Natural Hazards and Earth System Sciences (2003) 3: 321–332

c European Geosciences Union 2003

Natural Hazards
and Earth
System Sciences

Simulation of the trans-oceanic tsunami propagation due to the 1883

Krakatau volcanic eruption
B. H. Choi1 , E. Pelinovsky2 , K. O. Kim3 , and J. S. Lee1
1 Department of Civil and Environmental Engineering, Sungkyunkwan University, Suwon, 440-746, Korea
2 Laboratoryof Hydrophysics and Nonlinear Acoustics, Institute of Applied Physics, 46 Uljanov Street, 603950, Nizhny
Novgorod, Russia
3 Research Center for Disaster Environment, DPRI, Kyoto University, Gokasho, Uji, Kyoto, 611-0011, Japan

Received: 13 June 2002 – Revised: 16 September 2002 – Accepted: 8 October 2002

Abstract. The 1883 Krakatau volcanic eruption has gener- with seawater and generated a tsunami. At 06:36 LT, the
ated a destructive tsunami higher than 40 m on the Indone- 500 m-high peak of Danan (also does not exist) exploded
sian coast where more than 36 000 lives were lost. Sea level and collapsed, sending more seawater into the molten magma
oscillations related with this event have been reported on chamber of the eruption and producing another tsunami up to
significant distances from the source in the Indian, Atlantic 10 m. The third blast, at 09:58 LT, tore the remaining part of
and Pacific Oceans. Evidence of many manifestations of the the Krakatau Island (Rakata Island) apart. Including ejecta,
Krakatau tsunami was a subject of the intense discussion, and 9–10 km3 of solid rock was blown out of the volcano. About
it was suggested that some of them are not related with the 18–21 km3 of pyrolastic deposits spread out over 300 km2
direct propagation of the tsunami waves from the Krakatau to an average depth of 40 m. Fine ash spread over an area
volcanic eruption. Present paper analyzes the hydrodynamic of 2.8×106 km2 , and thick pumice rafts impeded navigation
part of the Krakatau event in details. The worldwide prop- in the region up to five months afterwards. A caldera with
agation of the tsunami waves generated by the Krakatau 6 km in diameter and 270 m deep formed where the central
volcanic eruption is studied numerically using two conven- island had once stood. The third blast was the largest sound
tional models: ray tracing method and two-dimensional lin- ever heard by humanity and was recorded 4800 km away.
ear shallow-water model. The results of the numerical simu- The atmospheric shock wave traveled around the world seven
lations are compared with available data of the tsunami reg- times. Barometers in Europe and the United States measured
istration. significant oscillations in pressure over nine days following
the blast. The total energy released by the third eruption was
equivalent to 200 megatons atomic bomb (8.4×1017 joules).
The two predawn blasts each generated tsunami that
1 Introduction
drowned thousands in the Sunda Strait. The third blast-
A most devastating tsunami occurred when the volcano on induced wave was cataclysmic and devastated the adjacent
the Krakatau Island, situated in the Sunda Strait between Java coastline of Java and Sumatra. The northern coastline of the
and Sumatra (Fig. 1) erupted in August 1883 (Simskin and eruption was struck by waves with a maximum runup height
Fiske, 1983; Bryant, 2001). Four largest explosions of this of 42 m (town of Merak, see Fig. 1), and tsunami penetrated
volcano in 26–27 August produced tsunami waves recorded 5 km inland over low-lying areas. Great amplification of the
on the coasts of the Sunda Strait. These waves are gener- tsunami waves in Merak (from 15 m to 42 m) is related due
ally attributed to submarine explosion, caldera’s collapse and to the tunnel-shaped bay. At least 36 000 people were killed,
pyrolastic flow entering the sea (Francis, 1985). On 26 Au- most by the sea waves, and 300 villages were destroyed. Sea
gust about 17:00 LT, loud explosions recurred at intervals of surface oscillations have been recorded worldwide: the In-
10 min, and a dense tephra cloud rose 25 km above the is- dian Ocean (Indonesia, India, Sri Lanka, Yemen, Mauritius,
land. Small tsunami waves 1–2 m in height swept the Sunda South Africa, Australia), the Pacific (USA, New Zealand,
Strait. On the morning of 27 August, three horrific explo- Japan and Chile), the Atlantic (France, Antilles, Panama and
sions occurred. The first explosion at 05:28 LT destroyed the South Georgia Islands), see Fig. 2. Various descriptions
130 m peak of Perboewatan on the Krakatau Island (now it of tsunami wave manifestation (both, eyewitness and tide-
does not exist), forming a caldera that immediately infilled gauge records) and corresponding scientific accounts are col-
lected by Simskin and Fiske (1983).
Correspondence to: E. Pelinovsky Numerical simulation of the tsunami propagation in the
(enpeli@hydro.appl.sci-nnov.ru) Sunda Strait around the Krakatau Island using the ray method
322 B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation

Fig. 2. The locations of the 1883 Krakatau tsunami recording.

