eontributions T,,T, , and T3, each
of them appearing twice with an opposite sign. Consequently, and contrary
t o the barotropic transfers of kinetic energy, the same scale can simul-
taneously receive available potential energy from another scale and feed the
third one.
4.4.2.3. The baroelinic conversions hetween wadable pkentdal energy and

kinetic energy. The interactions schematically shown in Fig. 7 lead to
concomitant changes in different components of K and A during energy
conversions. As for the transfers between different spectral components of
available potential energy, so the set of conversions between available
potential energy and kinetic energy can be partitioned into two subsets. The
first one concerns the zonal component of the first mode ( 1 , O ) and one single
eddy component (1, n) or (2, n ) of K and A . Using (4.83) and (4.84) and
substituting for wi its values given by (4.46)-(4.51), we obtain for the

adiabatic frictionless baroclinic (BC) rate of change of kinetic energy the
following expressions:
(4.109) (aK/at)Yo= cfE(1 + uo)F1el, o(fl1. n #i. n - 6 ; . n # 1 , n)

-
-
- fa4w::o
(4.110) = - ~ f & [ l+ (1
(LJK/at)y,cn + n2))ao]-1(2n2+ 1)
x ’31, o(01, n 16;. n - K ,n 161. n)
= (8A/at)Y,cn
resulting from the interaction between scales (1,0) and (1, n) and
(4.111) (aK/at)Yo=cfE'(1 + u o ) - l f l i . o ( O z , n $ ; , n -&,n$z,n)
= -(aAjat):g
resulting from the interaction between scales (1, 0) and (2, n). The meaning
of Eqs. (4.109)-(4.112) should be well understood. Equations (4.109) and
(4.111) express conversion rates between A l , and Kl, o . However, each of
these processes does not operate individually but is necessarily coupled with
another conversion taking place respectively [Eqs. (4.110) and (4.112)]
between the components (1, n) and (2, n) of A and R.Let us also emphasize
that according to (4.109)-(4.112),an increase of kinetic energy a t one scale
implies a decrease of kinetic energy at the associated scale. I n fact the
following conservation law governs the baroclinic kinetic energy production
in this case:
(4.113) [2(m2 +n2)- 1][1 + (m2+na)uoI-' aE1, ,/at

+ ( I +cro)-'aE,,n/at=o
where E is either K or A and m = 1 [Eqs. (4.109), (4.llO)lor 2 [Eqs. (4.111),

4.112)]. Equation (4.113)is a particular case of a more general conservation
law governing the baroclinic energy conversions in the model (Quinet, 1973a).
Let us also mention that, according to (4.109)-(4.112) and (4.68), the growth
of baroclinic disturbances implies a poleward transportation of heat.
The second subset of associated energy conversions involves the zonal
components of the second mode i = (2, 0) and the two eddy components
j = (1, n)and k = (2, n) of A and K and results from the interactions between
148 A. QUINET
scales (2, 0), (1, n), and (2, n) (see Fig. 7). Following (4.83) and (4.84), these
conversion rates assume the form
(4.114) (aK/at)E:o = cf&'(l + 4u0)-'
x {(akz + a f z - a j Z ) T l + +ai2 -ak2)TZ}
- -(aA/at)yO
(4.115) (aK/t3t)$',cn
= @d'[l + (1 + n ' ) ~ o ] - '
X {-(ak2+Uj2-U,')T, $ ( a ; +a,2-Uk2)T,}
= -(aA/at)ya
(4.116) (aK/at);c,, = cf~''[l + (4 + n')uo]-l

x {-(ak2 + -u,')T~ - (akZ + alz - aj2)!Z'3}
- -(aA/at)!yn
where T,, T , , and T 3 have been defined in (4.106)-(4.108).I n this case, the
following conservation law governs the baroclinic production of kinetic
energy within the three nonlinearly associated components of K .
4.5. Comments 0% the Various Physical Processes

A by-product of Sections 4.3 and 4.4 is the provision of a common basis
t o the various possible interpretations of the time evolution of the model.
Different, physically equivalent points of view have been adopted, all of them
being classical. Thus, we can conceive the need for a poleward transport of
heat (4.67)as the driving mechanism of the flow, and it becomes apparent that
this transport can be accomplished by disturbances superimposed on the
zonal flow (4.68). Generally speaking, following (4.64), the eddies are then
able to accomplish also a meridional transport of zonal momentum which
favors [Eq. (4.65)] the generation of narrow jets. On the other hand, if the
point of view of the energetics is adopted, the transfers of zonal kinetic
energy toward eddy kinetic energy r(4.96)-(4.98)] or the conversions of
potential into kinetic energy [(4.109)-(4.112) and (4.114)-(4.116)] will
appear as the energy sources of the eddy motion. Clearly, this point of view
invites the concept of dynamic instability and the meridional momentum or
heat transports appear then as consequences of the instability of the zonal
flow. It should be kept in mind that these transports can progressively
A NUMERIVtlL STUDY O F VACILLATION 149
(re)build new instability condit'ions. For instance, t'he momentum conver-

gence associated with the growth of baroclinic disturbances could lead to
barotropically unstable zonal wind profiles.
All the energy processes can receive an int,crpretation in t'erms of flow
configuration through (4.17) and (4.54). The description of the link between
dynamic and energetic processes is simple as long as only one zonal scale and
one eddy scale are envisaged [(4.101), (4.102);(4.109)-(4.112)]. The energet,ic
interpretation of the effects of int'eractioiis bet,ween three scales of motion is
less simple except for the barot>ropickinetic energy transfers [(4.96)-(4.98)].
For the baroclinic case, the main interest in formulas (4,103)-(4.105), and
(4.114)-(4.116) is to provide processes relating only two scales of motion as
identified by the TI, T 2 ,and T,tderms.
5 . TIIE NUMERICAL STLrDY O F VACILLATION
5.1. The Nwmesicnl Procedure

The model presented in the previous sections has been used by Lorenz
(1863) to discuss some aspects of vacillating flows. In particular, Lorenz
has shown that vacillation arises when the instability conditions of the st'eady
Rossby regime prevail. We shall be concerned here with a detailed descrip-
tion of the properties of vacillat,ion, presenting first its dynamical character-
istics and then paying more at,t,ent,iont o itjs energetics.
I n t'he numerical model, the single longit>udinalwave of the first two modes
has been chosen, more or less arbitrarily, t o be a t wavenumber n = 3.
Accordingly the dependent variables of the model have, at most, the spectral
components (0, 0), (1, 0),(2,0),(1, 3), and ( 2 , 3). The numerical integration
of the equations has been performed by the Runge-Kutta method which
possesses sufficient numerical stabilit,y (Young, 1968) as to allow the identi-
fication of periodic flows within reasonable limits of confidence. It should be
recalled that Fultz' experiments are performed in such a way that, the fluid
is brought first in solid rotat,ion with t,he vessel; then the temperature con-
t,rast.is very slowly established. Since our purpose has never been to reproduce
the laboratory conditions of appearance of vacillation (obviously the geom-
etry of the model is not suited t o t,his), the t.herma1 forcing and the rotation
rate are both direct'ly applied wit'h tJheirfinal values. The equations are such
that if no wave is init,ially present,, no wave can develop in the syst,em [see
Eqs. (4.29)-(4.42)]. Consequent,ly, in order t o allow the generation of dis-
turbances, weak waves are initially superimposed on a weak zonal flow. This
numerical expedient simulates the random disturbances ine~it~ably present
in the laboratory fluid whcn the experiment is started.
150 A. QUWET
We have seen that the equations of the model are invariant for a change of
sign of the variables of the second mode, the variables of the first mode being
left unchanged. Consequently, when one solution is obtained, a second
solution always exists which is similar to the first, except for the sign of the
variables of the second mode. The initial conditions of integration decide
which of the solutions develops.
I n order to choose adequately the numerical experiments to be performed
for simulating the two kinds of vacillation delineated by Pfeffer and Chiang
(1967), it is useful to recall the specific characteristics of potential energy
and kinetic energy vacillations. It has been seen that the tilt of the waves
undergoes only minor oscillations in a potential energy vacillation (see
Section 2 ) . Accordingly, the barotropic process of momentum transport
(4.64) can be considered as being of secondary importance in this kind of
flow. However, a “ tilted trough ” vacillation necessarily implies considerable
fluctuations of the meridional transport of west to east momentum and
consequently a marked intervention of a nonsteady barotropic mechanism.
Now it can quite safely be stated that the dominant mechanism of generation
of the waves is of baroclinic character. On the other hand, it has been noted
(Quinet, 1973b) that the higher the rotation rate, the ,relatively greater
becomes the influence of the barotropic kinetic energy transfers. Thus, a t a
given thermal forcing, and assumiDg that with this thermal forcing both kinds
of vacillation are possible, the potential energy vacillation should precede the
kinetic energy vacillation when the rotation rate is gradually increased.
The origin ofthe temperature scale is chosen so that e,*, = 0 and the value
0:. , = 0.25 is used. With the wave n = 3 for the first and second modes, this
thermal forcing warrants the possibility of generating different kinds of
vacillating flows (Quinet, 1973b). Following the arguments of the above
discussion, we shall examine the kind of flow generated a t a ‘‘ low ” rotation
rate and determine if it compares reasonably well with a potential energy
vacillation. Then it will be seen that kinetic energy vacillation occurs at
higher rotation rates.
Let us recall that, according to the definition of k (k= Ff - I ) , the rotation
rate is proportional to k-l.
5.2. The Potential Energy Vacillation
5.2.1. The Dynamics of Potential Energy Vacillation. The values of o,

$z.o, #;,,, h3, $k3,el,o,ez.o, 0;. 3 , e;,3, and o0 com-
puted for 8 ~ , o = 0 . 2 5and k=0.2925 are listed in Table I for every five
nondimensional time units to =ft. The waves are given by their amplitude
and their phase difference with respect to the wave (1, 3) of the +field.
Basically, the flow is viewed in a coordinate system moving with the (1, 3)
I. Numerical integration of the model for k = 0.2925, el, = 0.25"
TABLE
L
0 0.2191 0.0147 0.0537 0.0221, 41" 0.2196 -0.0 0.0222, -52" 0.0095, 11" 0.0778
5 0.2199 0.0136 0.0526 0.0222, 38" 0.2203 -0.0006 0.0218, -53" 0.0095, 7" 0.0772
10 0.2204 0.0117 0.0530 0,0210, 34" 0.2206 -0.0013 0.0221, -53" 0.0091, 4O 0.0770
15 0.2204 0.0100 0.0549 0.0190. 32" 0 2203 -0 00x7 0 0229, -53" 0 0083, 3" 0.0773
20 0.2201 0.0087 0.0573 0.0170, 32" 0.2197 -0.0017 0.0238, -52" 0.0074, 4" 0.0779
25 0.2194 0.6983 0.0594 0.0153, 34" 0.2189 -0.0014 0.0246, -51" 0.0067, 6" 0.0787
30 0.2187 0.0086 0.0607 0.0145, 38" 0.2182 -0.0010 0.0250, -51" 0.0064, 9" 0.0793
35 0.2181 0.0096 0.0611 0.0148, 41 ' 0.2179 -0.0005 0.0251, -50" 0.0065, 12" 0.0796
40 0.2179 0.0111 0.0605 0.0160, 44" 0.2178 -0.0001 0.0248, -50" 0.0070, 15" 0.0796
45 0.2179 0.0128 0.0589 0.0179, 45" 0.2181 0.0002 0.0242, -51' 0.0078, 15' 0.0793
50 0.2183 0.0141 0.0568 0.0200, 45" 0.2187 0.0003 0.0233, -51" 0.0087, 14" 0.0788
55 0.2189 0.0145 0.0545 0.0216, 42" 0.2194 0.0 0.0225, -52" 0.0093, 11" 0.0780
56 0.2191 0.0145 0.0541 0.0218, 41" 0.2196 -0.0001 0.0224, - 52" 0.0094, 11" 0.0779
~ _____
a Variables are obsarved in a coordinate system moving with wave (1.3) of the #-field.
152 A. QUINET
-0~2E8-D~288-D~288-D.ZBB-0.Z18-0.2SE-D.ZEE-D.2E8-D~28S-D~288-D~2~S-D.28E,-D.2SO-D.288 -0.0 0.000 -0.000
-L1.256-J~271-J~293-0.298-U.203-0~Z61-0~256-0~Z71-0~293-0~29~-0~283-0.261-0.255 0.071 -0.000 -0.003
0.lb2 -0.002 -0.008

0.211 -0.005 -0.011
0.2b9 -0.008 -0.008

0-309 -0.009 -0.000
0.318 -0.008 0.008
0.289 -0.005 0.011
0.221 -0.002 0.008
0.120 -0.000 0.003

0.0 -0.0 0.0
1
-0.0 0.000 -0.000
0.01b -0.000 -0.002
0.150 -0.002 -0.006
0.219 -0.004 -0.009
0.215 -0.001 -@.DO6
0.311 -0.008 -0.000
0.315 -0.007 0.006
028b -0.006 0.009
D.2M -0.002 0.006
0.116 -0.000 0.002
1.0 -0.0 0.0
-~.298-0~298-0~298-0.298-0.29S-0~298-0~29S-0~29E-0~298-0~29E-0.298-0~298-0~298 -0.0 0.000 -0.000
- 0 . 2 9 7 - 0 ~ 2 7 ~ - 0 ~ Z ~ 5 - 0 . 2 1 3 - 0 . 2 ~ 0 ~ 2 9 7 - 0 ~ 2 7 6 - 0 ~ 2 b ~ - 0 ~ 2 7 3 - 0 . 2 ~ 0 . 20.081
9~ -0.000 -0.002
0.150 -0.001 -0.005
0.227 -0.003 -0.007
0.280 -0.005 -0.00’)
0.311 -0.ODb -0.000

0.311 -0-005 0.005
0.276 -0.003 0.007
0-207 -0-001 0.0W

0.111 -0.000 0.002
0.0 -0.0 0.0
3
-0.L91-0.297-0.291-0.297-0.297-0.297-0.297-0~297-0~297-0~291-0.297-0~297-0~291 -0.0 0.000 -o.mo
0.082 -OIDOO -0.002
0.159 -0.001 -0.005
0.228 -0.001 -0.001
0.280 -0.005 -0.005
0.309 -0-ODb -0.000
0.308 -0.DOS 0.005
02?2 -0.003 0.007
0.201 -0.001 0.005

0.109 -0.000 0.002
0.0 -0.0 0.0
4
FIG.8. See facing page for legend.
153
A NUMERICAL STUDY OF VACILLATION
-0.0 0.000 -0.000

0.079 -0.000 -0.002
0.154 -0.001 -0.006
0.223 -0.OOb -0.001
0.ZTb -0.006 -0.006
0.301 -0.001 -b.OOO
0.309 -0.006 0.006
0.215 -0.OOb 0.008
0.201 -0.001 0.006
0.Ill -0.000 0.002
I .o -0.0 0.0
-0.0 0.000 -0.000
0.01b -0.000 -0.002
0.146 -0.002 -0.008
0.214 -0.005 -0.010
0.211 -0.001 -0.008
0.301 -0.009 -0.000
0.31I -0.008 0.001
0.283 -0.005 0.010
0.215 -0.002 0.001
0.116 -0.000 0.002

0.0 -0.0 0.0
-0.0 0.000 -0.Doo

0.011 -0.000 -0.003
0.1b2 -0.002 -0.008
0.211 -0.005 -0.011
0.269 -0.001 -0.008
0.309 -0.010 -0.000
0.311 -0.001 a.oon

0.208 -0.005 0.011
0.220 -0.002 0.008
0.119 -0.000 0.003

0.0 -0.0 0.0
7
FIG.8. Synoptic representation of the $-field during the potential energy vacillation
observed for k = 0.2925, &, = 0.26. Spacing between isolines is 0.1 in L2f units. The
time interval between two successive maps is 9 nondimensional time units to =ft. The
zonal wind in Lf units, the zonally averaged northward W-E momentum transport in
L2faunits, and momentum transport convergence in LJa units are respectively printed
in the three columns at the right of the maps.
154 A. QUINET
wave of the $-field. The periodic character of the flow can be illustrated
by comparing the first and last row of Table I, so that the period T of the
phenomenon is 56 nondimensional time units. It should, however, be men-
tioned that in a fixed reference frame, there is a systematic phase shift on
each wave component of the $- and 8-fields between the beginning and the
end of the period. Accordingly, there is some ambiguity regarding the
definition of the period of the flow as this period depends on the coordinate
system chosen. When only the physical mechanisms of evolution are taken
into consideration, this distinction seems irrelevant. Hence, and in order to
avoid ambiguity, the vacillation period will be considered as given by the
period of a nonwavelike component, namely $2e .z
Table I shows weak fluctuations of the component (0.2179< 2
0.2204)and also rather limited fluctuations of the component (0.0083<
$2, <0.0147). Such a periodic flow where t,b2, oscillates without changing
sign has been called unsymmetric vacillation by Lorenz (1963).
The amplitudes of the waves exhibit cyclic variations, these being more
pronounced for the #-waves, for which they may reach 50 yoofthe amplitude
ofthe (2,3)component. The waves progress one wavelength in approximately
8to,and their relative positions with respect to each other undergo only minor
alterations during the cycle. The 8-field lags behind the $-field by 50 to 53 deg
for the first mode and by 28 to 31 deg for the second mode. As expected, this
configuration points t o a baroclinic origin of the eddies ((4.109)-(4.112)).The
phase difference between the two modes of the wave of the $-field varies
between 32 and 45 deg, the (2,3)wave preceding the (1,3)wave. Accordingly
[(4.54)],the wave pattern tilts SE-NW and there is a systematic southward
momentum transport (4.64) during the whole cycle. The amplitudes of the
wave (1, 3) of the $- and 8-fields reach their extreme values nearly simul-
taneously, as do the amplitudes ofthe wave (2,3) ofthe $- and 8-fields. As a
rule, the lowest values of the amplitude of the waves (1, 3) are attained at
nearly the same time as the maximum amplitude of waves (2, 3), and con-
versely. Consequently, the fluctuation of the amplitude of the global wave 3
+
[waves (1,3) (2,3)]of the $ and 8 fields is reduced to about its lowest value.
The synoptic representation of the stream field a t the mean level is given in
Fig. 8. The spacing of stream lines is 0.10 nondimensional units and the time
interval between two successive maps is 9 time units, the uppermost map
corresponding t o the first TOW of Table I. Except for an eastward translation,
the last map is a repetition of the first. The zonal wind, the zonally averaged
eddy momentum transport, and momentum transport convergence, expressed
respectively in Lf,L2f =,and Lf units, a t latitude y / L = 18'12, with 7t = 0,
It should also be noted that the small phase shift could be due to the numerical
integration scheme.
1, . . . , 10, corresponding to the different printed rows of the +-values, are

listed on the right side of the maps. Clearly the tilt of the trough and ridge
lines is practically constant during the cycle. The zonal wind U being given by
(5.1) u =( v N ~ % , +
s i n ( y / ~ ) 242+,, 0 sin(zy/l))
its maximum remains south of the center of the channel since $2, >0.
The maximum zonal wind oscillates in the narrow latitude belt y/L = 76" a t
t = 0 (or t = T)and y/L = 82" at t = T / 2 , corresponding respectively t o the
maximum (0.0147) and minimum (0.0084) values of o . The small vari-
ations of the zonal wind profile are represented in Fig. 9 where the figures on
c
a
-I 54
U
L
0.1 0.2 0.3
FIG.9. Meridional profile of the zonal wind Corresponding to maps 1 and 4 of Fig. 8.
Latitudes are given in degrees according to (4.1).
the curves refer to the corresponding maps in Fig. 8. The largest zonal wind
maximum is observed on map 1 (and 7) when this maximum occupies its
lowest latitude (76"),and the smallest maximum appears on map 4 where it
reaches its highest latitude (82").
Owing t o the presence of the second mode of wave 3, a distinction should
be made when comparing the flow patterns in the northern and southern
parts of the different maps. It appears then that in the northern part, the
troughs deepen from map 1 to map 4, where well defined closed lows are
observed. I n this respect, map 1 could be considered as characteristic of a
high index circulation and map 4 as corresponding to a low index circulation.
This comparison is consistent with the fluctuations of the zonal wind, the
closed cyclones being associated with relatively weak jets. The southern
part undergoes a similar pulsation, the most intense ridge being observed on
map 1 and then decaying down to map 4.A detailed comparison would be
156 A. QUINET
needed in order t o establish definitely the analogy of the flow presented here
with the potential energy vacillation emphasized by Pfeffer and Chiang
(1967). Let it be simply stated that the nearly steady character of the wave
tilt during its amplitude fluctuation and the fluctuations of the intensity of
the zonal jet during the cycle are important characteristics common to both
phenomena.
The southward transport of west to east momentum reported in the second
column of Fig. 8 is evidently associated with the NW-SE tilt of the waves.
The maximum southward transport occurs on map 1 and then decays,
approximately domi to map 4, where the northern cyclone has developed.
The momentum transport during the cycle is given for the three latitudes
yIL = 54", go", and 126", from bottom to top in Fig. 10a, where the figures
[ v'uq
128' o
-0.01 -0.01
aoi
90" 0
-0ai
aoi
-0.01 -0.01
a C
FIG.10. Time evolution of the zonally averaged (a) northward W-E eddy momentum
transport [v'u'] in Lzfa units; (b) W-E eddy momentum transport convergence
-a[w'u']/ay in Lfa units; and (c) eddy heat transport [w'fl'] in BL3fa units, at the
three latitudes y / L = 54", go", and 126", during the potential energy vacillation observed
a t k = 0.2926, 0:. ,, = 0.25. The figures along the z-axis refer t o the maps of Fig. 8.
along the x-axis refer to the maps of Fig. 8. As imposed by (4.64), there is
maximum southward momentum transport a t the central latitude and
identical weaker values a t the two other latitudes.
The third column of Fig. 8 indicates that there is momentum divergence
north of the center of the channel and, as a consequence of (4.65),momentum
convergence in the south. Accordingly, the southward export rate of west to
east momentum leads to a reduction of the zonal flow in the northern part
and creates a zonal jet in the southern part. Figure 10b gives the time
A NUMERICAL S T U D Y O F VACILLATION 157
variation of -a[v’u‘]/ay during the cycle. The maximum values of the low
latitude momentum transport convergence and of its high latitude diver-
gence arise a t time t = 0 (and t = T), when the zonal wind maximum occupies
its lowest latitude y I L = 76” and reaches its largest value. At time t = TIZ,
the low latitude convergence is minimum, corresponding to the weakest jet
a t the highest latitude y / L = 82”. Accordingly, the zonal jet fluctuations are
strongly correlated with the momentum transport and its convergence.
The heat transport at the three above mentioned latitudes is given in Fig.
1Oc. It is toward the north a t all latitudes and during the whole cycle. The
largest values occur in the region of maximum zonal wind. They are approxi-
mately the same a t 54” and 90” latitude, and roughly half as much a t
y / L = 126”. The heat transport is nearly constant a t low latitude during the
cycle but exhibits fluctuations of approximately the same amplitudes a t the
central and high latitudes, where the fluctuations are in phase. The maximum
heat transport corresponds to map 5, i.e. somewhat later than the time of
occurrence of low index circulation, when the zonal flow is minimum. It
should be noted that the evolution of [u‘e‘] does not show any asymmetry
with respect to the extreme values while Pfeffer et al. (1965)have reported the
existence of such asymmetries in the potential energy vacillation.
The occurrence in time of the extreme values of U , [v‘u’], -a[v’u’]/ay,
’ ] summarized in Table 11, where M stands for “maximum ” and
and [ v ’ I ~is
m for “minimum.” Since, a t a given latitude, these quantities do not change
their sign during the cycle, the modulus only has been considered for labeling
the extreme values.
TABLE11. Distribution in time of the occurrence of extreme

values (M = maximum, m = minimum) of zonal wind u,zonally
averaged relative S-N eddy momentum transport [w u ]and
momentum transport convergence - a[w u ]/ayand zonally
averaged northward heat transport [w &] during the potential
energy vacillation cycle observed a t k = 0.2925, Of, = 0.25
Maps 1 2 3 4 5 6 7
~
U M m M
[w u l M m M
-a[v u ]/ay M m M
[w’e’~ m M
5.2.2. The Energetics of the Potential Energy Vacillation. Figure 11 shows

the evolution of the available potential energy A and the kinetic energy K ,
expressed in c = 2r2pOg-’L4f2 units, K being approximat.ely one half of A
t
aS9
0.68
5
10 20 30 40 50
FIG.11. Time evolution of the available potential energy A and of the kinetic energy
K , expressed in 2n2p0 g-lL4f a units, during the potential energy vacillation observed
at k = 0.2925, 8:. = 0.25.
1.53
1.52
1.51
0.01
0.21
020
0.19
FIG.12. Time evolution of the zonal A , and eddy A , available potential energy and
of the zonal K , and eddy K , kinetic energy, expressed in 2rrapo g-'Lty a units, during
the potential energy vacillation observed at k = 0.2925, 8:. = 0.25.
A NUMERICAL STUDY O F VACCILLATION 159
during the cycle. The striking feature is the nearly perfect phase opposition
between the two curves although nonadiabatic processes are taking place.
The fluctuations of A are of greater amplitude than those of K .
Figure 12 gives the graphic representation of A , , A , , K , , and K , during
the cycle. A , and K , undergo the major oscillations while A , is nearly
constant. As a consequence of the very intense zonal thermal forcing, A , is
two orders of magnitude greater than A,; K , is approximately one third of
A , and K , one order of magnitude greater than A , . I n opposition t o the
curves of A and K , the A , and K , curves or the A , and K , curves are nearly
in phase, so that any increase of A , (AE)is accompanied by a simultaneous
increase of K , ( K , ) .
I n order to investigate the nature of the energetics of the flow, Fig. 13
gives the evolution of the various terms of the energy budgets (4.79)-(4.82).
Clearly C, is the dominant conversion term, so that the zonal kinetic energy
is essentially maintained against dissipation by conversion from A , . I n lesser
proportion A , is also transferred into A , , by C, and C,, , which, in turn, is
converted (C, >0 ) into K Ea t an important rate. Accordingly, there must be
a considerable generation of A , by heating, G, , feeding both C, and C, CAR.+
Evidently, without any longitudinally distributed heat source, GE < 0 [see
(4.88)]. The kinetic energy transfers are toward the zonal flow ( C , < 0) but
the contribution of this term is two orders of magnitude less than C, so that,
as expected, the barotropic kinetic energy transfers are extremely weak.
FIG.13. Time evolution of the different components of the energy budgets [see Eqs.
(4.79)-(4.82) for definitions], expressed in 2napOg-lL4f3units, during the potential
energy vacillation observed at k = 0.2926, 52., = 0.26.
160 A. QUINET
C , ,CA,C A R , C , , and BEare nearly in phase but in phase opposition with

C , and D,. Thus, weak values of 8, are associated with high values of C ,
and vice versa, a situation which leads to large 0uctuations of A,. These are
-+
however partly reduced by CA CAR. In contrast, neglecting the 0uctuations
+
of G, , the nearly perfect correlation between CA CAR and C, tends to re-
duce to zero the 0uctuations of A , . Finally, the 0uctuations of C , C A R +
being more intense and in phase opposition with those of C,, the zonal avail-
able potential energy A, and kinetic energy K, increase or decrease simul-
taneously while the evolution of KE is essentially governed by C, .
Figure 14 shows the evolution of T,defined in (4.100),which governs the
barotropic kinetic energy transfers arising between scales (2, 0), (1, 31, and
(2, 3), and the evolution of TI,T,, T,,defined in (4.106)-(4.108),which
FIG.14. Time evolution of T, TI,T z , and T3[see Eqs. (4.100) and (4.106)-(4.108)
for defitions] during the potential energy vacillation observed at k = 0.2926, a:, =
0.26.
governs the available potential energy transfers and conversions between

the same scales. The essential fact to note is the weakness of these terms as
compared with the global conversion and transfer rates. Thus, it can be con-
sidered that the nonlinear character of the potential energy vacillation is
restricted to the interactions of the two modes (1, 3)and (2, 3) of waves with
the zonal field (1, 0). Since these self-interactions do not influence the baro-
tropic kinetic energy transfers (see Section 4.4.2), the potential energy
vacillation results only from the baroclinic transfers and conversions of
available potential energy. The strong correlation between C A , C,, and C,
appears then to result from the fact that only the self-interaction terms
(4.101), (4.102),and (4.109)-(4.112) come into play in the energy processes.
The term T being positive, there is a permanent transfer from Kl,
toward K z , o and K l , 3 . On the other hand, A l , 3 is transferred toward
,
A,. ( T1<0) but chiefly toward A z , (T,> 0) while the transfers between
A z , and A,, are extremely small (T, 2: 0). Let us finally note that, re-
garding the role of the (2, 0 ) , (1, 3), and (2, 3) interactions, the barotropic
processes, although very weak in comparison with the self-interaction
contributions, are the dominant ones.
Figure 15 describes schematically the energy cycle of the potential energy
vacillation; the values indicated on the arrows represent mean values over
one period.
14.1
0.2
t -cK
1
- / G ~ + IG 1.1
~ ~
FIG.15. Energy diagram averaged over one period of the potential energy vacillation
observed at k = 0.2925, 0:. , = 0.26. Energies are expressed in c = 2m2p,g-1L4f units,
the conversion and transfer rates are expressed in cf units.
One of the major deficiencies of the simulated flow with respect to labor-
atory potential energy vacillation is the symmetric character of the heat
transport during the cycle and the relatively weak fluctuations of the ampli-
tudes of the waves. Once it has been recognized that potential energy vacil-
lation involves essentially baroclinic energy exchanges, it is possible to
obtain intense potential energy vacillittion a t a high rotation rate by simply
not including the barotropic energy exchanges in the model. Such a model,
with the correct cylindrical annular geometry, has been dealt with by
Merilees (1972). If high amplitude potential energy vacillation can indeed be
simulated in this way, the heat transport nevertheless remains symmetric
during the cycle. The introduction of viscosity and heat conductivity into
the numerical model would improve the results. It has indeed been observed
that amplitude vacillation is most common with highly viscous fluids (Fowlis
and Hide, 1965). However, whether these physical factors will introduce in
the model the specific effects capable of leading t o asymmetries in the heat
transfer cycle remains an open question.
152 A. QUINET
5.2.3. A Physical Interpretation of the Potential Energy Vacillation. We will

now try t o give a simple physical interpretation of the flow. Taking into
account the results of subsection 5.2.2, one need only deal with the dominant
baroclinic processes related to the self-interactions of the waves with the
zonal thermal field of the first mode 8,. o .
The generation of A , consists of a cooling of cold air in the north and a
warming of warm air in the south. Accordingly, f9:.o being shady, G , is B
minimum when the zonal temperature contrast between the southern and
northern part of the fluid is the largest (elm a maximum). I n this case, the
intensity of the direct Hadley cell generated by the N-S thermal contrast
into the fluid, i.e. the conversion rate C, of A , into K , is maximum (see
Table I and Fig. 13). The associated relative excess of heat transport by a n
intense Hadley cell allows a reduction of the eddies which are then generated
a t their minimum rate C,. However, according to the linear baroclinic
instability theory, this situation of maximum N-S temperature contrast
and, consequently, of maximum vertical wind shear, favors the generation
of disturbances. This is re0ected by the subsequent increase of C,. The
amplifying disturbances are then able to assume a supplementary part of the
heat transport. Hence, the N-S temperature contrast into the fluid is re-
duced, entailing a decrease of Cz . The rate C, (and C,) goes on growing until
it reaches its maximum value, which occurs a t the time of minimum C, .
I n this situation, under the influence of the thermal forcing, C , attains its
maximum value since el, is a minimum (see Table I and Fig. 13). After the
time of minimum value of €I ,,,,and in accordance with linear baroclinic
instability theory, the baroclinic generation rate of the eddies CE decreases.
The associated deficit of heat transport is then taken up by an intensification
of the Hadley cell ((7,) which is, however, insufficient to avoid the reappear-
ance of a strong thermal contrast between the southern and northern part of
the fluid. When Ola0 is maximum, G , and C, are minimum while C, is max-
imum (see Table I and Fig. 13) and the cycle starts again.
The most striking aspect of the potential energy vacillation energetics is
that the maximum baroclinic generation rate of the eddies C, appears a t the
minimum N-S thermal gradient. This shows how the nonlinear mechanism
of the interplay between the zonal fields and the eddies can lead t o situations
radically different from those suggested by the theory of linear perturbations.
5.3. The Kinetic Energy Vacillation
5.3.1. The Dynamics ofthe Kinetic Energy VmiUation. We now describe the
results of a numerical experiment performed with the same thermal forcing
6;. =0.25 as in the previous case, but a t a higher rotation rate, k = 0.275.
Table 111lists the values of the flow variabbles for every five time steps in a
TABLE111. Numerical integration of the model for k = 0.275, 0;. = 0.25a
P
to 41.0 *% 0 41.3 4 2 ,3 81.0 e2. el. e2. 0 0
w
a
0 0.2186 -0.0294 0.0458 0.0336, -122" 0.2188 -0.0053 0.0188, -55' 0.0136, -156" 0.0782 3
*3
5 0.2203 -0.0225 0.0339 0.0357, -151" 0.2223 -0.0014 0.0138, -58" 0.0146, 175' 0.0746
10 0.2230 -0.0084 0.0366 0.0296, -177" 0.2246 0.0035 0.0156, -57" 0.0120, 150' 0.0726
15 0.2241 0.0030 0.0470 0.0211, 165" 0.2239 0.0067 0.0206, -56" 0.0080, 135" 0.0738 F
20 0.2225 0.0125 0.0572 0.0156, 126" 0.2206 0.0073 0.0245, -54" 0.0052, 95" 0.0774
!
25
30
0.2192
0.2184
0.0248
0.0296
0.0570
0.0462
0.0249,
0.0337,
77'
58'
0.2176
0.2186
0.0065
0.0054
0.0237, -53"
0.0190, -55"
0.0095,
0.0136,
41"
24"
0.0798
0.0784 *
U
0
35 0.2201 0.0227 0.0339 0.0359, 29" 0.2222 0.0015 0.0137, -58" 0.0147, -5' 0.0747 w
40 0.2230 0.0085 0.0365 0.0296, 3" 0.2246 -0.0035 0,0155, -57" 0.0120, -30" 0.0726 2
45
50
0.2241
0.2226
-0.0031
-0.0126
0.0469
0.0570
0.0212, -15"
0.0158, -54"
0.2239
0.2207
-0.0067
-0.0073
0.0205, -56"
0.0245, -54"
0.0080, -45'
0.0052, -85'
0.0737
0.0773
8
55 0.2197 -0.0230 0.0581 0.0227, -98" 0.2179 -0.0067 0.0243, -53" 0.0085, -134' 0.0797 F-
60 0.2185 -0.0296 0.0458 0.0338, -122" 0.2187 -0.0054 0.0188, -55" 0.0136, -156' 0.0783
!+-?I
Variables are observed in a coordimte system moving with wave ( 1 , 3) of the #-field.
164 A. QUWET
-0.350-0.350-3.350-~.35~-~~.35~-0.350-0.350-0~350-0.350-0.350-0.350-0.350-0.350 -0.0 -0.000 0.000

0.144 0.000 0.001
0.260 0.003 0.013
0.328 0.008 0.0111
0.312 0.014 0.013
0.308 D.Olb 0.000
0.244 0.014 -0.013
0.171 0.008 -0.018

0.102 0.003 -0.013
4.246 0.219 0.291 0.215 0.240 0.225 0.215 0.219 0.2P1 0.215 0.260 0.225 0.265 0.041 0.000 -0.004
2.261 0.267 G.261 0.2bl 0.261 0 . 2 6 1 0.2b7 0.261 0.2b1 0.261 0.261 O.2bl D.2bl 0.0 -0.0 0.0
1
-0.0 -O.DOO 0.000
0.111 0.000 0.000
0.201 0.000 O.Wl
0.217 0.000 0.001

0.313 0.001 0.001
0.314 0.001 0.000

0.285 0.001 -0.001
#.I89 0.185 u.104 0.188 0.192 9 0.105 0.184 0.188 0.192 0.193 0.189 0.232 0.000 -0.001
D.230 0.241 U.261 0.264 0.254 0.2kl 0.231 0.241 0.261 0.264 0.256 0.241 0.231 0.162 0.000 -0.001
0.083 0.000 -0.000

0.3D3 0.303 0 . 0.0 -0.0 0.0
2
-0.0 0.000 -0.000
L07b -0.000 -0.002
0.151 4.002 -0.001

0.220 -0.005 -0.010
0.218 -0.001 -0.007
0.314 -0.009 -0.000
0.319 -0.001 0.007
0.287 -0.001 0.010
0.218 -0.DD2 0.007
0.118 -0.000 0.002

4.331 0.331 0.331 0.331 (1.331 0.331 0.331 0.331 0.331 0.331 0.331 0.331 0.331 0.0 -0.0 0.0
3
-0.2b6-3.266-~.266-0.266-6.266-0.266-0.266-0~2b6-0~266-0.266-0.266-0~266-0~~66 -0.0 0.000 -0.000
-0.2b2-0.231-0.228-0.256-6.28U-0.290-0.262-0.231-0.22~-0.25b-0.2~8-0~290-0~262 0.016 -0.000 -0.004

0.102 -0.003 -0.013
0.110 -0.008 -O.OL@

0.214 4.011 -0.013
0.108 -0.016 -0.000
0.342 -0.014 0.013
0.329 -0.008 0.018
0.260 -0.003 0.013
0-114 -0.000 0.001
0.0 -0.0 0.0
4
FIG.16. See facing page for legend
-0.0 0.000 -0.0w

0.003 -0.000 -0.000
O.lb2 -0.000 -DAD1
0.231 -0.000 -0.001
0.289 -0.001 -0.001
0.314 -0.001 -0.000
0.313 -0.Wl 0.001

0.277 -0.000 0.001
0.208 -0.000 0.001
O.llI -0.000 0.000
0.0 -0.0 0.0
5
-~.332-~.332-5.332-J.332-L.332-0.332-0.30~1~~-B~332-0~332 -0.0 -0.000 0.000
0.118 0.000 0.002
0.218 0.002 0.007
0.288 0.005 0.010

0.319 0.008 0.001
0+31k 0.009 0.000

0.Zll 0.008 -0.001
0.220 0.005 -0.010
0.151 0.002 -0.001
0.01b 0.000 -0.002
0.296 0.296 0.296 0.296 0.296 0.29b 0.296 0.296 D.29b O.29b 0.2Pb 0.296 0.296 0.0 -0.0 0.0
6
-0.0 -0.000 0.000
0.1.4 0.000 0.00k
al.2bO 0.003 0.013
0.329 0.008 0.011
0.3k2 0.Olk 0.013
0.108 O-Olb 0.000

