JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 99, NO.

B10, PAGES 19,617-19,634, OCTOBER 10, 1994

On the derivation of coseismicdisplacementfields using

differential radar interferometry: The Landers earthquake
Howard A. Zebker, Paul A. Rosen, Richard M. Goldstein,
Andrew Gabriel, and Charles L. Werner
Jet Propulsion Laboratory, California Institute of Technology, Pasadena

Abstract. We present a map of the coseismicdisplacementfield resultingfrom the

Landers, California, June 28, 1992, earthquake derived using data acquired from an
orbiting high-resolutionradar system. We achieve results more accurate than previous
spacestudiesand similar in accuracyto those obtainedby conventionalfield survey
techniques.Data from the ERS 1 synthetic aperture radar instrument acquired in April,
July, and August 1992 are used to generate a high-resolution,wide area map of the
displacements.The data representthe motion in the direction of the radar line of sight
to centimeter level precision of each 30-m resolution element in a 113 km by 90 km
image. Our coseismicdisplacementcontour map gives a lobed pattern consistentwith
theoretical models of the displacementfield from the earthquake. Fine structure
observedas displacementtiling in regions several kilometers from the fault appears to
be the result of local surfacefracturing. Comparisonof these data with Global
PositioningSystem and electronic distancemeasurementsurvey data yield a correlation
of 0.96; thus the radar measurementsare a means to extend the point measurements
acquiredby traditional techniquesto an area map format. The techniquewe use is (1)
more automatic, (2) more precise, and (3) better validated than previous similar
applicationsof differential radar interferometry. Since we require only remotely sensed
satellite data with no additional requirementsfor ancillary information, the technique is
well suited for global seismic monitoring and analysis.

Introduction derived from the European Space Agency (ESA) ERS 1

satellite data for analysis of the magnitude 7.3 earthquake
Interferometric radar techniques for the generation of centered near Landers, California, on June 28, 1992. In this
highly accurate digital elevation models (DEMs) by now study a single interferogram which contained phase signals
have been well documented in the literature [Zebker and from the local topographyand from the earthquakedisplace-
Goldstein, 1986; Goldstein et al., 1988; Prati et al., 1990; ments was subtractedfrom a manipulated U.S. Geological
Zebker et al., 1992; Evans et al., 1992; Madsen et al., 1993, Survey (USGS) 15 arc min DEM of the area. The residual
also personal communication, 1993; H. A. Zebker et al., phaseswere interpreted as ground displacementsfrom the
personal communication, 1993]. A related application of event. The interferogram, when corrected for topographic
such techniquesallows the measurementof the motion of all effects, shows a displaced dual-lobed pattern of fringes
resolved points in a remotely sensedimage [Goldstein and emanatingfrom the fault zone, where each fringe represents
Zebker, 1987; Goldstein et al., 1989]. These similar tech- about 2.8 cm of motion in the radar line of sight direction.
niques both follow from analysis and interpretation of inter- They alsoderive a theoreticalfringe pattern from a model of
ferograms, which consist of the phase differencesbetween the earthquake motion which matches the observations
two radar images of the same scene acquired at separate fairly closely.
locations or times: a sensorlocation change gives sensitivity Despiteits success,there are severalimportantlimitations
to topography and a sensor temporal change gives motion in the techniqueusedfor the above study. Although a USGS
sensitivity. A combination of the two approaches, denoted 90-m spacingDEM was availablefor this site, for many sites
differential radar interferometry since the phase measure- in the world, no DEM exists. In addition, an existing DEM
ments of interest result from the difference of two interfer-
may not be sufficientlyaccurate to yield the desired preci-
ograms,has previously been used by Gabriel et al. [1989] to
sion. DEMs typically contain errors and distortions of the
map the changes in surface elevation of agricultural fields
order of the phenomenabeing investigated.The CNES team
over a large area to centimeter-level sensitivities.
estimatesa precisionin their measurementsof about 2.8 cm
More recently, there has been activity by at least two
in the radar line of sightmotion, limited mainly by impreci-
groups applying the capabilities of radar interferometry to
sion in the USGS DEM plus radar system noise. Also, the
the study of seismic phenomena. Massonet et al. [1993] of
DEM must be precisely coregistered to the radar image,
Centre National d'Etudes Spatiales (CNES) in Toulouse,
which itself may be a difficult task. (S. N. Madsen et al.
France, used an interferometric digital elevation model
(personal communication, 1993) provide more on errors
Copyright 1994 by the American Geophysical Union. induced by DEM misregistration.)Finally, since the inter-
Paper number 94JB01179. ferogramphasesare all measuredmodulo 2,r, the absolute,
0148-0227/94/94JB-01179505.00 or even relative, phaserelationshipbetween arbitrary points

19,617
19,618 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR

in the scene is difficult to determine. Thus it is virtually examine the case where no ground movement between radar
impossible to fit continuous two-dimensional models of the observations occurs. Consider two radar systems observing
displacement field to the observations. the same ground swath from two positions A1 and A2,
These limitations aside, it is important to realize that the respectively, as illustrated in Figure 1. The measured phase
phase displacements due to motion in an interferometric at each point in each of the two radar images may be taken
DEM can be hundreds of times more sensitive than simply as equal to the sum of a propagation part proportional to the
differencing the actual height measurementsbefore and after round-trip distance traveled and a scattering part due to the
an event (see below). More complete use of phase informa- interaction of the wave with the ground. If each resolution
tion allows the interferometric approach to map centimeter element on the ground behaves the same for each observa-
scale distortions over a region many tens of kilometers in tion (see more on this important condition below), then
size at a resolution of a few meters. calculatingthe difference in the phasesremoves dependence
In this paper, we approach the Landers analysis differ- on the scattering mechanism and gives a quantity dependent
ently from Massonet et al. by utilizing only data acquired by only on geometry. If the two path lengths are taken to be p
the ERS 1 satellite. Our approach overcomes the aforemen- and p + 8p, the measured phase difference 4' will be
tioned limitations and hence is more readily quantifiable
given the radar system parameters, and the quality of the
result can be measured "up front." Specifically, imprecision
introduced by the USGS DEM in the CNES study is not
present, coregistration occurs automatically in forming the or 2,r times the round-trip distance difference in wave-
interferograms, and the entire usable phase field is "un- lengths. The law of cosinespermits solutionfor 8p in terms
wrapped," meaning that the displacement at each point is of the imaging geometry as follows. Then
known digitally in an absolute sense. Unwrapping renders
(p + 8p)2= p2 + B2 - 2pB sin(0 - a) (2)
the displacement field more amenable to computer modeling
and analysis and permits the precision of the technique to be where the baseline length is B, the range to a point on the
increased from the 2.8-cm radar line of sight reported by groundis p, the look angleis 0, and the angleof the baseline
Massonet et al. to about 0.2 cm obtained here. Further, we with respect to horizontal at the sensor is a. Neglecting the
verify the accuracy of the measurementsby comparing to a termof order(Sp)2 yields
displacement field derived from conventional surveying
techniques. These survey data were derived from a combi- B2
8p • B sin (0- a) +--. (3)
nation of electronic distance measurement (EDM) lines and 2p
Global Positioning System (GPS) satellite receivers. The
methods and results presented here can serve as a baseline For simplicity in describing the approach we used, we can
for the design of a seismic monitoring program. make a second approximation, although it is not necessary
The structure of the paper is as follows. We begin by for the analysespresented below. In the case of spaceborne
summarizing the theory of radar interferometry and differ- geometries we can ignore the second term on the fight-hand
ential interferometry, error sources, and expectations of side of (3) and obtain
performance for seismic studies. Next, we present the set of
differential radar interferometric observations of the
8p • B sin (0 - a) (4)
Landers earthquake and discusstheir accuracy. Finally, we or