Fig. 1. Location of the Krakatau Island. The present paper considers the direct tsunami wave prop-
agation from the Krakatau volcanic eruption in the frame-
work of the shallow water theory to analyze the contribu-
has been performed by Yokoyama (1981). The wave heights tion of this mechanism for an explanation of the observed
were calculated using the Green’s law based on the energy sea level disturbances. Main goal of this study is to per-
flux conservation. Then Nakamura (1984) repeat these cal- form the numerical simulation of the tsunami waves on globe
culations using the finite-difference scheme. He simulated using the detailed bathymetry of the World Ocean. It can
also the tsunami wave propagation in the adjacent part of obtain more rigorous estimates of the tsunami travel time,
the Indian Ocean and the comparison with the observed clarify the role of the tsunami pathways through the Antarc-
data leads to the estimated depth in the equivalent tsunami tic basin, and compare the wave characteristics in different
source in 700 m. The Krakatau tsunami was also numer- areas. The material is organized as follows. The available
ically simulated by Kawamata et al. (1992) for the region data of the tsunami waves from the Krakatau events is briefly
of outside the Sunda Strait by assuming the caldera forma- summarized in Sect. 2. The ray tracing method allowed to
tion, which makes the surrounding water rush into the cavity. calculate the pathways on the globe is described in Sect. 3
Nomanbhoy and Satake (1995) investigated the mechanism and it is applied to determine the travel time of the lead-
of the tsunami generation at the Krakatau volcanic eruption. ing wave. The direct numerical simulation of the tsunami
Based on the numerical simulation of the near field (Java and wave propagation is performed in the framework of two-
Sumatra) they concluded that the submarine explosion model dimensional linear shallow-water theory; see Sect. 4. The
as the source of the largest tsunami is favored. hydrodynamic model is applied to describe characteristics
The arrival time of the tsunami in the numerous ports of the tide-gauge records, and to improve computed values
of World Ocean has been estimated in several years after for tsunami travel time. The applicability of hydrodynamic
the Krakatau event based on the existing tide-gauge records model to describe the observed sealevel disturbances related
(35) and eyewitness accounts, and, very roughly, on the long with tsunami waves generated at the Krakatau volcanic erup-
wave theory for the wave speed (cited on Simskin and Fiske, tion is discussed in conclusion.
1983). In many places the sea level disturbance were weak,
“no precise or close comparison between various tidal dia-
grams can be made, and this doubt of the identification of 2 Tsunami wave descriptions
any particular wave at different places, causes much uncer-
tainty in the result, as far as it relates to the speed of waves”. Various data of the 1883 tsunami manifestation are col-
This concerns mainly Pacific and Atlantic data, which prob- lected in the books by Simskin and Fiske (1983), Murty
ably had no connection with the Krakatau event. (1977), Bryant (2001). Also some data are presented in
Ewing and Press (1995) analyzing historic data have no- Russian Navy Atlas (USSR Navy, 1974) and in unpublished
ticed, ”tide-gauge disturbances at distant stations correlate manuscript by Soloviev and Go (1974). The locations of
in time with the first aerial wave arriving at the station” and places where tsunami waves were observed are presented in
suggested new hypothesis of the origin of the sea level dis- Fig. 2. On the coasts of the Sunda Strait the wave height
turbances at distant stations based on coupling between the reached 15 m in average with maximum up to 42 m. Our
barometric disturbance and the ocean surface wave. Then focus is to study the far-field tsunami propagation from the
Press and Harkrider (1966) and Garret (1976) developed this Krakatau volcanic eruption only. Data of the tsunami obser-
mechanism and obtained the satisfactory agreement for wave vations on long distances from the Krakatau is briefly repro-
amplitudes in the USA coastal stations. Probably, the sea duced, revised and discussed, as follows:
level disturbances in New Zealand (in particular, in Lake On Andaman Islands, India (Port Blair, 2440 km from the
Taupo not connected to the ocean) were also related with the Krakatau, Fig. 3), the weak tsunami was recorded on the
atmospheric pressure waves (Bryant, 2001). tide-gauge station and no data of tsunami manifestation on
B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation 323

slow behind in their action, for they washed away a belt of

land about 132-198 feet (40.23–60.35 m) in extent, including
the burial ground situated on the coast to the south-west of
the bay compelling the inhabitants to seek shelter in a neigh-
boring cocoa-nut garden”. Sixteen recessions were counted
between noon and 03:00 LT on 27 and the rushing water pro-
duced what was described as a hissing sound. At Arugam
Bay (southeastern coast, Fig. 3), “Three moorwomen, three
children, and a man were crossing the bar about 03:00 LT.
A big wave came up from the sea over the bar and washed
them inland. Soon after the water returned to the sea. The
man said that the water came up to his chest: he is a tall man.
These people were tumbling about in the water, but were res-
cued by people in boats who were fishing in the Kalapuwa
(inland estuary). The lost the paddy they were carrying, and
one of the women died two days after of her injuries”.
From these descriptions we can conclude that tsunami on
Sri Lanka presents the group from 16 waves with average
period of 11 min, and its crest height is more than 1 m. The
reported first wave is positive or negative depending from the
location. The wave has approached to the coast 5–7 h after
origin.
Tide-gauges records at many ports in India show the sea
level disturbances. In particular, tide-guages records at
Madras and Negapatam (see, Fig. 3) have a period of about
Fig. 3. The locations of tsunami observations in India and Sri 1.0 to 1.5 h. The wave amplitude at Negapatam station was
Lanka. 0.56 m. An extraordinary phenomenon of tides was wit-
nessed at Bandora near Bombay (4500 km and 11 h from the
Krakatau), on the morning of 28 August (there is probably
the coast. The quality of the tide-gauge record is not very the mistake on p. 147 of the book by Simskin and Fiske
high to determinate the wave characteristics. According to (1983), because the Krakatau time of this event is 21:00 LT
received opinion, the wave height is reached 0.2 m. The first on 27 August as cited in the same book on p. 43. The local
wave arrived at 14:00 LT (15:30 of the Krakatau time), and time should be 19:00 LT on 27 August) by those who were
the characteristic wave period is about 1 h. The travel time at the time on the seashore. “The tide came in, at its usual
can be estimated as 4 h. We will show this tide-gauge record time and in a proper way. After some time, the reflux of the
in Sect. 4 comparing with computing data. tide went to the sea in an abrupt manner and with great impe-
On Sri Lanka (3113 km from the Krakatau) two descrip- tus, and the fish, not having sufficient time to retire with the
tions of tsunami on 27 August are available. In Galle (south- waves, remained scattered on the seashore and dry places,
western coast, see Fig. 3), “an extraordinary occurrence was and the fishermen, young and old, had a good and very easy
witnessed at the wharf at about 01:30 LT (15:00–15:30 of the task to perform in capturing good-sized and palatable fish,
Krakatau time). The sea receded at far as the landing stage without the least trouble or difficulty, to their hearths’ con-
on the jetty. The boats and canoes moored along the shore tent, being an extraordinary event never seen or heard of be-
were left high and dry for about three minutes. A great num- fore by the old men; but suddenly the flux came with a great
ber of prawns and fishes were taken up by the coolies and current of water, more swift than horse’s running. The tide
stragglers about the place before the water returned. Since was full as before, and this flux and reflux continued two
the above was written, the sea has receded twice throughout or three times, and at least returned by degrees as usual”.
the harbour”. At Negombo (north from Colombo, Fig. 3), at So, tsunami in India began from the fast ebb, and two-three
03:00 LT (16:30–17:00 of the Krakatau time). “The rise of waves with steeper slopes were reported; the period of such
the tide was so much above the usual water-mark that many waves should be 10 min or more, looking on description of
of the low morasses lying in close proximity to the seaside the people behavior.
were replete with water that flowed into them. However, “The great tidal disturbance” at Aden occurred 12 h after
the water thus accumulated did not remain long, but, form- the Krakatau volcanic eruption, on distance of 3800 nauti-
ing into a stream, wended its course in a southerly direction, cal miles (7038 km). The observed wave amplitude on tide-
through low lands, to a distance of nearly a quarter of a mile, gauge record is 0.24 m.
and found a passage back to the sea; thus the water that had On Rodriguez Island in the Indian Ocean (4653 km from
so abruptly covered up such an extent of land did not take the Krakatau and 1600 km east of Madagascar) the tsunami
many days in draining off.” “The receding waters were not effects were observed starting from 27 August 13:30 LT
324 B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation

(16:30 of the Krakatau time). “It was ebb tide, and most shown in Sect. 4. The travel time of the first wave is esti-
of the boats were aground. The sea looked like water boil- mated as 13.5 h.
ing heavily in a pot, and the boats, which were afloat, were In Australia, within four hours of the final eruption, a
swinging in all directions. The disturbance appeared quite tsunami arrived at North-West Cape, 2100 km away (Bryant,
suddenly, lasted about half an hour, and ceased as suddenly 2001). The wave swept through gaps in the Ningaloo Reef
as it had commenced. At 14:20 LT a similar disturbance be- and penetrated 1 km inland over sand dunes. Nakamura
gan; the tide all of a sudden rose 5 feet 11 inches (1.8 m), with (1984) pointed out, the sudden sea waves with maximum dis-
a current of about 10 knots an hour (1.85 km/h) to the west- turbance of 1.5 m came to Cossack (northwestern part) on 27
ward, floating all boats, which had been aground, and tearing August at 16:00 LT (15:15 of the Krakatau time). At Gerald-
them from their moorings. All this happened in a few min- ton (southwestern part), the sea level drew down to 1.8 m
utes. The tide then turned with equal force to the eastward, suddenly at 20:00 LT (19:00 of the Krakatau time) on the
leaving the boats which were close in-shore dry on the beach, same day. In Williamstown (near Melbourne) the wave ap-
and dragging the Government boat (a large decked pinnace) proached at 04:40 LT on 28 August (01:40 of the Krakatau
from heavy moorings, and leaving it dry on the reefs. The time). In Tasmania, the fast arriving of the tidal wave in the
inner harbour was almost dry. The water in the channel was Huon River (south of Hobart) arrived with large speed and
several feets below the line of reefs; and, owing to the sudden force on August 28 and 29; this flow lifted the rubbish on the
disappearance of the water, the reefs looked like islands ris- height 1 m (Soloviev and Go, 1974). The wave approached
ing out of the sea. The tides continued to rise and fall about to the northwestern shelf of Australia for 5 h and then propa-
every half hour, but not so high, or with the same force, as gated along the Australian coast for 10 h.
the first tide. By noon on the 29 August, the tide was about
On New Zealand (7767 km from the Krakatau), in Auck-
its usual height”.
land Harbor at 04:00 LT on 29 August (23:00 on 28 August
So, tsunami presents the wave group with the period of
of the Krakatau time), “in a few minutes the tide rose fully
30 min and the total duration about two days. Several waves
1.8 m and as suddenly receded again, leaving the vessels in
have the same amplitudes and at least the first waves have
port high and dry. Throughout the day the way was felt sev-
steep fronts (duration of a few minutes) like shock waves.
eral times with equal strength”. Additional information is
The crest height is 1.8 m and the through amplitude is ap-
in the unpublished manuscript by Soloviev and Go (1974).
proximately the same, therefore, the wave height is 3.6 m.
In Timaru (eastern coast of the South Island), several waves
Wave velocities were about 5 m/s and this corresponds to
were pointed in the morning of 29 August. Also the water
the relation from linear theory for the depth about 2 m. The
rose and fell twice for 40 min in the Lake Taupo (center of
tsunami waves have approached 6–7 h after origin.
the North Island) with amplitude 0.5 m. If the date, August
On Mauritius Island in the Indian Ocean (5445 km from
29 is correct, the travel time is about of 37 h.
the Krakatau) tsunami was observed at the same time. “At
about 13:30 LT the water came with a swirl round the point The tide-gauge record at Le Havre, France (17960 km
of the sea wall and in about a couple of minutes returned from the Krakatau) shows the sea level disturbances at 21:35
with the same speed. This took place several times, the wa- on 28 August (the Krakatau time), 32.5 h after the Krakatau
ter on one occasion rising 2.5 feet (0.76 m)”. In another volcanic eruption. The wave height is week, about 12 mm.
place (funnel-shaped harbor) ”the water, which was then un- In USA the sea level disturbances can be detected on the
usually low, suddenly rushed in with great violence, rising tide-gauges at San Francisco, Kodiak and Honolulu, which
fully 3 feet (0.91 m) above the former level. An alternate ebb will be given in Sect. 4. At Sacramento (San Francisco), the
and flow then continued till nearly 19 h, and the interval be- first oscillations appeared at 11:00 LT on 27 August (02:00
tween high and low water were about 15 min. There was no on 28 August of the Krakatau time), and the maximal wave
high wave or billow, but strong currents, the estimated ve- up to 0.1 m – 2 h later, at 13:00 LT (04:00 on 28 August of
locity of which was about 18 knots an hour. Vessels moored the Krakatau time), and the travel time of the first wave can
near the Dry Docks swayed much and at about 18:30 LT one be estimated as 16 h. At St. Paul’s (Kodiak), small waves
of the hawsers of the Touared, 10 inches in circumference, (0.1 m) were recorded between 17:00 LT and 23:00 LT on 27
parted. Buoys in the neighborhood were at times seen spin- August (09:00–15:00 on 28 August of the Krakatau time).
ning round like tops. Disturbances were observed on 28 Au- The travel time of the first wave is estimated as 23 h. At Hon-
gust also, and there unusual currents even on the 29 August”. olulu, sea level disturbances began at 03:00 LT on 27 August
So, characteristics of tsunami waves on the Mauritius and (20:00 on the same day of the Krakatau time) and contin-
Rodriguez Islands (arriving time, the wave period, wave am- ued two days. Maximal wave approached 2 h later and had
plitudes) are almost the same. a height of 0.25 m. The travel time of the first wave can be
On South Africa, Port Elizabeth (7546 km from the estimated as 10 h. On the Caribbean Sea, Virgin Islands, at
Krakatau), the first tsunami wave has approached approxi- St. Thomas, “A tidal wave occurred on 27 August, the wa-
mately at 16:00 LT (23:30 of the Krakatau time) on 28 Au- ter receded from the shore three times” (Lander et al, 2002).
gust and the maximal wave 4 h later with height about of At Colon (Panama Canal), the tide-gauge record shows wave
1.4 m. The sea level oscillations with visible period of 1 h disturbances with the height up to 0.4 m. The first negative
continued during next day. This tide-gauge record will be wave approached at 15:00 LT on 27 August (03:30 on 28 Au-
B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation 325