0.2kk 0.Olk -0.013
0.170 0.008 -o.oie

0.102 0.003 -0.013
..ZIU 0.293 J.283 J.Zk9 ~ . 2 2 >0.236 ( - 2 1 3 0.291 0.203 D.Zk9 0 . 1 2 5 0.23b 0.21) 0.0kb 0.000 -0.00k
1.266 J.Zb6 0.2bh 3.1bb ~ . 2 h b 3.2hh 0.2bb 0.266 0.261. D.266 0.2bb 0.Zbb 0 . 2 b S 0.0 -0.0 0.0
FIG.16. Synoptic representation of the #-field during the kinetic energy vacillation
observed a t k = 0.275, 8: = 0.25. Spacement between isolines is 0.1 in L2f units. The
time interval between two successive maps is 10 nondimensional time units to = j t . The
zonal wind in Lf units, the zonally averaged S-N momentum transport in L2j2units,
and the momentum transport convergence in Lf2 units respectively, ere, printed on the
three columns at the right of the maps.
166 A. QUINET
coordinate system moving with the wave (1, 3) of the #-field. The flow is
still periodic, the period being 60t0.The increase of the rotation rate entails a
considerable increase of the fluctuations of the variables in comparison with
the previous case. The component #2, reaches extreme symmetric values of
-0.0294 (to = 0, to = 2') and t0.0296 (to= 30) so that this kind of flow has
been termed by Lorenz (1963) a symmetric vacillation. The waves (1, 3)
and (2, 3) of the $-field progress one wavelength in approximately 74t0 and
89t0 respectively, but these periods undergo small fluctuations during the
cycle, especially for the (2, 3) component. Again, a t each mode, the &field
lags behind the #-field. The phase shift between the $- and 8-waves has been
increased and varies for the (I, 3) and (2, 3) waves between 53" and 58" and
between 28" and 37", respectively. The amplitudes of the (1, 3) and (2, 3)
components of the $-field are still negatively correlated but their phase shift
may now vary over 277. Accordingly there are intense fluctuations of the
barotropic processes, especially of the momentum transport (4.64) which
changes sign during the cycle.
The synoptic representation of the stream field a t the mean level is given
in Fig. 16. Spacing between stream lines is still 0.1 nondimensional units and
the time interval between two successive maps is 10 nondimensional time
units. Except for an eastward translation of approximately one sixth of the
wavelength, map 7 is a repetition of map 1. It can be seen now that the trough
and ridge lines undergo substantial oscillations, characteristic of a '' tilted
trough" or kinetic energy vacillation. The wave tilts from SW to NE on
maps 1 (and 7) and 6 and tilts from SE t o NW on maps 3 and 4 while the
waves are symmetric on maps 2 and 5. Let us already point out that the
symmetric configuration precedes the maximum tilt (maps 1 and 4) by 20
time units and follows it by 10 time units only.
The zonal wind, the zonally averaged eddy momentum transport, and the
eddy momentum transport convergence are given on the right side of the
maps of Fig. 16 as in Fig. 8. The maximum zonal wind occurs to the north of
the center of the channel on map 1 (and 7 ) when the waves are tilting north-
east and when there is a northward transport of W-E momentum through
the whole width of the channel. At this time there is momentum transport
convergence to the north of the center of the channel and momentum trans-
port divergence to the south. The situation is reversed on map 4, the south-
ward momentum transport converging south of the center of the channel
where the maximum zonal wind is located. Between these two situations, but
not midway, the momentum transport and the momentum transport con-
vergence vanish a t all latitudes (maps 2 and 5).
The negative values of the momentum transport observed after map 2 are
associated with the deepening of the waves, generating high latitude cyclones
which are established on map 3, somewhat before the time of the maximum
zonal jet of map 4.Similarly, positive values of the momentum transport are
associated with amplifying low latitude highs.
The zonal flow profile is given in Fig. 17 where the figures on the different
curves refer t o the corresponding map of Fig. 16. The extreme latitudes of the
maximum zonal wind are respectively 112" on map 1 (and 7) and 67" on
map 4,when the waves reach their maximum tilt. At these times, the zonal
flow is barotropically unstable for small linear perturbations, the zonal wind
profile presenting an inflection point on Fig. 17. This is not without suggesting
an explanation for the rapid reestablishment of the symmetric configuration
of maps 2 and 5 after the time of maximum tilt.
180
t
162
-p 144
126
-
0
0
108
0 90
3
.=
4-
72
m
4 54
36
18
0.1 0.2 0.3 0.4
FIQ.17. Meridional profile of the zonal wind corresponding to maps 1 to 7 of Fig. 16.
Latitudes are given in degrees according to (4.1);the inflection point is at 63" and 117'
on curves 1 and 4, respectively.
The fast evolution after the occurrence of a n unstable zonal wind profile
is particularly well represented in Fig. 18 which is the analog of Fig. 10.
Northward or southward eddy momentum transport is simultaneously
observed at the three latitudes of 54",go", and 126" while high latitude
momentum transport convergence is balanced by low latitude divergence,
and conversely. As noted by Fultz, the maximum value of [u’v’] is located
a t the middle latitude. I n the model considered here, this circumstance is
necessarily a consequence of (4.64).
There is considerable asymmetry on each of the curves of Figs. 18a and
lBb, the extreme values, corresponding to maps 1 and 4,being slowly reached
and then rapidly left. This allows one to give the following interpretation of
the momentum evolution during the cycle: the momentum transport con-
vergence associated with the tilting of the waves tends to remove the jet
from the center of the channel, where the thermal forcing would locate it.
168 A. QUINET
FIG.18. Time evolution of the zonally averaged (a)northward W-E eddy momentum
transport [ v ' d ] in Lafa units; (b) W-E eddy momentum transport convergence
-a[v'~']/& in Lfa units; and (c) eddy heat transport [v’e] in BL3f units, at the three
latitudes v / L = 64', go", and 126" during the kinetic energy vacillation observed at
k = 0.276, el,,, = 0.26. The f i p e s on the z-axis refer to the corresponding maps of
Fig. 16.
Before the maximum wind reaches the latitude of maximum momentum

convergence the mechanism of barotropic instability comes into play and
rapidly redistributes the momentum within the zonal flow, releasing
barotropic instability.
The heat transport shown in Fig. 18c exhibits important differences at the
three latitudes and differs rdicblly from the previous case (Fig. 1Oc). The
curves are out of phase at the two extreme latitudes, but the amounts of
heat transported across these latitudes during the complete cycle are now of
equal importance. There are two strong maxima and minima at the central
latitude, while there exist only one well defined maximum and one well
defined minimum, separated by a weak secondary minimum and maximum,
on the other curves. Thus, at latitudes of 64" and 126", the heat transport
slowly increases and then rapidly decays. The maximum high latitude heat
transport occurs near the time of appearance of the closed cyclones at those
latitudes. An analogous situation occurs at low latitude, at the time of
appearance of the closed highs. It will be established in the next section
that, as suggested by these correlations, the cellular circulations arise from
intense transformation rates of eddy available potential energy into eddy
kinetic energy. Consistent with Fultz' observation the minima of heat
transport at 54" and 126" latitude occur near the time of maximum zonal
wind in the northern and southern part of the channel, respectively, when the
momentum transport and the momentum transport convergence through
the whole width vanish. Likewise, as in laboratory experiments, the minimum
heat transport at a given latitude follows the minimum momentum conver-
gence at this latitude. As for the potential energy vacillation, the relative
positions of the times of occurrence of the extreme values of U , [u'u'],
-a[u'u']/ay, and [u'e'] are schematically shown in Table IV. M stands for
TABLEIV. Distribution in time of the occurrence of extreme values (M = maximum,

m = minimum) of the zonal wind U,the zonally averaged S-N momentum transport
[w'u'] end momentum transport convergence - a[w'u?/ay, and zonally averaged heat
transport [ w ' 8 ] during the kinetic energy vacillation cycle observed at k = 0.275,
87, = 0.25"
Maps 1 2 3 4 5 6 7
The region of location of the extreme value, north (N) or south (S) of the center of
the channel, is also indicated.
positive maximum and m stands for (negative) minimum; N (north) and

S (south) indicate the region of location of the associated extreme value with
respect to the center of the channel.
5.3.2. The Energetics of the Kinetic Energy Vacillation. E'igure 19 shows

the evolution of A and K expressed in nondimensional units. Once again,
any increase of A is accompanied by a simultaneous decrease of K , and vice
versa. The orders of magnitude of A and K are respectively the same as in
Fig. 11 but, noting the scale differences along the y-axis in Figs. 19 and 11,
the fluctuations are now increased and amount t o approximately 10 yoof the
total values. There is no noticeable asymmetry in the curves.
Figure 20, the analog of Fig. 12, now shows the presence of small phase
shifts between A , , A,, K,) and K , . The curve of K, is asymmetric with
respect to the extreme values. As discussed in the previous section, this
characteristic can be ascribed to the barotropic instability conditions pre-
vailing a t certain times (maps 1 and 4) in the fluid. On the contrary, the
curve of K , appears symmetric with respect to the extreme values. There is
also some asymmetry on the curve of A,. This will be discussed later.
, to
0.16
0.14
0.48
10 20 30 40 50 60
FIG.20. Time evolution of the zonal A , and eddy AE available potential energy and
of the zonal K , and eddy K Ekinetic energy, expressed in 2+po g-IL4ja units, during
the kinetic energy vacillation observed at k = 0.276, 8
:. = 0.25.
A NUMEXICAL STUDY OF VACILLATION 171
The conversion rate C , is still the dominant term in Fig. 21, which corres-
ponds t o Fig. 13. Each curve now exhibits two sequences of maximum and
minimum, the first sequence corresponding t o -0.0294 J / z , < +0.0296
and the second one to 0.0296 2 i+hz. 2 -0.0296. Except for an increase of
the fluctuations, there is great qualitative similarity between Fig. 13 and
Fig. 21 regarding the behavior of Gz , Cz, C, , C A, CAR,Dz, and D, . Accord-
ingly, as in the potential energy vacillation, so the self-interactions of the
waves (1,3)and (2,3)with the zonal thermal field (1,O)remain the dominant
contributions to baroclinic energy processes in the kinetic energy vacillation.
FIG.21. Time evolution of the different components of the energy budgets [see Eqs.
(4.79)-(4.82) for definitions], expressed in 2rrap0g- lL4f3 units, during the kinetic
energy vacillation observed at Ic = 0.275, 8;. = 0.25.
In contrast, the barotropic energy exchanges CK now play a n important

role. CK being negative, this term represents a barotropic stabilization of the
eddies. I n this respect, we note that, paradoxically, the maximum stabiliza-
tion (minimum values of C,) occurs a t time 30 corresponding t o the curve
labeled 4 in Fig. 17, when the zonal flow is definitely barotropically unstable
for a small perturbation.
To understand this paradox, the nonlinear character of the flow should be
invoked by considering the threefold kinetic energy transfers defined in
(4.96)-(4.98)where, according to C , , T should be positive. Let us first recall
that the only energy source in the model is a t scale (1,0), namely O:, o . The
kinetic energy a t scale (2,O) has then necessarily t o come via the [(2,0),(1,3),
(2, 3)] interactions since there is no possibility of a direct supply of K z , o
from K l , o . Now, it has already been mentioned, and it is also observed in
this case, that the baroclinic transfers associated with the [(2,0), ( 1 , 3 ) , (2, 3)]
interactions are quite small. Accordingly, the major source of kinetic energy
172 A. QrJINET
,
K 2 , is a barotropic energy transfer from K l , . This actually is the reason
why T is positive; in this sense, the flow is barotropically stable. Thus the
rate of increase or decrease of K z , or, considered a t the mean level, of t,b2.
is mainly governed by the respective increase or decrease of T.
It is easily seen that the linear condition of barotropic instability
d2U/dya= 0 assumes the form
so that the relative increase of t,b2,0 leads t o barotropic instability. The

maximum value of +z, is precisely observed when K z , is fed a t the maxi-
mum rate by the eddies. After that time, we do not observe a transfer of
K 2 , towards K , , but a rapid reduction of the kinetic energy transfer from
the eddy flow toward the zonal flow of the second mode. This transfer nearly
falls to zero in 10 time units and then slowly increases again to reach its new
maximum value 20 time units later.
The evolution of K , nevertheless remains dominated by C , as indicated by
the strong positive correlation between C, and K , . The curves C , and C ,
being also nearly in phase, the relative deficit of conversion of A , into K ,
tends to be compensated by an increase of the barotropic transfer from K ,
towards K , . As a result the fluctuations of K , are much weaker than the
fluctuations of A , .
I n contrast with C , , the baroclinic energy transfer and conversion are
symmetric with respect t o their extreme values. Surprisingly it is observed
that the maximum generation rate of the eddies C , occurs near the time of
maximum static stability. The fluctuations of rz, being of the same order of
magnitude as the fluctuations of r l , , the vertical zonal wind shear oscillation
is controlled chiefly by r z .o . It can be observed that the maximum values of
C , are attained near the time of extreme (positive or negative) values of
rz.o . On the other hand, the time of minimum C , nearly coincides with that
of zero r 2 .o . Accordingly, and owing to its nonlinear nature, the baroclinic
behavior of the flow is subjected t o the antagonistic influences of an increase
of static stability, proceeding with an intensification of the vertical zonal wind
shear associated with the increase of rZ,o . Presumably such an interplay
between two factors is responsible for the symmetric character of the
baroclinic process curves, contrasting with the strong asymmetry of C , . Let
us also mention that, as for the potential energy vacillation, the maximum
value of C, occurs when el, is relatively small, and vice versa.
Figure 22 shows the evolution of T [(4.100)]and TI, T 2 ,and T,[(4.106)-
(4.108)]during the vacillation cycle. With respect to the analogue, Fig. 14, all
these terms have been increased, the term T remaining one order of magni-
tude greater than the others. Accordingly as long as only the influence of the
[(2, 0 ) , (1, 3), (2, 3)] interactions is considered, the barotropic kinetic energy
FIQ.22. Time evolution of T, T,,T,, and T3[see Eqs. (4.100) and (4.106)-(4.108)
for definitions] during the kinetic energy vacillation observed at k = 0.275, 6:. a = 0.25.
transfers are still the dominant energetic processes. I n contrast with the self-
interaction terms, the terms T and particularly T I ,T,, and T,are extremely
weak. Nevertheless, their contributions are now sufficient to introduce the
observed shifts between the curves of A , , A , , K , , and K , of Fig. 20. Owing
to the weak values of A , , indicating that the dominant self-interaction
transfers from A , are nearly entirely converted into K , , the influence of
T,,T,, and T 3 is most marked on the A , curve which exhibits noticeable
asymmetries.
Figure 23, where the numbers on the arrows represent mean values over
one period, schematically describes the energy cycle of the flow. Actually,
there is little difference between Fig. 15 and Fig. 23. This shows how the
averaging process can smooth important aspects of the energy cycle since
we know that the role of the C , term is quite different in both cases.
13.4
FIG.23. Energy diagram averaged over one period of the kinetic vacillation observed
at k = 0.275, 6:. a = 0.25. Energies are given in c = 27r2pag-’L4f2 units; the conversion
and transfer rates are expressed in cf units.
174 A. QUINET
The generated flow exhibits remarkable similarity to the laboratory kinetic

energy vacillation. This confirms indirectly that the neglected effecqs of
viscosity and thermal conductivity should be of minor importance in this
kind of flow. We have seen that this does not seem to be entirely true for the
potential energy vacillation.
5.3.3. A Physical Interpretation of thx Kinetic Energy Vacillation. The

great similarity between Figs. 13 and 21 suggests that we may use our
description of the potential energy vacillation as a starting point for the
interpretation of the kinetic energy vacillation. Indeed, recalling that, as
in the potential energy vacillation case, the maximum (minimum) value of
C, occurs near the time of minimum (maximum) value of r l , = o , the
kinetic energy vacillation heat transport cycle has an important component
similar t o the potential energy vacillation heat transport cycle. Nevertheless,
differences now occur under the influence of the intense fluctuations of the
(2, 0) variables. Thus, the physical interpretation of the existence and
influences of these fluctuations becomes the keystone of the interpretation
of the kinetic energy vacillation.
As the thermal forcing is unchanged, we have to associate the modifications
of the flow with the increase of the rotation rate. It is well known (Lorenz
1963) that the increase of the rotation rate destabilizes the second mode
variables. On the other hand, the relative increase of #2,0 [(5.2)] leads to
zonal wind profiles which are barotropically unstable with respect to small
perturbations. Hence, a t sufficiently high rotation rates, the zonal wind
profile may exhibit a point of inflection. The question is now to see how the
concept of linear barotropic instability can account for the flow behavior.
Consider, at the mean level, a meridional profile of the zonal wind which is
symmetric with respect to the axis of the channel (&, ,, = 0). At the rotation
rate used here, k = 0.275, the value #2. = 0 is unstable and may become
sufficiently large as to lead to linearly unstable zonal wind profiles. The time
of maximum value of # 2 , 0 is also the one of maximum tilt of the #-wave
(see Table I11 and maps 1 and 4 of Fig. 16) and corresponds to extreme values
of meridional momentum transport and of momentum transport convergence
(see Figs. 18a and 18b). After this time a rapid reduction of all these baro-
tropic processes, as well as of the transfer of kinetic energy from the eddies
toward the zonal flow, is observed (curve C, of Fig. 21). Hence, the concomi-
tant decrease of #2, can be considered as resulting from the release of the
(linear) barotropic instability of the zonal flow.
If an examination of the zonal wind profile can provide some criterion for
the appearance of barotropic instability in the model, the subsequent process
of redistribution of the kinetic energy between the zonal flow and the eddies is
governed by (4.96)-(4.98).Note that the influence of this mechanism on the
waves of the stream field is hidden by the dominant baroclinic processes.
Indeed, the increase of K l , and the decrease of K z . 3 , associated with the

barotropic reduction of Kz. o , is not reflected in the amplitudes of the waves
(1, 3) and (2, 3) of the +-field following extreme values of &,o .
It should be emphasized that neither nor are small when $,,,
reaches its extreme values. I n fact, the unstable zonal wind profile results
precisely from the existence of the eddies which concentrate zonal momentum
a t some particular latitude. The release of barotropic instability is then
operated through a reduction of energy transfer from the eddies toward the
zonal flow.
Obviously, the increase of the rotation rate influences also the baroclinic
processes. The most important qualitative effect is observed on r 2 .o .
When rz, is large and positive (negative) a strong vertical wind shear pre-
vails to the south (north) of the center of the channel. The extreme values of
r2, occur a t the time when the flow pattern a t the midlevel exhibits closed
circulation cells. The northern cell is associated with a maximum southern
vertical shear and vice versa (see Table I11 and Fig. 16). On the other hand,
the first maximum of C , (Fig. 21) corresponds approximately t o map 3 of
Fig. 16 (high latitude low), where the high latitude eddy heat transport is
maximum (Fig. 18c);the second one corresponds to map 6 (low latitude high),
where the low latitude eddy heat transport is maximum. Accordingly, the
intense eddy transport of heat allows an intense conversion of eddy available
potential energy into eddy kinetic energy. This conversion reduces the vertical
wind shear a t the latitude where the process is taking place but, at the same
time, increases the shear in the other half of the channel, where conditions of
strong baroclinic instability are then created. As a result, the associated
closed circulation cells alternately disappear and reappear north and south
of the center of the channel. At the same time, the zonally averaged S-N
eddy heat transport (Fig. 18c) varies considerably along the S-N direction.
Actually the heat transport curves a t latitudes 54" and 126" (Fig. 18c) are out
of phase, the small low latitude values being associated with large high-
latitude values, and vice versa.
5.4. The Various Vacillation Cycles

When numerical experiments are performed a t various rotation rates and
thermal forcings, different kinds of vacillation can be generated. I n order to
easily distinguish between these circulations, we represent the flow in the
subspace of the ( Oe+h1,y&. o ) variables, the supercript 0 t o the left indicating
that the coordinate system is moving with the wave (1.3)of the +-field.
At high thermal forcing and relatively low rotation rates, the model
exhibits unsymmetric vacillation. This kind of flow has been discussed in
Section 5.2 and is represented in Fig. 24a corresponding to t3T,o=0.25,
k = 0.2925. The period of vacillation is 56t,. Recalling that the equations of
176 A. QUmET
-0.01 t
( aI
FIG.24. Representation in the 3, 4%.

o)-planeof: (a) an unsymmetric vacillation
(k = 0.2926, 8:. = 0.26); (b) a symmetric vacillation (k = 0.275, Of, = 0.25); (c) an
asymmetric vacillation (k = 0.226, 8:. = 0.26); (d) a double asymmetric vacillation
(k = 0.2, 09, = 0.26); (e) a double symmetric vacillation (k = 0.26, &, = 0.175)
generated in the model.
FIG24 d, e . See facing page for legend.
the model are invariant for a change of sign of the second mode variables, the
variables of the first mode being left unchanged, there also exists an other
unsymmetric vacillation cycle whose representation would be the image of
Fig. 24a with respect t o the ot,41s ,-axis. The initial conditions of integration
decide which one of the possible circulations is ultimately realized. The
same, of course, applies to each curve of Fig. 24.
When the rotation rate is increased, the symmetric vacillation, discussed in
Section 5.3, develops. The flow represented in Pig.' 24b corresponds to
,
0:. = 0.25 and k = 0.275 and has a period of BOt, .
At still higher rotation rates, the model performs an asymmetric vacillation
,
showninFig. 24c, corresponding to 0:. = 0.25, k = 0.225, andaperiod of 40t,.
I n this flow, t,h2, changes sign during the cycle but reaches extreme values
of different moduli. The characteristics of the curves associated with the
energy processes of the asymmetric vacillation are similar to those of the
symmetric vacillation except that there are now two different maxima
(and two different minima) on each curve.
For 0:,,=0.25 and k=0.2 the model exhibits still another type of
vacillation, shown in Fig. 24d, which may be called a double asymmetric
178 A. QUINET
,
vacillation, the period of which is 67t,. At 0:. = 0.25 and k = 0.15, the
single asymmetric vacillation has reappeared (not shown in Fig. 24).
In the course of this investigation, keeping the thermal forcing constant
el,, = 0.25, the amplitude of the vacillation increases with increasing
rotation rate. On the other hand, the period of vacillation, referred to one
loop, decreases when the rotation rate increases.
For the three experiments performed at k = 0.25, k = 0.225, and k = 0.2
,
with a weaker thermal forcing 0;. =0.175, the model exhibits a do&&
symmetric vacillation shown in Fig. 24e, which corresponds to k = 0.25. The
period of vacillation is 1396,. Comparing this period with the corresponding
,
case at 0:. = 0.25 indicates that, as could be expected, the reduction of the
heat to be transported associated with the reduction of the thermal contrast
at the lateral boundaries induces a longer vacillation period.
At still lower thermal forcing, the numerical experiments using the wave
?a = 3 would become questionable since experimentation and theory both
show that the flow is then controlled by shorter waves.
To sum up, the model is able to simulate at least five different kinds of
vacillation: (a) unsymmetric vacillation, (b) symmetric vacillation, (c)
asymmetric vacillation, (d) double asymmetric vacillation, (e) double sym-
metric vacillation. The analytic description of tliese flows, if ever possible,
would presumably be increasingly complex. The driving energy mechanisms,
however, can be reasonably believed to be of the same kind. It should
nevertheless be kept in mind that the relative as well as the absolute impor-
tance of the various terms of the energy budgets (4.79)-(4.82) is modified
when varying the thermal forcing and the rotation rate. I n this respect, it is
interesting to note that the mean value flz of C, during one vacillation period
drops considerably with increasing rotation rates. For instance, at the given
,
thermal forcing 0: = 0.25, flz at k = 0.15 is only one half of fl, at k = 0.275.
This confirmsthe qualitative argument mentioned in Section 2, that the higher
the rotation rate, the more geostrophic will be the flow. Moreover, it is
observed that the influence of the nonlinear interactions between the scales
(2.0), (1, 3)) and (2, 3) increases with the rotation rate. Accordingly, at high
rotation rates, the curves of A , , A , , K , , and K , become less correlated
with each other, and the respective times of occurrenceof their extreme values
exhibit more or less important shifts.
6. VACILLATION
IN TEE ATMOSPHERE
6.1. Potential Energy Vaeilhtion i n the Atmosphere

Pfeffer et al. (1969) have published striking analogs between atmospheric
and laboratory flows, especially regarding the index circulation changes.
Figure 25, borrowed from the Pfeffer et al. (1969) paper, reproduces the
High index circulation
DAY 1
Low index circulation
DAY 15
FIG.25. Comparison between an atmospheric index circulation change (after Winston
and Krueger, 1961) and the extreme stages of an amplitude vacillation generated in a
rotating annulus experiment (courtesy of Pfeffer and colleagues, 1969).
500 mbar charts of 26 December 1959 and 10 January 1960, characteristic

of high and low index circulation, respectively. The extreme stages of an
amplitude vacillation in a rotating annulus experiment are shown to the right
side of these charts. The interval between the two situations, either in the
atmosphere or in the annulus, is 14 revolutions. The complete modification
of the flow, from nearly symmetrical to strongly lobed waves, appears dis-
tinctly both in the atmosphere and in the laboratory experiment.
180 A. QUINET
6.2. Kinetic Energy Vwillation in M e Atmosphere

The routine northern hemisphere weather maps a t the 500 mbar level
during the period 4 February 1969-22 February 1969 suggest the existence
of some vacillating behavior on a well-defined four-wave pattern of the
geopotential field. I n order to further investigate this situation the geo-
potential data were analyzed by the least squares spectral method (El'iasen
and Mackenhauer, 1965)using even spherical harmonics up to (m= 9, n = 9).
The maps reproduced in Fig. 26 were then drawn by retaining the zonal
components (2, 0 ) , (4, 0 ) , and (6, 0) and only the four-wave harmonics
(4, 4), (6, a), and (8, 4) from the set of scales involved in the analysis. The
interval between two successive isolines in Fig. 26 is 80 geopotential meters.
The altitudes of the low (L) and high (H) pressure centers are also shown.
There is a striking similarity between these maps and those of a vacillation
cycle. The " square '' four-wave of the first map (4February 1969)transforms
into a well marked " lobed " pattern on the 12th and is reestablished on the
last map (22 February 1969) so that the period of the cycle is 18 days. At the
same time the axis of the wave, initially oriented approximately in the S-N
direction (no tilt north of 50" latitude) becomes SE-NW (westward tilt)
with a maximum tilt on the loth, then changes to SW-NE (eastward tilt)
with a maximum tilt on the 16th and, finally, the S-N direction of the wave
axis is reestablished again on the 22nd. During this time, the wave progresses
westward by approximately three quarters of the wavelength (-70" of
longitude). It can also be easily recognized that the high index circulation of
4 February 1969 exhibits more intense midlatitude zonal flow than the low
index circulation prevailing on 12 February 1969. On the other hand, the
circulation is dominantly zonal both on 4 February 1969 and on 22 February
1969 but weaker on the 22nd, after which the cycle does not reappear.
It would be extremely useful to pursue the large-scale synoptic, aerological
analysis of the flow patterns of Fig. 26 and, in particular, t o identify s i g d -
cant characteristic features on the first map (4 February 1969) in order t o
define initial condition criteria leading to such an evolution in the atmos-
phere, This would have very interesting practical implications. Indeed,
although it is not possible to derive precise synoptic information from such
very large-scale patterns, it should then be possible to forecast, in a very
economic way, the main features of the large-scale synoptic evolution for a
rather extended period of time. This, for instance, would allow one to know
the sequence of weather types that would prevail during two weeks or more
(18 days in the above given example).
The difference between the flow patterns on 4 February 1969 and 22
February 1969 and between their respective further evolution seems t o
indicate that the appearance of vacillation in the atmosphere depends
Ill W.0Z.IZ OD1
FIG.26. Maps drawn in retaining the zonal harmonic components (2, 0 ) , (4,0 ) , (6, 0 ) ,
and the four-harmonics (4, 4),( 6 , 4 ) , and (8,4), of the 500 mbar geopotential field during
the period 4 February-22 February 1969 and showing the existence of an atmospheric
kinetic energy vacillation cycle.
182 A, QUINET
critically upon the intensity of the zonal flow. Returning to Section 5 of this
monograph, there is no doubt that the meridional profile of the zonal wind is
also of importance in this respect.
6.3.Conclusion
The two examples of Figs. 25 and 26 amply demonstrate that vacillation
is not only a characteristic laboratory flow closely related to the large-scale
atmospheric circulation but has also an intrinsic pragmatic aspect. A clear
synoptic identification of initial conditions leading to atmospheric vacillation
would be a first step in making economic long range weather predictions, a t
least in favorable circumstances. Moreover, it is believed that a highly
truncated numerical weather prediction model would provide, in a rather
short computation time, a meaningful objective long range forecast of the
large-scale atmospheric flow.
ACKNOWLEDQMENTS
I am indebted to Professor J. Van Mieghem for suggesting writing this monograph

and for careful reading and commenting on the manuscript. Let me also acknowledge
the benefit received from many helpful discussions with Professors P. Defrise and J.
Van Isacker. The maps of Fig. 26 have been drawn by using a subroutine program
provided by the author's colleague, W. Struylaert.
I am most grateful to Professors D. Fultz and R. Pfeffer and to Dr. W. Fowlis, as
well as to their publishers, who have granted me permission to reproduce previously
published illustrations.
LISTOF SYMBOLS
A Available potential energy Transfer rate of K, into K,
of the system Conversionrate of A, into K Z
A , Eddy available potential 2rap,g-'L4fa; is used as
energy energy unit
A , Zonal available potential Specific heat of dry air at
energy constant pressure
+
a [(3/4)" (1/4)K]/2N 0.797 Interaction coefficient [see
- a t a Eigenvalue of the Laplacian Eq. (4.20)]
operator in Cartesian Dissipation rate of K E by
coordinates friction
B b-lc-lf Dissipation rate of Kz by
b [(3/4)' - (1/4)E]/2N 0.124 friction
C A Transfer rate of A z into A, Contrav&riant and covariant
resulting from sensible heat components of the velocity
advection of the reference body with
CAR Transfer rate of Az into A, respect to an absolute
resulting from changes of reference body (entrain-
static stability in time ment velocity)
CE Conversion rate of A, into K , e-symbol [see Eq. (3.36)]
2P Coefficient of friction a t of the reference body with

surface po respect to the coordinate
F' Coefficient of friction at system
surface pz Temperature or nondimen-
F, Basic function of the spectral sional contribution of scale-
representation interaction effects to baro-
f Coriolis parameter (vertical tropic energy transfer rates
component of the vorticity [see Eq. (4.100)]
of the absolute motion of Nondimensional contribu-
the reference body) tions of scale-interaction
a Generation rate of total effects to baroclinic energy
potential energy conversion and transfer
GE Generation rate of A , by rates [see Eqs. (4.106)-
external heating (4.lOS)]
G, Generation rate of A, by Temperature at the inner
external heating wall of a n annulus
GER Generation rate o f A , due Temperature at the outer
to the change of the wall of an annulus
reference state in time Time
GZR Generation rate of A, due Nondimensional time unit
to the change of the to =ft
reference state in time Zonal wind
9 Acceleration of gravity x-component of the isobaric
9 0 9 9" Metric tensor and its reci- wind
procal Relative velocity vector of a
2H Coefficient of heating at the material point with respect
lower layer to a reference body
H' Coefficient of heating a t the Contravariant and covariant
upper layer components of V
I Internal energy of the system y-component of the isobaric
K Relative kinetic energy, per wind
unit mass in Section 3.2, Contravariant and covariant
for the whole system else- components of the velooity
where of a material point with
KE Eddy kinetic energy respect to the coordinate
Kz Zonal kinetic energy system
k Ff -1, and is used as a Space coordinate (a= 1, 2, 3)
measure of rotation rate Cartesian coordinates in a
L Length scale plane surface
m Meridional wavenumber or Determinant of qn
mode Isobaric divergence .of the
n Longitudinal wavenumber wind
P Potential energy of the Kronecker tensor [see Eq.
system (3.1111
P Pressure Vertical component of the
R Specific gas constant for dry relative vorticity
air or the reference body Vertical component of the
Ro Rossby number absolute vorticity (7= 5
ra, r , Contravariant and covariant +f)
components of the velocity Potential temperature
184 A. QUINET
Temperature of the under- OPERATORS

lying surface apx= = a, Partial derivative with res-
K Rlc, pect to the space coordin-
ate xa
P Density of the fluid
Partial derivative with res-
U A measure of the static
stability [see Eq. (3.77)] pect to time
Derivative with respect t o
7 Stream function of the
time along the flow
vertical wind shear
1 Integration symbol
c Potential function of the
body forces
dm Mass element of the fluid
Element of horizontal area
9Jm. n Basic function of the spectral Laplacian of X [see Eq.
representation (3.4211
X Velocity potential of the Jacobian of X and Y [see
isobaric wind Eq. (3.43)]
a Angular velocity of rotation Differential Q-operator [see
of the reference body Eq. (3.44)]
S ( X , Y ) Differential S-operator [see
w dP/dt Eq. (3.46)]
wi Nondimemional spectral s Local derivative of X with
component of Vax [see respect to to (2= axlato)
Eq. (4.12')] x Horizontal average of X
2w.4 Vorticity tensor of the abso- [XI Zonal average of X
lute motion of the reference X' Fluctuation of X with respect
body to [XI (X’= x - [XI)
INDICES
I n the tensorial expressions, the Roman indices take the values 1 and 2, the Greek
indices take the values 1 to 3. They are summation indices when repeated in the same
term.
In Section 3, subscripts 0, 2, and 4 refer respectively to levels p, = 100 cbar, p, = po/2,
end p4 = 0 cbar; subscripts 1 and 3 refer to the intermediate levels p, = 3po/4 and p, =
po/4, respectively.
Subscripts to the scalar dependent variables #, T , 0, u denote spectral components
[see Eqs. (4.7)-(4.10)].
REFERENCES
Bolin, B. (1962). Studies of the general circulation of the atmosphere. Advan. &ophys.
1, 87-118.
Charney, J. (1947). The dynamics of long waves in a baroclinic westerly current. J.
Meteorol. 4, 136-163.
Davies, T. (1963). The forced flow of a rotating viscous liquid which is heated from
below. Phil. Trans. Roy. SOC.London, Ser. A 246, 81-112.
Davies, T. (1969). On the forced motion due to heating of a deep rotating liquid in an
annulus. J . Fluid Meoh. 5, 693-621.
Defrise, P. (1964). Tensor calculus in atmospheric mechanics. Advan. Geophya. 10,
261-314.
D6i)s, B. (1969). The influence of the large-scale heat sources on the dynamics of the
ultra-long waves. Tellue 21, 26-39
Eady, E. (1949). Long waves and cyclone waves. Tellw, 1, 35-52.

Eliasen, E., and Mackenhauer, B, (1965). A study of the fluctuations of the atmospheric
flow patterns represented by spherical harmonics. Tellus 17, 22&238.
Fjortoft, R. (1953). On the changes in the spectral distribution of kinetic energy of two-
dimensional non-divergent flow. Tellus I, 225-230.
Fowlis, W., and Hide, R. (1965). Thermal convection in a rotating annulus of liquid:
Effect of viscosity on the transition between axisymmetric and non-axisymmetric
flow regimes. J. Atmoa. Sci. 22, 642-558.
Fultz, D. (1949). A preliminary report on experiments with thermally produced lateral
mixing in a rotating hemispherical shell of liquid. J . Metwrol. 6 , 17-33.
Fultz, D., Long, R., Owens, G., Bohan, W., Kaylor, R., and Weil, J. (1959). Studies of
thermal convection in a rotating cylinder with some implications for large scale
atmospheric motion. Meteorol. M a o g r . 4, 1-104.
Fultz, D., Kaiser, J., Fain, M., Kaylor, R. E., and Weil, J. (1964). Experimental investi-
gations of the spectrum of thermal convective motions in a rotating annulus. I n “ Re-
search on Hydrodynamic Analogues of Large-Scale Meteorological Phenomena,”
pp. 2B1-2B89. Final report, Dept. Geophys. Sci., Univ. of Chicago, Chicago,
Illinois.
Hide, R. (1953). Some experiments on thermal convection in a rotating liquid. Q m r t .
J. Roy. Meteorol. Soc. 79, 161.
Hide, R. (1958). An experimental study of thermal convection in a rotating fluid. Phil.
Trans. Roy. SOC.London, Ser. A 250, 441-478.
Holton, J. (1972). “An Introduction to Dynamic Meteorology,” Int. Geophys. Ser.
Academic Press, New York.
Krueger, A., Winston, J., and Haines, D. (1965). Computation of atmospheric energy
and its transformation for the northern hemisphere for a recent five-year period.
Kuo, H. L. (1949). Dynamic instability of two-dimensional non-divergent flow in a
barotropic atmosphere. J . Meteorol. 6 , 105-122.
Kuo, H. L. (1954). Symmetrical disturbances in a thin layer of fluid subject to a hori-
zontal temperature gradient and rotation. J . Meteorol. 11, 399-41 1.
Kuo, H. L. (1967). Further studies of thermally driven motion in a rotating fluid. J .
Meteorol. 14, 553-558.
Langlois, W., and Kwok, H. (1969). “Description of the Mintz-Arakawa Numerical
General Circulation Model,” Tech. Rep. No. 3. Dept. Meteorol., University of
California, Los Angeles.
Lorenz, E., (1953). A proposed explanation for the existence of two regimes of flow in
a rotating symmetrically heated cylindrical vessel. I n “Fluid Models in Geophysics,”
pp. 73-80. Johns Hopkins Press, Baltimore, Maryland.
Lorenz, E. (1955). Available potential energy and the maintenance of the general circula-
tion. Tellus 7 , 157-167.
Lorenz, E. (1960). Energy and numerical weather prediction. Tellus 12, 364-373.
Lorenz, E. (1962). Simplified dynamic equations applied to the rotating-basin experi-
ments. J. Atmos. Sci. 19,39-51.
Lorenz, E. (1963). The mechanics of vacillation. J . Atmos. Sci. 20, 448-464.
Margules, M. (1903). Uber die Energie der Stiirme. “Jahrb. kais.-kon. Zent. Meteorol.”
[transl. by C. Abbe insmithson. Misc. Collect. 51,633-595 (1910)l.
Merilees, P. (1968). On the transition from axisymmetric and non-axisymmetric flow
in a rotating annulus. J. Atmoa. Bci. 25, 1003-1014.
Merilees, P. (1972). On the periods of amplitude vacillation. J. Meteorol. SOC.J a p . [2]
50, 214-225.
186 A. QTJINET
Namias, J. (1954). Quasi-periodic cyclogenesis in relation to the general circulation.