compare our results with those of the earlier study and with
the in situ measurements made by GPS techniques.

Summary of Theory
In this section we derive the equations needed for calcu-
lating ground displacement fields from interferometric syn-
thetic aperture radar measurements. Here we assume that
the reader has a general knowledge of radar remote sensing
systems. The interested reader may consult a general text on
radar remote sensing such as that by Elachi [1988] or by
Curlander and McDonough [1991] for questions on radar
system operation and processing. As for information on the
technique of radar interferometry, much of the work is still
too new for general textbooks, and thus the technical liter-
ature is the only source available. We cite the major relevant
papers in this text, and the reader may consult these when
appropriate.
A side-looking spaceborne synthetic aperture radar sys-
tem may map a continuous swath many tens of kilometers in
Figure 1. Radar imaging geometry. The solid lines show
width as the satellite progressesalong its orbit track, yielding that radar signalpaths for the first interferogram pair formed
measurements of the amplitude and phase of radar echoes by antennasat A1 and A2. Dashed lines show signalpath for
associated with independent patches on the ground perhaps second interferogram acquired over the same site but with
10 m in size: this size is the resolution of the radar. We first antennas located at A1 and A2'.
 ZEBKER ET AL.' COSEISMIC DISPLACEMENT FROM RADAR 19,619

3p • Bii. (5) is a function of the angle 0, which depends both on the

illumination geometry and also the topography at each point
B II - B sin(0 - a) is simplythecomponent of thebaseline in the radar image. To evaluate (9) via (10) directly, we must
parallel to the look direction. This is the parallel-ray approx- solve for the topographic map of the area of interest from the
imation used by Zebker and Goldstein [1986] in their initial interferometric data or obtain the elevation data from an-
paper on topographic mapping. other source. In the interests of simplicity and accuracy, we
Equations (1) and (4) show that the measured phase of an have devised an indirect approach for which it is not
interferometer is the component of the interferometer base- necessaryto implement the step of either topographic solu-
line parallel to the look direction to a given point on the tion or registration of dissimilar data sets.
surface measured in wavelengths, multiplied by two for We remove from the interferogram phase a term that
round-trip travel. We note that the height sensitivity of the would exist even in the absence of topography on a spherical
instrument enters through the dependence of the exact look Earth. The phase corrected for the "curved Earth" effect,
angle 0 on the altitude z - h - p cos 0, where h is the height denoted &nat,is given by
of the sensor above the reference surface.
471-
If a second (denoted by a prime) interferogram is acquired
(11)
over the same area, sharing one orbit with the previous pair 0flat
= T [Bsin(0- ,) - B sin(00- -)],
so that p and 0 are unchanged (dashed lines in Figure 1), we
can compare the interferogram phases with each other. Thiswhere 00 is the look angle to each point in the image
second interferogram is acquired with a different baseline B'
assuming zero local height. The interferogram phase after
andbaseline orientation
a', thusa differentBil. Combining this correction represents the distortion of the interference
(1) and (4) above, we obtain grating pattern due to topographic variation relative to a
spherical surface and displacements due to motion in the
4rr
scene. The interferograms shown throughout this paper have
' = • Bil. (6) been "flattened" according to (11).
Noting that the deviation of the exact 0 from 00 is small,
Examination of the ratio of the two phases yields we can expand the first term on the right-hand side of (11),
leading to
0/0' = Bll/Bil. (7)
471-
In other words, the ratio of the phasesis equal to the ratio of Offat= • 80B cos (0o- a), (12)
the parallel components of the baseline, independent of the
topography.
Now consider the situation of two interferograms acquired where 30 = 0 - 00. Numerically, t•flat is equal to the product
over the same region as before but in this case an earthquake of the perpendicular component of the baseline B ñ, assum-
has displaced each resolution element between observations ing no topography is present on the surface, and the topo-
for the primed interferogram. The displacements are as- graphicangulardistortion 30. Thus the ratio 0nat/0hatis now
sumed small with respect to a resolution cell so that the radar in terms of 00 rather than 0 and dependsonly on the viewing
echoes remain correlated. Here in addition to the phase geometry and the baseline. If we now restate the differential
dependence on topography there is a phase change due to phaseequation (9) above in terms of the flattened phase &nat,
we obtain
the radar line of sight component of the displacementAp. In
this interferogram the phase 0' will be given by

t••at--• d)flat
=TAp. (13)
0' = --•-(eil+Ap). (8)
With this function, we can now solve directly for the
The displacement term Ap adds to the topographic phase displacement Ap without requiring the exact values of 0, and
term, creating confusion in the interpretation of the result. hence the topographic information, at an intermediate step.
However, if the data from the initial unprimed interferogram We have used this procedure (equation (13)) for the reduc-
are scaled by the ratio of the parallel components of the tion of the data presented in this paper.
baseline and subtracted from the primed interferogram, we We note that if the baseline used in the flattening operation
can obtain a solution dependent only on the displacement of (equation (11)) is not exactly the true baseline value, (12) will
the surface, as follows contain error terms and the subsequent displacement maps
will be distorted. This condition is described in detail in the
471-
appendix; for the rest of this paper we will assume that the
Bil
O=-•'-
ill AP. (9) correct baseline values are used.
We have shown that the phase in radar interferograms
Since the quantity on the left is determined entirely by the depends both on the local topography and on any motion
phases of the interferograms and the orbit geometries, the that may occur between viewing instances. We may com-
line of sight component of the displacementAp, is measur- pare the sensitivity of the phase measurement to the phe-
able for each point in the scene. nomena of topography and displacement, which may be
The ratio derived by differentiating (8) with respect to height through
B IIanddisplacement.
In thefirstcase,usingdz - p sin0 dO,
Bil B' sin (0 - ,') obtained from the dependence of height on angle described
(10)
Bii B sin (0- a) above, we find
19,620 ZEBKER ET AL ß COSEISMIC DISPLACEMENT FROM RADAR

