You are on page 1of 11

IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 40, NO.

1, JANUARY 2015 37

Peer-Reviewed Technical Communication


Forecast System for Offshore Water Surface Elevation With
Inundation Map Integrated for Tsunami Early Warning
Guan-Yu Chen, Member, IEEE, Chin-Chu Liu, and Cheng-Chung Yao

Abstract—In this study, a forecast system for tsunami early tsunami should be evaluated promptly so that a proper coun-
warning in Taiwan is constructed by integrating the source model termeasure can be employed and both casualties and property
that generates the initial tsunami elevation based on earthquake losses can be reduced.
fault parameters, the propagation model that calculates the
offshore sea surface quickly by precomputed reciprocal Green’s In Taiwan, there are a few tsunami events which have hap-
functions, and the nearshore module that interpolates the inun- pened in the past [1]. For the southwest coast of Taiwan, a
dated distance to the shoreline based on precomputed inundation tsunami was reported in 1781. Some detailed descriptions of
simulations. This system is integrated and executed by a Matlab this event suggest the event was hazardous. The record has it
graphic utility interface (GUI). In this interface system, the rup- that “after the water flowed back to the sea, people were stuck on
ture parameters such as epicenter location, focal depth, and fault
angles are used to compute the initial tsunami wave elevation. top of tall bamboo trying to find others to help them get down to
These rupture parameters can be retrieved from the website of the ground.” When the fishermen came back after fishing, “they
the U.S. Geological Survey or the F-net Broadband Seismograph found the houses were submerged and the fishing rafts could sail
Network of Japan for the most updated earthquake parameters. over bamboo.” These descriptions suggest that the water depth
For the highly nonlinear coastal inundation, a bell-shaped wave is was higher than 4 m. The fishing rafts went out to sea before
used to generate the inundation map, which is then transformed
into a Google Earth format and demonstrated via this geographic the tsunami came; therefore, it was a fair day, and, hence, this
information system. This system has been applied in both 2010 tsunami was not a disguised storm surge.
Chilean Tsunami and 2011 Great Eastern Japan Tsunami. In In [1], two more tsunamis were reported in the southwestern
both cases, the result has order-of-magnitude accuracy; thus, this and northern coasts of Taiwan: In January 1661, a large tsunami
system provides a quick first estimation for an early warning of a assailed the Tainan area and flooded lots of huts. In December
tsunami when a submarine earthquake occurs.
1867, hundreds of people drowned in Keelung when seawater
Index Terms—Geographic information system, inundation map, deluged the land. These historical events are important because
reciprocal Green’s function, tsunami early warning. if hazardous tsunami occurred in the past, they can also occur
in the future.
In a broader view than the tsunami events of far away or long
I. INTRODUCTION ago, large earthquakes frequently occur in the area surrounding
Taiwan. In the Chi-Chi earthquake of 1999 in the central area of

S UBSEQUENT to the Indian Ocean Tsunami of December Taiwan, the largest fault displacement on the ground reached 10
2004, there have been quite a few tsunami events around m. This scale of vertical bottom movement can generate giant
the world. More recently, the Samoan Tsunami in 2009, tsunami if it occurs in the sea. Tsunami events in the neigh-
Tsunami incidents in Chile and Indonesia in 2010, and the dev- boring countries such as Japan, the 1993 tsunami of Okushiri,
astating Great Eastern Japan (Tohoku) tsunami in March 2011 Hokkaido, and the Moro Gulf Tsunami, 1976, in the Philippines,
all highlight the risk of tsunami in coastal zones. To protect have claimed thousands of lives in the last century. Tsunami
coastal residents from a tsunami, the hazard of an upcoming generated in these countries may also affect the coast of Taiwan.
Thus, a tsunami warning system is necessary based on both the
historical events and the geological structure nearby.
Manuscript received September 16, 2012; revised June 10, 2013 and De-
cember 04, 2013; accepted December 12, 2013. Date of publication January Since most tsunamis are generated by submarine earthquakes,
30, 2014; date of current version January 09, 2015. usually a tsunami alarm is announced after an earthquake is
Associate Editor: N. R. Chapman.
detected and its magnitude is determined. Most big submarine
G.-Y. Chen and C.-C. Liu are with the Department of Oceanography, Na-
tional Sun Yat-sen University, Kaohsiung 80424, Taiwan (e-mail:guanyu@fac- earthquakes do not generate large tsunami, and, hence, most of
ulty.nsysu.edu.tw; caite8001@hotmail.com). the tsunami warnings turn into false alarms.
C.-C. Yao is with the Department of Civil Engineering, Chienkuo Technology
To make a reliable alarm, a quicker and more convenient early
University, Changhua 500, Taiwan. (e-mail: molie@cc.ctu.edu.tw).
This paper includes the supplementary downloadable material available at warning system with more accurate numerical simulations is es-
http://ieeexplore.ieee.org, provided by the authors. The material presents a com- sential. However, even after collecting the required earthquake
plete video demonstration of the interface system explained in this paper. Con-
parameters, numerical tsunami simulations take hours before it
tact G.-Y. Chen at guanyu@faculty.nsysu.edu.tw for questions regarding the
multimedia material. is feasible to forecast a tsunami. If the simulation is to be exe-
Digital Object Identifier 10.1109/JOE.2013.2295948 cuted beforehand, the number of combinations for all possible