Tell- 6, 8-22.
Pfeffer, R., and Chiang, Y. (1967). Two kinds of vacillation in rotating laboratory
experiments. Man. Weather Rev. 96, 75-82.
Pfeffer, R., and Fowlis, W. (1968). Wave dispersion in a rotating, differentially heated
cylindrical annulus of fluid. J . Atmoa. Sci. 25, 361-371.
Pfeffer, R., Mardon, D., Serbenz, P., and Fowlis, W. (1966). “A new Concept of Avail-
able Potential Energy,” Rep. No. 66-1. Dept. Meteorol., University of Florida,
Tabhassee.
Pfeffer, R., Fowlis, W., Buzyna, G., Fein, J., and Buckley, J. (1969). Laboratory studies
of the global atmospheric circulation, Proc. Stanstead Semin. 8th. Meteorol. 89,
89-91.
Phillips, N. (1965). The equations of motion for a shallow rotating atmosphere and the
“traditional approximation.” J . Atmos. Sci. 23, 626-628.
Quinet, A. (1973a). The structure of non-linear processes. Tdlwr 25, 636-544.
Quinet, A. (1973b). Non-Linear mechanisms in a nonconservative quasi-geostrophic
flow which possesses 30 degrees of freedom. Tellw, 25, 545-559.
Roeaby, C. (1926). On the solution of problem of atmospheric motion by means of
model experiments. Mon.’ Weather Rev. 54, 237-240.
Rossby, C. (1947). O n the distribution of angular velocity in gaseous envelopes under
the influence of large-scale horizontal mixing process. Bull. Amer. Meteorol. SOC.
28, 53-68.
Starr, V. (1948). An essay on the general circulation of the earth‘s atmosphere. J .
MetenraZ. 5, 39-48.
Statistiques Quinquennales. (1971). “ Observations &rologiques-station d’Uccle.”
Inst. Roy. Meteorol. Belg. 1971. Uccle.
Stessel, J. P. (1969). Le freinage de l’atmosphhre par 1’ocBan. Ph. D. Thesis, Univ.
libre de Bruxelles, Bruxelles.
Thompson, P.. (1961). “Numerical Weather Analysis and Forecasting,” Macmillan, New
York.
Van Iwker, J. (1963). Conservation de la rotationnelle absolue et de 1’6nergie dans les
modhlea atmosph6riques. Contpib. Int. Symp. Dyn. Large Scale Proceaa, 1963 (unpub-
lished).
Van Mieghem, J. (1952). Energy conversion in the atmosphere on the scale of the
general circulation. Tdlw 4, 334-351.
Van Mieghem, J. (1957). Energies potentielle et interne convertibles en energie cin6tique
dans I’atrnosphBre. Bear. Phys. Frei. Atmos. 30, 5-17.
Van Mieghem, J. (1973). “Atmospheric Energetics,” Oxford Monographs on Meteorology.
Oxford Univ. Press (Clarendon), London and New York.
Van Mieghem, J., and Vandenplas, A. (1950). Les Bquations de la dynamique atmos-
phBrique en coordonnBesgAnBralisqes. Application a u cas des coordonn&s sphbriques.
Inat. Roy. Meteorol. Belg., Mem. 41, 1-55.
Winston, J., and Krueger, A. (1961). Some aspects of a cycle of available potential
energy. Mon. Weather Rev. 89, 307-318.
Young, J. (1968). Comparative properties of some time differencing schemes for linear
and non linear oscillations. Mon. Weather Rev. 96. 357-364.
FILTER TECHNIQUES IN GRAVITY
INTERPRETATION
Frans De Meyer
Royal Belgian Meteorological Institute. Uccle. Belgium
.
1 Introduction .......................................................... 187
.
2 Convolution Filtering .................................................. 189
2.1. Definition of Convolution Filtering .................................. 189
2.2. The Frequency Response .......................................... 191
2.3. Linear. Discrete Filters ............................................ 192
2.4. Construction of Linear. Discrete Filters in the (2.y)-Plane ............ 195
2.5. Construction of Linear. Discrete Filters in the (u.w)-Plane .............. 200
3 . Upward and Downward Continuation of the Surface Gravity Effect .......... 203
3.1. Definition of the Surface Uravity Effect .............................. 203
3.2. The Upward Continuation a8 a Filtering Operation .................... 205
3.3. Construction of a Numerical Upward Continuation Filter .............. 210
3.4. The Downward Continuation as a Filtering Operation .................. 213
3.6. Construction of a Numerical Downward Continuation Filter ............ 219
4 . Frequency Filtering .................................................... 226
4.1. Procedure for Filtering in the Frequency Plane ........................ 226
4.2. Optimal Wiener Filtering .......................................... 228
4.3. Strakhov’s Method for Extraction of Potential Field Signal ............ 233
4.4. Digitization of a Continuous Field-Aliasing .......................... 237
4.5. Estimation of the Power Spectrum .................................. 240
4.6. Convolution Filtering versus Frequency Filtering ...................... 244
.
5 Calculation of Derivatives of Higher Order ................................ 248
Appendices ............................................................ 250
List of Symbols ........................................................ 254
References ............................................................ 256
1. INTRODUCTION
One of the most important problems in the interpretation of gravity meas-
urements is that of separating a surface field. derivable from a potential. into
independent components and of ascribing separate geological structures to
these parts . The geological interpretation of gravity anomalies therefore
consists largely of estimating the positions and shapes of the disturbing
masses in the upper part of the earth’s crust and of determining the density
structure of the area in which they are imbedded .
I n this respect gravity interpretation uses the whole of the following
procedures: (1) A gravity survey yields values of the gravitational field
over a limited part of the earth’s surface. (2) Bouguer corrections are applied
to allow for the topography of the survey area and for the attraction of the
187
188 FRANS DE MEYER
earth’s spheroid; the residue, the Bouguer anomaly, reflects the hetero-
geneous structure of the upper part of the crust. (3) The irregularly spaced
gravity data are interpolated onto a regdar (rectangular or square) grid,
for ready entry to a computer, and are automatically contoured. (4)A
regional-residualanalysisis applied, which separates the field into components
of larger and smaller extent. ( 5 ) Several interpretation techniques are used
for obtaining a better insight into the composition of the field, such as filtering
the surface field, calculationofthe vertical derivatives, upward and downward
continuation. (6) A model-fitting technique is applied in order to obtain a
geological model that could produce the observed part of the surface field.
It is a fact that a gravimeter, measuring the variations in the vertical
component of a potential field, is sensitive enough to register the local effects
of many types of geological configurations. Frequently the data are subjected
to elaborate reductions and inaccurate numerical approximation methods;
consequently they are not exact and are seldom precise. Finally we have a
map and a distribution of data on a regular grid, giving a more or less com-
plete picture of the behavior of the field a t the earth’s surface and of the in-
ternal mass distribution in the earth’s crust.
Even if the surface field were known very accurately, there remains an
inherent lack of determinacy of the source. Indeed, any gmvitational field
possesses two characteristics, preventing a unique interpretation. The finst
difficulty arises from the fact that the reduced anomaly at each observation
point undergoes the influence of a very complex mass distribution in the
upper part of the crust, and the attraction of local, shallow structures is
often seen as a small deviation in the broad picture of the regional features
of the map. The effects caused by these disturbing masses are of course in-
distinguishable, but usually we can assume that they are independent of one
another, stating implicitly that they are considered as being random, affecting
individually only a single observation point.
The second difficulty results from the property t>hatany gravity field,
derivable from a potential, implies the inherent ambiguity in defining the
source of the potential field (Skeels, 1947). The distribution of the field in free
space can be calculated uniquely, since the upward continuated field satisfies
Laplace’s equation, with known Dirichlet’s conditions in the earth’s surface.
However, if knowledge of the field at a given depth is desired, then we require
the solution of Poisson’s equation with unknown density function. It is well
known that the gravitational field is not in itself sufficient to define the mass
distribution which produces it: for a given gravity field, measured at the
earth’s surface, an infinite number of mass distributions can be found, all
accounting for that field. Even within the limits of some known physical
parameters an infinite variety of solutions is possible and no degree of pre-
cision or amount of data w ill remove this fundamental ambiguity.
FILTER TECHNIQUES IN QRAVPTY INTERPRETATION 189
Therefore it is evident that a less ambiguous interpretation of a measured

gravity field must rest on data of another kind, which are mostly geological
and seismological. However, this does not prevent us from trying to perfect
the interpretation techniques and to give the computational methods a more
rational basis.
2. CONVOLUTIONFILTERING
2.1. DeJinition of Convolution Filtering
We consider the class 8 of the continuous, indefinitely derivable functions
f (2,y) of the coordinates x and y, with the property that the following norm
is finite
(2.1) llflla= lim (1/4T,T,)ry If(x,y)Iadxdy<m fez

Tt. Ty+ m -Ty -TI
Note that the functions of 8 are not necessarily quadratically integrable, as
we see from the trivial case f (2,y) = 1.
A linear filter is, by definition, an operator which associates a function
g(x, y) of 8, called output or response, with any function f (2,y), input or
signal, of 8. Formally we write
(2.2) 9(G Y) = F{f(x, Y))
where .Fis a specified operation on the functions of 8.
Two properties are required €or the operator 9:
(1) Linearity or superposition principle: if for each y) the output of
the filter is
(2.3) gk(z, y) = g{fk(2, Y))
then we must have for any sequence of constants ak , 1 < k < n,
This simply states the fact that the output of the linear filter 9 from a sum
of inputs is equal to the sum of the outputs of the filter, with each input
applied separately. *
(2) The operator .F commutates with the translation operator r e ,11 defined
by its action on a function f (2,y) of 8
(2.5) TC. , f h Y) =f (z - & Y - rl)
which means that
(2.6) 9 7 ~n =. 7 t en .F
This property implies that the filter behavior is independent of the origin of
the (2,y)-plane: if g(x, y) is the response of the filter to a n input f (2,y),
190 FRANS DE MEYER
then g(x -6, y - q ) will be the response to the input f(x - [, y - q), for
any 6 and q. Indeed, applying both sides of Eq. (2.6) to the input f (x,y)
gives
9% )) = %{f (Z - I ,Y - $1
n{f(X, Y
and
Tt. R R{f(Z,Y)) =7t, ng(x3 Y) = 9(x - r, Y - q)
Equality yields
~b(f(Z--B,Y--rl)}=g(x’:6,Y -71)
I n the theory of gravitational interpretation the filter % is usually defined
by a two-dimensional convolution integral of the form
where the weighting function h(x, y) is called the impulse response or filter
function and it describes the behavior of the filter in the coordinate plane
(x,Y).
If we let the W a c delta distribution S(x,y) be a member of the class Y
and if we apply &x, y) as an input to the filter (2.7), then it immediately
follows that
g(z, Y) = IW
jmh(5,rl)
-m - w
- 6 Y - 9) d17 = h(X7 Y)
Therefore we conclude that the filter function h(x,y) may be regarded as the
response of the filter to the Dirac impulse function S(x, y).
It often happens that the impulse response is circularly symmetrical, so
that h(s,y) is a function of the radius variable r
(2-8) h(x, y) = h(r) where r2 = x2 + y2
Transforming t o polar coordinates (z- 6) + i( y - q) = rei6, Eq. (2.7) can be
written in the equivalent form
g(z, y) = fm fn h(r cos 8, r sin e)f(x - r cos 8, y - r sin 0)r dr de

0 0
since the convolution product is commutative. If the filter function is a

function of the radius variable only, it follows from (2.8) that
where
FILTER TECHNIQUES M GRAVITY INTERPRETATION 191
is the average of the functionf (x,y) in the coordinate plane (2, y) on the circle
of radius T with center a t the point (2,y).
2.2. The Frequency Response

An equivalent form of the input-output relation of a convolution filter
results from the use of the Fourier transform: i f f (2, y) is a member of the
class 9, then we assume that its Fourier transform exists and is given by
with inverse relation
(2.12) f (2, y) = (1/4.rra) lm

P(u, v)ef(Uf+uY)
-m
du dv
where 4271 and v/27r are the frequencies in cycles per unit length in the x-
and y-directions, respectively. If we assume that the Fourier transforms
F(u, v), G(u, v) and H(u, v) of the respective input, output, and fdter function
of the convolution filter (2.7) exist, i.e. if the functions of the class 9 satisfy
Dirichlet's conditions in any finite domain and are absolutely integrable,
then we obtain from the convolution theorem of Fourier transforms the
following relation between the signal and response
(2.13) G(u, W) = H(u, v)F(u,
V)
where
(2.14) ~ ( uv),= J-m

m
I
m
--a)
n(x,y)e-*(Uz+"u)dz dy
is known as the frequency response, characterizing the filter behavior in the

frequency plane (u, v); H(u, w ) tells us the portions of the Fourier spectrum
of the input which will be amplified or attenuated by the fdtering operation.
I n this connection we must remark that, in a strictly mathematical sense,
the Fourier representations of the input and output are not necessarily
meaningful. Indeed, in order t o have a definite Fourier transform these fields
need to be absolutely integrable over the (x,y)-plane; any function which
goes to zero for large x and y and which does not contain singularities is
absolutely integrable, but a gravity field does not vanish for large x and y.
The problem is apparent: since we suppose that the input field is a realization
of a two-dimensional stochastic process, we have the property that it does not
vanish outside a finite region of the (2,y)-plane. To overcome this mathe-
matical difficulty we must assume that the input field wiIl be defined over a
large, but bite area of the coordinate plane and that it becomes zero outside
192 FRANS DE MEYER
this domain. Otherwise it would be imperative to work with Wiener’s general-

ized Fourier transform,.thus making the mathematical expressions much more
complex.
It may be readily proved that the frequency response H(u,v) is a circularly
symmetrical function of the radial frequency p if the filter function depends
only on the radius variable r . Indeed, transforming to polar coordinates
+ +
x iy = refeand u iv = pel’, it follows from Eq. (2.14) that
Using the relation

(2.15) s,”” etpcos@ = 2T ~,(p)
(Sneddon, 1951, p. 515), where J , ( p ) is the zero-order Bessel function, we

obtain
m
(2.16) H ( p ) = 27r h(r)Jo(rp)rdr
0
with inverse relation
These relations are known as the zero-order Hankel transforms.
2.3. Linear, Discrete Filters

In the practical case, the data are sampled on a rectangular grid, with
sampling intervals A x and Ay in the x and y directions, respectively. For
regularly spaced observations,the convolution integral (2.7) is approximated
by the following sum
and the continuous filter function h(x, y) is now replaced by a discrete set of
filter weights hkg.In this case the frequency response of the discrete filter
(2.18) is given by the analogous form of (2.14)
M N
(2.19) H,(u, u)= A x A y 1 1- Nhk,e-t(kuAs+
k= -M 1=
IUAy)
Another point of view of discrete filtering results from the application of an

adequate numerical integration method t o Eq. (2.9);the output of the linear
filter is then approximated by a weighted sum of the form
N
(2.20)
FILTER TECHNIQUES IN GRAVITY INTERPRETATION 193
1 I 1 I I l ~ l l l - l - - t l i I I 1 1 1
FIG.1. Square integretion grid in the (5,y)-plane.
with weights h k , and R, the radii of circles passing through the points of
intersection of a square integration grid (see Fig. 1).
We now derive the explicit form of the frequency response of the discrete
filter, as defined by Eq. (2.20). Substituting (2.10) into (2.20) we obtain
Taking Fourier transforms of both sides of this expression we have
exp[ -iR,(u cos B + v sin B)] dB

+
Transforming t o polar coordinates u iv = pei" and using Eq. (2.15) we
obtain for the frequency response of the discrete filter (2.20)
N
(2.21) H,(u, V)= C hkJ0[Rk(u2+ v ~ ) " ~ ]
k=O
Consequently, the averaging process over the circle of radius r defines a

linear filter with frequency response
(2.22) H,(u, v) =Jo[r(u2 + v2)1’2]
194 FRANS DE MEYER
Since the frequency response (2.21)depends only on the radial frequency p,

the filtering effect of Hd(u,u ) is independent ofthe direction in the (u,v)-plane.
But this is an oversimplification of the filtering problem for data on a square
grid; indeed, in the derivation of the form (2.21) for the numerical frequency
response of the discrete filter (2.20) we have implicitly assumed that the
circle averages can be taken from an infinity of values, since these are given
by Eq. (2.10). However, the practical situation is such that the input field
f(x, y) is known on a regular grid and there is only a finite, discrete set of
observations available for the analysis; in consequence, the corresponding
frequency characteristic of the discrete filter will be a function of the number
of points on the integration circles and of the distribution of the grid points
on these circles.
Therefore we have to replace the expression (2.20) by
N 1 nk
where nk is the number of points on the circle of radius Rk , passing through

the grid points with coordinates (xfk,yjk),relative to the point (x,y). Hence, if
the points are distributed equidistantly over the circle of radius Rk , we have
(2.24) x j k = R, cos(2~j,/n,), yjk = Rk sin(2rjk/nk)
Taking Fourier transforms of both sides of Eq. (2.23) we find
(2.25) Q(u,v) = H*(% v) F(u,v)

where now Hd(u,v) is defined as the frequency response of the discrete filter
with weights hk , for a square integration grid with a finite number of grid
points on the integration circles,
(2.26)
For a square grid, nk is a n even number; every grid point on the circle of
radius Rk defines another point, having coordinates of opposite sign. There-
fore the sum in Eq. (2.26) can be written as
or,
(2.27)
I n conclusion, one can say that the frequency response of the fllter, that
consists of averaging f ( x , y) over a finite number n, of points on a circle of
radius r , is given by
nria
(2.28) H J u , v) = (2/n,)
f=l
+
cos(u~, vy,), +
xj2 y? = r2
The practical presentation of the problem already produces a n intrinsic

smoothing: the fact that the continuous field f (x,y) is sampled on a square
grid (Ax = Ay = so) eliminates the high frequency response of the input
field. Indeed, from the sampling theorem (Bendat, 1958) it follows that the
highest frequencies, left in the data after the digitalization of the continuous
field, are given by
(2.29) 1% I = 1% I = .rr/so
known as the Nyquist frequencies, associated with a sampling on a square
grid with spacing so, and the maximum radial frequency is then
(2.30) pN =2/5TlS0
These are the cutoff frequencies, induced by the digital profile and they are
controlled by the grid spacing.
Figure 2 shows the differences between the forms (2.22) and (2.28) in the
Nyquist domain 0 5 u 2 u N ,0 < v < v, of the frequency plane for sr= 4
and n, = 8 points, distributed equidistantly over the circle of radius r.
It was found that the differences are less than 0.01 % over the Nyquist
domain for n, = 12 and n, = 16 points on the circles and this result is nearly
independent of the distribution of the points on the circles. From Fig. 2 it
follows that the largest deviations for n, = 8 points is 1.7 o/o in the point (m, T)
and significant deviations occur a t four-point averaging (about 30 %). I n
consequence one may expect that averaging over eight and more points
yields a representative estimation of the true circle average; a, significant
deviation of the exact circle average occurs a t four-point averaging and these
deviations are mainly situated in the high frequency region of the Nyquist
domain.
2.4. Construction of Linear, Discrete Filters in the (2,y)-Plane
I n this section the construction of a discrete filter of the form (2.20)

with weights h, by an algorithm that works entirely in the coordinate plane
will be considered. Suppose that the frequency response of an ideal low-pass
filter
H(p)#O for O<p<p,
=O for p>pc
U
I 2 3 K
(a)
I
I I I
1 2
u 3 1
(b)
FIG.2. Deviations for (a) n, = 4 points, contour interval 2 %; (b) n, = 8 points, con-
tour interval 0.1 %.
is approximated by
m P ) = H(P) for OIpIp,,

(2.31) H(p)kcc(p) for PC < p 5 P k
=O for pk <p,
for some cutoff frequency pc; pk is the radial frequency where we want R(p)
to become zero. The introduction of the kernel k,(p) enables us to avoid the
discontinuity a t the cutoff frequency pc (see Fig. 3).
0 PC pk P
FIG.3. Hypothetical low-pass fllter.
For the function k,(p) one can use one of the following kernels:
FQjerkernel U p ) = [sin 4 P - P C ) b sin(p - PC)l2
Jackson kernel U p ) = [sin .(P - PC)b s i d p - Pc)l4
(2.32)
Abel-Poisson kernel k,(p) = 1/{1 + [a(p - pc)]'}
Weierstrass kernel k,(p) = eXp[-E(p - pc)']
The corresponding filter function is the Fourier transform of the frequency

response; using Eq. (2.17) we obtain
Pk
198 FRANS DE MEYER
where r2 = xa + y2. This can also be written as
It follows immediately that A(x,y) is only a function of the radial coordinate

r in the (x,y)-plane. This filter function is seen to be infinite in length; for
practical use it must be shortened to a finite length. Therefore we introduce
an even function d,(r), frequently called a data kernel or covariance kernel
in the theoryof the spectral analysis of time series, and subject to the restric-
tions that d,(O) = 1 and d , ( r )= 0 for r > R . The choice of the parameter R
will be discussed later. Hence the modified impulse response is defined
(2.36) +
M z ,y) = h(x,Y) d , [ ( s 2 Y2)1’21
+
having the property that hr(x, y) = 0 for (x2 y2)1/2> R. A summary of
some frequently used data kernels is:
Box car kernel do(r)= 1, r <R
=0, r>R
Bartlett kernel d1(r) = [1 - ( I r l /R)1 do(r)
(2.36)
Hanning kernel d&) = +2[1 + cos(.rrr/WIdO(4
Hamming kernel d3(r)= [0.54 + 0.46 cos(~r/R)]do(r)
Arsac kernel d , ( 4 = [1 - (ra/R2)1dO(4
I n order t o modify the filter function h(c,y) as little as possible in the inter-
val (0, R) one can use the kernel
(2.37) d 5 ( r )= 1, 0I r I0.9R
= ${l - c o s [ ~ ( R
- r)/O.lR]}, 0.9R < r < R
The modified frequency response corresponding to this filter function
h,(r) of finite length is then given by Eq. (2.16)
(2.38) R d P ) = cJ; W p ’ ) D , ( p ; P’)P’ dP‘
+ jODk-”H(p‘ + po)k:,(p’ + Pc)Di(P; pr + pc)(p’ +Pc) dp’ t
where the function D,(p;p’) is defined by

For the kernel do@)one can apply the following relations for the zero-order
Bessel function (Watson, 1958, p. 134)
(2.40) (p' f p) JJo(P'z)Jo*(PZ)z dz

=~ C p ' J , ( p ' ~ ) J o * ( p -
~ ) p J * ( p ' ~ ) J , * ( p ~ ) l / ( p '-
2 pa)
(2.41) (p' = p ) p , ( p z ) J o * ( p Z ) z dz
= i t ~ " ~ , ( f z ) J o * ( p ~-
) J -l(pz)Jl*(pz) - J ~ , ( p z ) J 1 ( p z ) l
for any complex number z, in order to obtain
(2.42)
(p' f p) Do(p; p ' ) =R[p'J,(Rp')Jo(Rp) - p J 0 ( ~ P ' ) J 1 ( ~ p ) l / ( p -' f2a )
(p' = p) D o h p ) =4 R " J o 2 ( ~ p )+J12(RP)l
The expression (2.9) for the output field is now approximated by
(2.43) g(z, Y) = 2n- loR

hr(r)f(z,Y,r)r 0%
I n order t o meet the property that the filter coefficients should add u p t o
unity, let R be the solution of the equation
which lies nearest to the outer integration circle of radius R, of the chosen
integration grid. Hence, substituting (2.34) and (2.35) into (2.44) one obtains
PC
(2.45) W )= J H(p)D,(p;0)p dp
0
+J:-" ~ ( +ppc)ka(p + +
p c ) ~ i ( p pc; o ) ( p + pc) dp - 1 =0
The box car kernel d o ( r )gives
(Sneddon, 1951, p. 513) and R is then the solution ofthe equation
(2.47) f(R) R
:u H ( p ) J , ( R p ) dp
200 FRANS DE MEYER
Applying the trapezoidal rule with interval A to the finite integral (2.43)
yields the weighted sum
(2.48)
where
The form (2.48) can then be converted into the equivalent expression
(2.50)
in which the radii R , have values so that the corresponding circles will pass
through certain grid points. The conversion from (2.48) to (2.50)is made by
applying an n-point interpolation formula to yield an expression for each
f (x,y, r,) in terms of valuesf (x,y, Rk)on the n nearest circles. By combining
the interpolation coefficients with the values of w(rl) one obtains the co-
efficients hk in Eq. (2.50).
An application of this method will be given in Section 3.5 for the process of
the downward continuation of a surface gravity field.
2.5. Construction of Linear, Discrete Filters in the (u, v)-Plane

. In order to construct a suitable numerical Hter for the continuous filter
(2.9), we want it to give the property that the corresponding frequency
response H,(u, v) must have a “good” approximation to the theoretical
frequency response H ( u , v) in the Nyquist domain of the frequency plane.
We want the difference between the ideal output g(x, y) of the theoretical
filter and the numerical output #(x,y) of the discrete filter to be a minimum,
according to a given norm. Therefore let us consider the following metrics:
(1) the C-metric on the space of the continuous functions, bounded in the
(2,y)-plane
(2.51)
(2) The L,-metric on the space of the absolutely integrable functions
(2.52)
(3) the L,-metric on the space of the quadratically integrable functions
(2.53)
The form a(x, y) represents the difference between the actual and numerical
outputs
(2.54) Y)= g(x, Y) -m, Y)
From (2.13) and (2.25) it immediately follows that the Fourier transform
of a(z, y) is given by
(2.55) A(u, W) = G(U, V ) -&(U, W ) = F(U, V)[H('U,W) -Hd(u, V ) ]
where Hd(u,w) is one of the forms (2.21) or (2.27). Applying the inverse
Fourier transform to both sides of (2.55)yields
Majoring the absolute value of the integral in (2.56) we find that
Similarly, using Parseval's theorem for double Fourier transforms

W
(2.58) la112, = IlAIK, = j" j-m -m

1 P(u,v)1 2[H(u,v) - BJu, v)12 du dv
If the Fourier transform of the input field is bounded, this leads to the in-
equalities
(2.59) llallc I c1 /Im

jmI 4 -m
H(u, I
- H d u , v) du dv
and
(2.60) la1122 <c2 J", j" [ H ( u ,v) -Ha(% v)I2 du dv

-m
where c1 and c2 are constants. According to the basic idea of minimizing the
difference between the true and numerical outputs, the filter coefficients h,
may be determined from the condition
W
(2.61)
1 -03
I~ ( uv),- ~ , ( u w) 1
, du dv = minimum
or from the L,-metric

m
(2.62) m:J J -m
[ ~ ( uv),- Hd(u, w)12 du dw = minimum
202 FRANS DE MEYER
Furthermore, it is known that the study of the frequency behavior of the

filters in question may be restricted to the Nyquist domain; hence, one has
the following criteria for determining the filtenveights hk
(2.63) [INrINI H ( u , v) - H,(u, v)I du dv = minimum
and
(2.64) [H(u,w) - H,(u, v)]" d u dv = minimum
The minimalization according to Eq. (2.M) leads naturally to the method of

least-squares. Indeed, writing I ( h ) for the left-hand side of Eq. (2.64), the
coefficients h, are obtained from the conditions
(2.66) aI(h)/ahj= 0, 0 <j 2 N

Writing H R k ( uv)
, for the frequency response of the averaging process over the
circle Rk,as given by Eq. (2.22) or Eq. (2.28), this leads to the equations
(a/ahj)s'"-VN -UN
[B(u, V)-
k=O
N
hkHRk(U,V )
l2 d u dv = 0
and to the following linear system
= rr",
from which the weights hk may be computed.
rN H ( u , v)H ra,(u,
w) d u dv, 0 5 j<N
The minimalization of the integral on the left-hand side of Eq. (2.63)

leads us to the method of linear programming which is an extremely useful
numerical aid to this kind of approximation problem. The algebraic pattern
of linear programming problems has the following general formulation
n
(2.67)
where m 2 n or m < n and there are a n infinite number of solutions. I n

order to choose an optimal solution from all possible solutions one also
introduces the condition
n
(2.68) z= C ck xk = minimum
k=l
where all the coefficients alk , bj , and ck are known and the xk have t o be
determined. One often deals with the additional condition
(2.69) xk<O, l<k<n

Among several numerical methods for solving this system of linear in-
equalities, the simplex algorithm of Dantzig (1963) is perhaps the most used.
This technique can be applied to the problem of the approximation of a
function by a finite series of the form
(2.70)
with given functions f k ( p ) . For m values p r of the variable p , let the general
formulation of the approximation be
n
(2.71) -
and the task is to find the coefficients hk such that E should be minimum,
without the restriction (2.69). The weighted sum in (2.70)will then provide a
numerical function which approaches g ( p ) in a band of width 2 ~This
. tech-
nique will be used in Section 3.3 for the construction of upward continuation
filters.
AND DOWNWARD
3. UPWARD CONTINUATIONOF THE SURFACE
GRAVITY
EFFECT
3.1. Definition of the Surface Gravity Effect

Thelpotential of the gravitational field of the earth a t a point P(r) on or
outside the surface is given by the well-known Newton's expression
with dr' the volume element of the volume v of the earth, r'thepositionvector
of an arbitrary point of the volume v, p(r') the density, Q! the gravitational
constant, and
1 r - r' I = r2 + rI2 - 2rr' cos y
The origin of the orthogonal coordinate system (2, y, z ) is a t the mass center
of the earth, with the z-axis coinciding with the rotation axis; y is the angle
between the directions of r and r',
204 FRANS DE MEYER
At points outside the earth, the integral in Eq. (3.1) is nonsingular and the
gravity potential satisfies Laplace’s equation
whereas for the interior points the integrand in (3.1) becomes singular at
r’ =r; therefore, inside the earth the potential is the solution of Poisson’s
equat,ion
(3.3) VzV(r’) = 4 d p ( r ’ )
The total potential W of the earth consists of two parts, the principal one,
V, caused by the Newton’s attraction, and the other by the earth’s rotation.
If the point P is on the earth’s surface, the gravity vector is defined as the
gradient of the total gravity potential on the surface
(3.4)
whose ajrection is called the vertical in P;the modulus g is known as the
gravitational acceleration or the gravitational field and the unit of g is the
gal = 1 cm/sec2.
Explicitly, the tidal variations in the vector g, in amplitude as well as in
direction, due to the changing positions of the sun and the moon with respect
to the earth are not considered and, therefore, it is assumed that the contri-
butions of the tidal effects to the gravity vector are eliminated from the
measurements, when performing an analysis of an observed surface field.
Indeed, the aim is to study the local deviations of the gravity field in the
earth’s surface, caused by subsurface masses. Consequentlyit will be assumed
that the earth is divided into a regular part, bounded by the earth spheroid,
and a deviation part. The regular part is supposed to have the same mass as
the earth as a whole and the earth spheroid to have the volume and flattening
of the geoid. Clearly it is implicitly assumed that the deviation part of the
earth’s field consists partly of negative masses, so that the total mass of the
deviation part is zero. The potential caused by these deviation masses of the
earth is denoted by U and it is the difference between the total gravity potent-
ial and the potential of the spheroidal earth.
Since the contributions of these local anomalies are inherent in the gravita-
tional field of the earth, a gravimeter can only measure the superimposed
influence of the gravity field of the earth spheroid and of the deviation
masses. The attraction of local anomalous bodies is often seen only rn a
minor distortion of the pattern due to some major structure. The gravimeter,
being leveled in the total gravitational field, can respond only to the vertical
component of the gravity field of the disturbing masses; this is called the
FILTER TECHNIQUES I N GRAVITY INTERPRETATION 205
“ gravity effect” by Grant and West (1965) and it describes in factthehetero-
geneous structure of the upper part of the earth’s crust. For a point P on the
surface we define the gravity effect by
(3.5) Ag(P) = a u ( P ) / a n
where n is the internal normal to the surface.
I n this connection, the assumption has been implicitly made that the direc-
tion of the gravity vector is not disturbed by the presence of these subsurface
masses, as we have supposed that Ag is measured in the direction of normal
<
gravity. The assumption that Ag g is physically justified by the fact that
changes in the densities ofthe crust within the anomalous region will produce
variations in g which will generally not exceed 10 mgal, so that one can say
that the ratio Aglg is of the order of
3.2. The Upward Continuation as a Filtering Operation

The fist boundary value problem of potential theory, also known as
Dirichlet’s problem, is stated as follows: i t is assumed that the earth’s surface
is replaced by a plane and that a local coordinate system (x,y, z ) has been
chosen, with the z-axis in the direction of the internal normal and the x-
and y-axis defining the horizontal plane z = 0. Let the boundary values be
the values of a function f (x,y, 0) =fo(x, y) in the plane z = 0, which has
continuous third derivatives on and above the horizontal plane. The problem
is t o find the solution of Laplace’s equation V z j = 0 which is continuous on
and above the surface and regular in the half-space z 0, and coincides with
fo(x, y) on the plane z= 0.
It can be proved that this solution, under the above given Dirichlet’s
conditions, exists and that it is unique. The rigorous proof of this statement
and the derivation of the form of the solution are based on the theory of the
Green function, corresponding t o the given boundary value problem (Morse
and Feshbach, 1953; Courant and Hilbert, 1962). It is certainly not the in-
tention here t o explain the general outline of the theory of the Laplace
equation; reference is therefore made to Kellog (1960) and Grant-West
(1965). From these works one obtains the solution of the first boundary
value problem for the gravity effect
showing that the distribution of the gravity effect above the earth’s surface is
completely determined by the integration of the surface data. Hence, the
problem of the upward continuation of the gravity effect is reduced to the
numerical integration of Eq. (3.6).
206 FRANS DE MEYER
Equation (3.6) is clearly a convolution integral of the form (2.7) and thus
defines a linear filter, with the surface effect Ag(x, y, 0) as input and as output
the desired field Ag(x, y, z ) a t height z, using a filter function
(3.7) h ( - ) ( x ,Y , Z ) = ( ~ Z ] / ~ TI )/- ( X ~ + ~ ~ + Z ~ ) ~ ’ ~
This intrinsic relation between the problem of upward continuation and the
general linear filter theory was first clearly pointed out by Dean (1958).
I n the frequency plane the equivalent form of Eq. (3.6) becomes
(3.8) Q(u, v, 2) = H‘-’(u, 0,z)C(u,v, 0)
in view of the convolution theorem of Fourier transforms. Q(u,v, z ) , Q(u,v, O ) ,

and H ( - ) ( u , v, z ) are the Fourier transforms of Ag(x, y, z), Ag(x, y, 0) and
h(-)(x,y, z), respectively. The explicit form of the frequency response of the
upward continuation filter follows from Eq. (2.16); H ( - ) ( u , v, z ) is then the
+
zero-order Hankel transform of the function I z I (r2 z ~ ) - ~Hence’ ~ . (Sneddon,
1951, p. 528)
(3.9) H ( - ) ( u ,v, ~ ) = e - ~ ~ ~ ~ = e x p [ - ~ z ~ ( u ~p 2+=vu~z +) v~z ~ ~ ] ,
This result can be easily interpreted if one notes that small masses mainly
contribute to the Fourier spectrum of the surface effect in the high frequency
region and that the broader structures induce waves of small frequency.
Combining (3.9) and (3.8) one obtains the fundamental relation
(3.10) +
G(u, v, z ) = exp[ - I z I (uz V ~ ) ~ ’ ~ ] v,
G 0)
(U,
The appearance of the decreasing exponential in Eq. (3.10) implies that the
field will be damped out with increasing height and that the high frequency
part of the Fourier spectrum will be smoothed out faster than the low fre-
quency region. The filter (3.10) clearly has the properties of a low-pass filter,
yet without a sharp cutoff: when the field is represented a t different heights,
first the fine details of local extent, corresponding to small shallow masses,
will disappear, leaving a t last a picture of the regional features of the field.
+
Transforming t o polar coordinates (x - 6) i(y - 7)= rete, Eq. (3.6) can
be written as
EILTER TECHNIQUES IN GRAVITY WTERPRETATION 207
is the average of the surface gravity effect over the circle of radius r, with cen-
ter a t the point (x,y). From Eq. (2.20)i t follows that the integral (3.11) can
be reduced to a weighted sum of the form
(3.13)
by using an adequate numerical integration method.

Without going into too much detail one can essentially state that the
methods of Peters (1949), reviewed by Choudhury (1972) and Henderson
-
(1960), mainly differ in the way the transition from (3.11)to (3.13)is made, as
well as in the coefficients h, of the weighted sum Ag(x, y, z).
Peters replaces the integral (3.11) by the sum
11 + 22)3121rdr
N Rk+l
(3.14) &?x, Y, 4 = IzI [A&, y, r ) / ( r 2
k=O Rk
and assumes that the value of Ag(x, y, r ) in the interval (&, R k + l )may be
approximated by the mean of the values a t the end points
w
Adx,y, r , !dAdX,
y, R k ) + y, R k + 1)1
for r in ( R k ,R k +l). Evaluation of the integrals in Eq. (3.14)leads to an expres-

sion of the form (3.13).
Henderson uses the following approximation for the field a t the height
z = -ms 0
where so is the mesh size of the square grid and m an integer. By expressing
Ag(x, y, r) in terms of polynomials of the second degree over the intervals
( R k , R k + J and using the formula for the gravity anomaly of the sphere,
Henderson arrives a t a weighted sum of the form (3.13)with coefficients shown
in Appendix 1.
Although it must be clearly kept in mind that a suitable numerical inte-
gration method (3.13) has to be a compromise between working with a large
amount of surface data (we ask for a small number of integration coefficients,
in order to reduce the calculation time and the edge effects, imposed by the
square grid of finite extent) and obtaining a sufficient accuracy for the cal-
culated field A ~ Xy,,x ) a t a given height, thisleads to the following conclusions:
208 F U N S DE MEYER
(1) From the point of view of the numerical treatment of the surface data,
both approximations are of the same type as they amount to the computation
of a weighted sum of the form (3.13).
(2) Peters assumes linearity of the surface field over the integration inter-
vals (R,, R,, while Henderson constructs his coefficients using a physical
model, thus allowing for the nonlinear variation of the field over most
integration intervals. This seems to be a reasonable improvement of Peters’
method.
(3) Peters arbitrarily overweights his last coefficient so that the sum of the
coefficients is unity; Henderson’s method does not take this restriction into
account.
(4) Both authors use almost the same number of integration circles, but the
extent of the integration grid is different: R, = d l 2 5 11 units for Peters
and R, = 4 6 2 5 = 26 units for Henderson.
(5) The surface extent of both integration arrays is the same for the con-
tinuations at different heights and only the values of the coefficients change,
whereas, in fact, the extent of the integration grid should be proportional to
the height were one wants to compute the field.
(6) As a final conclusion, the construction of these weighting sequences is
based upon empirical, more or less physically justified assumptions about the
observed surface field; therefore the quantitative success of a given set of
coefficients will directly depend on the degree to which a measured field
meets these assumptions. Hence it would be preferable to constructthe weights
in a more ‘‘ objective ” way, that is on the basis of the mathematical mechan-
ism of the process of the upward continuation of a surface field.
From the above it is evident that the question of the “ best” set of coeffici-
ents for the numerical computation ofthe integral (3.11)is difficult to answer
by the absence of a mathematical norm for the construction of the weights.
Indeed, the accuracy in the upward calculated field will depend more on the
frequency st&cture of the surface field in question than on the use of any
particular weighting sequence. The grid spacing, the number of integration
circles, and the weights, d l empirically chosen, have an important bearing
on the configuration of the h a 1 map.
The differences between the theoretical upward frequency response H ( - )
(u,v, z ) and the discrete frequency characteristicH,(u, v), given by Eq. (2.27),
are shown in Fig. 4 for Peters’ and Henderson’s filters z = 1, .so = 1, h = 1.
Note the direction dependence of H,(u, v ) in the frequency plane, which
implies that the asymmetries in the calculated field at height h are not only
properties of the surface gravity data, but are aIso dependent upon the kind
of Glter one uses. In the low frequency region the deviations are of the order
of 10 yo and we conclude that Peters’ and Henderson’s filters deform sign&-
antly the features that may be ascribed to the larger anomalies of the gravity
FIQ.4. Deviations for (a) Peters’ filter z = 1, so = 1, h = 1 contour interval 4%: (b)
Henderson’s filter z = 1, so = 1 , h = 1, contour interval 4%.
210 FRANS DE MEYER
field. For the higher frequencies one notes Merences of about 15 yo;since
the high frequencies refer to the noise component of the surface field and to
the field of small shallow masses, these effects will be of very local extent in
the h a 1 maps. Nevertheless, they introduce unwanted high frequency ripples
in the smooth output of the upward continuation filter. As a conclusion we
can say that Peters’ and Henderson’s filters are frequency-wisenearly equiva-
lent.
3.3. Construction of a Numerical Upward Continuation Filter
At fist it seemed obvious to use the method proposed by Dean (1958): if

one equates the theoretical and numerical frequency responses at N 1 +
Merent values p , of the radial frequency, then the solution of the resulting
system of simultaneous equations
will give the filter weights. For the moment, the discrete frequency response
(2.21)will be used. However, while testing the numerical solution of the linear
system (3.16)it was noted that the accuracy of the approximationwas strongly
dependent on the distribution of the points p j on the frequency axis and, t o a
lesser degree, on the particular choice of the integration grid Rk .
In order to show the instability of the solution of this linear system, the
results are given for the following distribution of the points pr: we divide the
Nyquist interval (0, pN) into two parts of equal length I, = (0, $ p ~ )and
+
I, = (+pN, pN). For N 1 = 20 filter coefficients we place N , equidistant
points in I, and N , = N - N , + 1 equidistant points in I, and let N , vary
from 3 to 12. The root-mean-square differences between the theoretical and
numerical frequency responses thus obtained are shown in Table I for
z = 1, so = 1, h = 1 unit spacing and the integration grid in the first column
of Appendix 2.
TABLE
I
Root-mean-squaredifferences
N1 N2
3 0.026 8 1.016
4 0.032 9 11.944
6 0.023 10 38.817
6 0.107 11 26.390
7 0.264 12 11.660
FILTER TECHNIQUES IN GRAVITY INTERPRETATION 21 1
It is noted here that there is a considerable increase in the root-mean-square

differences when passing from N , = 5 to N , = 6 points. For N , < 5 we obtain
coefficient sets yielding a reasonable approximation t o the theoretical
frequency characteristic H ( - ) ( u , w, z ) ; for the higher frequencies p 2 1 the
curves of the theoretical and numerical frequency responses match very
closely, but the largest deviations occur in the low frequency range p 5 1.
For N , 2 8 points the numerical frequency response is completely different
from the theoretical curve. Because the larger masses mainly contribute t o the
low frequency part of the Fourier spectrum of the surface field, it is required
that the numerical frequency response H d ( u ,w) closely match the decreasing
exponential H ( - ) ( u , v, z ) for the lower frequencies in order t o be protected
against an undesirable deformation of the larger anomalies of the field when
projecting it upwards. Therefore, it is concluded that Dean’s method does not
give reliable results.
We now try t o construct the filter coefficients from the approximation of
the discrete frequency response H,(u, w), given by Eq. (2.27),to the theoretical
function H ( - ) ( u ,v, z), by the methods of Section 2.5. As the ideal frequency
characteristic H ( - ) ( u ,w, z ) is already a function of the radial frequency, we
choose the points ( u l ,w,) a t the intersections of a bundle of radials through the
origin of the frequency plane with a bundle of concentric circles around the
origin. Also note that the function Hd(u, w) in Eq. (2.27) is symmetric with
respect t o the bisector of the first quadrant of the (u,w)-plane, so that one
may restrict the approximation t o the triangle w <u, O<u<u, of the
Nyquist domain.
When using the method of least-squares and the method of linear program-
ming, one has to compute s vector
b , = e x p [ - ~ z ( ( ~ , ~ + w , ~ ) ~ / 1~ <] ,i < m
and a matrix
+
nk/2
afk= (2/nk) cos(Utxjk vi Yj,)t 1 <i m, 1 <k <n
jk=1
The coefficients obtained with these methods can be found in Appendix 2.