47/' this we conclude that the radar properties of the desert

dab'= • B cos (0 - a) dO (14) surface change little over months if weather and other
environmental factors are not altering the condition of the
and ground significantly.
The second limitation, more important for this study, is
dab' 4•r B cos (0 - a) that the phases must be "unwrapped" before data from one
• - . interferogram may be used to correct the second interfero-
dz A p sin 0
gram (equation (9)) to estimate the displacementphases. The
For the displacement case we have measurementsof each phase are known only modulo 2•r, and
various techniquesexist [Goldstein et al., 1988; Ghiglia and
dab' 4 •r Romero, 1993; A. Hiramatsu, personal communication,
• - . (16)
dAp 1992] to determine the absolute phase relationship between
all arbitrary points in a data set (that is, unwrapping). While
Since the distance p typically is very much greater than the not fully characterized in any of the existing literature, it is
baseline distance B, it is evident from (15) and (16) that a apparent that the ability to unwrap arbitrary phase fields
much more sensitive dependence of phase results from dependson two factors: the noise level in the system and the
displacements than from topographic variation. In other interferometric fringe spacing. For the July-August pair
words, the system is more sensitive in an absolute sense to described here in particular, the interferometric baseline is
surficial change than to the topography itself. Comparing the quite large, being 40% of the critical baseline at which no
two results numerically, for the April-August ERS 1 case correlation between signals is possible. (For a more com-
described here (see next section), 1 m of topography gives a plete discussion on baseline decorrelation, see Zebker and
phase signature of 4.3 ø, while for the same pass pair a 1-m Villasenor [1992]). Since the fringe rate depends on local
surface displacement yields a phase signature of 12800ø, or surface slope, typically it is more difficult to estimate phases
nearly 3000 times greater sensitivity. Thus, while radar reliably in rough terrain than in flat terrain if the fringe rate
interferometry can be used to measure topography to an is high to begin with. The result of this is that we were unable
accuracy of meters, displacementsmay be determined to the to obtain reliable phase estimates in the rougher regions, as
centimeter or millimeter level. will be seen in the data presented below.
This ratio of sensitivities illustrates the power of the
interferometric technique to detect small changes. If, for
example, we chose to map seismic displacementsby differ- ERS 1 Interferograms of the Landers
encing DEMs, whether acquired interferometrically or by Earthquake
conventional stereo photogrammetry, changes would only The ERS 1 radar system, operating at a wavelength of 5.67
be visible if they were significant in size compared to the cm, images Earth from an altitude of about 790 km and
uncertainty of the DEM measurement, which is typically producesradar backscatter maps of 100-km-wide swathsat a
meters. For the interferometric case in the previous para- resolution of about 25 m across track and 6 m along track.
graph, for example, system noise limits the useful signatures We obtained raw ERS 1 radar signal samples acquired over
to those causing a phase shift greater than about 20ø, or 4.6 the Landers region on April 24, July 3, and August 7, 1992.
m. While thus permitting topographic mapping with a verti- We combined these to form two interferograms, one from
cal precision of 4.6 m (H. A. Zebker et al. (personal the April-August pair and one from the July-August pair.
communication, 1993) give a discussion of ERS 1 DEMs The April-August pair spansthe June 28 earthquake and was
with this precision), a worthwhile result for many applica- chosen over the April-July pair which exhibited an excep-
tions, it is not particularly useful for the study of earth- tionally large baseline. No data were acquired on May 29
quakes. In contrast, if data are acquired with an interfero- when the satellite again passed over the site. Orbit recon-
metric pair that spansthe seismicevent, even 1 cm of line of structions provided by the European Space Agency (ESA)
sight displacement results in a signatureof 64ø, easily detect- enabled us to determine the geometrical parameters for the
able in ERS 1 data. pairs chosen as given in Table 1. The parallel baseline
There are, however, two very important limitations to the components given in Table 1 are for a look angle of 21ø.
interferometric technique. First, radar echoes acquired on Since the radar swath is quite wide, the actual look angle
the three passes must correlate with each other; that is, the varies from about 17ø to 23ø and the parallel components
signalsmust be substantially similar over a significantperiod vary somewhat.
of time. Physically, this translates to a requirement that the The Landers area is shown in Figures 2a and 2b, where the
ground scattering surface be relatively undisturbed at the faults shown, illustrated by heavy lines, are those affected by
radar wavelength scale between measurements. Several the Landers earthquake and imaged by the ERS 1 radar.
studies have addressed this phenomenon, both theoretically
[e.g., Li and Goldstein, 1990], and experimentally [e.g.,
Gray et al., 1993]. Zebker and Villasenor [1992] were able to Table 1. ERS 1 Landers Interferometer Baseline
model and quantify the temporal decorrelation process and Parameters
found that different surfaces decorrelate at different rates.
Parallel Perpendicular
This limits the applicability of the approach to areas that do Baseline Orientation Component Component
not change much with time. Some regions, such as desert Pair B, m a, deg Bll, m B•_,m
areas, may exhibit very little decorrelation over long peri-
April-August 146.1 152 110.3 95.8
ods. In the data presented here, correlation was usably high 175 220.5 452.2
July-August 503.1
even after 105 days, the longest time period examined. From
 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR 19,621

35 ø 30'

35 ø

..... p

34 ø 30'

:......
('.• ..:...,
...... ... -•,.
...... .: --.-.:
..... •.•..: • ......
:. .•: • .-.t
....:--•...•
- •,,..
...... . ,..}.- •-:•...•.:.'•
•. ::':.g-
..........