Fig. 4. Krakatau tsunami travel time chart computed from ray-

tracing tsunami model.

gust of the Krakatau time), and the maximal wave – 1.5 h

later. The travel time of the first wave is 17.5 h.

On South Georgia Islands (Moltke Harbour), tsunami be-

gan at 14:00 LT on 27 August (23:00–24:00 on the same
Fig. 5. Tsunami pathways through the circumpolar sea.
day of the Krakatau time), and the maximal wave arrived
12 h later, its height is about 0.3 m (tide-gauge record will be
given later). The travel time of the first wave is estimated as 3 Ray tracing model
13–14 h.
The ray tracing method is usually used as a short wavelength
The additional information of tsunami manifestation in the approximation in the theory of the long wave propagation
Pacific is presented on the map from the Russian Navy At- in the smooth inhomogeneous media. Ray method leads to
las (USSR Navy, 1974) prepared by Sergey Soloviev, these calculate directly the travel time and the wave amplitude us-
locations shown in Fig. 2). First of all, the observations of ing the Green’s law based on the energy flux conservation.
tsunami with the height of 1 m on the coast of Japan should For tsunami waves this method is actively applied, includ-
be pointed (Bryant, 2001 mentions that tide-gauges in Japan ing the 1883 Krakatau event (see, for instance, Simskin and
measured changes of 0.1 m). In Chile, Talcahuano, the dif- Fiske, 1983). Below the formulating of the ray method for
ference between the maximal and minimal water level dur- the spherical earth developed by Satake (1988) will be used.
ing 1 h reached 0.7 m at 08:00 LT, 28 August (20:00 of the Ray tracing equations are:
Krakatau time, or 34 h after the volcanic eruption). Time
is indicated in the unpublished manuscript by Soloviev and dθ cos ζ
= , (1)
Go (1974); in their catalogue there is only general informa- dt nR
tion that tsunami affected also to the sea level on the South dφ sin ζ
= , (2)
American coast, and the sea level disturbances were 0.5 to dt R sin θ
1 m (Soloviev and Go, 1974; Nakamura, 1984). On the Cape dζ sin ζ ∂n cos ζ ∂n sin ζ cot θ
= 2 + 2 − , (3)
Horn (southern cape of America), the wave amplitude on the dt n R ∂θ n R sin θ ∂ϕ nR
tide-gauge is 0.25 m. Ewing and Press (1995) gives arriving where θ and ϕ are latitude and longitude of the ray, n =
time for the first wave, 14:20 LT on 27 August (16.5 h af- (gh)−1/2 is the slowness, g is the gravity acceleration,
ter the volcanic eruption), and for the first great wave, 7.5 h h(θ, ϕ) is the water depth, R is the radius of the Earth,
later. Weak tidal phenomenon was pointed in 1883 on the and ζ is the ray direction measured counter-clockwise from
Islas Tres Marias, Mexico, and Soloviev and Go (1974) con- the south. The above equations are solved by the Runge-
sider this as tsunami from the Krakatau. Kutta-Gill method. Integration is performed by the mid-
Russian Navy Atlas (USSR Navy, 1974) summarizes also point method, using interpolated velocities; the detail of
data of tsunami observations in Java Sea (up to 3 m, near the method parallels that of Sobel and von Seggern (1978).
the Sulawesi Island, 0.5 m) and Cocos Islands, 1.5 m, but we Visualization of numerical results is discussed by Choi et
were not able to find descriptions of these observations in the al. (1997).
literature. The location of the source in the numerical simulations al-
most coincides with the location of the Krakatau volcano, its
In the next sections numerical models will be used for ex- coordinates are 6◦ 100 S, 105◦ 100 E, and depth in the source
planation of the observed data (travel time, height) of the is 400 m. The number of rays is 3600 and they are initially
tsunami origin at the Krakatau volcanic eruption. distributed uniformly. For correct calculation of the travel
326 B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation

15
Table 1. Travel time of the first tsunami wave

Point Observed time, Calculated time,

calculated travel time, hr

Port Elizabeth
h h (South Africa)
10 Aden

Port Blair, Andaman Islands 4 4.5 Bombay, India

Galle, Sri Lanka 5 4.5 Mauritius Is.
Rodriguez Is.
Negombo, Sri Lanka 7 5
5 Galle, Sri Lanka
Bombay, India 11 8.6 Negembo, Sri Lanka
Port Blair Geraldton, Australia
Aden 12 10.17
Rodriguez Islands 6.5 6.67 Cossack, Australia
Mauritius Islands 6.5 7.3
Port Elisabeth, South Africa 13.5 12 0
Cossack, Australia 5.25 3 0 5 10 15
observed travel time, hr
Geraldton, Australia 9.25 4
Williamstown, Australia 15.6 8.5
Auckland, New Zealand 37 12.7 Fig. 6. Comparison of the observed and calculated travel times of
Havre, France 31.5 31.7 the first wave in the basin of the Indian Ocean.
San Francisco, USA 16 26
Kodiak, Alaska, USA 23 27.7
Honolulu, Hawaii, USA 10 23 served values for the first (leading) wave, given in Sect. 2.
Colon, Panama 17.5 17 Calculated and observed travel times are summarized in Ta-
Port Mollke, South Georgia Is. 13.5 17
ble 1. First of all the comparison can be done for the coast
Talcahuano, Chile 34 21.5
of the Indian Ocean (Fig. 6). The agreement is quite well,
mainly for points, where the tsunami waves approach on the
frontal direction (southern coast of Sri Lanka, Mauritius and
Rodriguez Islands, Aden, Port Elizabeth). The western coast
time some additional rays were used. The bathymetry is of Sri Lanka and India is in the “dark” zone of the ray pat-
taken from the 2-min ETOPO2 dataset (Smith and Sandwell, tern. Here another scenario of the tsunami waves propaga-
1997). tion in the form of the edge waves should be realized, and
Results of computing of the tsunami pathways are shown the actual arriving time should be increased, as it is observed
in Fig. 4. Tsunami waves affect a whole basin of the In- for Negombo and Bombay. The same effect have to be im-
dian Ocean, and also they pass into the Pacific and Atlantic portant for the Australian points where the observed travel
Oceans. These two pathways are well shown in Fig. 5; time is also large to compare with ray predictions. For in-
one way is the propagation to the Atlantic Ocean between stance, this difference reaches 7 h for the tsunami waves at
Antarctic and South Africa, and another one is the propaga- Williamstown (near Melbourne), see, Table 1. Our calcula-
tion to the Pacific between Antarctic and Australia and also tions in the framework of the ray tracing method do not take
between Antarctica and Southern America. Most of energy into account the wave propagation in the form of the edge
passed to the Atlantic Ocean to compare with the Pacific waves leading to increase the tsunami travel time. Formally,
Ocean. Early such pathways were not calculated because the large travel time to compare with prediction of the ray
Antarctic waters were unsurveyed. Also the transmission theory is no contradiction because sometimes the first waves
of the tsunami energy over these paths is doubtful (Ewing are not visible.
and Press, 1995; Simskin and Fiske, 1983). Using numerical For transoceanic tsunami propagation the comparison of
modeling with good sea-floor topography the tsunami path- observed and calculated travel times is presented in Fig. 7.
ways can be calculated with high accuracy. The results of the Surprisingly, the calculated arrived time for Le Havre, France
performed calculations show that the tsunami waves affect is the same as the observed one, meanwhile the wave passed
the coasts of the Indian Ocean, and the waves approach to the almost 18 000 km. Relative small disagreement is for Port
South America, both, Atlantic and Pacific coasts. The North Molke, South Georgia Islands and for St. Paul’s, Kodiak,
America is more protected from the Krakatau tsunami, and approximately 3–4 h. Maximal disagreement is for the New
the ray density determined the wave amplitude is relatively Zealand data where observed time is too large to compare
small. Other areas in the Pacific (western coast) and Atlantic with calculated one. Also the tsunami waves should arrive
(European coast) are almost not affected by tsunami. Also, it in Talcahuano (Chile) 12 h early that it is observed. In Hon-
can be noted that waves do not pass significantly through the olulu, San Francisco and Panama the tsunami waves came
Indonesian archipelago into the Pacific, and this is a natural significantly early that it is expected according to the ray
barrier for tsunami. theory. It means that such “early” disturbances cannot be
The ray tracing method allows calculating the tsunami connected with the direct sea wave propagation from the
travel time; these isochrones with 1 h interval are shown in Krakatau Island. This was concluded early using the rough
Fig. 4. The calculated travel times can be compared with ob- estimation of the tsunami travel time (Simskin and Fiske,
B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation 327

40

35
Havre, France
calculated travel time, hr

30 Colon, Panama
Kodiak
25 San Francisco
Honolulu
20 Talcahuano, Chile
South Georgia Is.
15 Auckland, New Zealand
Port Elizabeth, South Africa
10
Williamstown, Australia
5
Geraldton, Australia

0
0 5 10 15 20 25 30 35 40
observed travel time, hr

Fig. 7. Comparison of the observed and calculated travel times of

the first wave in globe.

1983) and here confirmed using detailed numerical simula-

tions on the 2 min bathymetry. The existing of such distur-
bances is assumed with the impact of the significant acoustic- Fig. 8. Snapshots of tsunami propagation with interval of 6 h.
gravity waves in the atmosphere generated at the volcano ex-
plosion on the sea surface and possible coupling between two
wave systems (Ewing and Press, 1995; Press and Harkrider, integral characteristics (period, amplitude) on the waves go-
1966; Garret, 1976; Lander and Lockridge, 1989). ing out the Sunda Strait only, not by the details of tsunami
generation. In such cases the conception of the equivalent
source can be used, when the real process of the tsunami
4 Hydrodynamic long wave modeling generating is parametrized by the initial displacement and
zero flow distribution. Then the hydrodynamic equations are
As it is known the ray tracing method is effective only to
solved as the Cauchy problem. Such model has been already
calculate the travel time, but not the wave amplitude. For
used for the Krakatau tsunami by Nakamura (1984), and his
evaluation of the tsunami parameters, its transoceanic propa-
source is a square of about 44×44 km2 (four grid points) and
gation from the Krakatau volcanic eruption is studied in the
depression 0.1 m (in fact, his numerical computation shows
framework of the linear shallow water theory on the spherical
that initial disturbance should be characterized by 700 m to
earth,
explain the observed data in the Sunda Strait). According to
 
∂η 1 ∂M ∂ Bryant (2001) during the third explosion Krakatau collapsed
+ + (N cos θ ) = 0, (4) in on itself, forming a caldera about 270 m deep and with
∂t R cos θ ∂ϕ ∂θ
∂M gh ∂η a volume of 11.5 km3 . In our computations, the equivalent
+ = f N, (5) source has dimensions 7.2×7.2 km2 (nine grid points) with
∂t R cos θ ∂ϕ
the center having coordinates: 6◦ 100 S and 105◦ 100 E (depth
∂N gh ∂η
+ = −f M. (6) is of about 400 m), and the initial depression of the shape
∂t R cos θ ∂θ form is 222 m. Such conditions provide the observed volume
where M(θ, ϕ, t) and N(θ, ϕ, t) are the flow discharges in 11.5 km3 , but not observed diameter in 6 km; this is related
meridional and zonal directions, η(θ, ϕ, t) is the water dis- with mesh resolution of the numerical scheme. In fact, the
placement and f is the Coriolis parameter. The total grid detailed features of the equivalent source should not be very
number on globe is 49 263 361 (10 801×4561) with the 2 min important for tsunami simulation on long distances.
mesh resolution (ETOPO2) and time step used was 5 s. On Snapshots of the tsunami propagation are presented in
the land boundaries the reflected condition is applied. Fig. 8 with the time interval 6 h. These computations are used
The modeling of the tsunami sources of the volcanic erup- to determine the characteristics of the wave field on globe.
tion is extremely difficult task and three hypotheses: caldera The characteristics of the relative energetic waves (with the
collapse, submarine explosion and pyrolastic flow are ac- crest amplitude more 0.1 m) are given in Table 2. First of all,
tively considered for the Krakatau event (Francis, 1985). Re- the results of the calculations of the travel time for the leading
cently Nomanbhoy and Satake (1995) concluded that the wave obtained by the ray tracing method and the hydrody-
submarine explosion model as the source of the largest namic modeling should be compared. Figure 9 shows a quite
tsunami is favored. The global behavior of the tsunami waves good agreement between two methods for the first wave ar-
during the 1883 Krakatau event should be determined by the riving. As it is expected, the hydrodynamic modeling should
328 B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation

Table 2. Computed travel time and the maximum crest amplitude

Arriving Time of Arriving Time of Maximal crest

Leading Wave, h Maximal Wave, h amplitude, m
Geraldton, Australia 4.3 7.5 1.00
Cossack, Australia 4.6 7.2 1.10
Perth, Australia 4.9 24.0 0.83
Batticaloa, Sri Lanka 5.0 14.6 0.47
Port Blair, Andaman Isl. 5.2 15.1 0.44
Colombo, Sri Lanka 5.3 38.0 0.47
Banjermasin, Kalimantan, Indonesia 6.6 11.1 0.62
Rodriguez lsl., Mauritius 6.8 7.0 0.16
Mauritius lsl. 7.6 7.8 0.77
Bombay, India 9.3 37.2 0.52
Sydney, Australia 11.1 34.8 0.16
Ho Chi Min. Vietnam 11.9 32.2 0.17
Port Elizabeth, South Africa 12.5 13.2 1.19
Auckland, New Zealand 13.3 33.0 0.19
Aden, Yemen 13.6 27.8 0.42
Cape Town, South Africa 13.7 16.9 0.66
Wellington, New Zealand 14.1 34.0 0.19
South Georgia lsl. 17.3 31.7 0.22
Luanda, Angola 18.7 32.0 0.19
Stanley, Falkland Isl. 20.1 33.2 0.16
Lagos, Nigeria 20.3 36.3 0.15
Montevideo, Uruguay 21.2 33.6 0.38
Rio de Janeiro, Brazil 22.0 23.1 0.18
Dakar, Senegal 23.2 31.4 0.13
Belem, Brazil 25.4 36.9 0.14
Halifax, Canada 29.7 36.5 0.13
St. Johns, Canada 29.8 32.9 0.11
New York, USA 30.6 35.8 0.22
Jacksonville, Florida, USA 31.0 35.4 0.11

give largest values for travel time than the ray method for
35
“dark” zones, where waves propagate as the edge waves. For
instance, the tsunami traveled to Bombay for 11 h according 30
to the observations; meanwhile the ray method estimated the
time, hr (hydro model)

travel time as 8.6 h. The hydrodynamic modeling gives 9.3 h, 25

close to the observed value. Another example is the tsunami 20

propagation along the Australian coast; the travel time is in-
creased in the hydrodynamic model on 1 h comparing with 15
predictions of the ray theory (but it is not enough for explain
10
the observed travel time).
5
According to calculations, the first wave is not maxi-
mal wave, which arrived significantly later (expect a few 0
of coastal locations: Mauritius and Rodriguez Islands, Port 0 5 10 15 20 25 30 35
Elizabeth). In some locations, the maximal wave arrives a time, hr (ray method)
day later, confirming that the seiche oscillations induced by
tsunami may continue 1–2 days. Applied numerical model Fig. 9. The comparison of the travel time of the first wave computed
does not include the bottom friction and wave breaking on by two methods: ray tracing and direct simulation.
the beach (the reflected boundary condition is used for all
coastal lines) and this can influence on wave dynamics in-
creasing wave amplitudes for large times as well as the total investigated.
duration of the tide-gauge record. This important problem Computed wave amplitudes exceed 0.1 m in many loca-
of appropriate “land” boundary conditions is not quite well tions of the World Ocean, not only in the Indian Ocean
B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation 329

Table 3. Computed and observed tsunami wave amplitudes 1

Mauritius Island

Observed wave Calculated crest 0.5

amplitude, m amplitude, m

elevation, m
Cossack, Australia 1.5 1.1
Port Blair, Andaman Isl. 0.2 (tide-gauge) 0.4 0
Colombo, Sri Lanka 1 (runup) 0.5
Rodriguez lsl., Mauritius 1.8 (runup) 0.2
-0.5
Mauritius lsl. 0.9 (runup) 0.8
Bombay, India 0.56 (tide-gauge) 0.5
Port Elizabeth, South Africa 0.7 (tide-gauge) 1.2
-1
Auckland, New Zealand 1.5 (runup) 0.2
6 7 8 9 10 11 12
Aden, Yemen 0.24 (tide-gauge) 0.4 time, hr
South Georgia Islands 0.3 (tide-gauge) 0.2
Fig. 11. Computed tide-gauge record for Mauritius Island.