It is evident that the accuracy of this approximation will be better for
larger values of n. I n our case this implies that the calculated fieldatadesired
height will be quantitatively better for a finer or larger integration grid R,, or
both. But a large number of integration circles entails a considerable increase
in the computer time for the numerical integration (3.13); a broader inte-
gration grid implies knowledge of an extensive part of the surface field and a
considerable loss of accuracy a t the edges of the observed part of the field.
There has t o be a compromise between obtaining sufficient accuracy with a
relative small number of integration coefficients by limiting the extent of the
integration grid and reducing the calculation time.
212 FRANS DE MEYER
As the work is done with a finite weighted sum in Eq. (3.13), the question
of the truncation error resulting from disregarding the field beyond the last
integration circle must be examined. As
(3.17) (1 z 1 /'W Jm
-m
Srn df d ~ / [ (-x 8’ + (Y
-m
- 7)’ + ~ ' 1 =~ 1' ~
one can write for the difference between the true field value at height z and
its value at the surface
(3.18)
m
Ag(z, Y,4 - A d s , Y,0) = I z 1 j ([Ag(s, Y,r ) - Ag(x, Y, O)l/(r2 +

0
z 2 ) 3 / 2 } p dr
using Eqs. (3.11) and (3.12).Now, applying the numerical integration method
as in Eq. (3.13) to the right-hand side of (3.18) one obtains
or
Let the circle of radius R represent the area taken into accountin the numer-
ical integration of the right-hand side of Eq. (3.18).Then it follows that the
true field at height z may be written as
+ I I 1 [Adx, Y,r)/(r"+
m
Adz, Y,4 -N &k
-
(3.20) Y,2 ) z2)312174dr
N Ag@, Y,4 + Adz, Y, 0) I z I / ( R 2+ z ~ ) " ~

using Ag(x, y, 0) as a measure of the average of the surface field outside the
circle of radius R .
Combining Eqs. (3.19) and (3.20) leads to the conclusion that the con-
tribution of the surface field within the circle of radius R is represented by
w
Ag(x, y, z) and the difference between the sum of the coefficients and unity,
multiplied by the center point value Ag(x, y, 0), may be attributed to the
contribution from the part of the surface field outside the circle of radius R .
Hence, it is noted that one possibility of reducing the truncation error is
to make the sum of the coefficients equal to unity, as in Peters' method.
Otherwise, if the filter weights do not add up to unity, the surface value
Ag(x, y, 0) must be multiplied by the difference between the sum of the GO-
efficients and unity and then the result added to the computed value
G(X,y, 2).
Whether or not the sum of the filter coefficients should be made equal to
unity depends entirely upon personal judgment. However, if one requires
that the weights add up to unity, it is found to result in relatively large values
of E in the method of h e a r programming and it was difficult t o reduce the
value of this E . On the other hand, not taking this restriction into account
one obtains a faster convergence to an acceptable solution, with a minor loss
of accuracy (rounding-off errors) and in addition small values of E . I n con-
sequence preference is given to the second alternative.
The significance of this E must be kept clearly in mind: in the method of
least-squares, E means the root-mean-square differencebetween the theoretical
and numerical frequency responses, while in the method of linear program-
ming E stands for the maximum difference between both curves. Therefore,
one can compare the following results:
+
z = 1, a0 = 1, h = 1, N 3 = 20, R, = 4100, least-squares: E = 1.7 yo,
maximum difference 5.5%,
+
z = 1, so = 1, h = 1, N I = 20, R, = 4 1 0 0 , linear programming: E =
3.5 yo,maximum difference 3.5 yo.
Hence, we have constructed a discrete filter, simulating the process of
upward continuation, giving an approximation to the theoretical filter
H ( - ) ( u ,v, z ) for z = 1, with an accuracy of 3.5 yo over the Nyquist domain
of the frequency plane.
The reduction of the integration grid with R, = 4 2 7 4 16.5 units t o
a grid with R, = dl00 = 10 units results in a small increase in the value of
E : from 1.3 yo to 1.7 yo in the method of least-squares and from 3.2 yo to
3.5 % for the method of linear programming. This is an interesting result
since a limited integration grid enables us to reduce considerably the deforma-
tion of the field near the edges of the surface grid on which the field is known.
The considerable gain of this method must be found in the generality of its
principles: the coefficients are derived from the knowledge of the frequency
response function, describing the mathematical mechanism of the upward
continuation of a surface field in the frequency plane. The method does not use
any empirical assumption about the field, as was the case with Peters’ and
Henderson’s filters. Also the flexibility of the method of linear programming
in these approximation techniques is remarkable: the construction of the
weights h, is practically independent of the choice of the distribution of the
frequency points (u,, VJ, because one can take rn several times larger than n.
3.4. The Downward Continuation as a Filtering Operation

The following problem will now be considered: given the normal component
of a gravitational field in the earth’s surface, how may one determine the
value of the field a t a given depth. This calls for the solution of Poisson’s
214 FRANS DE MEYER
equation with unknown density function. The problem is obviously indeter-

minate because of the inherent ambiguity of inverse boundary value problems,
common t o all potential fields.
Skeels (1947) points out that, for a given surface field, a n infinite number of
mass distributions can be found, which all account for that field. It is true
that a measured field generally introduces several physical parameters, such
as maximum and minimum depths of the anomalous masses, largest areal
extent, maximum density changes, etc., but even within these limits an
infinite variety of solutions is possible. This ambiguity does not depend upon
the spacing or precision of the observations in the earth's surface and no
degree of precision or amount of data will remove these fundamental un-
certainties.
To this are added the facts that one may expect the observations to show
errors, that the measured field is the superposition of the anomalous fields
of a very complicated ensemble of structures, that the measurements undergo
rough reductions (such as Bouguer corrections), and that the number of
observations is limited and that they are known on a regular grid of finite
extent. Hence it is evident that a quantitative interpretation of gravitational
anomalies can only give partial information about the form, location, depth,
and density of the subsurface masses.
It is essential to avoid dealing with the Poisson equation, inasmuch as
the mass distribution is unknown; consequently one has t o rely upon more or
less physically justified assumptions about the mechanism of the process of
downward continuation of a surface field.
The distribution of the field in a plane at depth z is to be determined; Grant
and West (1965) suppose that no disturbing masses intervene between the
surface and the level z, assuming implicitly the validity of Laplace's equation
between these two levels. The distribution of the field a t the surface is now
defined as the result of the upward continuation of the field a t depth z ; using
Eq. (3.6) we postulate the following relationship between these fields
(3.21) A d z , Y,0) = ( 4 2 ~ )
x I:m
Srn Ad[, 7,z)/[(x t)'+ (Y
-w
- - 7)’ +z ' I ~ ' d~ t dq, >0
This hypothesis leads to an inevitable divergence in the mathematical
expressions, describing the downward continuation process, because the
earth's surface is a real discontinuity plane and anomalous masses do occur
between the two levels.
The expression (3.21) is an integral equation for the field a t depth z ; with
the same notations as in Eq. (3.8) and applying the convolution theorem
of Fourier transforms we obtain
(3.22) G(u, V , 0) = H'-'(u, V , z)G(u, V , X )
FILTER TECHNIQUES I GRAVITY INTERPRETATION 215
Defining the frequency response of the downward continuation filter as

(3.23) H‘+’(u, 21, 2) = H‘-’(u, v, z)-1
+
= exp[z(u2 v ~ ) ~ ~ ] ,z > 0
it follows that
(3.24) G(u, v, z ) = Q(u,v, O)exp[z(u2 + 212)1/2]
Consequently, the process of downward continuation defines a linear filter

with as input the surface effect Ag(x, y, 0) and as output the field Ag(x, y, z )
a t depth z, using the frequency characteristic H‘+)(u,v, z ) . I n the divergence
of this frequency response lies the explanation of the fact that downward
continuation of a field, derivable from a potential, is so intractable. This
divergent nature of the theoretical frequency response for increasing depth
and high frequencies prevents us from obtaining a finite filter function in the
coordinate plane: only in the case where the surface field has a Fourier spect-
rum that attenuates with the larger frequencies more rapidly than the expon-
ential rises, will the downward continuation integral (3.21) converge. The
divergence of H ( + ) ( uv,
, z ) finds its cause in the intrinsic amplification of the
field toward its source, in the existence of small, shallow disturbing masses,
and in the presence of the field of random errors, which is inherent to the
observed field, whether they come from measurement, digitilization, or re-
duction of the data. Therefore it is imperative to eliminate the high frequency
contributions from the Fourier spectrum of the surface field and to amplify
the low frequency waves, building up the larger masses of the field in a way
that is consistent with Eq. (3.23).
Peters (1949) and Henderson (1960) give coefficient sets, simulating the
downward continuation filter. Both methods can be compared on the basis of
their approximation to the theoretical frequency response H ( + ) ( u ,v, z ) .
Although both algorithms are equivalent from the point of view of practical
application, they differ mainly in the principles on which they depend:
Peters sees the downward projected field in the form of a Maclaurin series;
Henderson, however, first computes the field a t different heights by the
expression (3.15) and extrapolates the distributions a t these levels, to-
gether with the surface field, into depth. Using the analogous form of (2.21)
for the frequency response of the numerical downward projection filter
(ho h, , . . . *h N )
3
the curves illustrated in Fig. 5 are obtained. The corresponding coefficients

are given in Appendix 3.
If we take for a norm the best approximation of the numerical frequency
response HL+)(u,v) t o H ( + ) ( u v,
, z ) , then Henderson’s method is more satis-
216 FRANS DE MEYER
PIG.5. Frequency response for (a) Peters’ Glter z = 1, a0 = 1, d = 1; (b) Peters’

filter z = 1, a,, = 1, d = 2; (c) Henderson’s filter z = 1, a0 = 1, d = 1; (d) Henderson’s
filter z = 1, a,, = 2, d = 2.
factory than Peters’ algorithm. On the other hand, the frequency decay is
sharper in Peters’ method and Henderson’s coefficients more greatly amplify
the high frequency waves in the input field, so that we may expect larger
fluctuations in the projected field than with Peters’ weights. The conclusion is
obvious: we need to construct a filter somewhere between these two methods,
that is, a filter matching H ( + ) ( uv,, z ) very closely in the low frequency range
and cutting off rather sharply.
Although the divergence of the theoretical frequency response is restricted

to the Nyquist domain of the frequency plane, we still have the problem that
the high frequency waves, corresponding to uninteresting features of the
surface effect, are more greatly amplified than the low frequencies, so that the
frequencies in the neighborhood of the Nyquist frequencies, that is waves
with periods of the order of 2s0, greatly deform the pattern of the field
of the broader anomalies in the output maps.
The expression (3.21) has the form of the general integral equation of the
tspe
(3.26)
where v is a measured field, k an analytic function, and $ is to be determined.

We further assume that q~ and t,b are bounded and that the kernel k belongs
to the class L,. By definition, a function f (XI, . . . , xn) is L, if its p-norm is
bounded
(3.27) Ilfllp= [i“
-m . - a !
m
-m
1/11
If(~l,...,z.)~pdxl~..dxn]< c o
Kreisel (1948) showed that $ cannot be determined uniquely from measure-
ments of q~ alone: if $&+, . . . , zn)is a solution of the integral equation (3.26),
then
also satisfies (3.26), where A may be taken arbitrarily large provided that
+ --
pla - +pnZi s also sufficiently large. Therefore $ is not determined con-
tinuously from measurements of q ~ ,because rapidly oscillating $ produce a
small q ~ .I n our case of gravitational interpretation this means that itisalways
possible to add t o any solution Ag(x, y, z ) a harmonic mass distribution of
arbitrarily great amplitude provided that the frequencies, induced by the
added distribution, are sufficiently large. Therefore the observations a t the
earth’s surface can give no information about changes in density occurring in a
horizontal distance small compared with the depth, and the solution required
is one that does not contain these rapid oscillations.
Accordingly one has t o look for partial information about # which is in-
sensitive to rapid oscillations in $. For this reason Kreisel defines the smoothed
version a€$
(3.28)
where w is a normalized filter function of a suitable low-pass filter.

218 FRANS DE MEYER
If K and W are the Fourier transforms of the analytic functions k and w,

respectively, and if (1) k and w belong to the class L,, (2)A = W / K is defined
t o be zero if W = 0, and A is L,, (3) the inverse Fourier transform X of A is
L,,and (4) I+% is bounded and integrable in any finite domain, then Kreisel
shows that (1) q~ is bounded and integrable in any finite domain, and (2) the
smoothed version of hz, is given by the expression
(3.29)
Applying this theorem t o the integral equation (3.21) we define the smoothed
version of the field a t depth z by
and from (3.29) it then follows that
with
(3.32) fL(+)(z,y, z ) = (1/4v2) f m

-m
Im
-a
["(a, v ) / K ( u ,v)]ei(ur+uy)dudv
Using (3.23) we obtain
(3.33) h(+)(z,
y, z) = (1/4ra) 1 m
-m
m
f
-m
, z)W(u,v)ei(ur+uy)dzl
H ( + ) ( uv, dw
Substitution of (3.33) into (3.31) yields the following frequency relation

between the Fourier spectrum of the surface field and the Fourier spectrum
of the smoothed field a t depth z
(3.34) B(u, 21, Z) = B(+)(u,
V , z)G(u, V , 0 )
where o(u, v, z ) is the Fourier transform ofhg(z, y, 2). We conclude that

Eq. (3.34) defines a modified downward continuation process by specifying a
linear filter with impulse response h ( + ) ( x , y, z ) and frequency response
R(+)(u, v, z). The smoothed field a t depth z has a bounded Fourier spectrum
for all frequencies provided that
(3.35) v,z ) = W(u,v)exp[z(uua+ v ~ ) ~ ’ ~ ] ,
R(+)(a, z >O
belongs to the class L,. Only in the case where the smoothing function w(x, y)
fulfils this condition can we apply Kreisel's theorem, in order to find a solution
FILTER TECHNIQUES IN’ GRAVITY INTERPRETATION 219
of the inverse potential problem, not for the field a t depth z itself, but for the
components of the field which are not eliminated from the Fourier spectrum
of the surface field by the frequency response W ( u ,w).
3.5. Construction of a Numerical Downward Continuation Filter

Bullard-Cooper (1949) and Grant-West (1965) considered the mathe-
matical filter whose weights are proportional t o the ordinates of the normal
distribution
(3.36) w(x, y) = ( 1 / 4 ~ y ) e - ( ~ ~ + Y ~ ) ' * Y
with corresponding frequency response
(3.37) w(u,w) = e-Y(Z12+U2)
thus defining the modified downward continuation filter
(3.38) w,z ) = exp[z(u2+ v2)lI2 - y ( G
R(+)(u, + w2)]
y is a constant and determines the degree of smoothing.
This analytic filter can be simulated by a discrete filter of the form (3.25);
the differences between the theoretical and numerical frequency responses
can be seen in Fig. 6 for the filters of Grant-West (Appendix 4). As a criticism
of this algorithm we could say that the cutoff is not sharp, that the deviation
in the low frequency range is considerable, and that the high frequency res-
ponse is relatively high.
FIG. 6. Frequency response for (a,) Grant's filter z = 1, so = 1, d = 1, y = 1/6;

(b) Grant's filter z = 1, so = 2, d = 2, y = 1/6.
220 F R B N S DE MEYER
I n order to avoid the amplification of the high frequency waves by the

factor H(+)(u,v, z ) we introduce the low-pass filter
(3.39) W P )=1 for O<p<pc, p2=u2+v2
=k a b ) PC<PsPk
=O Pk <P
for some cutoff frequency pc; pk is the radial frequency when we want W ( p )
t o become zero. The kernel k,(p) is one of the forms in Eq. (2.32).
From Eq. (3.35) i t follows that we now work with the modified downward
continuation filter
(3.40) B(+)(p, z ) = ezD for O<plpc
= erDk,(p) PC < P < P k
=O Pk <P
This function is illustrated schematically in Fig. 7 .
FIG.7. The functions W ( p )and a(+)(,, 2).
The results of Section 2.4 may then be applied directly, if we substitute

the function e s p for the general frequency response H ( p ) in Eq. (2.31). Using
Eq. (2.34) this immediately yields the filter function corresponding to the
frequency response B(+)(u, v, z )
which must be shortened t o a finite length, defining the filter function

(3.42) &+)(z, +
y, Z) = h f + ) (y,~ Z, ) di[(s2 Y ~ ) “ ~ ]
as in Eq. (2.35). The frequency response corresponding to this impulse

response of finite length may be obtained from Eq. (2.38)
(3.43)
DC
)(p, z ) = j ezp‘Dl(p; p‘)p’ dp’
J:-” +
0
+ exp[z(p’ PC)lka(P’ + P c ) ~ t ( pp’; + pc)(p’ +pC)dp’

where the function D i ( p ; p‘) is defined by Eq. (2.39)for the data kernel d i ( r ) .
The value R for the radial coordinate r where d o ( r )becomes zero is obtained
from the nonlinear equation in (2.47)
(3.44) I ( R )= R
- PC
+1 + PC)JI[R(P + Pc)l dP) -1
Pk
x {Jooc eZDJ,(Rp) dp ezco+Dc)k
a (p =0
0
Using the results in Eqs. (2.48)and (2.49)one finally obtains the following
weighted sum for the smoothed field at depth z
- N
(3.45) Ag(x, ?/,2) 2: hk Adz, Rk)
k=O
with weights hk .
There is a problem in defining the parameters pc , pk , and a. The choice of
pk and a is rather arbitrary and depends upon the kernel one decides to use.
If ka(p) is a FBjer kernel, then one can define pk as the first zero of k,(p) and we
obtain the relation
(3.46) - pk) == 7T
If k,(p) is a Weierstrass kernel, then one defines Pk as the freqwlicy where the
frequency response R j + ) ( p ,z ) is equal t o a small number 6
exp[2pk - a(pk - pC)21=

which yields the relation
(3.47) cr(pk- f C l 2 =‘pk + 1 In 1

The choice of the cutoff frequency pc , however, is fundamental and will
depend largely on the Fourier spectrum of the observed part of the surface
field. Therefore, the choice of pc will remain fairly subjective. I n this con-
nection one might argue for two alternatives:
(1) If it happens that the Fourier spectrum of the observed field falls off
sharply a t a certain radal frequency, then we can use this frequency as the
222 FRANS DE MEYER
definition of pc and we know that the components of the field, building up

the low frequency structure of the Fourier spectrum, are more or less signifi-
cantly projected downward.
(2) I n the opposite case one might choose a priori a value for p c , thus
defining “signal” and ‘Lnoise”components in the surface field, and then
one knows the part of the field that is continuated downward and which part
is left a t the surface.
However, there is no mathematical criterion allowing a choice between these
alternatives, as this can be immediately related to the impossibility of solving
uniquely the inverse potential problem with unknown density function.
Using this method the following approximations are arrived at:
Test (1): z = 1, pc = 1, R = 5.98725, A = 0 . 1 , N = 13, 3-point Lagrange
interpolation, Weierstrass kernel with u = 6.61 and pk = 2.
Test (2): z = 1, pc = 1, R = 9.00739, A = 0.1, N = 20, 4-point Lagrange
interpolation, Weierstrass kernel, with a = 6.61 and pk = 2.
interpolation, FBjer kernel with u = 2~ and pk = 1.5.
interpolation, FBjer kernel with u = 2~ and pk = 1.5.
The corresponding coefficients can be found in Appendix 5 and the fre-
quency responses are illustrated in Fig. 8.
It immediately follows that the choice of the particular kernel k,(p) is
irrelevant. I n the low frequency range 0 2 p < pc the approximation N = 20
is evidently better than N = 13 and the cutoff is sharper; in the high fre-
quency interval p > pc all curves show oscillations of the same magnitude (of
the order of 0.3) so that these filters have an almost equivalent effect in this
interval. However, from the fact that the outer integration circle has a value
of d 5 0 g 7.1 units for N = 13, as opposed to 4 1 0 0 = 10 units for N = 20, it
is concluded that the filters for N = 13 are qualitatively equivalent to the
filters for N = 20. The oscillations above the cutoff frequencypc are somewhat
misleading: indeed, in this approximation technique these oscillations cannot
be avoided and appear naturally, but the considerable gain of this form of low-
pass filtering is found in the fact that the amplitudes of these deviations of the
zero level are relatively small, compared to the increase of the exponential
H ( + ) ( p z, ) in the Nyquist interval, attaining a value of about 77 in the point
(n, T ) for z = 1. This implies that the noise part of the surface effect can be
kept sufficiently under control when projecting the field downwards.
It is also possible to simulate the low-pass downward continuation Glter
R ( + ) ( uu,, z ) by using the method of linear programming, as discussed in
Section 2.5. Although it followed from the numerical frequency response that
the function R(+)(u, v, z ) was relatively we11 approximated by the discrete
filters we obtained with this method, the results were not significant. Indeed,
I I 3 4
(a1
, ......... I...
I
. . _ . . I. . . . . . . . . . I . . . . . . . . .I . . . . . f
2 3 4
I . . . . . . . . . I.......-.L..,.,
I
. . . .I . . . . . . . . . l . . . . . P . . .
2 3 4
(C) (dl
FIG.8. Frequency response corresponding to (a) Test (1); (b) Test (2); (c) Test (3);
(d) Test (4).
we found large coefficients(some of the order of lo6)bearing alternating signs;

the existence of observational errors in the measured field, constituting
a realization of a random field, prevents recognition of any meaning to the
output of these filters, because we compute weighted sums of circle averages
multiplied by these coefficients. A test on an artificial field, consisting of the
superposition ofthe gravitational anomalies of a series of spheres and a random
field, confirmed this presumption.
224 FRANS DE MEYER
Consequently i t was necessary t o limit the filter weights in amplitude. This

was possible by the generality of the method of linear programming. Up t o
this point we tried to h d the optimal solution of a system of linear in-
equalities of the form
n
(3.48) -&< xajkxk-bj<&, l<j<nZ
k=l
Several extensions of this system are possible.

If one requires that
(3.49) lxkl <y for l<k<n
for a specific y, then this requirement can be written as
(3.50) IXkI/yi<-E with Yi=Y/-E>O

adding the following system of linear inequalities to (3.48)
n
with
and
(3.53) bm+j=O, 1<j<n
Further one can demand that the sum of the coefficients should differ
slightly from unity, adding to the system (3.48) the condition
(3.54)
or
n
(3.55) ~ a m + n + l , k x k - b m + n + l ~ E
k=l
with
(3.56) am+n+i,k=p, 1Sk<n
bm+n+1 =B
Finally it is possible t o introduce approximation degrees over the Nyquist
interval. Indeed, suppose that we divide the interval J = (1 j <m> into
subintervals J , =(rnf< j< m,+ 1}, 1 < i 2 s, m, = 1, and m,,, = m. If
we now want to obtain the accuracies elat in the intervals J i then we can
write the system
(3.57) -
if we define
(3.58) a;k= a , a j k ,h,’ = a , h , , 1 < i <s, 1 5 k < n
mt< j<W*+l
Among the several filters obtained with this method, are the weights
given in Appendix 6, for the following choice of parameters:
Test (1): z = 1, pc = 1, N = 13 filter weights, Weierstrass kernel with
pic= 2 and ci = 6.61, y1 = 14, /3 = 4.
Test (2): z = 1, pc = 1, N = 13 filter weights, Weierstrass kernel with
pk = 2 and u = 6.61, y1 = 14, = 4, and approximation degrees
;E in the interval 0 pc pk
58 in the interval pc < p 2 pk
e in the interval Pk < p
The corresponding frequency responses are illustrated in Fig. 9.
The curve corresponding to Test (1) shows that the maximum deviation
from the modified downward continuation frequency response is about 14.4 yo
in the low frequency range of the Nyquist interval. The Test (2) yields a
value for E of about 10.7y0, which must be interpreted as follows: in the low
frequency range 0 < p < pc we have an approximation of about $ E = 5.3 yo
1 3 4
(01 (bl
FIQ.9. Frequenoy responee corresponding to (a) Test 0 ) ;(b) Test (2).

226 FRANS DE MEYER
and in the high frequency range p > p k the oscillations of the frequency
response have amplitudes of E = 10.7 yo.Since the introduction of the kernel
k,(p) enables us t o avoid the sharp cutoff a t the point p c , it is immaterial
what the accuracy of the approximation will be in the interval pG < p < pk .
As a final conclusion we can state that the method of linear programming
yields coefficients, simulating sufficiently the modified downward continua-
tion process, and constructs discrete filters which are d e h e d as a function
of the observed surface field. The only parameter t o be estimated is the cutoff
frequency p c , that is the point where we want the frequency response t o
become essentially zero. This technique enables us t o derive filters as functions
of the Fourier spectrum of the surface effect, in opposition to the methods of
Peters and Henderson. I n contrast with the upward continuation, which
defines a stable filter, it is now desirable to separate the input field into signal
and noise components. For this reason we remark that the coefficientsobtained
are only indicative, but the method is general enough to compute readily
numerical filters, once the cutoff frequency is estimated.
4. FREQUENCY FILTERING
4.1. Procedure for Filtering in the Frequency Plane

As a norm for the construction of a suitable weighting sequence h,,
describing the effect of a numerical filter by the discrete convolution (2.20),
the approximation of the numerical frequency response of the filter h, to
the theoretical frequency characteristic has been defined. Section 2.2 discussed
the fact that the representation ofthe input field in the frequency plane (u,v)
is not a t all evident, because the fields are almost never periodic, so that a
Fourier series description is not necessarily meaningful. I n spite of the non-
periodic nature of most two-dimensional data, the practical situation requires
that the observed fields be known only on a regular grid of finite extent, and
it seems intuitively plausible to suppose the measured part of the input field
t o be a realization of a periodic process. Indeed, when one works with measured
fields it is natural to think in terms of signal” and “noise” components,
implying that the possibility exists of separating these components on the
basis of their frequency structure.
It seems that the most interesting point in treating the data under the form
of their frequency description must be found in the fact that taking the con-
volution of two functions f and h is equivalent to the multiplication of their
Fourier transforms
FILTER TECHNIQUES I N GRAVITY INTERPRETATION 227
As a consequence, no calculation of a discrete filter coefficient set has to be

performed, as the filtering can be done entirely in the frequency plane by
specifying the theoretical frequency response.
I n this connection the following procedure for the calculation of the upward
continuated field is plausible: (1) Calculate the Fourier transform G(u, u, 0)
of the measured part of the surface field Ag(x, y, 0). (2) Multiply G(u, u, 0)
with the theoretical frequency response H ( - ) ( u ,u, z ) over the Nyquist domain
of the frequency plane. (3) Inverse Fourier transformation of the product
H ( - ) ( u ,u, z)G(u, u, 0) yields the field a t height z.
Filtering by multiplication in the frequency plane is often preferred to
conventional convolution because no filter coefficients need to be calculated,
very complicated filters are readily applied by specifying the frequency
response, no data are lost a t the edges of the data grid, and a fast Fourier
transforming algorithm is available (Cooley and Tukey, 1965). I n spite of the
advantages of this procedure it is dangerous to rely upon it unconditionally.
It is a fact that only a finite part of the surface field is known or can be treated
with a computer; in addition, working with a Fourier series means that this
part is considered to be extended periodically over the (5, y)-plane, but the
true frequency structure of the field as a whole is generally much more compli-
cated than this finite portion would indicate. This provides the key t o the
structural problems we deal with later on when we perform the filtering in
the frequency domain.
In view of the fast divergence of the theoretical frequency response
H ( + ) ( uu,, z ) it is obvious that filtering in the frequency plane is meaningless
in the process of downward continuation, since the inverse Fourier transform
of the product H ( + ) ( u u, , z)G(u, v, 0) does not exist, in general. I n addition,
we remark that the signal component will be superposed by a random noise
field, finding its origin in observational errors, digitalization and numerical
treatment of the data, and in the composition of the upper part of the earth’s
crust itself (small, shallow masses). As the modern measuring instruments
are almost stable, one may assume the instrumental noise to be small, imply-
ing that the larger part of the noise is caused by the medium and digitalization
of the surface field.
For the problem of the separation of an observed field into signal and noise
components, the origin of this noise is immaterial, though the knowledge of
the statistical description of the noise is necessary. Therefore one assumes
that the random noise is homogeneous (stationary) and Gaussian; this hypo-
thesis is only acceptable when the space of the noise generating masses is
large compared to the space of the signal producing structures. The presence
of this noise leads to divergences in the mathematical expressions, governing
the downward continuation process. In this respect we shall apply the Wiener
filter technique for the optimalization of the downward continuation filter.
228 FRANS DE MEYER
4.2. Optimal Wiener Filtering

We assume that the input field fi(x, y) consists of a deterministic signal
s(x, y), superposed by an additive noise component n(x, y)
(4.2) fAX, Y) = 4x, Y) + n(x, Y)
The field n(x, y) is thought t o be a realization of a stationary and ergodic
random process.
Our aim is to extract the signal s(x, y) from the total message fi(s,y) by
means of a linear filter, specified by the impulse response h(z, y), with mini-
mum error in accordance with a chosen norm of error. Because of the presence
of this noise whose statistical properties generally are unknown, it is expected
that complete recovery of the signal by the linear filter transformation will
be impossible, unless the situation is an exceptional one. Nevertheless, we
define a desired output fd(x, y), which may be any operation on the signal
component of the input process.
The difference between the actual output, given by the convolution of the
input field with the filter function h(x, y)
and the desired output, is the system error
(4.4) eh(x, Y) =fO(z3 Y) -fd(x, Y)

These fields are usually interpreted as being realizations of stochastic pro-
cesses in two dimensions, obeying a probability law, so that the mathematical
expectation operator E is defined for the processes in question. Wiener (1950)
determines the optimal filter function hopt(x,y) by minimizing the mean-square
system error, defined by
(4.5) - W b 2 ) = E{[fo(x, Y) -fa@, Y)17
I n the case of stationary and ergodic random processes, this expected value
is often interpreted as being defined by
for any field g(x, y).

Substitution of Eq. (4.3) into (4.5)yields
FILTER TECHNIQUES KN GRAVITY INTERPRETATION 229
Here it is assumed that all the fields considered have zero mean values; if
this is not true for an observed field, we suppose that the arithmetical mean is
substracted from the field values.
Note how naturally correlation functions arise in this type of problem. It
must be said, however, that the validity of the expression (4.6)as the definition
of the expected value of a field g(x, y) rests on two assumptions: the first is
the hypothesis of ergodicity, stating that the required statistical information
about the stochastic process, of which g(x, y) is a specific sample function,
may be extracted from one typical realization of the random process. The
second assumption is the hypothesis of stationarity of the field g(x, y).
implying that the statistical quantities are independent of the particular
choice of the origin in the coordinate plane. As a result of these assumptions
we conclude that all statistical properties of the processes in question may be
evaluated by appropriate integrals over the (5, y)-plane, thus avoiding the
necessity of defining them by ensemble averages over the realizations of the
stochastic process.
The form (4.7)gives the mean-square system error for a particular filter,
as specified by its weighting function h(x, y). The quantity E{eh2)is now min-
imized as a function of the filter function h(x, y), in order to find that linear
filter hopt(z,y), giving the least possible mean-square system error. The
existence and uniqueness of this optimum filter are tacitly assumed, but will
be usually guaranteed from the physical presentation of the problem.
It may be shown (Bendat, 1958) that, a necessary and sufficient condition
that hopt(x.y) be the filter function of the optimal filter, given by the minimal-
ization of the mean-square system error (4.7), is that it is the solution of
the Wiener-Hopf integral equation
By substituting (4.10) into (4.7) we find that the minimum mean-square

system error resulting from this optimal filter function is given by
230 FRAKS DE MEYER
I n the case of gravity interpretation let us define the filter input to be the
surface field Ag(x, y, 0) and then assume that it consists of a signal component
s(x, y) and an additive noise n(x, y)
(4.12) Ag(x, yt 0) = 4x9 Y) + 4x3 y)

The desired field now becomes the signal component of the input field
(4.13) f , ( X , Y ) = 4x9 Y)
According to Eq. (4.10) the optimal impulse response hopt(x,y) of the linear
filter that passes s(x, y) and rejects n(x, y) satisfies the Wiener-Hopf equation
where Ago is a shorthand notation for the surface gravity effect, yago(x,y) is
the autocovariance of the surface field, and yAgo,s(x, y) is the covariance
between the input field and the signal component.
Applying the convolution theorem of Fourier transforms, the optimal
frequency response HOpt(u,w) follows directly from Eq. (4.14)
(4.15) HOpt(U, = pAgo, s(u, v)/pAgo(u, v,

where P,,,(u, w) is the power spectrum of the surface field, defmed as the
Fourier transform of the autocovariance yaso(x,y)
(4.16) PAg0(u,V ) = 1
m
-m
m
-m
u ~dy
yAgo(x,y ) e - i ( U z +d~ )
and Pdgo, s(u,v) is the cross-power spectrum between Ag(x, y, 0) and s(x, y),
that is the Fourier transform of yAgo. s(x,y).
The minimum mean-square system error may be expressed in terms of
power spectra. With the definition of the function
m m
(4*17) x(xt y) = Y S ( ~y)
, -
-m -m
hopt(c, 7)YAgo. - 5,!/ - 7)d6 dq
i t is easy to see that
(4.18)
But the Fourier transform of x(x, y) is given by the convolution theorem of

Fourier transforms
FILTER TECHNIQUES IN GRAVITY INTERPRETATION 23 1
Hence (inversion formula)
From (4.12) and (4.8) it follows immediately that
(4.20) Yago(”, Y) = ~ s ( xY)

, + ~ s n ( ~Y)
, + ~ n s ( xY), + Yn(Z, Y)
If it is assumed that the signal and noise are uncorrelated
Ysn(x, Y) = 0,
that is,
P s n ( u ,w) = 0 for any x , y, u, w,
we then have that
(4.21) YAgo(”? Y) = Ys(u, v, + Yn(u, w,
or in terms of frequency
(4.22) pAgo(u? v, = p s ( u , v, + Pn(u, v,
We also find that
(4.23) YAgo, S(, Y) = Y s ( x , y) Or pAgo. s(u, v) = ps(u,v)
Combining (4.22), (4.23), and (4.15) one finally obtains for the optimal
frequency response
(4.24) Hopt(w v) = PA% v)/[Ps(u,
v) + Pn(u, 41
and the minimum mean-square system error is given by (4.19)
(4.25)
E { ~ ; ~ ~(1/4n2)
J = Jm
-m
Sm--m
ps(U, + pn(u,
v ) p n ( u , v ) / [ ~ s ( v)
u, V)I . du dv
Now we define the modified downward continuation process by calculating
the downward projected field of the signal component of the surface field.
Hence, the field a t a desired depth will be the output of two linear filters in
series: the first filter extracts the signal component optimally from the input
field, and the second filter amplifies the signal part toward its source,
according to the theoretical filter H ( + ) ( u ,v, 2). This is just the procedure
already mentioned in Eq. (3.35), where now the frequency response W ( u ,w) is
replaced by the optimum frequency characteristic Hopt(u,w) of the Wiener
filter.
232 FRANS DE MEYER
Suppose that fi(x, y) is the input to a first filter with filter function h,(s, y);
the outputf,’(x, y) of this filter serves as input to a second filter with weighting
function h,(x, y). Then it is easy t o see that the filter function of the compound
filter is the convolution of both weighting functions
whence
(4.27) H(% 9) = HI(%,vfH,(%,4
Indeed, according to Eq. (2.7),the output of the compound filter is
if the filter function of the compound filter is defined by Eq. (4.26).