:'2'."'
"-
.....
/.:,•,•.•'..'.'
...•.".
•....'....-.'.
....
.•.:,,1.•,•..•.•,...
'...........
.............
. . ......
,:•.h?'!'-'•.•.•'•...'"'
...........
....:-
. .,...,.,
...'
....... •-•.:S•.•.•
......,,,,,,,,,,•.,,,.::.,,.....-...
,.j,-•...••
' ••••.••'•'••'•:.•"!,•,.•,•
•:?.•
©......
.. .,
ß
- : ' -.i .'• .. ß "-, ...'..' ' ':'.1.
•:,.'".2
2....":•
•::•::.::...•
"- .•..•

•:: •
.............. ,5.:•:.:•?
":-•V:.,:
::'•' '"' • ;• """'"'":•
.......
":•'"'•
•: .•.:,:.,.-..:.'..
,, ß.•.•:.:•::•:....•..,..,,
-,-,: '" •'"'"'• :'.
,,....%.•:•::::.•:•.•?.•
.•..:•:•
:;•:::•.•:..,..:.•:.:.:{•
-• ............
•, ,. • .. ß•e' '":"
' '"ß.......
..•e.•.•..• '•
:•........
• •:;:•.--
-""•'
..... '•" • '•'
.,:.:..:•:.. 1'•.............
•::..:.,...•. •'..... ......••'•"::----•
,•,•...:..•::..•::. :•'- • :••:..•.
.-:-•' '•?"
:::•:.:::..•.:•::
- .
34 ø
-118' -117' -116 ø
LONGITUDE

Figure 2•. Shaded relief map delved from USGS DEM with geographicfeatures shown for reference.
The inset rectangleis the region of the ERS 1 radar swath analyzed here. Also noted are the approximate
position of the Camp Rock-Emerson and Homestead Valley faults. The cities of Barstow, Victorville, and
also Lucerne Valley area are shown for reference.

Figure 2a is a shaded relief representation of the region: sion Laboratory (JPL) using a software processor con-
limits of the ERS 1 data we analyzed are indicated by the structed specifically by us for ERS 1 interferometric appli-
rectangle. Figure 2b is the radar backscatter image with two cations. The data were processed using a range-Doppler
further areas denoted in addition to the faults. It represents algorithm, but the range-compressedsignalswere filtered for
an area roughly 113 km by 90 km. These data, as well as the the July-August pair using the method suggested by F.
radar images below, are in a radar slant range and along- Gatelli et al. (personal communication, 1993) to reduce
track direction coordinate system. Radar slant range, de- baseline decorrelation. We found that this approach yielded
noted by p in Figure 1, here means that the across-track about 5-10% greater correlation in some regions at the
distancesgiven are in terms of line of sight distance of each expense of a slight reduction in range resolution. More
point to the radar rather than that distance projected on the information on radar processingapproachescan be found in
ground. That is, the data have not been geocoded, or placed the general radar textbooks described previously.
in map coordinates.We have preservedthe "natural" spac- The interferograms obtainedin thisprocessare Shownin
ing of the data points in order to maintain the highest Plate 1, with the corresponding correlation coefficient maps
possiblesignalfidelity throughoutthe processingprocedure. shown in Plate 2. The top image in each case represents the
However, we do apply a geocoding transformation before April-August interferogram, while the bottom image shows
comparison with the field survey results described in the data from the July-August pair. The June 28 earthquake
following section. effects are found in the April-August pair. In these plots the
We processedthe radar signal samples at the Jet Propul- fringe signature of a curved Earth surface, as described in
19,622 ZEBKER ET AL.' COSEISMIC DISPLACEMENT FROM RADAR

Figure 2b. Radar image of the Landers area, where the fault locationsare illustratedby the heavy lines.
The radar image covers an area on the ground approximately 113km by 90 km. Insets of an irrigated region
as well as a fault zone are shown for later reference. These data, as well as the radar images in the
remaining figures, are in a radar slant range and along track direction coordinate system.

the section on theory above, has been removed from the extension of the method first presented by Goldstein et al.
interferograms for clearer display. Note that the very high [1988].
fringe rates, and corresponding loss of correlation, in the Finally, the differential interferogram was calculated by
mountainous regions for the July-August pair, lead to our scaling the July-August measurement by the ratio of the
inability to unwrap the phase in these regions. Also note in parallel baseline components for each look angle and sub-
the April-August pair a similar loss of correlation in the fault tracting that value from the corresponding value in the
zone, presumably due to (1) very high fringe rates of greater April-August pair. The result is a map of the displacements
than one cycle per resolution element, (2) large ground shifts of the ground in the radar line of sight direction (equation
resulting in lack of precision alignment of the pixels from (9)), shown in Plate 3, where the shift is coded by color and
pass to pass, and (3) rearrangement of the surface at the the brightnessat each point is the radar image brightness. In
wavelength scale from the earthquake itself. addition, contour lines representing line of sight displace-
These interferograms were filtered using a spatially vari- ments spaced every 5 cm are shown.
able bandpass filter that selected the optimal fringe rate It must be noted that the earthquake is not the only
passband in each 32 by 32 pixel subregion in the interfero- process affecting the phase measurementsin this region of
gram. In this process we also identified areas of low fringe the Mojave. Plate 4 is an enlargement of the April-August
visibility to serve as a mask in the final product, eliminating interferogram plus the correlation coefficientsfor the region
regionswhere we felt we could not trust the phaseestimates. east of Barstow indicated in Figure 2 where center pivot
The data were then unwrapped using the method of A. irrigation has been employed. The irrigated circles, and
Hiramatsu (personal communication, 1992), which is an some other agricultural fields, show a clear loss of correla-
 ZEBKER ET AL ß COSEISMIC DISPLACEMENT FROM RADAR 19,623

PHASE

Plate 1. Interferograms of the Landers area. (top) April-August interferogram; (bottom) July-August
pair. The June 28 earthquake effects are found in the April-August pair. The fringe signature of a curved
Earth surface has been removed from the interferograms for clearer display. Note the very high fringe
rates in the mountainousregions for the July-August pair, leading to our inability to unwrap the phase in
these regions.
19,624 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR

CORRELATION 0 ll.0

Plate 2. Corresponding correlation coefficient maps to interferograms of Plate 1. (top) April-August;

(bottom) July-August. Note in the April-August pair a loss of correlation in the fault zone, presumably
due to (1) very high fringe rates of greater than one cycle per resolution element, (2) large ground shifts
resulting in lack of precision alignmentof the pixels from pass to pass, and (3) stirring up of the surface
at the wavelength scale from the earthquake itself.
 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR 19,625

Plate 3. Differential interferogram of the Landers earthquake region. Radar line of sight displacements
are coded in color, ranging from -70 to 70 cm, while the radar reflectivity of the surface is shown as
brightness. Contours indicating each 5 cm of displacement are drawn in black.