0.6
Negombo
0.4

0.2
elevation, m

-0.2

-0.4

-0.6 Fig. 12. Comparison of the computed and observed tide-gauge

5 6 7 8 9 records for Port Blair, Andaman Islands.
time, hr

Fig. 10. Computed tide-gauge record for Negombo, Sri Lanka. and tide-gauge records. First of all, the witness reports will
be analyzed. According to the observations (Sect. 2), there
are about 16 waves during three hours, and the first wave was
(Table 2). For instance, on American Atlantic coast tsunami negative. The computed tide-gauge record for Negombo, Sri
the amplitudes are 0.1–0.4 m in Canada, USA, Brazil and Lanka presented in Fig. 10 shows the same features of sea
Uruguay with local highest amplitude 0.38 m in Montevideo. level oscillations during this period. Observed runup height
All these locations are in the zone of frontal approaching is about 1 m, meanwhile our calculations in the last “sea”
according to the ray pattern (Fig. 4). Tsunami waves with point gives 0.4 m only. Such disagreement is typical for com-
heights of 0.1–0.2 m are found in calculation for the Atlantic parison of the tide-gauge and runup data. More important is
African coast (Senegal, Nigeria and Angola). These loca- the coinciding of the characteristic wave periods, which is of
tions are in the “dark” zone of the wave rays, and tsunami 10–15 min. The same conclusion can be done for Mauritius
waves propagate here mainly as the edge waves. It is in- Island (Fig. 11), characteristic period is of 15 min. So, the
teresting to mention that the “computed” waves pass through hydrodynamic model explains the observed wave period at
Sunda Strait and have significant amplitudes on the Kaliman- different locations in the Indian Ocean.
tan Island (0.62 m). According to the observations tsunami Simskin and Fiske (1983) have collected tide-gauge
was recorded on many Indonesian Islands in the Java Sea records of the Krakatau tsunamis for two locations at the
with the height about 0.5–1 m correlated with computed data. Indian Ocean (Port Blair, India and Port Elizabeth, South
The wave reached also Vietnam (0.17 m) but we have no ob- Africa), two locations at the Atlantic Ocean (Moltke Harbor,
servations for this coast. The rough comparison of the calcu- South Georgia Islands and Colon, Panama) and three loca-
lated and observed wave amplitudes at several coastal loca- tions at the Pacific Ocean (Honolulu, San Francisco and Ko-
tions is given in Table 3. In average, computed wave ampli- dial, USA). The quality of these reproduced “paper” records
tudes are in reasonable agreement with observed tide-gauge is not quite good to digitize and to analyze. Nevertheless, all
data, and they are less in several times than the runup heights. these records were digitized and the tide component (calcu-
More detail comparison can be done for coastal locations lated by smoothing of real record) was eliminated for com-
with detail information about tsunami waves: witness reports parison with the computed time series. Figure 12 shows re-
330 B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation

Fig. 13. Comparison of the computed and observed tide-gauge Fig. 14. Comparison of the computed and observed tide-gauge
records for Port Elizabeth, South Africa. records for Port Moltke, South Georgia Islands.

sults of comparison for Port Bair, Andaman Islands. The

character of the first (leading) wave in observation and com-
puting is the same (form, arriving time and crest amplitude),
but totally the curves are different. First of all, observed
tide-gauge record has the characteristic period of 1 h, mean-
while the computed wave record is about 15 min. The large
difference in the computed and observed periods needs to
be explained. According to the calculations by Nomanbhoy
and Satake (1995), the characteristics period of the first two
tsunami waves near the Krakatau Island, in the Sunda Strait
is about 1 h (but the wave form is far from simplest sine), Fig. 15. Comparison of the computed and observed tide-gauge
and, therefore, this low-frequency spectral component should records for San Francisco, USA.
be in the tsunami spectrum on any distance from the source,
including the Port Blair. The computed tide-gauge record
has the wide spectrum, and this low-frequency component is coastal locations on American coast are induced by the air-
weak. The characteristic period (15,min) obtained in com- pressure waves generated at the Krakatau eruption in the at-
puting corresponds to the witness reports in all areas of the mosphere. Our calculations in the framework of the hydrody-
Indian Ocean, and, therefore, this high-frequency spectral namic theory with more accurate bathymetry for all locations
component is also a part of the real spectrum of tsunami. mentioned by Ewin and Press (1955) do not rebut their point
Perhaps, the tide-gauge station eliminates the high-frequency of view.
components due to its frequency characteristics. The same
difference can be pointed out for Port Elizabeth, South Africa 5 Discussion and conclusions
(Fig. 13). The high-frequency computed record has the in-
tense leading group and relative weak seiche oscillations, The numerical simulation of the tsunami propagated from
whereas the observed record contains large low-frequency the Krakatau Island has been done with very high accurate
oscillations during one day. The comparison of the time se- bathymetry from the 2 min ETOPO2 dataset in the frame-
ries with different spectra is impossible. We suggest investi- work of the ray tracing method and the direct hydrody-
gating the influence of frequency filtration (depending from namic theory based on shallow-water equations. Both mod-
its transfer function) on the waveform for some model and els demonstrate that tsunami wave penetrate in the Atlantic
real situations in future. and Pacific Oceans from the Indian Ocean, and the results
Comparison of the computed and observed tide-gauge of calculations of the travel time within both theories are in
records for Port Moltke, South Georgia Islands Additional good agreement between them. The observed travel times
problem to compare the computed time series and observed for coastal locations in the Indian Ocean are explained by
tide-gauge records for the coastal locations in the Atlantic the computing very well. The observed travel times for
and Pacific Oceans is related with large difference between other areas of the World Ocean are in contradiction with
travel times, see Fig. 14 (Port Moltke, South Georgia Islands) the theoretical predictions, except Le Havre, France (about
and Fig. 15 (San Francisco). Tsunami waves approached to 18,000 km from Krakatau). Observed sea level disturbances
most of coastal locations significantly early than it is pre- in New Zealand occurred significantly later (one day) than
dicted by the hydrodynamic theory (except Le Havre, France, predicted. Because the water level oscillations observed at
where the comparison is quite well). Almost 50 years ago, this time also in the Lake Taupo (center of the North Is-
based on rough estimates of the travel time, Ewing and Press land) closed from the sea, our calculations confirm the pre-
(1955) suggested that observed sea level disturbances at the vious conclusion by Bryant (2001) than these disturbances
B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation 331