Combining (4.26), (4.24), and (3.23) one sees that the frequency response
of the modified downward continuation filter is given by
(4.29) H$;i(u, w)/[P,(u,

w, z ) = Ps(u, + +
v) P,(u, w)] - exp[z(ua w ~ ) ~ ~ ]
and we conclude that the optimal frequency response of the modified down-
ward continuation is the product of the theoretical frequency response
H(+)(u, w, z ) of the downward continuation filter with the frequency character-
istic of the optimum Wiener filter t o pass the signal s ( ~y), and reject the noise
n(x, Y).
I n conclusion, the following procedure is arrived at, based on the optimal
Wiener filtering technique: (1) Calculate the Fourier transform C(u, w, 0) of
the surface field Ag(x, y, 0). (2) Compute the power spectrum P&,(U,w) of
the surface field. (3) Separate PAgo(u, w) into signal and noise components
P,(u,w) and P,(u,w). (4) Multiply Q(u,TI, 0)by Hi;:(u, v,z) over the Nyquist
domain. ( 6 ) Inverse Fourier transformation of the product H:;:(u, v, z )
G(u,v, 0)yields the downward continuated field at depth z.
When stating Eq. (4.29)i t was implicitly assumed that the output of the
filter (4.24)defines a band-limited process, that is, that the E’ourier transform
of the residue left in the surface field after Wiener filtering vanishes outside a
finite region of the frequency plane.
FILTER TECHNIQUES IN GRAVITY LNTERPRETATION 233
I n order to apply the Wiener filtering method, several restrictions need t o be

considered. The optimal downward continuation filter is constructed as a
function of the observed surface field and this is in conformity with our
conclusion a t the end of Section 3.5. But the fundamental problem in applying
Eq. (4.29)is to obtain the necessary statistical description of signal and noise,
that is, the separation of the estimated power spectrum of the surface field into
spectra corresponding t o the signal and noise components. This leads to the
adoption of a statistical model for the random noise. I n the absence of any
statistical information, the easiest (and perhaps the best physically motivated)
assumption is that of white noise, and then estimate the white noise level
from the power spectrum of the surface field. However, such a separation will
remain subjective, as it is a priori impossible to estimate the noise term
correctly without knowing explicitly all of its statistical properties. This can
be immediately related to the ambiguity of the inverse potential problem,
excluding an unambiguous separation of the input field.
Another weak point in the Wiener filtering method is the hypothesis that
the noise and the signal are uncorrelated. This is a strong requirement. For
gravity fields, derivable from a potential, this assumption could be made
acceptable by defining the noise as the part of the surface field that finds its
cause in random observational errors, digitalization, and the distribution of
small, shallow disturbing masses.
The Wiener filter theory is justified in the case where the fields in question
are realizations of Gaussian stochastic processes. Only in this case will a
linear filter be able to minimize the mean-square system error, as defined by
Eq. (4.5) (Clarke, 1969; Van Trees, 1968). However, in the absence of any
statistical information the Wiener filter theory offers the best method for the
available information.
We also have assumed that all statistical quantities, such as covariance
functions, may be calculated by averaging a single sample function of the
stochastic process over the (2,y)-plane. This corresponds to the ergodicity
hypothesis of stationary random processes. We think that more weight ought
to be given t o the question whether or not as measured field belongs to a
stationary and ergodic random process, although admittedly this problem is
an extremely difficult one.
4.3. Strakhov's Method for Extraction of Potential Field Signal

Let Ag(x, y, 0) be continuously defined over the horizontal plane z = 0
and let it be expressible in the form
(4.30)
where s(2, y) is some deterministic (nonrandom) field, called the signal,
234 FRANS DE MEYER
and n(x, y) a two-dimensional random function. We shall make the following

assumptions regarding these components:
(1) .s(x, y) is absolutely and square integrable
and
(4.32)
(2) n(x, y) is a realization of a stationary random process with zero mean
(4.33) E{n}= 0
and autocovariance
(4.34) m(5, 7) = E{n(x, Y)n(x+ 4, Y + 7))
The physical origin of the noise is immaterial; unfortunately, the statistical
characteristics of the noise are not known beforehand and this is a serious
disadvantage of Wiener's method. This difficulty is partially overcome by
Strakhov's method (1964, a, b), reviewed by Naidu (1966, 1967).
Suppose that the observed values in (4.30) are smoothed by a linear filter
with weights hk,, -M < k M , - N < I < N ; then one can write for the
output of this filter
(4.36)
and
(4.37)
the smoothed versions of the signal and the noise; Ax and Ay stand for the
grid spacings in the x- and y-directions, respectively.
The mean of the filtered noise is zero, since
(4.38)
in view of the assumption (4.33), and the autocovariance is given by
We attempt t o determine the filter coefficients h,, such that (1) the dis-
torted signal B(z, y) is as close to ~ ( 2y), as possible (in some metric), and
(2) the variance us;of the random process, of which ii(x, y) is a realization, is
as small as possible.
The distortion of the signal caused by the linear operator h,, may be esti-
mated by the &-metric. Using Parseval's theorem for double Fourier trans-
forms, we obtain
(4.40) JIs -Sllz, = 11 X -81IL, - Sm

-m
[S(u, v) -8(u, v)], du dv
where X(zc, v) and 8(u, v) are the Fourier transforms of s(x, y) and 9(z, y),
respectively. From (4.36)it immediately follows that
(4.41) B(u,8 ) = H ( u , v)S(u,v)
where H(u, v) is the frequency response of the discrete filter h,, ,
Combining (4.41) and (4.40)yields
Polya (1929) shows that every potential field signal has a unique smallest
convex region containing all singularities of the potential field. If we let d
stand for the depth of the uppermost singularity we have the following
inequality
(4.44) I Q(u,v, 011 I
(1/2r)llAg(x,y, d - E ) I I L ~exp[-(d - &)(u2+ +')1'21
where E is a small, positive number. We also note that lJAg(s,y, d - &)/IL1 is
a constant and may be interpreted as the total excess mass within the
anomalous region. Introducing Polya's inequality into (4.43)results in
s, I,
m m
exp[ -2d(u2 + - H ( u , v)], du dv
The following condition minimizes the signal distortion on the L,-metric
(4.46) Srn I
-m
m
-m
exp[-2d(u2 + va)1/2][1 - H ( u , v)], du dv = minimum
236 FRANS DE MEYER
“he simple minimization of the form (4.46)is not sufficient, as the filter co-
efficients could then be easily determined from the condition that H ( u , v ) = 1
over the frequency plane. Indeed, the condition that the variance of the
reduced noise be small must still be satisfied. Equation (4.39)readily gives for
the variance of the reduced noise
Dividing both sides of this expression by the variance yn(O,0) of the noise
n(x, y) it follows that
If the noise is approximately random we obtain the simple form
(4.48)
and this has been called the “prescribed noise reduction factor” by Naidu
(1967).
In practice it is impossible to estimate the autocovariance of the noise with
a sufficient accuracy; therefore, all the terms in Eq. (4.47),except the fist,
are indefinite. Strakhov’s method consists of minimizing the form (4.46),
with the condition that the noise should be reduced by a prescribed factor h2.
The remaining terms in Eq. (4.47)may then be interpreted as correction
factors t o be applied to this prescribed noise reduction factor for a given
statistical structure of the noise n(x, y). The filter coefficients h,, can be
found from the conditions (4.46)and (4.48)by a modification of the method
of steepest descent; a computer program in FORTRAN I1 language can be
found in the paper by Naidu (1967).The single parameter to be estimated
from the surface field is the depth d of the uppermost singularity of the poten-
tial field.
It is interesting that Strakhov’s method does not explicitly require know-
ledge of the statistical properties of the signal, neither of the noise. However,
Naidu’s introduction of a prescribed noise reduction factor is based upon a
white noise assumption, and this is very similar t o what has been said about
Wiener’s method. I n Strakhov’s method the signal distortion and variance
FILTER TECHNIQUES W GRAVITY INTERPRETATION 237
of the noise are minimized separately, while in Wiener's method the sum of
signal distortion and random noise is minimized. Wiener's method has the
advantage that the filtering is performed entirely in the frequency plane;
the method of Strakhov f i s t supplies the filter coefficients, which yield a
smooth output of the surface field by discrete convolution, and then this
filtered signal may be transformed to the frequency plane. It is difficult to
say which method is superior; the choice is left to the interpreter.
4.4. Digitization of a Continuous li'ieield-Aliasing

The analogy between Section 3.5 and Section 4.2 is quite remarkable.
When the downward continuation is performed by the conventional con-
volution procedure we have to simulate the downward continuation frequency
response by a linear, discrete filter, but the calculations are carried out in the
(x, y)-plane. The main problem then consists of obtaining a sufficient approxi-
mation to the modified downward continuation filter. If, however, the filtering
is performed in the frequency plane we can use the theoretical downward
continuation filter H ( + ) ( uv,, z ) , but the resulting problems are related t o the
calculation of the Fourier transform of the digitized surface field of finite
extent. Therefore we will now concentrate on the relation between the Fourier
transform of the continuous field as a whole and the Fourier spectrum of the
observed part of the field.
We suppose that the surface field f (x, y), distributed continuously over
the (2, y)-plane, is sampled on a rectangular grid, with spacings Ax and Ay
in the x- and y-directions, respectively. This yields the digitized field
(4.49) f (k Ax, 1 AY) = [do(X,Y) Y)f (2, Y)l*=kAz

Y=lAy
1 5 k < rn, 1 < 1 < n, where S(x, y) is the infinite Dirac comb in two dimen-
sions (Blackman and Tukey, 1958, p. 71)
m w
(4.50) 6 ( ~y), = A X Ay 1 C
a= - r = -
w m
S(S - q Ax) S(y - T Ay)
and
(4.51) do( x,y) = 1, A x < x < rn Ax, Ay< y < n Ay
= 0, otherwise
The aliased, finite version of the Fourier spectrum, corresponding to the

observations f (kAx, 1 Ay), is now defined by
m a-
P J u , v) = C C
k=-m I=-m
f (k Ax, 1 Ay)exp[ - i(u kAx + vl Ay)] Ax Ay
238 FRANS DE MEYER
which can be written as
Using the convolution theorem for Fourier transforms, this expression becomes
(4.52) Fa(u,W ) = D,(u, V) * V ( UV), * F(u, V)
where * denotes convolution and D,(u, v), V ( u ,v), and F(u,w ) are the Fourier
transforms of do@, y), 6(z, y), and f(z, y), respectively. Since the Fourier
transform of the infinite Dirac comb 6(x, y) is an infinite Dirac comb in
frequency
m
(4.53) V(u,v) = 1
q z -m r= -m
6(u - (2nqlAx)) 6(w - (2~rrlAy))
the form (4.52) becomes
with
(4.55) D,(u, v) = 4mn Ax Ay(sin nu Ax/nu Ax)(sin mVAy/mv Ay)
The expression (4.54) must be interpreted as follows: suppose that the
sampled version off(z, y) is available all over the (5, y)-plane. Then Eq. (4.49)
becomes
f(k AX, 2 Ay) =[A(z,y)f(X, Y)],=,AI., - 00 Ik 5 a,- 00 I25 00

y = 1Ay
and (4.54) reduces to

m
(4.56) 1 f
P,(u, v) = q = - m T=-w
P(u - (2nq/Ax),v - (2nr/Ay))
I f the sampling rates Ax and Ay are so chosen that f (x, y) contains no waves
of frequencies outside the Nyquist domain
(4.57) --n/Ax u n/Ax, -n/Ay Iv < r / A y
then it is completely determined by giving its ordinates a t the grid points.
As no information about f (2, y) is available between the grid points, there is
no means of directly estimating the amplitudes of frequencies outside the
Nyquist domain. I n other words, Eq. (4.56)shows that the sampled Fourier
spectrum Pa(u,w ) is obtained by folding the true Fourier spectrum F(u, v)
into the Nyquist domain and adding these contributions inside the low frequ-
ency region. This spectrum distortion has been called “aliasing” by J. W.
Tukey. If f(x, y) has the property that its Fourier transform F(u,v) is
essentially zero outside the Nyquist domain, then the terms for q # 0 and
r f O in (4.56) do not contribute to the sampled Fourier spectrum and the
digitization process loses no information. Since the Nyquist frequencies
uN = rlAx and vN = n1Ay are controlled by the grid spacings, we conclude
that the practical application of the concept of the sampled Fourier spectrum
consists of choosing grid spacings that are small enough to reduce the aliasing
to an acceptable level.
Furthermore we note from Eq. (4.64)that iff(x, y) is sampled on a rect-
angular grid of finite dimensions, then the aliased Fourier spectrum Fa(u,v)
will be the convolution of the sampled Fourier spectrum P,(u, w) with the
Fourier transform D,(u, v) of the function d,(x, y), thus introducing a second
kind of frequency distortion. Even if aliasing were absent, the true Fourier
spectrum F(u,v) would be blurred out by the frequency kernel D,(u,v).
The power spectrum may be treated in a completely analogous way.
Indeed, the best one can obtain from the sampled field is the knowledge of
the ‘‘ aliased )’ autocovariance
(4.58) ya(k AX,2 AY)=[ddx, Y) &x, Y)Y(X, ~)lz=rcAz
Y = IAY
where y ( x ,y) is the true autocovariance off (x,y). It immediately follows that
the sampled, finite version of the power spectrum P(u,v) is given by
and the above results hold for P,(u,w) and Pa(u,v).

Another relation between Pa(u, v) and P(u,v) may be derived if we note
that Eq. (4.58) can be written as
(4.60) ya(k AX, AY)= [arnn(z, Y)Y(Z, ~ ) I z = kAz

Y = IAY
where Smn(x,y) is the finite Dirac comb in two dimensions
(4.61) amn( 2, Y) = %(x) %(Y)

and
(4.62) Srn(s)= 4Az[S(z + rn Ax) + S(x -m Ax)] + Ax mfl

q=-m+l
6(z - q Az)
240 FaANS DE MEYER
In complete analogy with Eq. (4.59) the aliased, finite power spectrum is
given by the convolution
(4.63) PJu, v) = Jm Jm Vmn(u

- u', v - v')P(u',v’) du' dv'
- w -m
Substituting (4.62) into the expression
and using the fact that 6(x - xo) and e - * " + O are Fourier transform pairs,
together with the trigonometric relation
m
(4.64)
k=O
cos ku = cos )(m + l ) u sin imulsin ?p
we find that
(4.65) v) = A x A y cotg t u A x cots +v A y sin nu A x sin mu Ay
Vmn(u,
Hence, the aliased power spectrum may also be regarded as the convolution
v).
of the true power spectrum with the aliased kernel Vmn(u,
4.5. Estimation of the Power 8pectrum
When writing Eq.(4.58) it was explicitly mentioned that the best one could
obtain from the sampled field f (k Ax, 1 Ay), 1 k <m, 1 I n, are the<<
values y ( k Ax, I Ay), - m +
1 < k < m - 1, - n +
1 < I < n - 1 ofthe true
autocovariancey(x, y) of the surface field as a whole. This surely is an optimi-
stic statement. Indeed, the indirect method of power spectrum estimation
starta with the computation of the following estimator of the autocovariance
(4.66)
OIlLn-1
where it is assumed that the mean of the surface field is zero.Parzen (1957a)
advocates the use of the estimator
FILTER TECHNIQUES IN QRAVITY WTERPRETATION 241
It is well known that jjo(k, I) is an unbiased estimate of the autocovariance

of f(x, y), with the expected value given by
(4.68)
€or any finite m and n, but fo(k, I) is biased for finite m and n and asymptoti-
cally unbiased, since
(4.69) E{fdk, 4 ) = hm(k)hn(l)y(k,1)
with
hm(k)=l-(lkl/~), -m+l<k<m-l
(4.70)
= 0, Ik( > r n - 1
Bias, however, is not the only, and certainly not the most important criterion
for a "good" estimate and a preference for the biased estimate fo(k, 1) is
sometimes voiced on the grounds that i t has, in general, smaller mean-square
error than the unbiased estimate jjo(k, 1) (Parzen, 1957b).
These estimates are computable for values of k and 1 up to 1 k 1 = m - 1
and I1 I = n - 1, but jjo(m- 1, n - 1)is then estimated by the single product
f(1, l)f(m, n). Hence we cannot estimate y(k, 1) beyond Ikl = m - 1 ,
12 I = n - 1. It was empirically observed by M. G. Kendall that the sample
autocovariances fail t o damp down to zero for increasing values of k and 1
(rn and n), although the true autocovariance does damp down to zero €or
x + 03 and y + 03. To avoid the discontinuity of the estimates a t the end
points, one introduces a n even function d(k, Z), called a covariance kernel (or
covariance window), and subject to the restrictions that d(0, 0) = 1 and
d(k,Z)=Ofor I k ] > m ' - l a n d 111 > n ' - l , w h e r e m ' < m a n d n ' < n a r e
integers. The modified autocovariance estimates are then defined by
jje(k, I ) = d(k, l)vo(k, 11, fe(k9 1 ) =d(k, Z)fo(k, 1)
(4.71)
-m' + 1< k I m ' - 1, -n' + 1st I n ' - 1
and i t is interesting to see that jje(k, I) and fe(k, I) are defined for all k and 1
and vanish for I kl >m' - 1 and 121 > n' - 1, although the estimates
jjo(k, I) and f,(k, I) are not defined there.
I n consequence, the modified power spectrum is defhed by
(4.72)
or
(4.73)
242 FRAXS DE MEYER
It may be shown that these estimates of the power spectrum are asympto-
tically unbiased since, for example, the expected value of PJu, w) is given by
(4.74) E{P,(u, w)} = D(u, W ) * D,(u, V ) * V ( u ,V ) * P(u, W )
where D(u, v) is the Fourier transform of d ( k , I ) , and D,(u, v) and V ( u ,v)
are given by Eqs. (4.55) and (4.63), respectively.
Because the variance of Pe(u,w) is approximately given by
1 1
(4.75) variP,(u, v)) 21 (m‘/m)(n’/n)Pz((u,w) J”
-1
J -1
~ ( sy),dx dy
for large m and n, where the function h is defined by

4%y) = h(z/m’,yfn’),
i t may be expected that P,(u, w) is a consistent spectral estimate if we
choose the parameters m’ and n’ in such a way that m’lm -+0 and n’/n -+ 0
as m --f m and n -+ co.Otherwise one obtains a reduction in the variance by a
factor that depends upon the ratios m’/m and n‘ln and the covariance kernel
used.
The starting point in the direct method of power spectrum estimation is
the Schuster periodogram, which can be defined in the form
(4.76)
Elementary algebraic manipulation immediately yields that
(4.77) Pmn(u,w) =
m-1
c
k = -m+l
n-1
1 X,(k)h,(Z)~,(k, Z)e-t(ku+iu)
I= -n+l
from which the expected value of the periodogram estimates follows

m-1 n-1
E{Pmn(u,
w)} = C C
k = - m + l I = -?It1
X,(k)X,(Z)y(k, Z)e-icku+Iu)
or
(4.78) E{Pmn(u,w)) = Sn sn
-n -n
A,,(u - u ,u - v’)P(u’, v’) du’ dw
with Amn(u,u) the two-dimensional Fhjer kernel, associated with the periodo-
gram estimate of P(u, w)
(4.79) Amn(u,v) = (l/rnn)(sin&mu/sin&)“sin +nw/sin + w ) ~
The expected value of Pmn(u,w) is not P(u, v), but a weighted integral of
the power spectrum and, therefore one has to conclude that the periodogram
is a biased estimate of the density spectrum of the continuous field for finite
m and n. But in the limit as rn -+ ca and n + co,the FBjer kernel h,,(u, v)

approaches the two-dimensional Dirac distribution and the periodogram
becomes an asymptotically unbiased estimate of P(u,v). However, it is not a
consistent estimate since the variance of the periodogram estimates is given
approximately by
lim var{Pm,(u, v)} N P2(u,v), u,v # 0 or f7r
(4.80) ::,"
!x2Pa(u,v), a,v = O or &7r
Consequently, if we want to use this approximation for the power spectrum,

we are forced t o look for methods t o reduce the variance of this periodogram.
This lack of consistency is not due to any fundamental inefficiency of P,,
(u,v), since the periodogram can, in the case of a normal stochastic process,
be regarded as the maximum likely estimate ofthe power spectrum and should
hardly be capable of improvement from the purely statistical point of view.
There are several methods of reducing the variance of the periodogram.
One of them consists of smoothing the periodogram values Pmn(r, s),1 Ir m,
1 5 s <n by averaging over adjacent values; hence we define the smoothed
periodogram by
(4.81)
where the coefficients W(h,p ) are the weights of a suitable filter. It may be
shown that the expected value of the smoothed periodogram will be given by
P q
(4.82) lim E{P,,,,(r, 8 ) ) N P(2vr/m, 2mln)
m- m
1 1
A= - p fl= - q
W(h,p)
n- m
This relation produces the criterion by which the filter coefficients should add
up to unity, in order to obtain an asymptotically unbiased estimate of the
density spectrum.
The variance of the smoothed periodogram is given to a good approxima-
tion by
(4.83) lim var{f',,,,(r,

m- m
5)) N P2(27rr/rn,2ns/n)
a=
P
1 c W2(h,p)
q
- p fi= - 4
n- m
and one sees that the variance of the unsmoothed periodogram will be de-
creased by a factor
(4.84)
The variance (4.83) is minimized, for given p and q, by the sequence

W A , p ) = 1/(2P+ 1)Pq + I), -P <A I
P,-q p Iq
I
244 FRANS DE MEYER
giving a reduction factor
(4.85) R =a =f-,, 2 W P + 1I2(2q+

-4
=1 P P + 1)(2q+ 1 )
Hence, we conclude that, for averaging over nine adjacent ordinates ( p = 1,
q = I), we have that R = 1/9.
Another method consists of dividing the surface grid into K nonoverlapping
blocks of equal sue. For each block k the periodogram Pi?, an(^, v) is com-
puted and averaged over the K blocks, leading to the spectral estimate
K
(4.86) piz. An(% v, = ( l / K )
k=l
pgA. An(% v,
where m = M Am, n = N An, K = M N , and Am and An are the dimensions

of each block. From (4.86) i t immediately follows that the expected value of
this spectral estimate is approximated by
(4.87) EiPh?, An(U, v)) p(%v)
for large m and n and that the estimate is asymptotically unbiased.
As the variance is given by
(4.88) var(Ph%. an(^, w)) 2: (l/K)P"(u,
v), U, w # 0 or f r r
the variance of this spectral estimate is reduced by approximately a factor
1/K, if the blocks do not overlap.
Still another method consists of multiplying the observations f (k,I ) by a
data kernel, which is a function that is equal to unity for the larger part of
the surface grid, but vanishes outside this region. For further computational
details the reader is referred to the list of papers in the references.
4.6. Convolution Filtering versus Frequency Filtering

From the above sections i t is clear that the interpreter has two alternatives
whenever a atering operation on a large map has to be performed: either he
decides that the filtering ought to be done in the space domain, implying that
he needs a set of convolution coefficients, simulating the theoretical frequency
response, or the filtering may be carried out in the frequency plane by simple
multiplication of the Fourier transform of the map with the desired frequency
response. Let us now examine some advantages and disadvantages of both
methods.
4.6.1. Convolution Filtering. (1) The problem in conventional filter design

is that of obtaining the best fit between the frequency response of a discrete,
linear filter and the desired frequency characteristio with a minimum number
of coe%cients. A small coefficient set will minimize the number of arithmetic
operations involved in the discrete convolution.
FILTER TECHNIQUES IN GRAVITY lTTERPRETATION 245
(2) The frequency response is usually band-limited; hence the corresponding

filter function is infinite in length and needs t o be tailed off. However, a
steep cutoff for the frequency response is needed in order to minimize the
overlapping of the spectra of different anomalies in the surface field. Neverthe-
less it is a fact that several anomalies may contribute t o the same region of
the frequency domain and therefore i t is a priori impossible to separate
unambiguously a surface field into components, which may be assigned to
different structures. But the finite size of the set of filter coefficients limits
the steepness of the cutoff of the frequency response and unwanted compon-
ents may be passed by the finite filter.
(3) Another reason for requiring small filters may be found in the fact that
fewer points will be lost at the edges of the data grid in the convolution process.
In order to minimize fluctuations in the frequency response the coefficients
have t o be attenuated toward the edges of the atering grid, favoring again
a loss of steepness a t the cutoff.
(4) We have also seen that rectangular or square shapes of the data grid
and of the coefficient set introduce directional effects in the frequency domain.
These effects may be rather severe and ought to be treated accordingly.
This difficulty may be avoided by defining coefficient sets with a circular
symmetry.
(5) The main problem of conventional filter design is concerned with ap-
proximation techniques, but the data map will be left unchanged and no Fourier
transformations need t o be performed.
4.6.2. Frequency Filtering. (1) One advantage of frequency filtering is that

no coefhient array has to be calculated and the availability of fast multi-
dimensional Fourier transform methods indicates a routine in which filtering
can be performed by multiplication in the frequency domain, thus reducing
considerably the computing time. Very complicated filters are readily applied
and one works with the best possible frequency response.
(2) No data are lost a t the edges of the data grid. One way to reduce this
loss in the convolution method is to surround the data with zeros or to use the
measured data t o extrapolate the known part of the surface field. This extra-
polation may be done by fitting a polynomial surface t o the observations.
The procedure of filtering in the frequency plane reduces these edge effects
because it extrapolates the data by fitting a trigonometric series to the ob-
servations, and the measured part of the field may be thought to be extended
periodically over the space domain. It is also wise to extract a linear trend
from the data, so that the new data have zero mean and zero regional trend.
Indeed, a regional trend introduces spurious components a t all frequencies
and hence distorts the power spectrum in the Nyquist domain.
246 FRANS DE MEYER
(3) The main problem of frequency filtering is evidently concerned with the
fact that we work with a Fourier series of discrete data and the relation with
the true Fourier transform is always problematic, as shown in Section 4.4.
The practical computations of power spectra also impose several restrictions.
I n consequence it is not uncommon to observe distortion in the size and shape
of the anomalies of the transformed field. These deviations evidently arise
from the fundamental fact that the Fourier series is an incomplete approxi-
mation t o the exact Fourier transform of the field as a whole. I n this con-
nection the importance of the problem of aliasing must be recognized: if the
grid spacings are so chosen that the Fourier spectrum contains appreciable
power above the Nyquist frequencies, then we may expect that the calculated
spectra may differ considerably from the true spectra at low frequencies,
that is in the region where the frequency contributions of the broader anom-
alies are situated.
Considering these deficiencies of both methods i t is difficult to decide which
procedure should give the most reliable results. It is true that convolution
filtering is more time consuming, but frequency filtering supposes knowledge
of the desired output and involves individually designed filters for each
problem. A choice between both methods may be voiced on the grounds of
practical considerations regarding computer time and storage. Until the
advent of fast Fourier transform techniques, filtering in the frequency plane
demanded excessive computing time as compared with convolution. This
leads to a brief examination of the mechanics of Cooley-Tukey's method
(1965).
The discrete Fourier transform of a one-dimensional sample function f (k),
0 5 k 5 m - 1, is defined by
m-1
(4.89) F ( r )= C f (k)e-zn*kr/m, O r m -1
k=o
To solve for F ( r ) one would normally select a value for r and perform the
summation over k. Suppose now that m is factorable and that it may be ex-
pressed as
(4.90) m = mlmz
The indices k and r are now written as

k = k,m, + k o , 0I
ko <rn, - 1 , 0 2 k, 2 m, - 1
(4.91)
r = rlml + ro , 0 < r o < m1 - 1,O < r , <m2 - 1
With these notations Eq. (4.89) becomes
From the periodicity of the complex exponential it follows that

e-2nlkim2rirn - e-2ntkim2ro/m
and the inner sum over k, depends only on r o ; therefore one can define a new
array
(4.92)
and the result may be written as
(4.93)
This algorithm can be generalized if one can express m as a product

m = mi . - .mp and the above reasoning may be repeated for each step.
The fast Fourier transform of two-dimensional data may be written in
terms of one-dimensionaltransforms. Indeed, considering the two-dimensional
analog of Eq. (4.89)
(4.94)
we first calculate the transforms of the rows

m-1
Pl(k,s) = C j(k,
l=O
Z)e-2""S'"
and then take the transforms of the columns

m-1
F(r, s ) = C Pl(E,s ) e - 2 n i k r ' m
k=O
where F,(k, s) is the Fourier transform of the kth column.

One of the interesting properties of this method is the considerable re-
duction in computer time. I n the classical method the number of operations
required to compute a discrete Fourier transform of a N x N matrix is
approximately N4, whereas by the fast Fourier transform it is about 4N2
log, N (Cochran, 1967), if we use a radix 2 algorithm. Thus the number of
operations is reduced by a factor 4 log, NIN2. For N = 128 = 27 points,
this factor equals 28165536 2 112340 and the fast Fourier transform should
be 2340 times faster than the conventional method.
I n connection with filtering it is known that, for two-dimensional data with
N2points, the number of multiplicabions involved in filtering in the frequency
+
plane is about 4 N 2 log, N gNz; convolution filtering with M 2 weights
requires M 2 N 2 multiplications. For N = 266 and M = 20 we obtain a reduc-
tion factor of the order of 1/14 and frequency filteringshould be 14 times faster
than convolution filtering.
248 FRANS DE MEYER
5. CALCULATIONOF DERIVATIVES
OF HIGHER
ORDER
In the last twenty-five years much effort has been devoted to improving the
methods of calculation of derivatives of higher order of a surface field,
derivable from the gravity potential. The second vertical derivative in
particular is a very interesting aid in the interpretation of local features.
Because the anomalous gravity effect Ag(x, y, z ) satisfiesthe Laplace equation,
the following relation between the vertical second derivative and the two
horizontal derivatives in the directions x and y exists
(5.1) a2 Ag/az2 = -(a2 AglaX2 + a2 Ag/ag)

Therefore the second vertical derivative may be interpreted as a measure of
the total curvature of the anomaly surface, in two orthogonal directions. It
is clear that the curvature of the anomaly surface is greater for local than €or
broader structures; in this context the second vertical derivative may be
used to amplify the contributions of small, shallow masses to the surface
gravity map, and the method is based upon the fact that nearby sources have
greater influence on gravity gradients than on gravity itself.
Garland (1965) pointed out that the edges of an isolated structure are
often indicated on the anomaly surface by the inflection points of the surface;
as these are the points where the second vertical derivative becomes zero,
the zero contour of the second vertical derivative map may be interpreted as
showing roughly the edges of the anomalous masses. However, in view of the
fact that an observed gravity field is the result of the superposition of a very
complicated mass distribution, this zero contour w il
l only outline approxima-
tely the positions and the areal extent of the anomalous masses. Nevertheless,
one may conclude that, whenever the spacing of the surface grid is small and
the precision of the measurements is high, the second derivative method is
particularly interesting for resolving and sharpening anomalies of small areal
extent.
We will next show how the vertical derivatives define linear filters, amplify-
ing the Ecequency contributions of local effects. From Eq. (3.24)it follows that
the field at depth z may be written as
(6.2) Ag(x, y, z ) = (l/4ra) 1- 1- G(u, O)exp[z(u2+

m m
m
v, du dv
Differentiating both sides of Eq. (5.2)n times with respect to z results in
(6.3) Sm
(an1a.q Ag(x, y, 2) = (1/4~2)- m -m
zt, o)(u2+ V
a(%, 2 y
+
x exp[z(ua v2)1/2]ei(ut+uY)du dv
and in the limit as z --f 0

(5.4)
w
[ ( P / WAg(z,
) = ~(1/4r2)J w
y, z ) ] ~ =
-w
J- w
+
G(u, w, O)(u2 vz)nizei(us+uv)dudv
Writing G ( n ) ( v,
~ ,0) for the Fourier transform of the vertical derivative of
order n a t the surface we have the following frequency relation
(5.5) G(")(u,W, 0) = (uZ+ v')~/'G(u,V , 0)
Hence, the vertical derivative of order n defines a linear filter on the surface
gravity effect with frequency response
(5.6) H(")(u,w) = (u2 + v2p2
It is also obvious that the derivatives of higher order are independent of

the internal mass distribution and can be computed unambiguously, for
any observed surface field. The appearance of the function H(n)(u,v) in
Eq. (5.5) is the mathematical justification of the observation that small
anomalies and observational errors will be more amplified than the broader
features in the surface field.
I n particular, Elkins (1950) showed that the second vertical derivative may
be approximated by a discrete, linear filter of the form
N
(5.7) a2 &/az" 2: (1/so2)
k=O
1 hk Y, Blc)
where so is the grid spacing of the square grid. Several sets of coefficients h,
were derived by Elkins (1950), Rosenbach (1953), Henderson and Zietz
(1949),Peters (1949), and Henderson (1960). Mesk6 (1966) gave the frequency
responses of these discrete filters and indicated that Elkins' filters do not
give good estimates of the second vertical derivative, because the discrete
frequency response matches the theoretical frequency characteristic u2 w2 +
only in the low frequency region. However, although Elkins' filters are low-
pass filters and do not amplify the high frequency waves unnecessarily, they
give easily interpretable results, but the corresponding output maps may not
be interpreted as being second derivative maps. The other convolution co-
efficient sets yield better approximations to the second derivative frequency
response and therefore estimate the second derivative with higher accuracy,
but the structure of the output maps is often very complicated.
Anyhow it is important to amplify only the waves which are characteristic
for the major parts of the mass distribution and t o reject the noise field in the
surface data. Consequently we are led to consider, once again, low-pass filters,
cutting off rather sharply. Adequate coefficient arrays can be constructed
with the methods of Sections 2.4 and 2.5, where now the general frequency
250 FRANS DE MEYER
characteristic H ( u , v) has to be replaced by HCn)(u,

v). This requires a detailed
inspection of the power spectrum of the surface field and the interpreter must
decidewhich waveshe wants to amplifyandwhich partsof theEburierspectrum
he has to reject, thus gaining more control over the efficiency of his calcula-
tions. From Section 4.2 it is also evident that the Wiener filter method will
be very useful, because all we have to do is to replace the frequency response
H ( + ) ( uv,, z ) by HCn)(u,w).
APPENDIX1.
PETERS’
AND HENDERSON’S
FILTERS
FOR UPWARD
CONTINUATION
Peters Henderson
Rk
h=l h=2 h=l h=2 h=3 h=4 h=5
4 0 0.14645 0.05279 0.11193 0.04034 0.01961 0.01141 0.00742

2/1 0.21132 0.09175 0.32193 0.12988 0.06592 0.03908 0.02566
1/2 0.14943 0.11388 0.06062 0.07588 0.05260 0.03566 0.02509
4 5 0.12645 0.12541 0.15206 0.14559 0.10563 0.07450 0.05377
48 0.05335 0.07651 0.07146 0.05841 0.04611
48.5 0.08627 0.11512
2/13 0.06586 0.09902 0.10226 0.09173 0.07784
1/17 0.07771 0.12062
1/25 0.06650 0.11 100 0.12921 0.12915 0.11986
1/34 0.05276 0.09122
1/60 0.05635 0.10351 0.13635 0.15474 0.16159
458 0.03462 0.06369
499 0.02055 0.03896
d125 0.09459 0.18656
4136 0.03855 0.07379 0.10322 0.12565 0.14106
4274 0.02273 0.04464 0.08323 0.08323 0.09897
4625 0.03015 0.05998 0.08917 0.11744 0.14458
SUm 1 1 0.98003 0.96014 0.94043 0.92100 0.90195
APPENDIX2.
COEFFICIENTS
FOR THE CONTINUATION,
UPWARD h =1
1/0 0.13756 0.13353 1/0 0.13767 0.13852

1/1 0.23755 0.24163 2/1 0.23676 0.23594
1/2 0.13161 0.13578 2/2 0.13159 0.13346
4 4 0.04998 0.05134 4 4 0.05149 0.05508
4 5 0.07890 0.08909 d 5 0.0 7934 0.07335
2/8 0.02868 0.02379 2/8 0.02624 0.01975
4 9 0.02734 0.02922 4 9 0.02354 0.02050
d10 0.03527 0.04816 2/10 0.03495 0.03521
1/13 0.02339 0.03711 1/13 0.02829 0.03215
1/17 0.02736 0.03371 1/16 0.00825 0.00200
1/20 0.01711 0.00840 1/11 0.02091 0.02656
1/25 0.03324 0.00326 2/20 0.01493 0.01775
1/34 0.00426 0.01371 2/25 0.02984 0.00931
1/40 0.02336 0.01527 1/29 0.01124 0.02277
1/50 0.01388 0.02293 1/34 0.00193 0.01223
1/65 0.02096 0.01221 1/40 0.02277 0.03995
1/85 0.01488 0.02686 1/50 0.01665 0.01536
1/100 0.01433 0.01893 1/65 0.02694 0.01428
4125 0.02582 0.01045 1/85 0.02417 0.03928
4274 0.01756 0.01224 4100 0.02673 0.02147
Sum 0.96304 0.96760 Sum 0.95423 0.96493
E 0.013 0.032 E 0.017 0.035
a L, = method of least squares.

b L, = method of linear programming.
252 FRBNS DE MEYER
APPENDIX3.
PETERS’ AND HENDERSON’S
FILTERS
FOR THE DOWNWARD
CONTINUATION
Peters Henderson
d=l d=2 d=l d=2 d=3 d=4 d=5
40 0.3696 0.6434 4.8948 16.1087 41.7731 92.5362 183.2600

41 0.3025 0.5573 -3.0113 - 13.2209 -38.2716 -89.7403 -183.9380
42 0.3355 0.4889 0.0081 0.4027 1.7883 5.1388 11.8804
45 0.2746 0.3440 -0.5604 -1.9459 -4.7820 -9.9452 -18.6049
4 8 - 0.0376 0.0644 0.5367 1.7478 4.2324
48.5 0.2229 0.2085
1/13 -0.0689 -0.0596 0.1798 0.8908 2.4237
417 0.0346 -0.1061
1/25 -0.0605 -0.0522 0.1342 0.6656 1.7777
434 -0.2219 -0.5047
450 -0.0534 -0.0828 -0.0560 0.0718 0.3606
458 -0.3464 -0.6497
499 0.1134 0.2498
4125 -0.1120 -0.2315
4136 -0.0380 -0.0703 -0.0900 -0.0890 -0.0571
4274 -0.0227 -0.0443 -0.0639 -0.0802 -0.0921
4625 -0.0302 -0.0600 -00.0891 -0.1173 -0.1444
sum 1 1 1.0199 1.0393 1.0595 1.0790 1.0983
APPENDIX
4.
FILTERS
GRANT-WEST FOR DOWNWARD
CONTINUATION
2 = 0.75
Rk
y = 119 y = 1/6 y = 1/6 y = lj4 y = 1/4 y = lj3
40 2.1652 1.2361 2.5038 1.2024 1.3168 1.2609

41 0.3844 1.8444 1.4621 2.6644 3.4369 3.5492
1/2 -1.2647 -1.7333 -2.6018 -2.2727 -4.0842 -2.8342
1/5 -0.1081 -0.1712 -0.1531 -0.3683 -0.4216 -0.7242
1/S -0.0481 -0.0356 -0.0456 -0.0298 -0.0174 -0.0134
1/10 -0.0232 -0.0418 -0.0341 -0.0310 -0.0323 -0.0407
1/13 -0.0267 -0.0279 -0.0330 -0.0389 -0.0443 -0.0437
2/17 -0.0220 -0.0230 -0.0330 -0.0278 -0.0327 -0.0354
1/20 -0.0167 -0.0135 -0.0142 -0.0198 -0.0241 -0.0253
1/25 -0.0041 -0.0215 -0.0262 -0.0292 -0.0343 -0.0335
1/34 -0.0145 -0.0237 -0.0216 -0.0267 -0.0297
d40.5 -0.0018 -0.0013 -0.0151 -0.0196 -0.0164
450 -0.0126 -0.0164 -0.0199
d64.5 -0.0206
APPENDIX5.
FOR TEE DOWNWARD
COEFFICIENTS d = 1, APPROXIMATION
CONTINUATION v)-
IN THE (u,
PLANE
Weierstrass kernel FBjer kernel

Rlc
z = 1, po = 1, p r = 2, o! = 6.61 Z= 1, pc = 1, pc = 1.5, = 2~
do 0.24497 0.19155 0.18407 0.14391

2/1 1.32371 1.72299 0.99415 1.34005
2/2 0.91269 0.23966 0.90935 0.34145
1/4 0.62357 0.62636
45 0.35727 - 0.00880 0.59457 0.20118
1/s -0.29751 - 0.41586 0.05064 0.03488
2/9 0.33522 0.10951
2/10 -0.57614 -0.80877 -0.37375 -0,46622
2/13 -0.66704 - 0.50324 - 0.53409 -0.41646
2/ 16 - 0.57096 -0.57087
1/17 - 0.57360 - 0.06802 - 0.59571 -0.04116
2/20 -0.28797 -0.31652 -0.56818 -0.64196
v'25 0.09675 -0.02038 -0.35135 -0.30824
2/29 0.18419 -0.05637
2/34 0.47108 0.29053 0.11223 0.10694
2/ 40 - 0.00413 0.46039 0.53983 0.52919
2/ 50 0.14494 0.03834 0.70833
2/65 - 0.29889 -0.05065
2/ 85 - 0.24677 -0.61816
2/lOO 0.03518 0.02832
SUm 1.00006 1.00001 1.00010 1.00002
254 FRANS DE MEYER
6.
APPENDIX
CONTINUATIONd
COEFFICIENTSFOR TEE DOWNWARD PROGIRAMMINQ
= 1, LINEAR
METHOD
R, Test (1) Test (2)