tion presumably due to crop growth and phase shifts which occur in the July-August pair, and therefore the April-
are due to motion, not topography (it is a flat area). Gabriel August discontinuity must be due to a displacement of the
et al. [1989] found similar surface displacements of several surface where one piece moved more than the other. This
centimeters in fields that had been irrigated over a 9-day cracking effect is more pronounced in the region denoted B,
period. The motions observed in this image as well could be shown enlarged again in Plate 6, where the cracking is so
caused by changes in the surface elevations from pumping extensive that it seems the ground has been broken into
underground water or other hydrologic effects. many tiles each several hundred meters across. These data
Since one of the strengthsof this technique is its intrinsic are shown in unwrapped form. The phase unwrapping algo-
high spatial resolution, we also show in Plate $ an enlarge- rithm we use must identify phase discontinuities before
ment in the April-August interferogram of the region around calculating the absolute phase values; the locations of cuts
the fault zone shown in Figure 2b. The phases in an determined automatically by our algorithm are shown in
interferogram are not unwrapped and so should not exhibit black. Presumably, these phase discontinuities are represen-
discontinuities except in regions of severe layover (where tations of centimeter-scale displacement discontinuities on
the surface slopes are greater than or equal to the radar the surface resulting from the earthquake. It would be an
incidence angle, resulting in severe image distortion) unless interesting field exercise to compare the computer generated
spatially discontinuousmotions (breaks) occurred during the cuts with any visible surface scars.
period spanned by the interferogram pair. Nevertheless, We also present in Plate 7 a perspective view of the entire
Plate $ shows clear discontinuities in relatively flat areas. area shown in Figure 2b where the vertical scale is propor-
For example, the region denoted A in Plate $ shows a clear tional to the displacement in the radar line of sight of the
break in the phase measurements. A similar break does not surface. As usual, the brightness at each point is related to
19,626 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR

CORRELATION

PHASE
 ZEBKER ET AL.' COSEISMIC DISPLACEMENT FROM RADAR 19,627

Plate 5. Enlargement in the April-August interferogram of a region around the fault zone. The region
denoted A shows a clear break in the phase measurements,which must be due to a displacement of the
surface where one piece was displaced more than the other. More extensive cracking is found in region B
(see Plate 6).

radar reflectivity, while the color is the displacement

Plate 4. (opposite) Enlargement of the April-August inter- mapped into a repeating color table to accentuate the visi-
ferogram plus the correlation coefficientsfor a region east of
bility of the changes to produce a contour-like map. From
Barstow where center pivot irrigation has been employed.
(top) Radar reflectivity, (middle) correlation coefficient, and this view one can see that the displacement increases as the
(bottom) unwrapped interferogram. Black spots in the lower fault is approached at which point there is an abrupt break in
image are where correlation was insufficient for reliable the surface; from this point hence the surface displacement
phase estimates. The irrigated circles show a clear loss of is of opposite sign.
correlation, presumably due to crop growth, and phase shifts We assess the internal consistency and accuracy of the
which are due to motion, not topography. Examination of measurements presented here by three separate calcula-
the mountains at the top left of the image shows that a
tions. First, we calculate the expected errors due to statis-
topographic change of over 150 m is necessaryto cause a one
cycle change in phase, and the area in question shows less tical variation of the phase estimates. Assuming a radar
than 30 m topographicvariation. These phase changescould signal to noise ratio of 6 dB for the flat desert surfaces, our
be causedby changesin the surface elevations from pumping 20 equivalent look processing (20 resolution elements are
underground water or other hydrologic effects. spatially averaged to reduce statistical noise) yields a stan-
19,628 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR

2 km
Plate 6. Region B from Plate 5, showingphase data in unwrapped form. The phase unwrapping algorithm
we use identifies phase discontinuitiesbefore calculatingthe absolute phase values, and the locations of
cuts determined automatically by our algorithm are shown in black. These phasejumps likely correspond
to ground discontinuities at the cm level that appeared between April and July 1992, probably coincident
in time with the earthquake. Cracking is so extensive that it seemsthe ground has been broken into many
tiles each several hundred meters across.

dard deviation of 9.5ø in the phase for the geometry of the to be an underestimate as it does not take into account any
April-August interferogram and 14.5ø for the July-August temporal decorrelation due to surface disturbances or addi-
interferogram; these values follow from using a target radar tional processingartifacts such as misregistration or other
cross section of -17 dB and accounting for losses accruing samplingand interpolation errors.
from illuminating the ground off the boresightof the antenna. Second, we empirically determined statistical variations
Combining these yields an expected phase error of 10ø rms by measuring the observed phase standard deviations and
for the differential interferogram, equivalent to a horizontal converting the result to horizontal displacement errors.
displacement noise due to finite signal to noise ratio and Choosingboxes correspondingto about 400 m by 400 m of
baseline decorrelation of 0.2 cm. We would expect this value the surface in areas of little seismic variation yielded an
ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR 19,629
19,630 ZEBKER ET AL.' COSEISMIC DISPLACEMENT FROM RADAR

Table 2. Comparison of Radar and GPS Motion Estimates

Horizontal GPS
Displacement Vector in
for Observed Radar
Latitude, Longitude, Radar Direction, Difference,
Site deg deg Motion, cm cm cm

6052 34.52 - 116.84 47.8 33.2 14.6

6056 34.37 - 116.65 18.1 21.9 -3.8
7000 34.68 - 116.72 36.8 91.1 -54.3
7001 34.56 - 116.47 -37.1 -70.2 33.1
HECT 34.79 - 116.42 9.7 -5.2 14.9
LAZY 34.34 - 116.51 62.9 49.4 13.5
LUCS 34.44 - 116.88 26.4 20.7 5.7
POIN 34.45 -117.07 13.4 9.5 3.9
SOAP 34.90 - 116.98 12.3 1.7 10.6
STIM 34.54 - 117.24 7.8 7.4 0.4
FLASH 34.82 - 117.02 14.1 12.1 2.0
HARVARD 34.94 - 116.67 7.5 -0.4 7.9
BOULDER 34.51 - 116.56 176.1 210.8 -34.7
FRY 34.50 - 116.72 66.0 74.6 -8.6
MEANS 34.41 -116.55 82.2 69.8 12.4
OLD WOMN 34.39 -116.75 25.0 17.1 7.9
ORD 34.68 -116.81 44.4 48.3 -3.9
ROCK 34.54 - 116.77 63.9 69.1 -5.2