were not induced by the Krakatau eruption. Sea level dis- should arrive significantly early than observed sea level dis-
turbances on the American coast (Honolulu, San Francisco, turbances. Our calculations confirm the opinion by Bryant
Kodiak, Colon, Port Moltke), opposite, observed early than it (2001) that tsunami in New Zealand has another origin.
is predicted and, therefore, cannot be induced by the tsunami The last conclusion concerns the characteristic periods of
waves approaching from the Krakatau. The physical mech- the tsunami waves. In computing, the visible period is about
anism of their appearance is perhaps the “air” mechanism 15 min, and this coincides with observations made by wit-
discussed by Ewing and Press (1955), Press and Harkrider nesses. The observed tide-gauge records have the dominant
(1966) and Garret (1976). Sea level disturbances accord- period of 1 h. The real frequency characteristics of the tide-
ing to this hypothesis were transferred to the water from the gauge stations are unknown that to apply any filter procedure
barometric perturbations in the atmosphere. It is based on to cut parasitic frequencies. We suggest performing special
the almost the same time of wave arriving on barometers and experiments with filtration of tsunami waves in future.
tide-gauges on the American coast. The resonance mecha- Finally, we may say that the hydrodynamic model explains
nisms of the tsunami waves by the moving atmospheric dis- reasonably the tsunami waves from the Krakatau eruption in
turbances like cyclones are very popular to explain so called the Indian Ocean. Tsunami waves according to calculations
meteo tsunamis (Murty, 1977; Pelinovsky, 1996). Very in- may have amplitudes more 10 cm in the Atlantic Ocean, but
tense acoustic-gravity waves generated by the Krakatau vol- unfortunately there is no data of tsunami observation in se-
canic eruption may have the phase speed comparable with lected locations. Computed tsunami waves are too weak in
the water wave speed due to the stratification of the earth at- the Pacific Ocean, except New Zealand, that to induce visible
mosphere and as a result, there is a coupling between two sea level disturbances.
wave systems. On the shoreline and continental slope the
Acknowledgements. This work was supported by the Korean Sci-
large transformation between waves is possible providing the
ence and Engineering Foundation (KOSEF) through the Korean
local disturbances of sea level on tide-gauge stations. Earthquake Engineering Research Center (KEERC) at Seoul Na-
According to the calculations, the tsunami waves are sig- tional University. Particular support to EP was obtained through
nificant in the Indian Ocean having amplitudes exceeded RFBR grant 02-05-65107 and EGIBE Program. Thanks are also
1 m on the northwestern coast of Australia and southeast- due to F. Imamura, Tohoku University and K. Satake for providing
ern coast of Africa. According to the observations (Sect. 2) us with hydrodynamic and ray tracing codes respectively, and Ch.
the large tsunami waves were observed on all coasts of In- N. Go providing the unpublished manuscript by Soloviev and Go.
dian Ocean: Australia, India, Sri Lanka, Mauritius, and
South Africa. Computing explains the ytavel time, observed
group character of tsunami waves, and visible wave period References
(15 min) by witnesses, but not observed tide-gauge (see be- Bryant, T.: Tsunamis, Cambridge University Press. 2001.
low). Computed tsunami wave amplitudes in the Java Sea Choi, B. H., Kim, Y. K., and Kim, K. W.: Visualization of a tsunami
reached 70 cm correlating with observed data in 0.5–1 m. due to 1883 Krakatau eruption, Proceedings of Korean Computer
In the Atlantic Ocean, the computed wave amplitudes Graphics Society, 22–26, 1997.
exceed 10 cm in South America (South Georgia Islands, Ewing, M. and Press, F.: Tide-gauge disturbances from the Great
Uruguay, Brazil), USA coast (Florida, New York), Canada Eruption of Krakatoa, Transactions of the AGU, 36, 53–60, 1995.
(Halifax), and African coast (Angola, Nigeria and Senegal). Francis, P. W.: The origin of the 1883 Krakatau tsunamis, J. Vol-
Unfortunately, there is no data of tsunami observation on the canol. Geotherm. Res., 25, 349–369, 1985.
eastern coast of USA, Canada and Africa. For South Amer- Garret, C. J. R.: A theory of the Krakatoa tide-gauge disturbances,
Tellus, 22, 43–52, 1976.
ica, there is only one tide-gauge record is available (Port
Kawamata, S.; Imamura, F., and Shuto, N.: Numerical simulation of
Moltke, Fig. 14), and it shows that tsunami appears early on
the 1883 Krakatau tsunami, Proceedings of 20th IAHR, Tokyo,
several hours then predicted. Perhaps, the first sea-level dis- Japan, 1992.
turbances at Port Moltke were induced by air-pressure waves Lander, J. F. and Lockridge, P. A.: United States Tsunamis, 1690–
as it was suggested by Ewing and Press (1955), but the fol- 1988, National Geophysical Data Center, Boulder, 1989.
lowing sea level oscillations were related with approaching Lander, J. F., Whiteside, L. S., and Lockridge, P. A.: A brief history
of tsunami waves; observed and computed wave amplitudes of tsunamis in the Caribbean Sea, Science of Tsunami Hazards,
are in good agreement. For the Caribbean Islands and the 20, 57–94, 2002.
English Channel, where some unusual sea level disturbances Murty, T.: Seismic Sea Waves – Tsunamis, Bull. of Dep. Fisheries,
were observed, computations give weak amplitude waves, Canada, 1977.
less 10 cm. Nakamura, S.: A numerical tracking of the 1883 Krakatau tsunami,
Science of Tsunami Hazards, 2, 41–54, 1984.
In the Pacific Ocean, except New Zealand, the computed
Nomanbhoy, N. and Satake, K.: Generation mechanism of tsunamis
tsunami waves are too small to compare with observed sea
from the 1883 Krakatau eruption, Geophys. Res. Letters, 22,
level disturbances in USA and arrived significantly later than 509–512, 1995.
observed. As we mentioned, our calculations do not rebut the Pelinovsky, E.: Hydrodynamics of Tsunami Waves, Applied
hypothesis by Ewing and Press (1955) of another origin of Physics Institute Press, Nizhny Novgorod, 1996.
observed sea level disturbances. For New Zealand locations Press , F. and Harkrider, D.: Air-Sea Waves from the Explosion of
computed tsunami amplitude is about 20 cm, but these waves Kralatoa, Science, 154, 1325–1327, 1966.
332 B.H. Choi et al.: Simulation of the trans-oceanic tsunami propagation

Satake, K.: Effects of Bathymetry on Tsunami Propagation: Ap- Sobel, P. A. and von Seggern, D. H.: Application of surface-wave
plication of Ray Tracing to Tsunamis, PAGEOPH, 126, 27–36, ray tracing, Bull. of Seism. Soc. Am., 68, 1359–1380, 1978.
1988. Soloviev, S. L. and Go, Ch. N.: Catalog of tsunamis in western coast
Simskin, T. and Fiske, R. S.: Krakatau 1883 – the volcanic eruption of the Pacific Ocean, Moscow: Nauka, 1974.
and its effects, Smithsonian Institution Press, Washington, D. C., USSR Navy: Pacific Ocean (Series of the Atlases of the Oceans),
1983. Moscow, 1974.
Smith, W. H. F. and Sandwell, D. T.: Global Sea Floor Topography Yokoyama, I.: A geophysical interpretation of the 1883 Krakatau
from satellite altimetry and Ship depth soundings, Science,, 277, eruption, J. Volcanology and Geothermal Research, 9, 359–378,
5334, 1997. 1981.