~~
0.48901 0.47795
0.66810 0.87574
1.46550 1.32928
-0.21308 0.02968
1.09735 0.40451
-2.00891 - 1.49369
-0.52898 -0.71446
0.71144 0.44360
- 2.00891 - 1.49369
1.00588 1.11425
-0.34729 0.10756
1.21442 0.68057
-0.50866 -0.67464
1.03587 1.02668
0.1435 0.1067
ACKNOWLEDGMENTS
The author acknowledges his indebtednessto Prof. A. De Vuyst, whose encouragement
and interest were a constant stimulation t o accomplish this work;to Prof. J. van Isacker
for offering helpful suggestions and for the use of the computer a t the Royal Meteoro-
logical Institute; and to Prof. J. van Mieghem for presenting the manuscript.
LISTOF SYMBOLS
d,(.r) Covariance kernel
-
A d z , Y, Calculated field a t
4
6(z, y) Infinite Dirac comb depth z
a,,@, y) Finite Dirac comb 9 Linear filter operator
V(u,u) Fourier transform of f (x, Y) General two-dimen-
infinite Dirac comb sional function
Vmn(u,u ) Fourier transform of qu,v ) Fourier transform of
Gnite Dirac comb f (29Y)
A d z , y, 4 Average of Ag(x, Y. 0) f’@, Y) Input to linear filter
over the circle of fO(% Y) Output of linear fitter
radius r Y) Desired output of Wie-
Ag(x,y,z) Gravity effect a t the
- level z
Ag(z, y, z ) Calculated field at
Fa(%
ner filter
4 Fourier transform of
sampled field
height z 0 Gravitational constant
Gravity vector a(+)(,,

w, 2) Modified downward
Output of linear, con- continuation
tinuous filter frequency response
Output of linear, dis- RI +’(u,a,2) Frequency response
crete filter corresponding t o
Fourier transform of hl+% Y, z)
Ag(x,Y, 2) Frequency response of
Covariance function averaging process on
Covariance function of the circle of radius r
sampled field Frequency response of
Covariance estimates optimal Wiener filter
Filter function of linear Frequency response of
continuous filter vertical derivative of
Frequency response of order n
linear, continuous General kernel
filter Two-dimensional FBjer
Frequency response of kernel
Noise component of
linear, discrete filter
Filter function of modi- surface field
fied low-pass filter Power spectrum
Frequency response of Power spectrum of
modified low-pass filter sampled field
Power spectrum esti-
Finite filter function of
mates
modified low-pass
Periodogram
filter
Smoothed version of a
Frequency response
function I)
corresponding to
Density function
hr(x,Y) Radial Nyquist fre-
hkl Filter coefficients of
quency
two-dimensional
Translation operator
filter
SignaI component of
Upward continuation
surface field
filter function
Fourier transform of
Upward continuation
frequency response
4 2 , Y)
Grid spacing of square
Discrete upward con-
grid
tinuation frequency
U Variance
response
Potential of the devia-
H' + ) ( uw,
, z ) Downward continuation masses of the
tion frequency res- earth
ponse Frequencies correspon-
H i +)(u,w) Discrete downward ding t o the x, y-di-
continuation fre- rections
quency response Nyquist frequencies
Filter function of modi- Newton's potential of
fied continuation the earth
process General weighting
Finite filter function of function
modified downward Fourier transform of
continuation process w(x,Y)
256 FRANS DE MEYER
REFERENCES
A. General worka
Anderson, T. W. (1971). “The Statistical Analysis of Time Series.” Wiley, New York.
Arsac, J. (1961). “Transformation de Fourier et th6orie des distributions.” Dunod,
France.
Bartlett, M. S. (1962). An Introduction t o Stochastic Processes.” Cambridge Univ.
Press, London and New York.
Bendat, J. S. (1968). “Principles and Applications of Random Noise Theory.” Wiley,
New York.
Blaokman, R. B., and Tukey, J. W. (1968). “The Measurement of Power Spectra.”
Dover, New York.
Courant, R., and Hilbert, D. (1962). “Methods of Mathematical Physics.” Wiley,
New York.
Dantzig, G. B. (1963). Linear Programming and Extensions.” Princeton Univ. Press,
Prinoeton, New Jersey.
Dobrin, M. B. (1960). “Introduction to Geophysical Prospecting.” McGraw-Hill, New
York.
Doob, J. L. (1968). “Stochastic Processes.” Wiley, New York.
Garland, G. D. (1966). “The Earth’s Shape and Gravity.” Pergamon, Oxford.
Grant, F. S., and West, G. F. (1965). “Interpretation Theory in Applied Geophysics.”
McGraw-Hill, New York.
Grenander, U., and Rosenblatt, M. (1957). “Statistical Analysis of Stationary Time
Series.” Wiley, New York.
Hannan, E. J. (1960). “Time Series Analysis.” Wiley, New York.
Heiskanen, W. A., and Meinesz, F. A. (1968). “The Earth and its Gravity Field.”
Holloway, J. L. (1968). “ Smoothing and filtering of time series and space fields. Advan
aeophya. 4,361-389.
Kellog, 0. D. (1900). “Foundations of Potential Theory.” Dover, New York.
Kopal, 2. (1961). “Numerical Analysis.” Chapman & Hall, London.
Lee, Y. W. (1960). “Statistical Theory of Communioations ” Wiley, New York.
Morse, P. M., and Feshbaoh, H. (1953). “Methods of Theoretical Physics.” McGraw-Hill,
New York.
Papoulis, A. (1965). “ Probability, Random Variables and Stochastic Processes.”
Rice, J. R. (1964). “The Approximation of Functions.” Addison-Wesley, Reading,
Massachusetts.
Sneddon, I. N. (1961). “Fourier Transforms.” McGraw-Hill, New York.
Todd, J. (1962). “Survey of Numerical Analysis.” McGraw-Hill, New York.
Van Trees, H. L. (1968). “Detection, Estimation and Modulation Theory.” Wiley,
New York.
Watson, G. N. (1968). “A Treatise on the Theory of Bessel Functions.” Cambridge
Univ. Press, London and New York.
Wiener, N. (1960). “ Extrapolation, Interpolation and Smoothing of Stationary Time
Series.” Wiley, New York.
B. Upward continuation
Henderson, R. G. (1960). A comprehensive system of automatic computation in magnetic
and gravity interpretation. Geophyaica 23, 669-686.
FILTER TECHNIQUES W GRAVITY INTERPRETATION 257
Henderson, R. G., and Zietz, I. (1949). The upward continuation of anomalies in total
magnetic intensity fields. Geophysics 14, 51 7-534.
Nettleton, L. L., and Cannon, J. R. (1962). Investigation of upward continuation
systems. Geophysics 27, 796-806.
Peters, L. J. (1949). The direct approach to magnetic interpretation and its practical
application. Geophysics 14, 290-320.
C. Downward eontinuation
Bullard, E. C., and Cooper, R. I. B. (1948). Determination of the masses necessary t o
produce a given gravitational field. Proc. Roy. SOC.,Ser. A 194, 332-347.
Ku, C. C., Telford, W. M., and Lim, S. H. (1971). The use of linear filtering in gravity
problems. Geophysics 36, 1174-1203.
Nedelkov, I. P., and Burnev, P. H. (1962). Determination of gravitational fields in
depth. Geophys. Prospect. 10, 1-18.
Negi, J. G. (1967). Convergence and divergence in downward continuation. Geophysics
32,867-871.
Roy, A. (1966). Downward continuation and its application to electromagnetic data
interpretation. Geophysics 31, 167-184.
Roy, A. (1967). Convergence in downward continuation for some simple geometries.
Geophysics 32, 853-866.
Rudman, A. J., Mead, J.,Whaley, J. F., and Blakely, R. F. (1971). Geophysical analysis
in central Indiana using potential field continuation. Geophysics 36, 878-890.
Trejo, C. A. (1954). A note on downward continuation of gravity. Geophysics 19, 71-75.
D. Convolution and frequency Jiltering

Bhattacharyya, B. K. (1972). Design of spatial filters and their application to high-
resolution aeromagnetic data. Geophysics 37, 68-91.
Black, D. I., and Schollar, I. (1969). Spatial filtering in the wave-vector domain. Geo-
physics 34, 916-923.
Byerly, P. E. (1965). Convolution filtering of gravity and magnetic maps. Geophysics 30,
281-284.
Chan, S . H., and Leong, L. S. (1972). Analysis of least squares smoothing operators in the
frequency domain. Geophys. Prospect. 20, 892-900.
Clarke, G. K. C. (1969). Optimum second derivative and downward continuation filters.
Geophyaics 34, 424-437..
Clarke, G. K. C. (1971). Linear filters to suppress terrain effects on geophysical maps.
Connes, J., and Nozal, V. (1971). Le filtrage math6matique dans la spectroscopie par
transformation de Fourier. J. Phys. Radium 22, 359-366.
Darby, E. K., and Davies, E. B. (1967). The analysisand design of two-dimensionalfilters
for two-dimensional data. Geophys. Prospect. 15, 383-406.
Dean, W. C. (1958). Frequency analysis for gravity and magnetic interpretation.
Fraser, D. C., Fuller, B. D., and Ward, S. H. (1960). Some numerical techniques for
application in mining exploration. Geophysics. 31, 1066-1077.
Gunn, P. J. (1972). Application of Wiener filters to transformations of gravity and
magnetic fields. Geophys. Prospect. 20, 860-871.
Lavin, P. M., and Devane, S. J. (1970). Direct design of two dimensional digital wave
number filters. Geophysics 15, 1073-1078.
258 FRANS DE MEYER
Mesk6, A. (1966).Some notes concerning the frequency analysis for gravity interpretation.
Qeophyd. Prospect. 13,476-488.
Mufti, I. Ic. (1972).Design of small operators for the continuation of potential field data.
aeophy& 37,488-606.
Naidu, P. S. (1966).Extraction of potential field signal from a background of random
noise by Strakhov’s method. J. aeophys. Res. 71, 6987-6996.
Naidu, P. S. (1967).Two dimensional Strakhov’s filter for extraction of potential field
signal. aeophys. P r o 8 p d . 15, 135-160.
Strakhov, V. N. (1964s). The smoothing of observed strengths of potential fields.
Part I. Bull. A d . Sci. USSR, Geophys. Ser. 10, 897-904.
Strakhov, V.N. (1964b).The smoothing of observed strengths of potential fields. Part 11.
Bull. Acad. SCi., USSR, 0eophy8. Ser. 11, 986-996.
Swartz, C. A. (1963).Some geometricalproperties of residual maps. Geophy&m 19,46-70.
Swartz, C. A. (1954).Filtering associated with selective sampling of geophysical data.
Treitel, 6. (1967).Principles of digital Wiener filtering. Geophys. Pro8pect. 15, 311-333.
Zurflueh, E. G. (1967). Application of two-dimensional linear wavelength atering.
D. Power epectruna estimation

Akcasu, A. 2. (1961).Measurement of noise power spectra by Fourier analysis. J . Appl.
Phys. 32, 666-668.
Bartlett, M. S. (1960).Periodogram analysis and continuous spectra. Biometrika 37,
1-16.
Cooley, J. W., Lewis, P. A. W., and Welch, P. D. (1967).“The Fast Fourier Transform
and its Applications,” Res. Pap. RC-1743.IBM Watson Res. Cent., New York.
Goodman, N. R. (1961).Some comments on spectral analysis of time series. Techno-
Wk&&C8 3, 221-228.
Grenander, U. (1968).Bandwidth and variance in estimation of the spectrum. J. Roy.
Statist. SOC.,Ser. B 20, 152-167.
Hinich, M. J., and Clay, C. S. (1968).The application of the discrete Fourier transform
in Ihe estimation of power spectra coherence and bispectra of geophysical data.
Rew. Geophys. 6 , No. 3, 347-362.
Horton, C. W., Hempkins, W.B., and HoBnan, A. J. (1963).Statistical analysis of some
aeromagneticmaps from the North-western Canadian shield. Geophy8ks 29,682-601.
Jenkins, G . M. (1961).General considerationsin the analysis of spectra. T e c h & & 3,
133-166.
Jenkins, G. M., and Priestley, M. B. (1967).The spectral analysis of time series. J . Roy.
Statist. Soc., Ser. B 19, 1-12.
Jones, R. H. (1965).A reappraisal of the periodogram in spectral analysis. Techno-
rnetrica 7, No. 4, 631-642.
Lomnicki, Z. A., and Zaremba, S. K. (1967).On estimating the spectral density function
of a stochastic process. J. Roy.Statist. SOL,Ser. B 19, 13-37.
Murthy, V. K. (1961).Estimation of the spectrum. Ann. Math. Stat&. 32, 730-738.
Naidu, P. S. (1968).Spectrum of the potential field due to randomly distributed sources.
C e ~ p h 33,
y ~ 337-345.
Naidu, P. S. (1969).Estimation of spectrum and cross-spectrum of aeromagnetic field
using Fast Digital Fourier transform (FDFT) techniques. Geophpka 34, 344-361.
Parzen, E. (1966).On consistent estimates of the spectral density of a stationary time
series. Proc. Nat. A d . Sci. U.S. 42, 164-167.
HILTER TECHNIQUES IN ORAWTY INTERPRETATION 259
Parzen, E. (1957s). On consistent estimates of the spectrum of a stationary time series.

Ann. Math. Statist. 28, 328-348.
Parzen, E. (1967b). On choosing an estimate of the spectral density function of a station-
ary time series. Ann. Math. Statist. 28, 921-932.
Parzen, E. (1968). On asymptotically efficient consistent estimates of the spectral
density function of a stationary time series. J . Roy.Statist. SOC.,Ser. B 10,303-322.
Parzen, E. (1961). Mathematical considerations in the estimation of spectra. Techno-
metrica. 3, 167-190.
Sax, R. L. (1966). Application of the filter theory and information theory to the interpre-
tation of gravity measurements. Geophysics 31, 670-576.
Tukey, J. W. (1961). Discussion, emphasizing the connection between analysis of vari-
ance and spectrum analysis. Technometrim 3, 191-219.
Whittle, P. (1957). Curve and periodogram smoothing. J . Roy.Statist. Soc., Ser. B. IS,
38-47.
Wonnacot, T. H. (1961). Spectral analysis combininga Bartlett window with an associated
inner window. Technometrica 3, 236-243.
F. Fast jourier transform algorithm

Bergland, G . D. (1967). The Fast Fourier Transform recursive equations for arbitrary
length records. Math. Comput. 21, 236-238.
Bingham, C., Godfrey, M. D., and Tukey, J. W. (1967). Modern techniques of power
spectrum estimation. IEEE Trans. Audio Electroacoustics, 15, No. 2, 66-66.
Bogert, B. P. (1967). Informal comments on the uses of power spectrum analysis. IEEE
Trans. Audio Electroacoustics. 15, No. 2, 74-75.
Cochran, W. T. (1967). What is the Fast Fourier Transform. Proc. IEEE 55, No. 10,
1664-1674.
Cooley, J. W., and Tukey, J. W. (1965). An algorithm for the machine calculation of
complex Fourier series. Math. Comput. 19, 297-301.
Cooley, J. W., Lewis, A. W., and Welch, P. D. (1967). Historical notes on the Fast
Fourier Transform. Proc. IEEE 55, 1676-1677.
Cooley, J. W., Lewis, A. W., and Welch, P. D. (1969). The Fast Fourier Transform and
its applications. IEEE Trans. Educ. 12, 27-34.
Cooley, J. W., Lewis, A. W., and Welch, P. D. (1969b). Application of the Fast Fourier
Transform t o computation of Fourier Integrals, Fourier series and convolution
integrals. IEEE TTans. Audio Electroacoustics 15, 79-84.
HARM. (1966). Harmonic Analysis Subroutine for IBM 7090,” SDA No. 3426. SHARE
‘L
Distribution Agency, Program Inform. Dept., IBM Corp., 40 Saw Mill Road,
Hawthorne, New York 10532.
Helms, H. D. (1967). Fast Fourier Transform method of computing difference equations
and simulating filters. IEEE Trans. Audio Electroacoustics 15, 86-90.
Singleton, R. C. (1967). A method of computing the Fast Fourier Transform with auxi-
liary memory and limited high-speed storage. IEEE T r a m . Audio Electroacoustics
15,91-98.
Welch, P. D. (1967). The use of the Fast Fourier Transforms for the estimation of power
spectra, a method based on time averaging over short, modified periodograms.
IEEE Trans. Audio Electroacoustics 15, 70-73.
G . Second derivative method
Agarwd, B. N. P., and Lal, T. (1972a). A generalized method of computing second
derivative of gravity field. Cieophya. Prospect. 20, 386-394.
260 FRANS DE MEYER
Agarwal, B. N. P., and Lal, T. (1972b). Calculation of the vertical gravity field using
the Fourier transform. Geophys. Prospect. 20, 44-57.
Danes, 2. F., and Oncley, I. A. (1962). An analysis of some second derivative methods.
Elkins, T. A. (1950). The second derivative method of gravity interpretation. Geophysics
16, 29-50.
Henderson, R. G., and Zietz, I. (1949). The computations of second vertical derivatives
of geomagnetia fields. Geophysics 14, 508-616.
Mesk6, C. A. (1966). Two-dimensional filtering and the second derivative method.
Rosenbach, 0. (1963). A contribution t o the computation of the second derivative from
gravity data. Geophysics 18, 894-912.
H. Related topics
Agarwal, R. G., and Kanasewich, E. R. (1971). Automatic trend analysis and intepre-
tation of potential field data. Geophysics 36, 339-348.
Agoes, V. B. (1951). Least squares residual anomaly determination. G e ~ p h y ~ i c16,
s . 686-
696.
Al-Chalabi, M. (1971). Some studies relating to nonuniqueness in gravity and magnetic
inverse problems. Geophysics 30, 835-855.
Al-Chalabi, M. (1972). Interpretation of gravity anomalies by non-linear optimisation.
Geophys. Prospect. 20, 1-16.
Botezatu, R., Vimrion, M., Seurtu, F., and Cucu, G. (1971). Approximation of the
gravitational attraction of geological bodies. Geophy8. Pmspect. 19, 218-227.
Bott, M. H. P., and Smith, R. A. (1958). T h e estimation of the limiting depth of gravi-
tating bodies. Geophys. Prospect. 0, 1-10.
Cordell, L., and Henderson, R. G. (1968). Iterative three-dimensional solution of gravity
anomaly data using a digital computer. 33, No. 4, 596-602.
Grant, F. S. (1951e). Three-dimensional interpretation of gravitational anomalies. Part
I. Geophysics 17, 344-364.
Grant, F. S. (1951b). Three-dimensional interpretation of gravitational anomaiies.
Part 11. Geophysics 17, 756-789.
Grant, F. 6. (1953). A theory for the regional correction of potential field data. Geophysics
19, 23-45.
Grant, F. S . (1957). A problem in the analysis of geophysicaldata. Geophysics 22,309-344.
Grant, F. S. (1972). Review of data processing and interpretation methods in gravity
and magnetics, 1964-1971. Geophysics 37, 647-661.
Griffin, W. R. (1948). Residual gravity in theory and practice. Geophysics 14, 39-56.
Henderson, R. G., and Cordell, L. (1971). Reduction of unevenly spaced potential field
data to a horizontal plane by means of finite harmonic series. Geaphysica 36,856866.
Kreisel, 0. (1948). Some remarks on integral equations with kernels, A ( [ , - xl,. ., .
.f,, - xn, a).Proc. Roy.SOC.Ser. A 107, 160-178.
Kunaratnam, K. (1972). An iterative method for the solution of a non-linear inverse
problem in magnetic interpretation. Geophys. Prospect. 20, 439-447.
La Porte, M. (1972). Elaboration rapide des cartes gravimQtriquesdbduites de l’anomalie
de Bouguer B I’aide d’une caloulatrice Qlectronique.Geophys. Proepect. 10,23&257.
Nagy, D. (1966). The gravitational attractiyn of a right rectangular prism. Geophysics 31,
362-371.
Naidu, P. S. (1967). Statistical properties of potential fields over a random medium.
Geophysks 32, 88-98.
Naidu, P. S. (1970). Statistical structure of aeromagnetic field. Geophysics 35, 279-292.

Nettleton, L. L. (1953). Regionals, residuals and structures. Geophysica 19, 1-22.
Paul, M. K. (1967). A method of computing residual anomalies from Bouguer gravity
map by applying relaxation technique. Geophysics 82, 708-7 19.
Polya, G. (1929). Untersuchung uber Liicken und Singularitiiten von Potentzreihen.
Math. 2. 29, 549-600.
Qureshi, I. R., and Idries, F. M. (1972).Two-dimensionalmass distributions from gravity
anomalies. Geophys. Proapect. 20, 106-108.
Qureshi, I. R., and Mula, H. G. (1971). Two-dimensionalmass distribution from gravity
anomalies: A computer method. Geophys. Prospect. 19, 180-191.
Simpson, S. M. (1953). Least squares polynomial fiitting to gravitational data and density
plotting by digital computers. Geophyaica 19, 255-269.
Skeels, D. C. (1947). Ambiguity in gravity interpretation. Geophysics 12, 43-66.
Spector, A., and Grant, F. S. (1970). Statistical models for interpreting aeromagnetic
data. Geophysics. 35, 293-302.
Syberg, F. J. R. (1972a). A Fourier method for the regional residual problem of potential
fields. Geophysical Prospect. 20, 47-75.
Syberg, F. J. R. (1972b). Potential field continuation between general surfaoes. Geophys.
Prospect. 20, 267-282.
Talwani, M., and Euring, M. (1960). Rapid computation of gravitational attraction of
three-dimensional bodies of arbitrary shape. Geophysics 25, 103-225.
This Page Intentionally Left Blank
AERIAL METHODS IN
GEOLOG ICAL-G EOG RAPHICAL
EXPLO RAT10NS*
B. V. Shilin and V. B. Komarov
Laboratory of Aeromethods,
Ministry of Geology, Leningrad, U.S.S.R.
1. Introduction .......................................................... 263

2. Radar Aerial Survey .................................................... 265
2.1. Procedures ........................................................ 265
2.2. Interpretation of Rock Composition ................................ 272
2.3. Interpretation of Structural-Tectonic Texture ........................ 277
2.4. Interpretation of Geological Structure in Closed Regions. ............... 279
3. Infrared Aerial Survey .................................................. 282
3.1. General .......................................................... 282
3.2. Infrared Aerial Survey of Active Volcanoes .......................... 282
3.3. Study of Regions of Strong Geothermal Activity ...................... 286
4. Aerogeochemical Survey: Remote Sensing of Gases and Vapors .............. 304
4.1. Method .......................................................... 304
4.2. Equipment and Procedure. ......................................... 305
4.3. Aerial Measurements over Fumarole Fields of Caldera Uson ............ 308
4.4. Active Volcanoes of Kamchetka .................................... 313
5. Conclusione ............................................................ 321
References ............................................................ 322
1. INTRODUCTION
When the f%st images o f the earth’s outer appearance were obtained from
orbital altitudes, specialists in various fields of research became aware of a
new, exceptionally effective means of studying the earth’s natural resources
and the processes taking place on its surface. Sciences dealing with the earth
were offered the possibility, on the one hand, to rise to a new level of extensive
generalizations regarding the geological and geographical structure of the
planet and, on the other hand, to obtain regional information enabling more
reliable regional assessments for the rational development and mastering of
natural resources.
The use of information obtained from the orbital altitudes was preceded
by the successful application of other aerial methods. The necessity for
rapidly acquiring data on the natural resources of the enormous territory
of the USSR for their efficient exploration and use required the development
of methods of remote sensing from aircraft, based on the use of the electro-
* Published in cooperation with Advances in Space Science and Technology, edited by

Frederick I. Ordway.
263
264 B. V. SHILIX AND V. B. KOMAROV
magnetic spectrum and other physical phenomena. In spite of the limited

spectral range used for classical aerial photographic surveys (400-1000
nm),this method is the most universal and efficient of all aerial methods
with respect to both the amount of information obtained and the range of its
application in economic and scientific research.
Aerial photographic surveys are especially efficient for various kinds of
mapping: topographic, geological, landscape, soil, geobotanical, forestry, etc.
They have been used for inventory of forests, land surveying, land reclama-
tion, planning of railways and highways, power lines and pipe lines, and
various explorations.
Aerial photographs are also widely used for: (1) geographical investigations
of all kinds, including a complex study of landscapes, their typology, rhythm,
and dynamics of their development, the realization of measures to protect
the environment against erosion, plant diseases, water and air pollution,
etc.; (2) geological research connected with the study of the earth’s crustal
structure and tectonics (including recent movements), the search €or mineral
deposits, geological study of shallowseas, the hydrogeologicaland engineering-
geological explorations; (3) the study of forests, soils, swamps, vegetation
(including submerged plants), pastures, etc.; (4) the study of hydrological
conditions of dry land, processes on shores and coasts of various reservoirs,
the course of the rivers and processes in the river beds; (6) the study of
sea streams and sea h e a ~ n g hydro-optical
, characteristics of the sea, and
processes in the estuaries; (7)glaciological research; (8) archeological in-
vestigations; (9) the study of cloud cover; (10) the study of fishery resources
(stock-taking of the number of animals and fish), their protection, and
proper exploitation (wildlife management). They are also used for many
other special tasks.
The utilization of aerial photography data in the above-mentioned fields
increases the amount of information and improves the accuracy of the data
on the objects under study. It also permits a lessening in the amount of
ground exploration and a general reduction of the expense of solving the
problems.
The increase of efficiency in aerial photographic methods is directed
toward: (1)the use of small-scaleaerial photographs which allow for sufficient
optical and photographic magnification and permit, while keeping sufficient
detail, the encompassing of large areas for obtaining in a number of cases new
information; (2) stipulation of new, more profound connections between the
objects and photo images of their indicators while obtaining quantitative
characteristics; (3) the development of algorithms to identify various land-
scape objects for the automation of photoreading and interpretation pro-
cesses; (4) the design of complex automated systems to handle the aerial
photographs by use of computers and peripheral equipment.
AERIAL GEOLOGICAGGEOGRAPHICAL EXPLORATIONS 265
It should be noted here that in spite of the achievements in the field of

cybernetics and automated devices, the interpretation process remains
one of the most complicated tasks. Therefore, the final analysis of the results
obtained while using the automated systems for handling the photographs
should still be performed by a specialist comparing them with the initial
data.
For geological mapping and in the search for minerals the usual aero-
geophysical methods are utilized. They are based on the recording of gamma
radiation, electromagnetic and gravitational fields, as well as the radio wave
portion of the electromagnetic spectrum. These methods allow the acquisition
of data on deep structures of the eart8h’scrust and on its types of rocks.
When used in combination with aerial photography, they considerably
increase the reliability of geological results.
The methods of aerial photography and the aerogeophysical methods,
used successfully for a long time for the purpose of studying natural
resources, are fairly well developed and described in the extensive special
literature.
I n recent years, ever increasing attention has been centered on the develop-
ment of the techniques of remote sensing and methods for studying natural
resources, especially their installation in space vehicles. However, the
miteria for using some remote sensing receivers developed very recently
(infrared, radar, and others) are not yet sufficiently established. The efficient
use of these receivers for studying the natural resources from orbital plat-
forms cannot be anticipated without carrying out this stage of investigation.
The present review concerns some results on that stage of research using
infrared, radar, and geochemical aerial surveys with emphasis on trends in
the application of these methods.
2. RADARAERIALSURVEY
2.1. Procedures
The advent of side-looking radar systems brought with it a considerable
increase in the resolution of the images obtained, enabling their use for
studying wider areas of the country. This was also stimulated by the
quest of specialists t o use the small-scale images for study of large
territories, an aim which cannot always be attained through aerial photo-
graphy because of meteorological obstacles. I n this case the small-scale images
served as a kind of bridge in passing from the materials of traditional aerial
photography to extra small-scale images obtained from space altitudes.
I n the USSR the side-looking radar system TOROS was developed for
the purpose of ice exploration, which in a very short time has found wide
application for both the evaluation and the prognosis of ice conditions. This
266 B. V. SRILJX AND V. B. ROMAROV
system functions under the complex meteorological conditions of the Arctic

and rapidly obtains images of large territories. These images have further
use for solving geological and other problems.
The TOROS system scans the earth’s surface by means of two narrow ver-
tical beams formed by parabolic antennas mounted on both sides along the
aircraft fuselage and directed normally to their axes.
Scanning along the ilight line is accomplished by the aircraft’s forward
movement. An image of the terrain is formed on the screens of cathode-ray
tubes and is simultaneously recorded onto photographic film 19 cm wide.
The speed of the film’smotion is proportional to the aircraft’s flight speed and
the survey scale.
Thus, an image is obtained on the film. The density of this image at any
point is a function of the reflecting properties of the irradiated terrain objects
scanned in the given wave range.
The centimeter wavelength range, the narrow directivity pattern in a
horizontal plane, and short pulses (fractions of a millisecond) enable the
TOROS system to obtain an image of terrain with a fairly high resolution
close to that of the small-scale aerial photographs.
The radar image tone determined by the reflected signal intensity depends
on the operating frequency of the station, the angles of radiation, the texture
of the surface, and physical properties of the irradiated objects. Depending
on the reflection properties of the irradiated surface, the diffusive dispersion,
the specular reflection, and radio-wave absorption will take place.
Most of the objects, because of their inherent surface texture and physical
properties, diffusively disperse the radio waves, contributing to the genera-
tion of various reflected signals and, accordingly, the contours of different
tone on the radar image.
When the texture of the surface has irregularities smaller than one-half
wavelength, a radio signal reflection will appear close to specular reflection.
This will result in its disappearance on the receiving device and the absence
of an image (more precisely-in the appearance of black tone on the positive
image). I n this way the water surfaces, regions of salt marshes, takyrs, high-
ways, and other even surfaces will be displayed. The maximum absorption of
radio waves and the dark tone of the image will be observed also when
objects or regions with great moisture content and, consequently, high
electric conductivity, are irradiated. These objects may be: freshly tilled
plow land, the regions of tectonic breaks with outlets of subterranean waters,
and objects that absorb moisture readily and are surveyed after rain. The
small angles of radiation especially emphasize the relief on the radar images,
creating a pronounced “ sculptural ” picture.
Owing to these specific properties of image production, as well as a certain
generalization at the expense of small scde, the radar aerial images of the
AERIAL. GEOLOGICAL-GEOGRAPHICAL EXPLORATIONS 267
TOROS system present new information beyond that obtained by aerial

photographs. They are widely used in the USSR for compiling ice charts
and in geological exploration.
The TOROS system permits the compilation of maps of general ice cover
distribution based on the results of ice surveys for carrying out navigational
reconnaissance and permits the compilation of detailed Iarge-scale maps of
ice distribution and condition. These are of direct use for navigation and
for conducting experiments concerned with hydraulic engineering, rescue
operations, and other work at sea.
The intensity of the radar signal reflected from the ice surface depends on
the relief and " roughness )'of the latter &s well as on the physical properties
and condition of the ice fields. These characteristics may vary within wide
limits. However, there exists a certain set of indications for each of the
ice categories, which relate to the particular features of the surface relief and
the configuration and size of the ice fields.
Contour details and distinctions in character of the micro-relief of the floe
surfaces, which cannot be detected either by visual air reconnaissance or in
aerial photographs, may be observed in radar images with enhanced contrast.
They are the edges of sea ice floes frozen together and covered with snow,
narrow frozen cracks, and the character of their surface relief.
The following characteristics of the sea ice cover may be determined by
radar images: degree of packing, age (using enlarged gradations), relative
amount of floes of different size, the degree of hummocking of ice cover, the
amount, size, and orientation of free water areas and channels, the speed and
direction of the ice drift (by repeated surveys), and in specific cases also the
ice compression.
The determination of the ice packing is the simplest and most exact
operation. The discrepancies between estimates of ice packing made with
radar images and aerial photographs amount to only 3 4 yo.
I n contrast, the estimation of ice age with radar images is the most compli-
cated process. However, as experience has shown, some of the ice categories
can be determined with sufficient accuracy. Among these are: (a) initial ice
such as slush ice, nilas, sludge; (b) young ice (gray and gray-white) differing
from nilas by greater brightness of its image and by its tendency to break in
motion; (c) first-year ice without classification by color, first-year ice of
medium thickness, and thick first-year ice; (d) old ice (its estimation becomes
considerably more complicated during the summer when melted snow
accumulates on the ice surface). Figure 1 presents an example of a radar
image for different ages of ice and the results of its interpretation; an aerial
photograph of the same region is added for comparison.
The radar images transmitted on board an ice breaker provide the most
objective information on the state of ice cover and permit a quick choice of
268 B. V. SHILIN AND V. B. KOMAROV
FIG.I.(a) Aerial radar image of ice cover. (b)Aerial photograph of the same part of
ice cover
AERIAL GEOLOGICAL-GEOGRAPHICAL EXPLORATIONS 269
FIG.l.(c) The scheme of aerial radar image interpretation results (composed by

V. S. Loshtshilov). 1-nilas and open water; 2-gray and gray-white ice; 3-thick
one-yearice.
route for a ship convoy. Figure 2 gives a radar image fixing the route of a
ship convoy through gray ice to open water.
The interpretation of radar images permits compilation of operational ice
charts along the whole route of a ship convoy. They are much more detailed
and accurate than those compiled from visual observations.
The extensive application of radar images in geological exploration was
prompted by the ‘‘ all-weather ” availability of the method and the possibility
FIa. 2. Aerial radar image of gray ice floe. White line-route of the caravan; white
bright points on open water and near the outlet from the ice-separate ships of the
caravan (V.S. Loshtshilov’s material).
of quickly obtaining small-scale images for vast meas (up to ten thousands
of square kilometers), as well as adequate representation of the main
indicators of geological structure-relief, hydrographic pattern, surface
texture, humidity, vegetation, and top soil-as used in usual aerial photo-
graphic interpretation.
In some instances the radar survey has an advantage over the photo-
graphic survey as a result of a peculiar " sculpturedness') defining various
elements of relief (the main indicator of geological structure) and the-gener-
alization of small immaterial details of landscape on radar images, which
are often a hindrance in studying the aerial photograph (Fig. 3).
The morphological features of each type of relief are determined by the
material composition of the rocks, their stability against weathering, the
AERIAL GEOLOGICAL-GEOGRAPHIC& EXPLORATIONS 27 1
FIQ.3. Aerial photograph (a) and aerial radar image (b) of a district of the Kam-
chatka Eastern volcanic belt. Due to lower resolution on radar image small landscape
details are smooth, particularly the small structure of lava flows and covers. Accordingly,
the regional break marked by the chain of small volcanic cones is shown somewhat more
distinctly. I n a number of cases the contrasts of some objects (e.g. bright bare sections
on the aerial photograph) are changed.
character of clefts, and the amplitude of the most recent tectonic movements
which are especially sharply defined on radar images.
Besides the morphological types of relief, its individual forms, such as
ridges, mounds, washout cones, beds of waterways, terraces, as well as
the elements of its forms (shelves, slopes, edges), are well shown on the radar
images. Each specific morphological type and subtype of relief is distin-

guishable and permits one to obtain a general concept of geomorphological
structure of the territory explored.
2.2. Interpretation of Rock Composition

The unequal response of rocks to radio wave diffusion which becomes
apparent in radar images through the changes in tone, makes it possible in
some instances to determine the rock type. Thus, the image of coame-grained
rocks will be of lighter tone than that of finegained rocks, all other things
FIQ4e. See facing page for legend.

AERIAL aEOLOGICAGCEOGRAPHICAL EXPLORATIONS 273
FIG.4. Aerial radar image of Middle Asia mountain region (a) and scheme of its
geological interpretation results (b) (composed by V. A. Starostin). 1-limestone, 2-
marl, 3---clay marl, 4--conglomerate, 5-sandstono, 6-aleurolite, 7-clay, 8-loam,
9-sandy 1 am, 10-faults, 11-boundaries of lithologo-stratigraphicdivisions.
being equal. I n some cases a sharp difference is observed in the tone range of
the objects on the aerial photographs and radar aerial images (Fig. 3).
Occasionally, when rocks lie horizontally and irregularities of their texture
are less than one-half wavelength, there will be a nearly specular reflection
of the latter and the rocks will be of dark tone in the image.
I n this way, with due consideration of the character of the surface texture,
the relief features (slope angles), cleft characteristics, the character of the
erosion pattern, the degree of humidity, and the development of vegetation,
it is possible t o distinguish between rocks of different composition. Clearly
defined border lines between the lithologic-stratigraphic subdivisions em-
phasized by the above-mentioned features enable one to determine the
geological structure of the region.
An example of interpreting a sedimentary series of rocks is presented in
Fig. 4. On the radar image one can easily identify the carbonate and terrigene
274 B. V. SHILW AND V. B. KOMAROV
FIG.5a. Aerial radar image of Karymski volcano and its environment.

AERIAL GEOLOCICAGGEOGRAPHICAL EXPLORATIONS 275
Fra. 5b. Various geological formations in the region of Karymski volcano (composed
by N. A. Gussev). 1-volcanic foundation, plateau composed of pyroclastic and lava
products of different thickness; 2-recent alluvial and lake deposits filling the most
depressed parts of relief; 3-tuff deposits of the southern complex bearing no relation
to the structure of Karymski volcano; 4-young lava and agglomerate flows of the
recent Karymski volcano; 5-Volcanic deposits of other eruption centers: a-mainly
lava, b-mainly pyroclastic; "recent ash-lava deposits of Karymski volcano; 7-
geological border lines: a-interpreted with assurance, be-interpreted with incertitude;
8-border of ancient Karymski volcano and Dvor volcano; 9 4 a l d e r a scarp of the main
structure and active volcano; 10-Lakes.
FIG. 5c. Structural-tectonic structure of the Karymski volcano (composed by

N. A. Gussev). 1-geological borders of most considerable subdivisions: *interpreted
with assurance, &interpreted with less assurance, c-supposed continuations of
geological borders; 2-some morphological linear elements (erosion valleys, slope
discontinuities, scarps); 3 4 a l d e r a scarp of the main volcanic structure and the active
crater; P c e n t e r s of extinct volanic activity; &recent lava and agglomerate flows of
contemporaneousKarymski volcano; f i t h e most notable erosional and tectonic scarps;
7-lakes; 8-fractures and splits: e m o r e significant, b-less significant and interpreted
with incertitude, c-supposed principal directions of tectonic breaks. A-Main tectonic
directions of the Karymski volcano area.
AERIAL GEOLOGICAGCEOGRAPHICAL EXPLORATIONS 277
rocks of chalk, paleogene, and neogen, while the border lines of individual
formations are traced more distinctly as compared to aerial photographs.
Friable Quarternary deposits, especially the takyr regions, and shores with
different degrees of salinity, moisture, and surface texture are well dis-
tinguished as regards their material composition.
The volcanogenic and intrusive rocks, especially dikes and dike belts,
stand out by the morphology of relief, the character of clefts, and other
features. It should be noted that the image contrast of the dikes and other
linear geological objects, including faults, depends upon their position
relative to the scanning beam. The dikes are especially well displayed when in
a perpendicular position relative to the direction of radiation and much
worse when oriented in parallel.
The interpretation of volcanic formations is exemplified by Fig. 5 , which
presents the varying composition of eruptive products for the active volcano
Karymski on Kamchatka [Fig. 5(b)]. Aerial radar surveys permit adequate
interpretation of the lava streams of the volcano’s main cone and their
relationships, and the flat surfaces characteristic of the areas of young lake
and river sediment deposits.
2.3. Interpretation of Structural- Tectonic Texture

Aerial radar surveys allow one, on the basis of different rock material
composition and the morphology of folds, to distinguish between various
structural-tectonic subdivisions : structural zones, specific folds of different
orders, ring structures, and zones of tectonic fractures.
Owing to the differences in the surface texture resulting from the relief
features, clefts, sheet jointings, and other indicators of geological structure,
the folded structures are well distinguished on the aerial radar images.
The change of intensity in radio waves reflected from the surfaces, which
differ in their physical properties, indicates, in principle, the possibility of
using the rocks with high moisture content and, consequently, high electric
conductivity as marker horizons.
The degree of interpretability on aerial radar images of structural elements
(marker horizons, fractures) depends not only upon the material composition
of the rocks, their distinctness in relief, or conditions of occurrence, but also
upon their orientation with respect to the direction of radiation.
It might be well to point out the feasibility of interpretation of tectonic
fractures of different orders on the aerial radar images and the possibility of
distinguishing additional fractures that are imperceptible or poorly noted on
aerial photographs.
I n the territory of Kazakhstan and Uzbekistan, the interpretation of aerial
radar surveys permitted the identification of a number of large-sue systems,
fractured zones, differing in age and significance, which might serve for ore
discovery. On the basis of the geological analysis of aerial radar images
schemes of tectonic fractures were plotted for these regions.
Comparison of the interpretation results for aerial radar images and
geophysical explorations revealed good agreement. This suggests the possi-
bility of using the aerial radar images for mapping the fractures which are not
sufficiently visible on the surface and which are inadequately h e d by the
usual geological methods, but a t the same time are noted during geophysical
observations.
A distinct interpretation of tectonic fractures on radar images suggests
their use for the study of block tectonic structure (Fig. 6). An example of
FIQ.66. See facing page for legend.