average horizontal displacement of 0.4 cm rms for the As stated previously, the radar technique is sensitive to
high-frequencycomponentof variations. the line of sight componentof motion. We therefore calcu-
Finally, we attempted to addresslarger-scalevariations by lated the component of the GPS motion vectors in the
measuringthe displacementat 10 widely separatedlocations direction of the projection on the groundof the radar sensor
far from the fault, and we determined their standard devia- boresight, the vector from the sensor to a point on the
tion. In this case the boxes were separatedby 10 km or so, Earth's surface. As for the radar measurements, since the
so that sensitivityto larger-scalevariationswould dominate. line of sight direction is not in the plane definedby the local
This calculation gave a horizontal displacementerror com- Earth surface, we derived the equivalent horizontal surface
ponent of 0.6 cm rms for thesemedium frequencyvariations, motion to yield the observed slant range displacementusing
where medium frequency here refers to irregularitiesoccur-
ring with a spatial frequency of several cycles across the Ap
Ay = (17)
radar image. sin Oinc

which relates the horizontal displacementAy to slant range

Comparison With Field Measurements displacementAp and the incidenceangle 0inc. This angle is
In this section wc discussthe accuracy of our measure- equalto the look angle(denoted0 in Figure1) for a flat Earth
ments and compare the results to those obtainedin the field approximation, and is approximately equal for a curved
using Global PositioningSatellite (GPS) and electronicdis- Earth model. We use a curved Earth model for its improved
tancc measurement (EDM) survey data. As a basis of accuracy. The resultsof both of thesecalculationsare shown
comparison we will use the coseismic displacement field in Table 2 and Figures 3 and 4.
solution as derived by J. Freymueller et al. (personalcom- Because the orbit of the ERS 1 satellite is known only
munication, 1993), data which were compiled by K. W. approximately, as discussedabove there are residual tilts in
Hudnut et al. (personalcommunication,1993). Hudnut et al. the derived radar displacement field. Therefore we have
alsoanalyzedthesedata and obtaineda slightlydifferent,but removed this distortion by solving, in a least squaressense,
consistent solution. These calculated displacementswere for the planar tilt that minimizes disagreementbetween the
derived from a combination of GPS data from several radar and GPS/EDM measurements.This nicely illustrates
sourcesand EDM line lengthsobtainedby the USGS (please one aspect of the complementary nature of the two tech-
see the above references for a more detailed description of niquesfor analyzing ground motions: the radar measuresa
the data sourcesand techniques). widespread displacement field while the GPS/EDM data
The area of overlap between the field survey and our provide accurate point measurementswhich are used to
image contains 18 points at which both field data and radar refine the radar estimates.
estimates of the motion are available. Three additional site The mean value of the differences in Table 2 is 0.9 cm, and
measurementsof field data exist in the overlap region, but the rms difference is 18.9 cm. The formal correlation of the
we were not able to obtain reliable radar phaseestimatesfor data is 0.96, which we illustrate in Figure 3, a scatter plot
them (they occur in the gray regionsof Plate 3). As can be comparing the radar and GPS/EDM measurements. Note
seen from Plate 3, however, the radar data are generally that the best fit through the data evidencesa slightbias.
valid over a wide area and shouldfuture surveysor analyses Figure 4 shows the same data of Table 2 presented
produceadditionalfield points, they may be easily compared graphically. For each survey site, denotedby a triangle, we
with the present analysis. illustratevectors correspondingto motion as determinedby
 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR 19,631

250 i i i i i i survey techniques (diamond-headed arrows) and as deter-

mined by the radar (cross-headed arrows). Note that the
,,' 200 radar vectors are all parallel to the edge of the radar image,
o
as only the component of motion in the line of sight is
z
measured.
'150
From each of these presentations it is apparent that at
most sites, with significant exception of sites 7000 and 7001,
•: 100 -
z the measurements are in rough agreement. The absolute
disagreement is also large at BOULDER, but the motion
z
m
z
50 - here is quite large, and on a relative scale the agreement is
o
comparable to the values for the remaining sites. It is
o o interestingto note that in the deviant cases a large motion is
observed by the GPS technique, while a smaller displace-
CORRELATION = 0.958
ment is visible by the radar technique. In each case where a
-50
small motion is detected by the field survey, a small motion
is measured by the radar interferometer. Figure 4 also
-100 I I I I
suggeststhat there is a degree of spatial correlation in the
-100 -50 0 50 100 150 200 250
regions of agreement, that is, the amount of agreement is
HORIZONTAL DISPLACEMENT FOR OBSERVED RADAR MOTION spatially dependent.
There are several possible causesfor the disagreementsin
Figure 3. Scatter plot of displacementmeasurementswith
GPS/EDM data on vertical axis and radar measurements on the measurements. First, the radar technique is highly sen-
horizontal axis. The correlation of the data sets is 0.96, sitive to vertical motions which are not expressedin the GPS
however a slight bias is observed as the slope of the line is displacement field. While this is likely to affect the differ-
not 1. ences on the centimeter scale, it is probably not a significant

35 o .............. i .............. i .............. I ..............

,.• H•,RVARD
SOAP
A SITE

ß GPS/EDMVECTOR f•
FLASH
' RADAR VECTOR

34 ø 45'

"" ORD
7000•
STIM
34 o 30' 6052 -- '
FRY BOULDER

i i
POIN
• LUCS
•'• • MEANS OLDWOMN
6056 •
I M DISPLACEMENT LAZY

34015, at" "tietit "tttl: tttt't' ttttttl "t'ttt" 'tt''tt Itttt" it"Jill II
-117 o 30' -117 o 15' -117 o -116 o 45' -116 o 30"

LONGITUDE

Figure 4. Displacementvectorsas measuredby GPS/EDM data and by radar interferometry.Each GPS

or EDM site is denotedby a triangle, and a vector endingwith a square(GPS/EDM measurement)and a
vector ending with a cross(radar measurement)are shown in the direction of motion. Note that for the
radar case only the componentin the radar line of sightdirection is determined and thus all measurements
are parallel. Vectors are correlated at 0.96 level and show that radar and field surveys are measuring
similar phenomena.
19,632 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR

factor in the radar underestimation of the motions. This only data acquired from an orbiting high-resolution radar
follows from the unlikelihood that vertical motions would system and to achieve results comparable in magnitude to
just happen to be in the direction with respect to the radar to those obtained by conventional field survey techniques.
cancel out any horizontal shifts. Data from the ERS 1 synthetic aperture radar instrument
The GPS sites, particularly the dual-frequency sites, in acquired at three separate instances of time are sufficient to
fact yield vertical components to the displacement. The rms generate a high-resolution, wide area map of the displace-
vertical displacement for the GPS sites is 17.2 cm, but most ments. Comparison of these data with GPS and EDM survey
of this is associated with site 7000; when this site is removed, data indicates a high degree of confidence in the radar
the remaining rms displacement is only 5.1 cm. Ignoring measurements. We are confident that the differences be-
vertical displacements, as was done in our analysis, results tween the radar and GPS measurements are reconcilable and
in a misinterpretation of lateral shift of magnitude equal to do not point to a fundamental limitation in the radar tech-
the true vertical movement divided by the tangent of the nique. Further work is needed along these lines however.
incidence angle. The errors in the above cases then become The power of the differential interferometry technique for
40.5 cm and 12.0 cm, respectively. However, we must note seismological applications lies in its centimeter-scale mea-
that for nine of the 10 sites the 1 - rr error in the vertical surement sensitivity of line of site displacementsover a wide
displacement is larger than that of the measurementitself, so area. The derived displacement fields can be used as a tight
these data must not be overly interpreted. constraint in the modeling of earthquake motion. The fine
The second cause for disagreement is error in the mea- accuracy, fine spatial resolution, and large areal coverage
surements. As discussed above, the radar data exhibit sta- will likely allow increasingly detailed models to be explored,
tistical errors less than 1 cm rms on both small and medium on both large and small spatial scales. The promise of a
scales and thus would be insignificant for this comparison. system to map small-scale fractures in the Earth's surface
However, large-scale warping of the radar image remains a over a wide region automatically with a remote sensing
possibility. We were able to remove most of these effectsby system will greatly facilitate field activities by permitting
minimizing the errors with the least squares removal of concentration in the most important areas.
planar tilts as described previously. That this correction was What of earthquake prediction? Current understandingof
approximately correct may be verified by examining the the behavior of earthquakes suggeststhat differential inter-
residual motion in the upper and left hand portions of the ferometry may not have the accuracy required to detect
radar image, those portions farthest from the fault. The precursory seismicmotions necessaryfor prediction. Calcu-
observed motion here is very small, as we would expect. If lations based on theoretical seismic deformation models
the ERS 1 coverage had been such that the fault was show small but steady deformation rates in fault zones with
positioned in the center of the radar swath, we could have a change in the rates occurring within a period of months to
verified the lack of displacement more accurately all the way years prior to a seismic event [Lorenzetti and Tullis, 1989;
around the image. The possibility of a long-scale error thus Stuart et al., 1985]. While the steady deformations have
still exists and may to some degree explain the observed been observed and are well studied, no precursory rate
spatial correlation of the errors. changeshave been measured. Furthermore, the steady rates
The errors in the GPS/EDM data themselves account for themselves are probably at or below the limits of detectabil-
part of the disagreement. The 18 sites listed in Table 2 ity by differential interferometry, perhaps 10 mm/yr, while
exhibit an rms error of 9.1 cm in the Stanford analysis, while the precursory signal is expected to be smaller. Wide area
the 10 GPS-only sites have a 7.7-cm rms error of J. W. mapping of the surface distribution of these small deforma-
Hudnut et al. (personal communication, 1993). tion rates afforded by differential interferometry may pro-
A third possibility is the existence of phase unwrapping vide new insights into local accumulation of strain close to
errors in the radar data. As each unwrap error results in a and along a fault, but the possibility that radar interferome-
one cycle phase error in one interferogram, these errors try can be used as a predictive tool now appears to be
would appear as A/2 errors in Ap, or 8 cm in horizontal shift remote. This is not to say that likely future technological
if it occurred in the April-August pair or 2 cm in the advances in spaceborne radar such as higher-resolution,
July-August pair. However, we have examined the data for increased signal to noise ratio, and multiple frequency oper-
signs of unwrapping errors and believe that the regions near ation will not close this gap and permit the sensitivities
the GPS sites are unwrapped correctly. In addition, it is required for millimeter-level surface characterization. None-
unlikely again that phase unwrapping mistakes would nearly theless, in addition to after the fact seismic event modeling,
correct for GPS-observed displacements. currently radar interferometry can aid in monitoring, fore-
Finally, the locations of the GPS sites are known only to a casting, and in some cases predicting a range of hazardous
few tens of meters in the radar image as the radar data are events. For example, volcanoes are known to bulge prior to
not accurately geocoded, thus leading to estimates at the eruption at a scale suitable for radar interferometry.
wrong places. However, we have analyzed the regions In the short run, existing and planned radar missionssuch
around the sites in the radar data and have determined that NASA's Shuttle Imaging Radar, the European Space Agen-
the displacement does not change rapidly in those areas. cy's ERS 2, Canada's Radarsat, and the Japanese JERS 1
Thus even a slight positional shift would not result in a system could be operated to emphasize repeat-pass obser-
significant error. vations at the largest acceptable incidence angles, providing
a very large suite of instruments collecting data that may be
processedfor change detection analysis. For the future one
Discussion
can envision a global seismic satellite mission designed to
We have shown that it is possible to map a coseismic detect and forecast earthquakes and other natural hazards: a
displacementfield resulting from a major earthquake using single satellite in a short repeat period orbit similar in design
 ZEBKER ET AL.' COSEISMIC DISPLACEMENT FROM RADAR 19,633

to that proposed by H. A. Zebker et al. (personal commu- are not nearly aligned with the look direction, the ratio in
nication, 1993) for global topographic mapping. The repeat (A7) may be expanded to give
cycle of the orbit should be short, of the order of 1 day, to
8B' 8B
minimize
theeffectsof temporal
decorrelation.
Precisesat-
1q + 8a' tan(Oo-a')
ellite ephemeris from GPS measurements can ensure auto- B' B
matic construction of interferograms and displacement
fields. Only three repeat periods of data need to be stored at
-/5a tan (0o- a) . (A8)
any time; the processingcan proceed in real time, and results
can be perused automatically for evidence of anomalous
displacements. Detailed design of the radar system and Using (A6) and (A8) to evaluate the displacement gives
orbital scenarioplus the establishmentof detection and false
alarm thresholds, must await interest by the global commu- A

nity. Given the enormouscost in lives and resourcesinflicted A/•--• [• hat

--• •>fiat]
by earthquakes, interest is sure to follow any evidence that
radar interferometry can be used predictively in assessing = Ap - /SB' sin (0o- a') + /sa'B' cos (00- a')
natural hazards such as earthquakes.
- y[-/SB' sin (0o-a')+/5a'B' cos (00-a')]

A
Appendix: Baseline-Induced Displacement Errors +/sa' tan (00-a')
--'Y•--•qbflat
B' B
Equation (13) is the displacement determined from the
flattened interferometric phase assumingperfect knowledge
of the baselines. Reiterating, -/sa tan (00- a) (A9)