AERIAL GEOLOGICAL4EOGRAPHICAL EXPLORATIONS 279
FIG.6. Aerial radar image of block st,ructure region of upper Paleozoic effusives in
Kazakh fold country (a) and fractures limiting the rocky blocks (b). Fractures are inter-
preted with certainty according to the orientation of the erosional pattern large elements,
zones of increased humidity, and zones of vegetation cover development.
compiling sufficiently detailed structural tectonic schemes is presented by

the results obtained for the volcano Karymski and for the adjacent region
[Fig. 5(c)] where the main directions of fractures whose intersection controls
the position of extinct and active eruption centers are displayed. The analysis
of data for a somewhat greater territory allowed the establishment of pre-
viously unknown relationships of volcanic structures and volcano-tectonics.
2.4. Interpretation of Geological Structure i n Closed Regions

It is common knowledge that the main indicators of the closed geological
structures on the surface of friable deposits are: (1) morphology of contem-
poraneous relief forms (their orientation, position in space), (2) changes of
material composition of deposits and orientation of the contours of their
facial differences, (3) change of soil-vegetation cover character, (4) changes of
physical condition of friable deposits (in particular, humidity) and orientation
of the contours of humidified lithological differences, and (5) orientation of
eIements of erosion pattern.
280 B. V. SHILW AND V. B. KOMAROV
(b)
FIG.7. Aerial radar image of the region of submountain proluvial plane (a) end
its structural-tectonic structure (b) (composed by V. A. Starostin). Designations:
1-fractures, 2-structural lines.
As stated above, these indicators are easily identified on the radar aerial
images and, using them, it is possible to interpret the hidden structures. I n
the regions of recent sinking and rising of foundation the tone differencesare
very important. In addition, the orientation of the relief and hydropattern
forms, in particular the sharp changes in direction of elements of the erosion
patterns, are also of vital importance.
The closed tectonic fractures along which the tectonic movements took
place are interpreted on the radar aerial images by the oriented contours of
the humidification zones, “ straightened ” portions of erosional pattern,
contacts of the contour configuration and structure differing in tone, and
pattern of the image.
AERIAL GEOLOGICAL4:EOGRAPHICAL EXPLORATIONS 281
The closed structures of West Turkmenia provide an example of inter-

preting buried structures on the radar images. The contours of two folds
divided by the break of northeastern bearing may be distinctly followed
on the radar image (Fig. 7). On the aerial photographs of the same area the
above-mentioned features of geological structure cannot be interpreted.
The conformity of the elements of structural-tectonic schemes, from inter-
pretation of radar aerial images, to the data of geophysical methods and
drilling corroborates the obvious potentialities of this new method for
geological explorations in closed territories.
The examples described above demonstrate wide possibilities of using
radar images for photogeological interpretation and for compiling structural-
tectonic and geomorphological maps, as well as for the interpretation of the
composition of rocks.
The possibility of obtaining quickly the information for vast territories,
regardless of the time and weather conditions, offers a means of completing
in a short time radar surveys in different areas of the country (different in
their geological structure and geographical location), and of conveying the
data to geologists for practical use.
Interesting results were also obtained by using radar images for soil and
field crop studies. One can differentiate plowed lands and those sown with
various crops by the degree of protective cover, the structure of the vegeta-
tion cover, and the moisture, which are manifested in the image through the
change in its tone (Fig. 8 ) .
m1
pJz
FIG.8. Aerial radar imago of agricultural region (a) and the results of its interprctation
(composed by N. N. Semyonova). Drsig~iations:1-fallow fields, 2--unsuitahle, erodcd,
old arable lands. The numerals on the scheme indicate: 1-wheat, 2-millct, 3-barlc.y,
4-maize.
Thus, science and research have gained a new highly efficient means for the
investigation of the earth’s surface which, together with other available means
for obtaining information (aerial photography, aerogeophysics, etc.), will
promote the extension of our knowledge of the natural resources.
3. INFRARED
AERIAL SURVEY
3.1. General
The development of this method for compiling thermal charts of the earth’s
surface is associated with the great success of infrared techniques in recent
years. The usefulness and extension of this application in studying natural
resources are predicated on the fact that many natural and artificial objects
and phenomena are accompanied by temperature contrasts. The landscape
elements with abnormally high temperatures are most readily studied by
infrared aerial surveys. Consequently, at the present time infrared aerial
survey has found a fairly wide and efficient application in the study of regions
of recent volcanism (Gussev et al., 1972; Shilin et al., 1969, 1971) and for
fighting forest fires.
Some results for the first of the above-mentioned applications are given
below. The specifics of using the infrared aerial surveys in this case require a
separate consideration of their use for studying the active volcanoes and the
associated regions of strong geothermal activity.
The infrared aerial survey was carried out from the aircraft Li-2 with the
aid of an infrared scanner, whose detailed description has been given by
Shilin et al. (1971).
3.2. Infrared Aerial Xurvey of Active Volcanoes

An example of the detailed study of active volcanoes by means of infrared
aerial survey is the Karymski volcano-one of the most active volcanoes in
the Kurilo Kamcliatka volcanic zone. It is a regular cone situated in a large
caldera. The diameter of the crater is about 250 m.
No information has been found in the literature about the arrangement
and intensity of this volcanic thermal field, although in recent years the
volcano was systematically investigated by volcanologists (Ivanov, 1970).
This is explained by the inaccessibility of the crater for a direct land-based
study because of the contant high activity.
The infrared aerial surveys were carried out in 1967 and 1972. During the
first period, the volcano was comparatively inactive-only a strong gas
emission by the crater fumarole fields was observed visually. During the
years that followed, the volcanic activity sharply increased, attaining its
maximum in 1971. Fresh lava streams descended the slopes of the volcano.
The period of gas-ash outbursts was reduced to a few minutes. I n 1971 the
activity decreased; during the surveys in October 1972, the outbursts
occurred every few hours, The crater configuration changed, compared to
1967: it had the shape of a slightly extended oval, but the direction of the
long axis changed from NW to SW. The results of the survey in 1967 are
given by Shilin et aZ. (1969).
I n 1972 the infrared aerial survey was carried out from two altitudes (400
and 800 m) and a t different levels of signal discrimination (Fig. 9), which
allowed the obtaining of extensive information on the crater structure. The
infrared image 9(b) was obtained in the spectral interval with a minimum
width of 4.65.3 p (enabling the complete elimination of the influence of the
sun) and with a small level of signal discrimination. As a result, the general
scheme of crater thermal fields and adjacent regions was revealed [Fig. 9(c)].
A similar, but slightly more detailed, scheme is provided by the interpretation
of the infrared image 9(d) obtained a t a somewhat higher level of discrimina-
tion. On that image the weakly heated peripheral areas of the anomalies have
disappeared and the infrared image 9(d) acquired a somewhat greater
distinctness as compared to the preceding one. The analysis of infrared
images demonstrates that the thermal fields of the crater (slightly nonuni-
form in their structure) have the shape of a figure eight extended in the
direction of the crater oval. Intense anomalies of arch-like form are observed
on the NW rim of the crater. A large but somewhat weaker anomaly is
noted on the southern slope, about 80-100 m below its edge. The visual
observations and analysis of the aerial photograph indicate that all these
anomalies are connected with exceptionally active fumarole fields and large
ascent channels over which an intensive gas emission is observed. The
chains of anomalies on the N W slope are due to the anomalous heating of a
young lava stream.
The thermal structure of the crater shown in Fig. 9(c) is confirmed and
shown with improved accuracy by the infrared images 9(e) and (f) obtained
from a lower altitude under close measuring conditions-an average level of
discrimination and narrow spectral limitations of 4.6-5.3 p in (e) and 4.2-
5.3 p in (f). I n accordance with the wider interval in the last case, the portions
which are accentuated on the infrared image have a slightly greater area than
the images of 9(e) due to the passing of the signal from the less heated parts
of the crater. The analysis of these infrared images demonstrates that in the
crater part of the volcano the most active and most heated is its SW area,
whose thermal structure has almost not changed as compared to that shown
in Figs. 9(b) and (d). The intensity of the anomaly is considerably reduced
in the NE area, where some local parts are accentuated. The anomaly chain
of arch-like shape is distinctly observed on the NW rim of the crater. The
FIG.9.(a) Plan of aerial photograph of the Karymski volcano crater. Scale 1: 5,000.
(b) Infrared image of the Karymski volcano crater. Scale 1: 20,000. Spectral interval
4.6-5.3 p. Flight altitude 800 m. Time 11: 10 a.m. Low level of signal discrimination. (c)
General scheme of the volcano crater and its environment’s geothermal fields. Scale
1: 20,000. (d) Infrared image. Scale 1: 20,000. Spectral interval 4.6-5.3 p. Flight altitude
800 m. Time 11: 12 a.m. Low level of signal discrimination. (e),(f) Infrared images.
Scale 1: 10,000. Spectral interval 4.6-5.3 p . Flight altitude 400 m. Time 11: 18 a.m.,
11: 22 a.m. and 11: 28 a.m. Medium level of signal discrimination. (g) Geothermal fields
of volcano crater from data of infrared images 9(e)and (f) interpretation. Scale 1: 10,000.
Medium level of signal discrimination allows the revelation of the geothermal fields of
high intensity. (h),(i) Infrared images. Scale 1: 10,000. Spectral interval $65.3 p.
Flight altitude 400 m. Time 11: 32 a.m. and 11: 38 a.m. High level of signal discrimination.
( j ) Volcanic crater ascent channels scale from data of infrared images 9(h) and (i) inter-
pretation. Scale 1: 10,000. High level of signal discrimination allows the revelation of the
crater ascent channels-the objects of the highest temperature and intensity of radiation.
284
285
286 B. V. SHILL" AND V. B. KOMAROV
anomaly of the southern slope proved t o be considerably weakened. Inter-

pretation of the infrared images 9(e) and 8(f) enables one to plot the scheme
of thermal sections of mean intensity for the crater and its environment
[Fig. 9(g)l.
The increase of the signal discrimination level [compared to the image
9(b) the radiation intensity is reduced about 25 times] while obtaining the
infrared images 9(i) and (j) made it possible t o compose a scheme of the
volcano crater ascent channels 9(k). The most active channels are located
on the NW rim of the crater (arch-like chain of channels-1) in the SSW
area, where they are separated into two nearly equal groups (2 and 3). The
two pairs of channels a t the northern crest of the crater (5 and 6) seem to be
equal in intensity and only a thorough comparison of infrared image 9(i)
and (j) (the latter has been obtained in a narrower spectral interval) reveals
a somewhat lesser activity of the channels located more to the east (6). The
same may be applied t o the central channel (4),which is very slightly per-
ceptible on the infrared image 9(j). All the images display a single channel on
the SSE crest of the crater (7).
I n this way, the method of infrared aerial survey over the active volcano
makes it possible to obtain a variety of information, ranging from the general
scheme of thermal field location t o the arrangement of the most active ascent
channels. It is also possible to conclude that the most active channels of the
volcano are drawn, essentially, t o the peripheral parts of the crater, a matter
of practical interest.
The comparison of the material obtained in the aerial surveys of 1967
(Shilin et al., 1969) and 1972 reveals that in this time interval sharply defined
changes occurredin the structure of the thermal fields of the crater. This seems
to be quite natural when one considers that the volcano passed through a
period of activity in 1970-72. The scheme of thermal field arrangement in the
NNE and SSE parts of the crater or its periphery has been preserved only
in its most general form.
3.3. Study of Regions of Strong Geothermal Activity

The possible application of infrared aerial surveys t o the caldera Uson on
Kamchatka (Svyatlovski, 1959), where thermal phenomena of various
types are known to exist over a vast territory was investigated. The caldera is
located in the central part of the eastern volcanic belt t o the south of Lake
Kronotski and constitutes an extensive (diam 10 km) relatively flat depression
cut 50 to 100 m deep into the surrounding plateau. To the central elevated
part of the caldera belong the main groups of active and extinct thermal
activities. The f i s t are represented by numerous hot and warm springs,
small lakes, baths, streams, gas and steam jets, mud pots, and small vol-
AERIAL GEOLOGICALCJEOGRAPHICAL EXPLORATIONS 287
canoes, as well as by thermal platforms (Gussev et al., 1972). The areas

closely adjacent to these objects are quite devoid of vegetation and are
composed of hydrothermally altered bright opalized rocks. The edges of
hydrothermally altered zones are brightly colored due to the presence of
various oxides. Accordingly, the extinct and active thermal phenomena are
often determined on the aerial photograph by the bright tone of hydro-
thermally altered rocks.
Near the active and partly extinct thermal activities sometimes arises
thick vegetation of high grass which, apparently, may be explained by the
presence of an anomalous heat flow favorable to vegetation. The temperature
of the active thermal phenomena varies from a few degrees over the back-
ground values to 96-98°C.
I n 1967 an infrared aerial survey was carried out from a n aircraft in
September in the daytime from an altitude of 800 m which produced infrared
images with an average scale of 1 : 18,000. Simultaneously, an aerial photo-
graphic survey was carried out for a combined interpretation with the
infrared images. The aerial photograph with a scale of about 1: 18,000 for
the central part of the caldera Uson and the mosaic of infrared images are
shown in Figs. 10 and 11, respectively.
On comparing the infrared image with the aerial photograph their com-
parability in contour and tone becomes evident. However, their contrasts
depend upon quite different qualities. In the aerial photographic survey
these are the differences in spectral reflectance of the landscape elements,
while in the infrared aerial survey the differences reside in thermal emissions
or in radiative temperatures. The contour similarity between the infrared
image and the aerial photograph is explained by the coincidence of certain
landscape elements as regards reflectance and thermal differentiation (e.g. the
contrasts of water and land, of sunlit and shaded slopes in conditions of
dismembered relief, etc.). An appreciable tonal differentiation of infrared
images results from a very wide range of temperatwes, characteristic of
various landscape elements in the daytime-from 5-7°C near the cold springs
and small rivers up to nearly 100°C a t the gas-steam jets and hot springs.
The interpretation of infrared images of the central part of the caldera
enables one to distinguish a large group of bright, anomalous objects of
irregular shape extending in an approximately northwestern direction
and a group of extended parallel anomalies of greater size (located in the
upper part of the infrared image). The limited dynamic range of infrared
equipment and the comparatively broad spectral interval of measurements
( 3 . 2 5 . 3 p ) makes it difficult to eliminate false insolation anomalies, which
greatly hamper the interpretation. The comparison of infrared images with
the aerial photographs, and especially with the stereocouples, enables one to
draw the conclusion that the second group is connected with facing the sun
FIQ.10. Plan of aerial photograph of the central part of caldera Uson. Scale 1 : 18,000.
AERIAL GEOLOGICAL-GEOQRAPHICAL EXPLORATIONS 289
FIG. 11. Mosaic of infrared images of the central part of caldera Uson. Spectral
interval 3.2-5.3 p. Flight altitude 800 m. Daytime, clear weather, separate clouds
Survey of 1967.
and well heated elements of relief. It has been established that in September
the contrast between shaded and sunlit slopes in this region may reach
15-18°C. However, the intensity of the apparent anomalies on the infrared
images makes it possible to overlook the anomalies caused by geothermal
phenomena, as further study has proved.
The first group of anomalies is interpreted more simply, as under the con-
ditions of a practically 0at relief the presence of apparent anomalies has been
reduced to a minimum. An important role is played by the shape and location
of these anomalies, their direct connection to the areas of bright (hydro-
thermally changed) rocks on the aerial photograph and-especially significant
-with the water surfaces of lakes, rivers, and streams. The combined inter-
pretation of large-scale aerial photographs (1: 5,000 1 : 8,000) with infrared
images allows the determination of some types of thermal phenomena
according to their specific indications (Gussev et al., 1972). Many of the
open thermal phenomena of the caldera (Fig. 14) may be determined by
means of indirect interpretation, after performing some field work a t the
test sites.
However, in the initial stage of analyzing the data obtained by aeromethods
in 1967, a scheme of thermal activities (Fig. 12) was compiled, and the
anomalies were divided into two groups: the white color designated the
anomalies which, with a great degree of certainty, may be connected with
geothermal objects; the spotted part indicated the anomalies requiring
ground exploration to determine their nature. The latter particularly concern
the areas of weak heating of landscape elements near intensive thermal
phenomena observed on infrared images.
Based on the results of 1967 and subsequent local checks, in 1972 the
infrared aerial survey of the central part of caldera Uson was carried out from
a lower altitude (600 m) and in the spectral interval of 4.2-5.3 p, which
allowed the infrared image of thermal phenomena to be obtained on a larger
scale and almost without the interference of insolation (Fig. 13). This, in
turn, allowed a scheme of caldera thermal phenomena (Fig. 14) to be plotted,
differing in detail and reliability from the data of 1967. It suffices t o say that
in the survey of 1972 only two thernial areas in the lower part are doubtful
(4 in Fig. 14).
I n Fig. 13 the structure of lakes and thermal fields is interpreted in more
detail; even small thermal objects, such as small single springs, baths, and
Vud pots are determined with assurance. Measurements of temperatures on
the terrain a t the moment of aerial survey provided the following results; a
hot lake (1 on Fig. 13) +38"C, a cold stream (2) +6"C, a mud bath (3) +52"C,
the lake Vosmyorka (4) +lS"C, the spring near the lake shore on the central
fumarole field (5) 45"C, soil in the inactive area (6) f12"C.
For a detailed test of the representativeness of the infrared aerial survey
FIU.12. Data of laboratory interpretation of aerial phohosurvey and infrared aerial

survey (composedby N. A. Gussev).The continuous tone white color denotes the anom-
alies connected with geothermal objects; the specks indicate the anomalies which need
to be checked in the field to reveal their nature.
292 B. V. SRILM AND V. B. KOMAROV
Fro. 13. Mosaic of infrared images of the caldera Uson central part. Scale 1: 12,000.
Spectral interval 4.2-5.3 p. Flight altitude 600 m. Daytime, clear weather, separate
clouds. Survey of 1972. Numerals and arrows indicate the sites of land temperature
measurements during the aerial survey.
.
\
4.
v
FIG.14. Geothermal activities in the central part of caldera Uson from data of infra-
red aerial survey of 1972. Designations: 1-geothermal objects, warm rivers and the most
heated parhs of the lakes; 2-parts of lakes with medium geothermal heating; 3-parts
of lakes with weak geothermal heating; 4-supposed anomalous regions requiring
additional field checkup; I I a and IIb-two actlve parts of central fumarole field. -
294 B. V. SHILIN AND V. B. KOMBROV
data while mapping the thermal activity of the area of anomalies revealed,
surface thermometrical measurements with a grid of 50 x 50 m, as well as
soil and geological observations, were conducted in 1968. The grid was
transferred to a more accurate photomosaic with a scale of about 1 : 5,000.
In the most interesting anomaly areas the grid was narrowed to 25 x 10 m.
All the open thermal activities (springs, gas-steam jets, mud pots, etc.)
were measured by contact mercury thermometers with a range of 0°C to
50°C subdivided into 0.5"C. For the open thermal activities maximum
thermometers were also used. The observation point temperatures were
determined, both on the surface and to depths of 20 and 50 cm.
The data a t the 50 cm depths were used since the influence there of daily
temperature fluctuations could be neglected, and they were therefore best
for comparison with the infrared aerial survey data. The values of the
measured temperatures were plotted on a photomosaic (scale 1 :5,000) with
the identification of the observation point on the aerial photograph. Iso-
therms were plotted separately for the ground and reservoirs in order t o
obtain a better description (Fig. 15).
Using the results of soil-geological observations a t the points where
temperature was measured, a map of open thermal activities was obtained
(Fig. 16), where the areas of strongest geothermal activity are accentuated,
namely the areas of the Eastern (Fig. 16, I ) and Central (Fig. 16,II)fumarole
fields as well as the area of Fumarole Lake-Central Lake.
A comparison of Figs. 12, 15, and 16 shows that all the anomalies revealed
by the infrared aerial survey coincide with the areas of strong geothermal
activity. As a rule, the anomalies of the first type (Fig. 12, white tone, and
Figs. 11, 13) correspond t o the open thermal activities or separate objects
(Fig. 16)-gas steam jets, hot water and mud pots, thermal springs, etc.,
i.e. areas with the highest temperature (Fig. 15), whereas anomalies of the
second type (Fig. 12, spotted part) correspond to the areas with lower
anomalous temperature (20-25°C) that border the open thermal activities.
The comparative analysis of infrared images, photo interpretation data,
thermometrical surveys, and the distribution of geothermal activities enables
one to draw the following conclusions. All data of the infrared surveys are
fully confimed by the results of field observations, the zones of anomalies
coinciding with the areas of strongest thermal activity (Fig. 16) and highest
temperature (Fig. 15), i.e. areas with the convective type of anomalous heat
flow (Shilin et al., 1969). The scheme of thermal activities in 1972 is much
more detailed than the corresponding data for 1967. I n this scheme it is
possible to divide the anomalies of the Central fumarole field into two active
areas (IIa and IIb, Fig. 14)and to note a whole series of thermal objects which
have not been determined by the infrared survey of 1967.
On the thermometrical representation, the isolines of which result from
considerable averaging of the true picture of the thermal field with respect
to space and time (the measurements were conducted during two months),
it is possible to identify only the general contours of the areas with a strong
geothermal activity without revealing the specific objects, their shape or
size. This information, which is of paramount importance for the further
geological exploration, may be obtained in considerable detail only by means
of infrared aerial surveys. As demonstrated in Shilin et al. (1969), the
numerous surveys a t different levels of registered signal discrimination in the
receptor channel of the equipment enable one t o reveal zones with various
anomalous heating (as was done for the Karymski volcano and for hot
streams on the slope of Kihpinych volcano), which greatly increases the scope
of geological information.
The results of laboratory and field interpretation of the aerial photo-
graphic surveys and infrared aerial surveys allow one to discern the presence
of two systems of breaks, in northwestern (principal) and northeastern
directions, that cut the central part of the caldera, but these are characteristic
for the whole caldera and the surrounding region as shown by the available
geological data. These breaks in both directions and especially the weakened
areas of their intersections are practically all associated with known geo-
thermal activities.
On the extension of the main zone of the N W strike (Fig. la), passing
through the Eastern and Central fumarole fields and Lake Fumarole an
exceptionally active Western fumarole field (studied by means of infrared
aerial survey in 1972) is also known to exist.
Another example of mapping an area with various great thermal activities
by means of aerial methods is presented in the results obtained for the well-
known Geyser Valley (Ustinova, 1955; Vinogradov, 1964) located a t the foot
of the volcano Kihpinych. The Geyser River flowing from the volcano cuts
into a powerful complex of volcanic rocks represented by lavas and tuffs of
andesite and dacitic composition. In the area of the main thermal activities
the river forms a narrow canyon cut t o a depth of more than 300 m a t the
absolute depth of its surrounding plateau of about 800-1000 m. Because of the
presence of the volcano in the vicinity of a high massif, the character of the
relief seems to be very unfavorable for getting the data from the large-scale
aerial surveys. However, in 1967 and in 1972 we succeeded in obtaining data
enabling us to plot sufficiently detailed schemes of thermal activities.
The thermal activities in the form of hot springs with various discharges,
steam jets, mud pots and small lakes, geysers, thermal platforms, etc. are
known over the whole length of the Geyser River valley, with the following
regularity being observed in their arrangement (Ustinova, 1955). Close to the
river the massive-yield hot springs are situated, a little higher-the geysers
and hot springs with small and varying discharge. In the upper part of the
FIG.18. Thermometric scheme of the central part of caldera Uson from data of
field temperature measurements (in "C).Scale 1 : 12000. (Composedby E. I. Vavilov and
B. V. Shilin.)
AERIAL GEOLOGICAGGEOCRAPHICAL EXPLORATIONS 297
!
I Gas-vapor jets
2 Hot water pots
3 Mud p o t s
4 S m a l l mud volcanoes
5 Hot a n d worm springs
6 A r e a o f strongest
termal activity
FIG. 16. Scheme of geothermal activities in the central part of caldera Uson
(I-activities in the Eastern fumarole field, 11-activities in the Central fumarole
field). (Composed by E. I. Vavilov.)
298 B. V. SRILIN AND V. B. KOWOV
slopes the steam wells, mud pots, and small lakes are to be noted, mainly on
the terrace in the middle part of the slope.
The aerial survey was carried out over the middle (and thermally most
active) part of the Geyser River. The results are shown in Figs. 17-20.
FIQ.17. Plan of aerial photograph of the most geothermally active part of Geyser
Valley. Soale 1 : 10,000,
Figure 17 is an aerial photograph to a scale of about 1 : 10,000; Fig. 18 is the

mosaic of infrared images to a scale of about 1: 15,000 (1972); Fig. 19 is the
arrangement of thermal activities of the same scale composed from the
interpretation data; Fig. 20 is the mosaic of infrared images to a scale of about
1 : 20,000 (1967).
The rather favorable conditions of surveying in the middle of October
1972, the weakly heated slopes exposed to the sun, and the presence of large
AERIAL GEOLOGICAGGEOORAPHICAL EXPLORATIONS 299
Fro. 18. Mosaic of infrared images of the mme part of the valley. Scale 1: 16,000.
Spectral interval 4.2-6.3 p Flight altitude 600 m. Daytime, clear weather, separate
clouds. Survey of 1972.
FIQ.19. Thermal activities of Geyser Valley from data of infrared aerial survey, 1972.
The numerals show geothermal regions. The thermal field VIII is shown shaded due to
the impossibility of detailed distinction between separate objects.
FIG.20. Mosaic of infrared images. Scale 1: 20,000. Spectral interval 3.2-5.3~.Flight

latitude 900 m. Daytime. Considerable cloudiness. Survey of 1967.
shaded areas, as well as heavy cloudiness during the flight of 1967, enabled
us to distinguish easily on the infrared images between the bright local
anomalies of thermal activities with sharp contours and pale broad anomalies
connected with solar heating.
Figure 21 shows the arrangement of thermal activities compiled by
former investigators according to field observations (Vinogradov, 1964). The
pattern includes the geysers (l),the main pulsing sources (2),and the so-called
thermal areas (3) which imply the accumulation of active areas of various
thermal activities and whose determination is very conventional.
200. 0 zoo 4oou
FIG.21. Scheme of geothermal objects in Geyser Valley from the data of field observ-
ations of the Institute for Volcanology. Designations: 1-geysers (1-Rozovi Konus, 2-
Horizontal, 3--Zhemchushni, PVelkian, 5-Nepostoyanni, 6-Dvoynoy, 7-Fontan,
8-Novi Fontan, g-shchel’, 10-Bolshoy, 1 I-Malyi, 12-Konus, 13-Bolshaya
Pechka) ; 2-main pulsating springs ( 1-Paryashchi, 2-Plachushchi, a-Nepreryvni,
4-Grot, 5-Malahitovyi grot, 6-Malaya Pechka, 7--Kamenka, 8-Skalisti, 9-Vorota
v Geyzernuju); 3-the most active geothermal regions.
The interpretation of aerial photographs, analyses of infrared images, and

patterns of thermal activities prove that the data of aerial methods are fully
confirmed by the data of field observot,ions, but they immeasurably surpass
the latter in the totality of information content.
I n the easternmost part of the thermal activities, where the Geyser River
forms two peculiar bends (I), anomalous heating of the river itself (due t o the
springs in its upper part) is observed as well as the areas where small warm
302 B. V. SHILM AND V. B. KOMAROV
springs situated in the vicinity discharge into the river. The latter are
interpreted on the infrared image as slightly stretched anomalies suggesting
" commas " by their shape. The heating of these objects essentially exceeds
the heating of the river, which enables one to identify them as independent
thermal activities. However, their direct connection with the river demon-
strates that they are warm springs, whose discharge is not great and does not
cause an appreciable increase in the temperature of the river water. Here
again, near the river, a little higher along its left bank, a few small point-like
thermal activities, apparently gas-steam jets, are revealed.
Further west occur anomalies of complex thermal structure (11), with the
southwestern part undoubtedly due to the intensive heat of the geysers
Velikan, Zhemchuzhni, Horizontalni, Rozovi Konus, the pulsating spring
Paryashtshi of small geysers, and hot water and steam discharges connected
with them. However, a considerable part of the thermal field of this district
is connected with the slope of the right bank of Geyser River.
An intensive group of anomalies stretched along the following rectilinear
river district (111)is connected with the geysers Dvoynoi, Nepostoyanni,
Fontan, Novi Fontan, the high-discharge pulsating springs Grot, Malahitovi
Grot, and others. A high, and comparatively constant, discharge of these
thermal objects resulted in a high intensity of infrared radiation of this group
of anomalies.
The ring of local anomalies (IV) is connected with the thermal activities
of the terrace on the left bank of the Geyser River and the valley of its small
warm effluent. The thermal activities are represented by numerous small warm
lakes, mud pots, gas steam jets, heated platforms, and in the effluent valley
also by hot springs and small geysers (Ustinova, 1955; Vinogradov, 1964).
The temperature of the objects varies from a few tens of degrees in small warm
lakes up to the boiling point in geysers and springs. The terrace consists
almost completely of hydrothermally changed warm, claylike, opalized
rocks covered with thick high grass and shrubbery. Only the areas of thermal
activity are free from vegetation. The data of the infrared aerial survey and
aerial photographic survey demonstrate that the western, more elevated
part of the terrace has no active areas, whereas a ring structure (of great
interest for further geological work) occupies its middle and eastern parts.
I n a similar region of Iceland this kind of structure, though much greater in
scale, has been identified through infrared surveys.
About 150-200 m to the west of the ring structure, the infrared aerial
survey reveals a group of three anomalies (V) connected with the hydro-
thermally changed and heated rocks in the upper part of the southern slope
of the Geyser River valley. There, although only small discharges of steam
jets are observed, the intensity of the anomaly indicates high intensity heat
flow, which had not been noted in prior investigations.
AERIAL GEOLOGICAGQEOGRAPHICAL EXPLORATIONS 303
The anomalous zone VI, which for some reason or other has not been
mentioned before (Ustinova, 1955; Vinogradov, 1964), should be particularly
identified, as the character of the heat anomalies proves that they are caused
not only by the warm water of the Geyser River, but also by thermal objects
located on the right bank. These, in all probability, are small springs; on the
infrared image they are connected with the river.
Downstream of the Geyser River, where the valley is a narrow canyon
with steep, up to 70°, slopes (Vinogradov, 1964), two thermal anomalies
had been previously identified (Fig. 21): the region of geysers Bolshoi and
Malyi a t the beginning of the canyon and the region of the geysers Konus
and Bolshaya Petchka. The analysis of the aerial photograph and the
infrared images gives an essentially new picture of the thermal structure of
this zone (VII): from the region of geysers Bolshoi and Malyi the zone is
divided into two parallel subzones. The northern subzone is connected with the
anomaly from the warm waters of the Geyser River flowing into it from the
adjacent small springs, from geysers Konus and Bolshaya Petchka, and the
pulsating springs Skalistyi and Kamenka. The southern subzone is con-
nected with the thermal discharges of the right slope in the upper part of the
river. Apparently, these discharges are not accompanied by outflow of warm
water, as on the infrared image no connection with the river is observed.
The eighth anomalous group (VIII) coincides with the region of thermal
activity located 500 m higher than the mouth of the stream flowing into the
Geyser River on its right bank, where the active region rises up to a height
of 50 m. Here many mud pots, gas steam jets, springs, etc. are situated,
whose water appreciably increases the temperature of the stream, a fact well
discernible on the infrared image.
I n summary, the infrared aerial survey distinctly reveals all the anomalous
zones of the main thermal region of the Geyser Valley and its inner structure,
and the mass of information largely exceeds the data of many years of field
observations. The data of infrared aerial surveys may serve as a basis for
subsequent detailed geological field investigations of the structure and
origin of thermal zones.
The comparison of the infrared aerial surveys of 1967 and 1972 proves that,
when the flight altitude is reduced, the detail of geothermal objects is some-
what increased, and the reduction of the spectral interval width (from 3.2 p
to 4.2 p ) practically eliminates the losses resulting from solar heating and
from reflected solar radiation. It should also be noted that no appreciable
changes took place in the structure of the geothermal activities of the
Geyser Valley during the last five years.
I n conclusion, i t may be said that the large-scale infrared aerial survey
permitted the study with sufficient detail of the geothermal fields and the
active volcanoes of the Eastern volcanic belt from the volcano Karymski in
the south up to Lake Kronotski in the north. The study of the regional
geological structure of this region undoubtedly indicates the presence of an
interconnection between these objects. However, this important inter-
connection is revealed only on the infrared aerial surveys of sufficiently
small scale (less than 1 : 100000) which were carried out many times under
favorable conditions at night different levels of signal discrimination.
4. AEROQEOCHEMICllL SURVEY: REMOTE

SENSINQOF GASES AND VAPORS
4.1. Method
The method to be considered now is based upon recording from aircraft
the volatile elements and compounds evaporated by surface objects into the
atmosphere. By analogy with the widely used geochemical surveys studying
the haloes of dispersion of chemical elements and compounds in soils and
rocks (Ginzburg, 1957), one may label these “gas haloes” or “evaporation
haloes” in the near-surface layer of the atmosphere and call such studies
aerogeochemical surveys.
To the volatile elements and compounds, which are of interest from the
geological-geochemicalpoint of view, belong in the first place the halogens,
sulfurous gas, hydrogen sulfide, a number of hydrocarbons, mercury (to a
certain extent), etc. The work of Berringer discusses the great potential
of using aerogeochemicalsurveys in geology for seeking sulfur-sulfide deposits
by SO, haloes, oil deposits by SO, and iodine haloes, and polymetallic
deposits by mercury haloes (Barringer, 1964; Barringer and Schork, 1966;
Barringer et al., 1968).
Among recent work, the report (Rouse and Stevens, 1970) on the study of
SO, haloes over the sulfide deposits merits consideration. According to the
geological conditions (mainly the thickness of overlying sediments, the
presence of faults, and weather factors), the SO, concentration varies from
0.03 to 1.0 mg/m3, when the measurements are taken from an automobile
and from a helicopter with flight speed up to 40 miles/hr.
The available general geological and geochemical research suggests a
potential use for aerogeochemical survey. However, the actual data are
evidently inadequate for organizing operations in the search for minerals. In
particular, there are practically no data on field measurements of gas con-
centration in the air over mining fields and over oil fields. Accordingly, it
seems more appropriate to conduct the experimental flights on objects with
exceptionally pronounced haloes of evaporation, namely the active vol-
canoes and regions of strong geothermal activity.
AERIAL GEOLOGICAL-CEOGRAPIIICAL EXPLORATIONS 305
4.2. Equipment and Procedure
The specific features of measuring in flight the haloes of evaporation from

the objects on the surface impose two demands on the equipment-a con-
tinuity of measurement and rapidity of action. A great number of high
precision methods for determining the gas concentration have been developed
in recent years; yet in most cases they display a n essential defect-the cycling
of measurements or a considerable time needed for one measurement.
Consequently, in the first stage of the research i t seems to be an optimal solu-
tion t o use the gas analyzers based on the coulomb-polarography principle
with a sensitivity of 0.03 mg/m3 (Alperin et al., 1968). Devices of this type
enable one t o take continuous measurements, although they display an
appreciable inertia. The latter restricts their use to helicopters.
During the summers of 1970-72, the gas analyzer was used not only for
SO,, but also for H,S measurements. The acquisition of data on the H,S
concentration is possible only in the presence of SO, absorption filters a t the
inlet of the gas analyzer. A cartridge with cotton was used as a n SO, filter,
which proved t o be quite satisfactory (Manganell, 1970).
When the measurements are made from the helicopter, distortions resulting
from the turbulent flow created by the rotor complicates matters. On the
basis of theoretical and experimental data on helicopter aerodynamics,
optimal conditions for measuring the gas haloes from a helicopter are provided
when the sensors are installed a t a distance of 1 m in front of the fuselage and
speeds of 50-70 km/hr, so that the influence of turbulent flow from the rotor
may be neglected.
However, in 1970-71, under the prevailing technical conditions, various
methods of sampling the gas specimens were used. I n 1970, the measurements
were made by directly sampling the gases during the flights in extended gas
plumes of an active volcano, or when the helicopter hovered a t a low height
over the fumarole fields. This permits one to neglect the inertness of the
device and the air intake tubes were extended through the first left illumina-
tor of the helicopter. The instantaneous intake of specimens into a specially
designed system of samplers was accomplished through the same tubes. The
system consisted of a few glass 1 liter ampoules with cocks a t their ends, a
vacuum pump, and a system of fluoroplastic tubes. When sampling was done
in flight, the ampoules were connected to the pump and evacuated. On
signal of the navigator, the cock of one of the ampoules connecting it with the
air intake tube was opened. The air was taken into the ampoule and its
subsequent analysis was performed in flight. After the measurement was
completed, the ampoule was evacuated for the next sampling. The test results
of this sampling system indicated that reliable results may be obtained only
from the first measurements. During the subsequent use of the same am-
poules, the gas concentration sharply decreased because of absorption by

water which settled on the walls during the first measurements as a result of
adiabatic expansion.
I n 1971-72 the measurements were performed using an intake device in
the form of a tube, 1 m long, installed in the front part of the helicopter. The
sample was collected into polyethylene sacks (45liters), which were filled in
1-2 sec.
All aerial surveys of 1970-71 were conducted over the active volcano
Karymski and the thermal activities of the caldera Uson which covers a large
area of thermal springs and gas-steam jets.
While working over the volcano the helicopter climbed to the level of the
gas-ash plume (Fig. 22). The navigator determined visually the thickest
FIG.22. Eruption of volcano Karymski. (Photo by N. A. Gussev.)
part of the plume and the flight proceeded normal to the plume, usually a t a
speed of 100 km/hr or less. On the signal of the navigator a recorder was
switched on and the polyethylene sacks were filled. Usually the flight time
through the plume lasted tens of seconds. During the turn of the helicopter
preceding the next approach to the plume, the gas concentrations in the sacks
were measured. I n the plume of an explosion only one or two passes were
made because the plume dispersed rapidly. Sometimes the sequence of
volcanic explosions was so rapid that the helicopter did not land in between.
I n only two years more than seventy measurements were made over the
voIcano Karymski. An example of the recording of the anomalous concen-
tration of gas in the plume is shown in Fig. 23.
I n flights over the thermal activities of the caldera Uson, the concentra-
tion of gases was measured under different flight regimes (altitude, speed) in
various meteorological conditions. The altitude varied between 5 and 40 m,
while the method of specimen sampling was nearly the same as that used
over the volcano, although the navigator had to be very careful while giving
the signals for sampling because of the extremely small size of the objects in
this case. Measurements made directly by the apparatus were done only
while hovering. Besides, all measurements, about 60 of them, were made only
in flights against the wind.
Fro. 23. Example of recording the anomalous concentration in the train of outbursts
of volcano Karymski.
Over the fumarole fields of the caldera Uson some measurements were
made in the regions of the most active exhalations of the Central and Eastern
fumarole fields, represented by sectors that are very close to thermal springs,
boiling baths, and gas-steam jets (Fig. 24) which noisily emit great quantities
of gas and steam. The smell of hydrogen sulfide in calm weather is noted a t a
distance of many meters. During field measurements over these objects, the
+
values of the total concentration of SO, H,S varied from 1.5 mg/m3 (with
the apparatus installed a t a height of 3.5 m) up to 4.5mg/m3 (for its position
near the edge of the boiling baths). The maximum concentration (of more
than 20 mg/m3) has been noticed when the intake tube was introduced
directly into the mouth of the gas discharge channels.
Thus a few field measurements have shown a very rapid decrease of gas
308 B. V. SHlLIN AND V. B. KOMAROV
Fra. 24. Fumarole fields of caldera Uson.
concentration on moving aside and upstream of the spring. These data are in
good agreement with the results of the measurements in air, discussed below,
which indicate that the gas content very quickly decreases with the altitude
due to its dispersion in the air.
4.3. Aerial Measurements over Fumarole Fields of Caldera Uson

The results of aerial measurements over the thermal activities of the
caldera Uson (Central and Eastern fumarole fields, Figs. 10 and 24) are
summarized in Table I. I n analyzing these data the following should be
noted.
All the thermal activities under study are distinctly b e d by the aero-
geochemical survey, while the results obtained are essentially dependent on
the measuring conditions. An appreciable dependence of the measuring
results on the flight altitude and speed as well as on the wind velocity is
observed. Maximum concentrations were noticed at the minimum height of
5-10 m, small helicopter speeds, and in calm weather (cases shown in Table
I-Serial Nos. 1, 3, 10, 11, 15, 16, 23, 24, 26, 28, 34, 40, 62, 53). When the
flight altitude is increased, a rapid decrease of gas concentration down to the
background values a t a height of 25-40 meters is observed (Nos. 1-7, 15-18,
2 6 3 2 , 4 0 4 5 ) .This is the consequence of the decrease of the gas concentration
by rapid dispersion in the atmosphere as the distance from the source
increases.
TABLEI. 'Data on aerogeochemical survey over caldera Uson thermal activities
Helicop- Average Content

Time of Flight
Serial ter wind of
No.
Region0 measure-
ment
altitude
speed speed
Course
+
SO, H 8
Remarks
(m) (km/W (WW (mg/m3)
b
1970 M
E
1 CFF 12 : 50 p.m. 4-5 40 25 350°-3600 0.15 Sunny day, separate clouds, $
air temperature 3-5 C 0
M
2 CFF 12: 53 p.m. 4-5 40 25 35Oo-36O0 0.08 Wind gust, the train "floats " o
3 CFF 12 : 55 p.m. 10 40 25 350"-360" 0.22 5:
4 CFF 12 : 56 p.m. 10 20 25 350"-360° 0.15 8
5
6
7
8
CFF
CFF
CFF
EFF, C
12 :58 p.m.
13 :02 p.m.
13 :04 p.m.
13 :06 p.m.
15
15
15
4-5
30
20-30
20-30
10
25
25
25
350"-360"
350'-360"
350"-360"
350°-3600
0.06
0.04
0.03
0.09
E
M
0
9 EFF, C 13 :08 p.m. 4-5 10 25 350"-360 0.11