A
Equation (A9) shows that in addition to the desired term Ap,
Ap=• [• hat-Tr•fiat], (A1) there are slowly varying (fraction of a cycle) sinusoidal
where artifacts acrossthe displacementfield and topographic resid-
uals dependenton baseline length and angle errors. Even if
A the slowly varying artifacts are removed empirically, accu-
{bna
t --B sin (0- a)- B sin (00- a), (A2) rate estimates of the displacements, to fractions of a cycle,
4z'
require fairly accurate baselineknowledge. We can estimate
B' cos (00- a') the scale of the topographic term as follows. Assume B • B'
•' = . (A3) and likewise for the uncertainties/SB and/sa. Then the final
B cos (00-a)
term in (A9) becomes
With imperfectknowledgeof the baselines
i9 and&, errors
A /sBnet
areintroduced
in boththephase,denoted •fiat,andthescale
factor, denoted •. Defining AbtøPø
4•rqbfiat
Bne-
•- (A10)
• = B + /SB (A4) where we have assumed a worse case •, = 1 and an
equivalent net baseline error /sBne
t including /SBand /saB.
& = a + /sa, (A5) Expanding ½•fiat(0)about00,
we have to first order
Aktop
o--•/sBnet/5
0,
A A
where/50 = z/p is the angulardeviation of the look direction
4w •flat
=Bsin(0- a)- i9sin(00- &)=• qbflat due to topography. Thus, to limit displacement errors due to
residual
topography,
Abtop
o,to say1/4wavelength,
theerror
-/SB sin (0 - a)- /saB cos (00- a) (A6) in /sBne
t must satisfy the inequality
and
/sBne
t < 0.25A P/Zmax
cos(0o- &') where Zmaxis the maximum topographic extent over the
cos(00- &) scene. For p = 800,000 m, A = 0.0566 m, Zmax-- 5000 m,
/sBne
t < 2.3 m.
B' cos (00- a') +/SB' cos (0o- a') +/sa'B' sin (00- a')
B cos(00 - a) +/SB cos (0o - a) +/saB sin (00 - a)
Acknowledgments. We would like to acknowledge Paul Segall
(A7) for supplying the GPS/EDM measurementsand for several useful
discussions regarding the intercomparison of the data sets. We
Note that for •, baseline length error and angle error are would also like to acknowledge discussionswith Ken Hudnut for
complementary: baseline length error is weighted highly discussionsprompting a reexamination of the comparison of our
results with the GPS/EDM data. The research described in this
when the baselineis orthogonalto the look direction(00 - paper was carried out by the Jet PropulsionLaboratory, California
ai = 0), whereas angle error is weighted highly with the Institute of Technology, under a contract with the National Aero-
baseline aligned with the look direction. For baselines that nautics and Space Administration.
19,634 ZEBKER ET AL.: COSEISMIC DISPLACEMENT FROM RADAR

References Madsen, S. N., H. A. Zebker, and J. Martin, Topographic mapping

using radar interferometry: Processingtechniques,IEEE Trans.
Curlander, J. C., and R. N. McDonough, SyntheticAperture Radar, Geosci. Remote Sens., 31,246-256, 1993.
Wiley-Interscience, New York, 1991. Massonnet, D., M. Rossi, C. Carmona, F. Adragna, G. Peltzer, K.
Elachi, C., SpaceborneRadar Remote Sensing:Applicationsand Feigl, and T. Rabaute, The displacementfield of the Landers
Techniques,IEEE Press, New York, 1987. earthquakemapped by radar interferometry, Nature, 364, 138-
Evans, D. L., T. G. Farr, H. A. Zebker, J. J. van Zyl, and P. J. 142, 1993.
Mouginis-Mark, Radar interferometric studies of the Earth's Prati, C., F. Rocca, A. Monti Guarnieri, and E. Damonti, Seismic
topography,Eos Trans. AGU, 73, 553,557-558, 1992. migrationfor SAR focusing:Interferometricalapplications,IEEE
Gabriel, A. G., R. M. Goldstein, and H. A. Zebker, Mapping small Trans. Geosci. Rem. Sens., 28, 627-640, 1990.
elevationchangesover large areas:Differentialradar interferom- Stuart, W. D., R. J. Archuleta, and A. G. Lindh, Forecastmodelfor
etry, J. Geophys. Res., 94, 9183-9191, 1989. moderate earthquakes near Parkfield, California, J. Geophys.
Ghiglia, D.C., and L. Romero,Robusttwo-dimensional weighted Res., 90, 592-604, 1985.
and unweighted phase unwrapping using fast transforms and Zebker, H. A., and R. M. Goldstein,Topographicmappingderived
iterative methods, J. Opt. Soc. Am., 11, 107-117, 1993. from syntheticaperture radar measurements,J. Geophys.Res.,
Goldstein, R. M., and H. A. Zebker, Interferometric radar measure- 91, 4993-4999, 1986.
ment of ocean surface currents, Nature, 328, 707-709, 1987. Zebker, H. A., and J. Villasenor, Decorrelation in interferometric
Goldstein, R. M., H. A. Zebker, and C. L. Werner, Satellite radar radar echoes, IEEE Trans. Geosci. Remote $ens., 30, 950-959,
interferometry:two dimensionalphase unwrapping,Radio Sci., 1992.
23, 713-720, 1988. Zebker, H. A., S. N. Madsen, J. Martin, K. B. Wheeler, T. Miller,
Goldstein,R. M., T. P. Barnett, and H. A. Zebker, Remotesensing Y. Lou, G. Alberti, S. Vetrella, and A. Cucci, The TOPSAR
of ocean currents, Science, 246, 1282-1285, 1989. interferometric radar topographic mapping instrument, IEEE
Gray, A. Laurence,andP. J. Farils-Manning,Two-passinterferom- Trans. Geosci. Rein. $ens., 30, 933-940, 1992.
etry with airbornesyntheticapertureradar, IEEE Trans. Geosci.
Remote Sens., 31, 180-191, 1993. A. Gabriel, R. M. Goldstein, P. Rosen, C. L. Werner, and H. A.
Li, F., and R. M. Goldstein, Studiesof multi-baselinespaceborne Zebker, MS 300-227,Jet PropulsionLaboratory, 4800 Oak Grove
interferometric synthetic aperture radars, IEEE Trans. Geosci. Drive, Pasadena, CA 91109.
Remote Sens., 28, 88-97, 1990.
Lorenzetti, E., and T. E. Tullis, Geodetic predictionsof a strike-
slip fault model: Implications for intermediate and short-
term earthquakeprediction,J. Geophys.Res., 94, 12,343-12,361, (Received December 6, 1993; revised April 21, 1994;
1989. acceptedApril 29, 1994.)