10
11
EFF, C
EFF, C
13 : 10 p.m.
13 : 12 p.m.
10
15
Hovering
Hovering
25
25
350"-360°
350"-360"
0.2
0.14
$
11
12 EFF, C 13 : 14 p.m. 20 30 25 350"-360" 0.04 Wind gust, the train "floats
13 EFF, C 13 :16 p.m. 20 30 25 350"-360" 0.11
14 EFF, C 13 : 18 p.m. 30 50 25 350"-360" 0.03
15 EFF, W 13 :22 p.m. 4-5 10 25 310"-320" 0.24
16 EFF, W 13 :26 p.m. 4-5 10 25 310°-320" 0.22
17 EFF, W 13 :26 p.m. 10 30 25 310°-320" 0.12
18 EFF, W 13 : 28 p.m. 20 30-40 25 310°-320" 0.06
19 EFF, W 13 :30 p.m. 5-7 70-80 25 310"-320" 0.14
2
20 EFF, W 13 :32 p.m. 15 100 25 310"-320° 0.05
21 EFF, W 13 : 34 p.m. 15 100 25 31Oo-32O0 0.04 Wind gust
22 EFF, W 13 : 37 p.m. 15 100 25 310°-320" 0.04
23 EFF, W 13 : 43 p.m. 5 10-20 25 310"-320" 0.3 0
continued
TABLEI-Continued
Helicop- Average Content Content

Time of Flight
Serial ter wind of of
Region" measure- altitude Course Remarks
NO.
ment
speed speed so2 H2S td
(4 (Whr) (h/W (mg/m3) (mg/m3) c
10.15.7 1 1971
- B
24 EFF, C 11 : 30 a.m. 5 20 10-20 240' 0.00 0.13 Sunny day, separate &
clouds, air tem- P
24
perature -2°C U
25
26
EFF, C
EFF, C
11: 32 a.m.
11: 33 a.m.
5
10
40
40
10-20
10-20
240'
240"
0.00
0.00
0.13
0.03
c
40 10-20 240' 0.00 0.01 Wind gust
PJ
27 EFF, C 11: 34 a.m. 15
p:
28 EFF, C 11: 36 a.m. 15 40 10-20 240" 0.04 0.11 0
29
30
31
EFF, C
EFF, C
EFF, C
11: 37 a.m.
11: 39 a.m.
ll:40a.m.
20
25
30
80
80
80
10-20
10-20
10-20
240'
240"
240'
0.00
0.00
0.00
0.09
0.09
0.01
E
0
d
32 EFF, C 11: 42 a.m. 40 80 10-20 240' 0.00 0.01
33 EFF 11: 44 a.m. 40 80 10-20 240' 0.00 0.00
34 CFF 11: 48 a.m. 5 40 10-20 280" 0.00 0.16
35 EFF 11: 51 a.m. 5 Hovering 10-20 280' 0.00 0.01
36 CFF 13: 10 p.m. 5 40 10-20 280" 0.00 0.06 The wind becomes
somewhat stronger
37 CFF 13 :10 p.m. 5 40 10-20 280" 0.00 0.06
38 EFF 13 : 15 p.m. 10 40 10-20 280" 0.06 0.03
39 CFF 13 : 16 pm. 10 40 10-20 280" 0.00 0.01
10.26.1971
40 EFF 12 : 37 p.m. 5 60 30 30"-40" 0.0 0.06 Sunny day, separate
clouds, air tem-
perature - 5°C
41 EFF 12 : 39 p.m. 10 60 30 30"-40" 0.0 0.03
42 EFF 12 :41 p.m. 15 60 30 30'-40' 0.0 0.03
43 EFF 12 :42 p.m. 20 60 30 30"-40" 0.0 0.03
44 EFF 12 :43 p.m. 30 60 30 30"-40" 0.0 0.01
45 EFF 12 : 44 p.m. 40 70 30 30"-40" 0.0 0.01
46 EFF 12 : 45 p.m. 30 60 30 30'-40" 0.05 0.01
47 EFF 12 :47 p.m. 20 60 30 30"-40" 0.0 0.02
48 EFF 12 :48 p.m. 15 60 30 30"-40" 0.0 0.02
49 EFF 12 : 50 p.m. 10 60 30 30"-40" 0.03 0.03
50 EFF 12 : 52 p.m. 5 60 30 30"-40" 0.0 0.08
51 CFF 12 :54 p.m. 5 40 20 30"-40" 0.0 0.06
52 CFF 12 :55 p m . 10 40 20 30"-40" 0.0 0.16
53 CFF 12 :57 p.m. 15 60 20 30"-40" 0.0 0.01
54 CFF 12 :58 p.m. 20 70 20 30"-40" 0.0 0.1
55 CFF 13 :00 p.m. 30 80 20 30'-40" 0.0 0.06
56 CFF 13 :01 p.m. 40 80 20 30"-40" 0.0 0.03
57 CFF 13 : 03 p.m. 30 70 20 30"-40" 0.0 0.07
58 CFF 13 :04 p.m. 20 80 20 30"-40° 0.0 0.03
59 CFF 13 :05 p.m. 15 80 20 30"-40" 0.0 0.06
60 CFF 13 :06 p.m. 10 60 20 30"-40° 0.0 0.02
61 CFF 13: 07 p.m. 5 40 20 30"-40° 0.0 0.07
-~
a Key t o abbreviations: CFF central fumarole field; EFF, eastern fumerole field, EFF, C eastern fumarole field, central part; EFF, W
eastern fumarole field, western part.
312 B. V. SHILJN AND V. B. KOMlLROV
The recorded value of the concentration decreases sharply when the speed
of the flight is increased. This becomes evident when comparing, e.g., the
values of Nos. 15, 16 and 19, 20, 9 and 10, 11, etc. In the last case, the in-
0uence of a small increase in speed is even more appreciable than the rise of
the flight altitude which is probably explained by the inertia of the gas
analyzer. This is confirmed by a good coincidence of the data obtained in
measurements from the hovering helicopter (Nos. 10, 11) with the use of the
sampler made of ampoules (Nos. 22, 23), where the intake of the gas for the
subsequent analysis is nearly instantaneous, and excludes the influence of
instrument lag. The disparity of the reading No. 21 (0.04 mg/m3)results from
the incorrect switching on of the sampler at the moment when the helicopter
has nearly ‘(slipped” over the gas halo. This was also the case when the
sample was taken into a polyethylene sack (with No. 24). Here only some
disparities of the values are noted in unsuccessful attempts to sample gas at
the edges of the halo (Nos. 26, 27, 60).
The low value of the concentration when the helicopter was hovering at a
low altitude (No. 35), was due to the unfavorable conditions of intake flow
caused by the helicopter rotor. However, when the helicopter is hovering
with a gas pickup through the intake tubes to the apparatus, high values are
observed which decrease somewhat according to the altitude (Nos. 10, 11).
At a low altitude (6 m) the helicopter practically hovers in the center of the
halo mixing the gas with its rotor and causing an (‘averaging ” of concen-
trations.
Various changes of the wind exert a great influence on the results of
aerial gas surveys. When the wind is strong and gusty, the gas halo “ creeps ”
along the earth, its upper edge drops, and measurements are hampered.
Thus, for example, at a constant altitude and speed of flight (Nos. 12, 13, 1,
and 2)) but with sharp gusty wind, the gas concentration is reduced by one
half as compared with calmer conditions. Evidently, this is also the reason
for the general drop of gas concentration over the Eastern fumarole field on
October 26, 1971 (Nos. 40-50)) as compared to the data obtained ten days
earlier (Nos. 24-33).
These conclusions indicate the changeability, of the gas halo and the
variability of the measured values due to the numerous conditions for
dispersal. This results in great difficulties for the development of such
surveying methods and makes only very generalized rules in the search for
objects of small dimensions possible:
1. The aerogeochemical survey should be carried out from a helicopter at
minimum possible altitudes and optimal speeds.
2. The aerogeochemical survey should be performed under conditions of
calm or steady wind with a speed of not over 10 km/hr. The direction of the
flight should only be upwind, when the conditions for sampling are most
favorable.
AERIAL GEOLOGICAIrGEOCRAPHICAL EXPLORATIONS 313
3. The lag of the equipment must be reduced and instantaneous sampling

must be improved or other principles for measurements are required.
As follows from the analysis of Table I (Nos. 24-61), in the gas haloes of
fumarole fields an appreciable predominance of hydrogen sulfide over
sulfurous anhydride (90-100 yo H,S, seldom 50-60 yo)is observed while in
the gas outbursts of the volcanoes quite an opposite picture occurs. Such a
difference in the final composition of gas haloes having a common or similar
primary deep volcanic focus is explained by the influence of processes of SO,
reduction while passing through thick deposits of various rocks. I n the case
of an active volcano the primary gases are ejected into the atmosphere with-
out considerable changes in their initial composition.
I n conclusion, we wish to emphasize the good agreement between the
results of field and aerial observations of 1970 and 1971, regardless of some
difference in the operational methods. Thus, for instance, the general charac-
ter of the gas halo over the Eastern fumarole field is virtually not altered:
the maximum concentrations amount t o 0.24 and 0.25 mg/m3, respectively;
the anomalous measurements of the value are recorded up t o a height of
20-25 m. This suggests a satisfactory correlation of the aerogeochemical
surveys of 1970 and 1971, as well as the absence of changes in the gas regime
of geothermal fields.
4.4. Active Volcanoes of Kamchatka
The most interesting results were obtained over the volcano Karymski.
During October, 1970, the volcano was active, with a period of gas-ash
explosions of 1-2 hr. The duration of an explosion usually did not exceed
1-2 min, the main outburst of ash and gas occurring in the first 5-10 sec.
After the cloud broke away from the crater, only a weak gas escape was
observed from the crater fumaroles. It was easy to notice visually the
changing of color in the outbursts connected with the change of their com-
position and ash content. After the outburst the propagation of the gas cloud
depended on the wind speed and direction, dispersing at a distance of 8-10
km from the volcano. During October 1971 the volcanic activity increased
sharply, the interval between explosions was reduced to a few minutes,
although the visual intensity of the strongest outbursts in our opinion hardly
exceeded the intensity of the similar explosions in 1970. The stable gas plume
extended from the volcano up to the Pacific Ocean coast, where near the
shoreananomalous concentration of SO,, 0.04 mg/m3 (No. 26), was observed.
I n October 1972, a considerable decrease of the volcanic activity was ob-
served; the interval between outbursts increased up to 4-6 hr. I n total,
during four working days of 1970, eight outbursts were investigated, in
1971-twenty nine, in 1 9 7 2 - 4 ~ 0The
. results of the measurements are given
in Table 11.
314 B . V. SHILIN AND V. B. KOMAROV
A comparison of the results obtained in 1970 and 1971 indicates that, as

compared to the previous season, the activity of the volcano increased con-
siderably. This is confirmed by a sharp increase of total SO, and H,S con-
centrations in the outbursts of 1971. Thus, for example, if in 1970 the maxi-
mum measured value amounted t o 2.4 mg/m3, in 1971 it exceeded 10 mg/m3.
I n the small outbursts (12 : 50 p.m., 12 : 24 p.m., etc.) very high concentra-
tions were observed which did not occur in 1970.
While analyzing the ratio of sulfurous gas and hydrogen sulfide in the
composition of volcanic gases, a sharp predominance of the former may be
observed. The content of hydrogen sulfide seldom amounts t o 50 yo of the
total sum and usually varies within the limits of0.2 to 2.0 mg/m3, reaching its
maximum sharply differing value of 5.7 mg/m3. The gas relationship occurring
in this case has already been explained above.
The idea of sulfurous anhydride being a sensitive indicator of growing
volcanic activity (Masayo, 1961) is again confirmed. The increase of SO,
concentration in volcanic gases attests to th > rise of temperature and pressure
in the volcanic focus, i.e. it is one of the first predictors of powerful eruptions.
A detailed examination of the data obtained in the aerial surveys of 1970
(Nos. 1-15) allows one to draw the conclusion that in the period of the survey
(October 16-October 24, 1970) the activity of the volcano was characterized
by the alternation of two types of explosions: gas-ash outbursts with a gas
content of 0.8 to 2.4 mg/m3 (10/16 a t 11 :08 a.m.; 10/21 a t 11 :07 a.m.,
14:32 p.m.; 10/23 a t 11 :45 a.m., 12:53 p.m.; 10/24 a t 11 :09 a.m.) and gas
outbursts with a gas content of less than 0.25 mg/m3 (10121 a t 12 : 38 p.m.,
11 :27 a.m.). Although the data are obviously insufficient to give the exact
sequence, they seem to be confirmed by visual observations.
I n 1971, during the intensive activity of the volcano, there were many
more measurements in gas outbursts, and appreciable fluctuations of con-
centration both within the limits of the data on one outburst and between
different outbursts were noticed. I n the tkst case these fluctuations were
somewhat less and may be attributed either to the imperfection of the
measuring methods, e.g. pilot errors (No. 48), or to the irregular gas
distribution in the halo, the latter being very notable (see, for instance,
Nos. 19-21, 22-25, 44-47, etc.).
As in the previous year, two types of explosions may be distinguished
according to the data of aerogeochemical surveys: weak purely gaseous
outbursts with a total gas content up to 2.5 mg/m3 (the most frequently
encountered is the value of about 1 mg/m3) and powerful ash-gas outbursts
with a total gas content up to 10 mg/m3 and more (the most frequent value
is 6-7 mg/m3). Also, it is important to note a general weak tendency to a
decrease of the quantity of sulfurous hydrogen in powerful explosions.
I n 1972 the anomalous concentration of SO, sharply decreased and the
TABLE11. Data on aerochemical survey over active volcanoes
Distance
to volcano
Heli- Content
Date and Time of Flight from Wind
copter of
Remarks
Serial
No.
time of
explosions
measure-
ment
altitude measure-
ment
speed
speed
Course
so2 + H2S
point
(4 (W (WW (WW (mg/m3)
1971
Volcano Karmski
1 10.16.1970 11:09 a.m. 1800 4.0 30 100 310 1.8 Clear weather, separate clouds
11: 00 a.m.
Volcano Malyi Semlyachek, crater lake
2 10.16.1970 11: 18 a.m. 0 0.5-0.7 20 100 100 0.4
Volcano Karimski
3 10.21.1970 11: 17 a.m. 2000 4 25 100 310 1.76 Clear weather, separate clouds
11 :00 a.m.
4 10.21.1970 11:23 a.m. 2000 5 25 100 310 1.3
11 :00 a.m.
5 10.21.1970 12 :45 p.m. 1800 5 25 100 310 0.25
I2 :38 p.m.
6 10.21.1970 12 :52 p.m. 1800 5 25 100 310 0.20
12 : 38 p.m.
7 10.21.1970 14: 46 p.m. 1700 5(9) 25 100 315 0.92
14 : 39 p.m.
~~
(continued)
TABLE11-Continued
Distance
Content of
Flight t o volcano
Serial
Date and
time of
Time of
alti- from Wind
Heli- +
SO2 H2S Content Content
measure- copter Course by of of Remarks
No. explo- tude measuring speed
sions ment point speed measuring SO2 Has
device
(m) (h/W (Whr)
8 10.23.1970 11:39a.m. 1200 2.0 20 120 290 0.2 Clear weather, separate clouds
11:27 a.m.
9 10.23.1970 11:49p.m. 1600 1.6 20 100 290 2.15
11: 45 a.m.
10 10.23.1970 11:53a.m. 1600 3.0 20 100 300 0.5
-
w
Q,
11
11: 45 a.m.
10.23.1970
12 : 53 p.m.
13 :00 p.m. 1700 1.0-1.5 20 100 290 1.75
12 10.23.1970 13: 04 p.m. 1900 5.0-6.0 20 100 290 2.4

12 :53 p.m.
13 10.23.1970 13 : 07 p.m. 2000 7.0-8.0 20 120 290 1.0
12 : 53 p.m.
14 10.24.1970 11: 17 a.m. 1600 5.0 30 100 320 0.8
11: 09 a.m.
15 10.24.1970 11:20 a.m. 2000 8.0 30 100 40 1.30
11:Oga.m.
1971
Volcano Karymski
16 10.12.1971 12: 05 p.m. 1800 4.0 40 100 230 2.0 0.98 0.25 Clear weather,
11:50a.m. separate
clouds
17 0.31 0.1
18 0.6 -
19 10.12.1971 12: 15 p.m. 2100 2.0 40 100 300 1.64 1.30 0.21
12 : 12 p.m.
20 2.15 0.25
21 0.03 0.09
22 10.12.1971 12: 34 p.m. 2100 3.0 40 100 320 4.75 1.35 0.42
12 :30 p.m.
23 4.75 0.7
24 0.24 0.2
25 0.14 0.05
26 10.15.1971 1300 25.0 40 120 340 0.04 0.04 - The train of the
10: 25 a.m. volcano floats
t o the shore
w
+ Volcano Malyi Semlyachek, crater lake
-3
27 10.15.1971 1600 0 40 100 320 0.1 0.05 -
10 : 35 a.m.
Volcano Karyrnski
10.20.1971
28 11: 43 a.m. 11: 47 a.m. 1400 2 20 120 360 - 0.32 0.01
29 11:47a.m. 11:51a.m. 1600 1.5 20 80 180 - 0.05 0.03
30 11: 47 a.m. 11:52 a.m. 1600 1.5 20 80 360 1.16 1.00 0.11
31 11:56a.m. 11:Ola.m. 1500 1.0 20 100 180 0.48 1.60 0.21
32 11: 56 a.m. 11: 03 a.m. 1500 1.5 20 100 360 3.0 1.80 0.21
33 12 :08 p.m. 12 : 11 p.m. 1700 1.5 20 100 260 0.9 0.26 0.06
34 12 :09 p.m. 12: 13 p.m. 1700 1.0 20 100 180 0.08 0.42 0.06
(continued)
TABLE
11-Continued
Distance
Content of
to volcano Heli-
Serial
Date and Time of
Flight
alti- from Wind copter
+
SOz HS2 Content Content
explo- measure- speed Course by of of Remarks
NO. tude measuring speed
sions ment point measuring SOz H2S
device
(m) (km) (WW (h/W
35 12:Ogp.m. 12 : 14 p.m. 1700 1.5 20 100 360 6.2 2.0

36 12:24p.m. 12 :26 p.m. 1700 1.5 20 100 190 5.4 0.6
37 12:24p.m. 12 : 27 p.m. 1800 1.5 20 100 260 3.25 0.38
38 12 :35p.m. 12 : 42 p.m. 2300 2.5-3 20 100 360 0.18 0.06
39 12:35p.m. 12 :43 p.m. 2300 3.5 20 100 180 0.56 0.11
40 12:35p.m. 12 :44 p.m. 2300 4.0 20 100 360 0.20 0.07
41 12:45pm.. 12 :48 p.m. 2000 2.5 20 100 260 2.80 0.3
42 12:45p.m. 12 :49 p.m. 2100 3 20 100 300 4.1 1.2
43 12:51 p.m. 12 : 53 p.m. 1700 2 20 100 300 3.9 2.0
44 12:56p.m. 13 :00 p.m. 2000 1.5 20 400 330 4.3 2.0
45 12:56p.m. 13 : 01 p.m. 2200 3 30 400 360 2.5 2.0
46 12:56p.m. 13: 03 p.m. 2500 5 30 100 200 4.5 1.4
10.22.197 1
47 13: 05 p.m. 2600 6 30 100 330 - 5.3 0.8
48 11: 55 a.m. 11:59 a.m. 2100 1 40 100 360 - 0.04 -
49 11: 55 a.m. 11: 59 a.m. 2100 1 30 100 360 - 0.71 0.13
50 11:55a.m. 12:OO 2000 2 30 100 180 0.45 0.92 0.22
51 11:55a.m. 12:Ol p.m. 2000 2 30 100 360 0.5 0.16 0.07
52 12 :06 p.m. 12 : 09 p.m. 2100 1 30 100 180 - 3.39 0.5
53 12 : 06 p.m. 12 : 10 p.m. 2100 1.5 30 100 360 - 3.01 0.2
54 12 :06 p.m. 12:11 p.m. 2100 2 30 100 180 - 0.28 0.28
55 12 : 15 p.m. 12 : 18 p.m. 2100 1 30 100 180 - 2.0 0.40

56 12 : 15 p.m. 12 : 19 p.m. 2100 1.5 30 100 360 - 0.58 0.18
57 12 :15 p.m. 12 :20 p.m. 2100 1.5 30 100 180 - 1.39 0.07
58 12 : 15 p.m. 12 :25 p.m. 2100 8 30 100 360 - 0.07 0.07 Train remote
from volcano,
human error
59 12 : 28 p.m. 12 : 31 p.m. 2000 1.5 30 100 180 - 0.06 0.05
60 12 : 28 p.m. 12 : 31 p.m. 2000 1.5 30 100 180 - 1.08 0.26
61 12 : 28 p m . 12 : 33 p.m. 2400 3 30 100 360 - 3.7 0.76
62 12 : 39 p.m. 12 :41 p.m. 1800 1.5 30 100 180 - 2.1 0.26
-
w
CD
63
10.29.1971
12 : 15 p.m. 12 : 20 p.m. 1700 0.5 8-10 30 270 0.75 0.5 ?

64 12 :25 p.m. 12 :29 p.m. 1700 0.5 8-10 80 90 1.0 - -
65 12 : 36 p.m. 12 :40 p.m. 1800 1.5 8-10 100 270 1.0 - -
66 12: 41 p.m. 12 :43 p.m. 1800 1.0 8-10 100 90 7.5 3.7 2
67 12 :45 p m . 12 : 50 p.m. 1900 1.0 8-10 100 270 0.5
68 12 :05 p.m. 12 : 54 p.m. 1900 2.0 %I0 100 90 0.1? 0.03 0.04
69 12 : 51 p.m. 12 :55 p.m. 1900 0.5 8-10 100 270 O.l? 0.27 0.15
70 12: 51 p.m. 12 :57 p.m. 1900 1.0 8-10 100 90 0.1 0.49 0.02
71 13 : 05 p.m. 13 :08 p.m. 1800 1.0 8-10 100 270 1 0.05 0.03
72 13 :09 p.m. 13 : 10 p.m. I800 2.0 8-10 100 90 0.35 0.25 0.1
73 13 : 09 p.m. 13 : 12 p.m. 1800 2.0 15 100 270 0.6 0.48 0.11
74 13 : 12 p.m. 13 : 13 p.m. 1800 1.0 15 100 90 7.5 7.23 0.09
maximum value amounted to 0.2 mg/m3. However, it should be noted that

the long period between adjacent outbursts did not permit the acquisition of
a su€Ecient number of statistical data.
The method of quantitative evaluation of the volcanic gas outburst effluent
(Shilin and Beryland, 1971) reveals, in conformance with the data of 1971,
that during the outburst the volcanic effluent amounts on the average t o
2 kglsec of sulfurous anhydride and 300 g/sec of hydrogen sulfide.
Thus, the following conclusions may be drawn:
1. I n the activity of the volcano during the period 1970 to 1971 a con-
siderable intensification not only in the increase of gas outburst frequency,
but also in a sharp increase of gas content, is observed.
2. Both in 1970 and 1971, two types of explosion with low and high gas
content can be distinguished, the latter being connected with a great quantity
of ash in the outburst. The presence of short-period changes in the composition
of volcanic gas outbursts are due to the specific processes in its depth focus.
3. In the composition of volcanic gases an appreciable prevalence of
sulfurous anhydride over the hydrogen sulfide is observed.
It is well known that the study of the volcanic gas composition enables one
to draw valuable conclusions on the composition of the mantle smelts,
specifics of the magma differentiation, etc., which will contribute to the
solution of the most important problems in geology and volcanology.
I n the case of fumarole fields of the caldera Uson type or volcanoes in a
static condition the study of the escaping gas can be performed by analysis
of samples collected on the ground. But during the eruption of a volcano
such measurements can only be made by aerogeochemical methods. Many
investigators have stressed the exceptional importance and at the same time
the practical difliculty of direct gas sampling from a volcano during eruption
(Ivanov, 1970; Masayo, 1961).
Investigations of the SO, concentrations are of especial interest because
this gas, as compared to the other components of volcanic outbursts, is
sensitive to variations of temperature and pressure. It can be surmised
(Masayo, 1961) that the character of SO, concentration change with time
may become one of the indices of growing volcano activity. The aerogeo-
chemical survey presents exceptional possibilities for speedy periodical
measurements over active volcanoes. It should be added to the method of
surveillance.
Further development of the aerogeochemical survey should be aimed at
improving the method, the accumulation of data in single-component
measurements (of SO, in the first place), and the measurements of some gas
concentrations, as well as the development of the theory of the generation of
evaporation haloes in the atmosphere and design of optical gas analyzers. The
objective experimental studies, aside from active volcanism, should also be
concerned with the discovery of mineral deposits.
5. CONCLUSIONS
The material presented above demonstrates the following facts 1
1. New means of remote sensing of the environment, e.g. radar, infrared,
and geochemical methods, used in surveys from aircraft allow one to obtain
information which greatly improves the accuracy of our concepts of various
objects and phenomena under study. I n some cases radically new ideas are
obtained. The experience in mounting the equipment which operates in the
visible range of the spectrum, as well as the infrared receivers, on
orbital platforms permits the conclusion that information can be obtaingd
from space carriers and by means of other receivers. The method of studyine
the objects in this case is different,, as on the one hand information can be
obtained periodically while on the other hand, the studies encompass simul-
taneously enormous areas with a considerable generalization of information
due to high altitudes and small scale. I n this case, the general approach in
processing the information and its interpretation is changed. The generaliza-
tion of information on the objects under study eliminates small, particular
details and encourages the establishment of general principles in the structure
of large regions. These distinctive features of information, obtained from
satellites, require the development of new methods for interpretation with
due regard for natural connections existing between the objects on a global
scale.
2. The most comprehensive study of an object or the earth’s surface is
obtained when several types of remote sensing receivers are used in combi-
nation. An example of this process is the study of the contemporaneous
volcanism (Kamchatka) by means of radar (regional geological structure),
infrared (the structure of geothermal zones), and geochemical (gas haloes of
volcanoes and geothermal zones) surveys.
It should be noted that in all investigations using new types of remote
sensing devices they usually supplement the principal and the most
informative kind-the photographic survey.
3. For the future, the application of new types of the environmental
remote sensing studies should concentrate on development of the methodolo-
gies of surveying and of simultaneous studying the corresponding physical
parameters on the ground. This will permit more reliable interpretation of
data obtained by remote sensing for solving the specific problems dealing
with natural resources by use of aerial and satellite carriers, and will provide
a basis for developing new methods of remote sensing.
ACKNO
WLEDQMENTS
The authors wish to express their gratitude to N. A. Gussev, V. S. Loshtshilov, N. N.

Semyonova, and V. A. Starostin for the courtesy of placing valuable data at their
disposal.
322 B. V. SHILIN LUVD V. B. KOMAROV
REFERENCES
Alperin, V. Z. et al. (1968). Automatic gas analyzer for continuous determination of
sulphurous gas in atmospheric air. Tr. Gl. Geofiz. Observ. No. 234.
Barringer, A. R. (1964). Proc. Symp. Remote Sensing Environ., 3rd, 1964, pp. 279-292.
Barringer, A. R., and Schork, J. R. (1966). Proc. Symp. Remote Sensing Environ., 4th,
1966, pp. 779-792.
Barringer, A. R., Newbury, B. C. and Moffat, A. J., (1968). Proc. Symp. Remote Sensing
Environ., 5th, 1968, ?p. 123-156.
Ginzburg, N. I. (1957). “An Experiment in Developing the Theoretical Principles of
Geophysical Search Methods.” Gosgeoltehizdat, Moscow.
Glushkov, V. M. et al. (1970). A new means for obtaining glaceological information.
Morsk. Plot. 99, No. 9.
Glushkov, V. M., Komarov, V. B., and Loshtshilov, V. S. (1972). Rep. Congr. I S P , 12th,-
1972, pp. 1-7.
Gussev, N. A., Karizhenski, E.Ya., and Shilin, B. V. (1972). Infrared aerial survey in
studying the regions of active geothermal activities. Sow. Geol. 1972, No. 1.
Ivanov, B. V. (1970). “Eruption of Volcano Karymski in 1962-65 and the Volcanoes of
Karymski Group.” Nauka, Moscow.
Komarov, V. B. (1971). Proc. Symp. RemoteSensing Environ., 7th, 1971, pp. 2275-2280.
Komarov, V. B., Starostin, V. A., and Nyavro, B. P. (1973). Radar aerial survey and its
significance in the complex of aerial and space methods of geological investigations.
Investigation of the natural environment by space means. Geologe and geomor-
phology. Izv. Akad. Nauk SSSR, Moscow pp. 111-119.
Manganell, R. M. (1970). Air Pollzct. Contr. 20, No. 7.
Masayo, M. (1961). Bull. Geol. Surv. Jap. 12, No. 8.
Palmason, J., and others. (1970). U S . Symp. Develop. Util. Geothem. Resour., I970
Vol. 2, Part 1.
Rouse, G. C., and Stevens, D. (1970). Mining Eng. (New Ywh)22, No. 12.
Shilin, B. V., and Berlyand M. E. (1971). Proc. Symp. Remote Sensing Environ., 7th,
1971, pp. 1733-1738.
Shilin, B. V., and others. (1969). PTOC.Symp. RemoteSensing Environ., 6th, 1969 pp. 175-
188.
Shilin, B. V., and others (1971). Proc. Symp. RemoteSensing Environ., 7th, 1971, pp. 133-
146.
Svyatlovski, A. E. (1959) “Atlas of Volcanoes of the U.S.S.R.” Izv. Akad. Nauk, SSSR,
Moscow.
Ustinova, T. I . (1955). “The Geysers of Kamchatka.” Geografgiz, Moscow.
Vinogradov, V. N. (1964). On t h e regime of Kamchatka geysers. Vop. Geogra. Kumchatki
No. 1.
SUBJECT INDEX
A linear, discrete filters for, 192-195

construction, 195-203
Aerial geological-geographic surveys, 263- Coriolis parameter, 119
322 Cumulonimbi
by radar, 265-282 distribution of, i n hurricane intensifica-
using infrared, 282-304 tion, 81-82
Aerogeochemical surveys, 304-320 in tropical meteorology, 32-26
equipment and procedure for, 305-308 Cyclostrophic balance, in hurricane
method for, 304 dynamics, 39
Africa sandstorms in, as tropical cyclone Cyclones (tropical), 1-100 (see also
sources, 7 Hurricanes)
Atmosphere, vacillation in, 178-182 definition of, 3
Atmospheric flow, laboratory simulation energetics of, 50-55
of, 102-113 generation of, 5-8
importance in global circulation, 13-17
B intensification of
time-dependent flowfield in, 82-86
Baguio, definition of, 3 theory, 75-86
Bowen ratio, 17 intensity of, relation to sea tempera-
Buoys, use in hurricane detection, 5 ture, 55-59
models of, 29-75
C Carrier, 29-50
moist adiabatic process theory and,
Carrier model of hurricane intensification, 89-91
76-77 path prediction for, 12-13
critique of, 78-79 speed estimate of, 35-45
Carrier model of tropical cyclone, 29-50 swirl-divergence relation and, 45-50
critique of, 33-35
maximum swirl speed estimate, 35-45
Christoffel symbol, 118
CISK, description and role in hurricane D
formation, 26-28
Computers, hurricane simulation on, Dirichlet’s problem, 205
66-70 Dust devils, definition of, 3
Convolution filtering
dekition of, 189-191
frequency filtering comparison to, 2 4 P
247 E
frequency response in, 191-195
in gravity studies, 189-203 Eye, of hurricane, 8
323
324 SUBJECT INDEX
F Heating effects, in vacillation, 129-131

Hurricane warnings, path predictions
Filtering techniques in, 3
convolution method, 189-203 Hurricane(s )
derivatives calculation of higher order, computer simulation of, 6 6 7 0
248-250 dangers and benefits of, 2
digitization of continuous field in, 237- definition of, 3
240 facts about, 3-5
downward continuation operation, kinetic energy and water content of,
2 13-21 9 87-88
construction, 219-226 properties of, 8-12
frequency type, 226-247 seeding of, implications of, 70-75
in gravity interpretation, 187-162 temperature measurements in, 63-66
power-spectrum estimation, 240-244 Hydrostatic approximation, 118
in surface gravity studies, 203-226
upward continuation type, 210-213
Frequency filtering
in gravity studies, 226-247 I
convolution filtering compared to,
244-247 Ice, TOROS exploration of, 265-272
optimal Wiener type, 228-233 Infrared aerial surveys, 282-304
procedure, 226-227 of geothermal areas, 286-304
Friction in vacillation, 129 of volcanoes, 282-286
Intertropical Convergence Zone (ITCZ),
hurriaane generation and, 4 6
G
Geological-geographical surveys (aerial),

263-322 K
Geophysical flows, simulated, 102-105
Geothermal activity, infrared aerial Karymski volcano
surveys of,286-304 aerial radar survey of, 274-277
Geyser Valley, infrared aerial surveys of, aerogeochemical surveys of, 313-321
295, 298-304 infrared aerial survey of, 282-286
Gravity, filter techniques in interpreta- Kamchatka volcanoes, aerogeochemical
tion of, 187-262 surveys of,313-321
surface type
definition, 203-205
upward and downward continuation
of, 203-226 M
Meteorology (tropical), 17-28

H CISK, 26-28
cumulonimbi, 22-26
Hadley cell mechanism, hurricanes and, stability, 17-22
15-16 tensor calculus in, 117-1 18
Hadley flow, Rossby flow compared to, vacillation importance in, 112-1 13
106-108 Moist adiabatic process, theory of, 88-91
Heat transport, northward sensible, 140- Motion field, dynamical properties of,
141 138-1 4 1
SUBJECT INDEX 325
P Tropical disturbance, definition of, 3

Tropical storm, definition of, 3
Pacific, tropical storms in, Atlantic Tropics meteorology of, 17-18
source of, 7 Trovado, definition of, 3
Papagallos, definition of, 3 Typhoon, definition of, 3
R U
Radar aerial surveys, 265-282 Uson caldera

of closed regions, 279-282 aerogeochemical studies on, 308-313
of ice, 265-272 infrared aerial surveys of, 286-292,
of rock composition, 272-277 296297
of structural-tectonic texture, 277-279
Rainfall, hurricane effects on, 2-3
RiehI-Malkus extra oceanic heat source
postulate, 26, 43-44, 53-55 v
critique of, 59-63
Rock, radar aerial survey of composition Vacillation, 101-186
Of, 272-277 atmospheric-flow simulation, 102-1 13
Rossby flow, Hadley flow compared to, characteristics of, 108-113
106-108 cycles of, 175-178
double symmetric, 178
equations of quasi-geostrophic flow in,
120-123
S friction and heating effects in, 129-131
Hadley and Rossby flows compared,
Sandstorms, hurricane genesis and, 7 106-108
Schuster periodogram, 242 kinetic energy type, 162-175
Seeding of hurricanes, implications of, in atmosphere, 180-182
70-75 dynamics, 162-169
Spectral dynamics, of vacillation, 131- energetics, 169-174
138 physical interpretation, 174-175
Stability, in tropical meteorology, 17-22 in laboratory flows, 108-111
Storm surges, from hurricanes, 2 meteorological interest in, 112-1 13
Strakhov’s method, for extraction of model atmosphere for study of, 113-
potential field signal, 233-237 131
numerical study of, 101-186
physical processes of model in, 128-139
potential energy type, 113, 151-162
T in atmosphere, 178-179
dynamics, 150-157
Temperature, of hurricanes, measure- energetics, 157-1 6 1
ment of, 63-65 physical interpretation, 162
Tensor calculus, general characteristics spectral dynamics of, 131-138
of, 114-117 baroclinic transfers, 145-148
Tornadoes, from hurricanes, 10 barotropic transfers, 144-145
TOROS, in ice exploration, 265-272 global forms, 141-144
Tropical depression, definition of, 3 individual forms, 144
326 SUBJECT INDEX
tensor formulation of quasi-hydrostatic w

flows in,1147120
theoretical aspects of, 113-1 14
two-layer model of Waterspouts, from hurricanes, 10
energetics, 125-128 Waves, from hurricanes, 2
equations, 123-125 Wiener filter theory, 232-233
unsymmetric, 175-176 Wiener-Hopf integral equation, 229
Volcanoes, infrared aerial surveys of, Willy-nilly, definition of, 3
282-286 Winds, from hurricanes, 2
A 4
9 5
C 6
D 7
~a
F 9
G O
H I
1 2
J 3

Advances in Geophisycs, H. E. Landsberg, 1974

Uploaded by

Copyright:

Available Formats

Advances in Geophisycs, H. E. Landsberg, 1974

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advances in Geophisycs, H. E. Landsberg, 1974

Uploaded by

Copyright:

Available Formats

ADVANCES IN

Editorial Advisory Committee

Academic Press New York San Francisco London

ACADEMIC PRESS, INC.

United Kingdom Edition published by

PRINTED IN THE UNITED STATES OF AMERICA

LIST OF CONTRDUTORS......................................... vii

A Numerical Study of Vacillation

Filter Techniques in Gravity Interpretation

Aerial Methods in Geological-Geographical Explorations

1.1. The Dangers and BeneJits of Hurricanes

1.2. Some Observational Facts on Hurricanes

manage to become a tropical storm has a good chance t o become a hurricane

1.3. Tropical Cyclone Generation

1.4. Properties of Mature Hurricanes

FIG.3. NOAA photograph of mature hurricane revealing lower-level rainbands

with respect t o rainfall, rainbands, and winds (Hawkins, 1971).As previously

2According to data collected on 30 typhoons for 1960-1970, tropical storms ap-

Grossman and Rodenhuis emphasize the uncertainty in numerical modeling

1.5. Path Prediction

1.6. The Importance of Tropical Cyclones in the Global Circulation

DIRECT FERREL CELL DIRECT HADLN

and turbulent mixing must currently be parameterized into computer simu-

where the numerically unassigned quantity in the denominator is the ambient

2.1. Stability in the Tropical Atmosphere

where q is the magnitude of the velocity vector (the wind speed).

EQUIVALENT POTENTIAL TEMPERATURE, ee

rather than H , . At sea level, the static temperature contribution T is about

At 50,000 ft the gravitational contribution (gzlc,) is about 150"K, or over

Fro. 5. (lop) The equivalent potential temperature 8. as a function of height z and

approaches the moist adiabat is there penetrative convection. For example,

2.2. Tropical Curnulonimbi

the rain areas are covered by active cumulonimbi. Agglomerations of cumulo-

The CISK process is the feeding of convective activity in a swirling flow

the release of the appreciable heat of condensation during moist adiabatic

the cyclone scale, in a cooperative interdependence. Such ideas suggest that

3.2. The Carrier Model

eyewall) for swirl distributions of practical interest, three-quarters of the

Carrier’s model thus pictures the tropical cyclone as a once-throughprocess

that convection dominates the slow ambient-sustaining processes, so for a t

3.3. Critique of the Carrier Model

3.4. Maximum Swirl Speed Estimate According to the Carrier Model

(1W P(T) = 6.1078 exp[a(T - 273.16)/(T - b)]

The [d(q2/2)/dp] contribution is negligible and is henceforth dropped. A

However, if one takes p e pa [so x = 1 in (12)] and p e pa, which incurs a n

pressure at which R H = 1, a t which point one switches to the moist adiabat

p ( z = 0) = p c ( < p e < p s ) and p(z = 0) = p,( < pe <ps)-the density discre-

(15) (Vlrnax sPA/P~I"~

and from the cyclostrophic balance

vations of a September 1947 hurricaneoff Tampa, Florida. PalmenandNewton

3.5. The Swirl-Divergence Relation for the Frictional Boundary Layer

Because the maximum speed achieved in a tropical cyclone is rarely much

(28) $41 +wdr + w2 - - $"/2x - (* - Y) - +n = 0

Matching of expansions gives

and a t 5 = 0 no-slip conditions are adopted:

(33) t,h = [Y(z)][l- c o ~ ( 2 - l / ~ ~ ) e x p ( - 2 - ~ ~ ~ ~ )

(35) w(r, 5 -+ 00) = W ( r )= 2-1’2Y,(x)

Numerical integration by finite-difference methods is formidable because

The reason weighted-residual calculations fail near the axis, as discovered