0364-9059 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
38 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 40, NO. 1, JANUARY 2015

earthquake magnitudes, hypocenter locations, and fault angles not generated because the tsunami height is too small to flood
is too large to be completely included. In practice, this approach the coast. Thus, this system provides a quick and effective first
needs lots of simulations over various seismic scenarios and estimation when a submarine earthquake occurs.
then interpolates between different orders of magnitude [2].
Another common approach utilizes the numerical Green’s II. FORECAST METHOD
function (GF), which is the response of unit elevation at any Tsunami simulation can be divided into three steps: the
point of interest based on linear equations for tsunami. Once the source, the propagation in the deep ocean, and the inundation
GF is computed for every point in the source zone, the response in the nearshore area. In this study, these three steps are carried
of any point in the computational domain can be obtained out separately by three modules that are integrated to forecast
simply by multiplication with the GF and the initial vertical both the offshore water surface elevation and the inundation
deformation, which is then superposed over the whole source map for a vulnerable city. The source model that generates the
zone [3]–[5]. Thus, with the aid of previously computed GFs, initial tsunami elevation based on earthquake fault parameters,
a tsunami warning can be issued quickly because no equation the propagation model that calculates the offshore sea surface
solving is involved. quickly by precomputed RGFs, and the nearshore module that
The GF approach is not perfect for the following three rea- interpolates the inundated distance to the shoreline based on
sons. First, the accuracy always depends on an accurate initial precomputed inundation simulations are integrated into a fore-
sea-bottom dislocation in the source zone, a common problem cast system for tsunami early warning. These three modules,
encountered in all tsunami simulations. In this study, the length with the integration under a graphic utility interface (GUI), are
and width of the fault plane are obtained from a widely discussed in detail in the following. It should be noted that the
used regression formula of Wells and Coppersmith [6], and a present system does not include in situ tsunami measurement
uniform dislocation is obtained based on the equivalence of the in its present form.
seismic moment and the mechanical work accomplished [7].
The sea-bottom elevation then is obtained by the elastic half- A. Tsunami Source Forecast
space dislocation model according to existing elasticity theo- For a thrust fault where the slip angle is between 45 and
ries of Mansinha and Smylie [8]. The details will be given in 135 , the most frequently used regression formulas of Wells and
Section II-A, and the results have at least order-of-magnitude Coppersmith [6], in terms of the moment scale magnitude of
accuracy. the earthquake , are employed. The surface rupture length
The second problem of the numerical GF approach is that (SRL) and rupture width (RW, both in kilometers) are used,
many computation cases, as many as the number of grid points respectively, to represent the length and width of the fault
in the computational domain, have to be executed to obtain all plane as
the required numerical GFs. Since the purpose of the GF ap-
proach is to forecast the tsunami wave height for a vulnerable SRL
coastal city, most of the computation and the requirements on RW (1)
computer memory are unnecessary. To save computation time
and reduce the memory requirement, reciprocal GF (RGF) [10] Similarly, for a normal fault where the slip angle is less than
is used. The tsunami forecast by RGF is as quick as the tradi- 45 or more than 135 , the regression formulas are
tional GF, but the computation is significantly reduced, as will SRL
be discussed in Section II-B.
The third problem of the GF approach is that a tsunami wave RW (2)
is highly nonlinear near shore; hence, the GF is not applicable For slip angle outside the above ranges, the fault is classified as
for the detailed fine-grid nearshore inundation simulation strike slip, and the regression formulas are
because of its nonlinear nature. Chen et al. [11] proposed
a methodology to quickly estimate the runup flow field and SRL
inundation distance from the shoreline based on the GF and RW (3)
an analytic solution of Carrier et al. [12]. However, although
the calculated runup height has reasonable agreement with the These three different fault types have different rupture mecha-
field survey, the inundation distance thus obtained varies sig- nism and stress distribution; hence, the corresponding regres-
nificantly, suggesting the dependency on the local topography, sion equations are different. By changing to the seismic
and, hence, a detailed inundation simulation is required. In moment via
this study, a bell-shaped wave is used to get detailed nearshore
inundation simulations for some specific wave heights and (4)
apply interpolation to obtain the inundation map for every
incident wave, as will be discussed in Section II-C. a uniform dislocation is obtained by equating the seismic
To verify the method, tsunami data in Taiwan obtained moment to the mechanical work released
during both 2010 Chilean tsunami and 2011 Great Eastern (5)
Japan Tsunami are used for validation in Section IV. The
forecast for the offshore tsunami elevation is shown to be where is the shear modulus of the crust and is usually assumed
accurate to the order of magnitude, while the inundation map is to be 3 10 N/km [7].
CHEN et al.: FORECAST SYSTEM FOR OFFSHORE WATER SURFACE ELEVATION WITH INUNDATION MAP INTEGRATED FOR TSUNAMI EARLY WARNING 39

The forecast is very quick because no equation solving is in-


volved.
The traditional numerical GF is the response of the whole
domain to the forcing at one source point. The wave propagates
from source to receiver . Its reciprocal (RGF) signifies the
response at for the unit elevated water surface (forcing) at
point . Loomis [10] noted that the traditional GF and the RGF
are symmetric with respect to and for free-surface elevation
(see also [15]). That is, . This reciprocity
property was proposed by Loomis as an efficient method for
tsunami forecast. The tsunami height then can be forecast in a
way similar to the traditional GF as
Fig. 1. Fault model that comprises three fault angles strike, dip, and slip.
(8)
The displacement at the sea bottom is then obtained by the
elastic half-space dislocation model of existing elasticity theory The concept is verified and extended to evaluate the efficiency
according to the fault plane model with three fault angles, as of the location of tsunami sensors [16]. For the Indian Ocean
explained in Fig. 1. The formulation is too lengthy to be shown Tsunami, the accuracy of the RGF method is also verified in
in this paper, and the readers are referred to [8]. This sea-bottom Fig. 2. The spatial resolution is 4 min, and the RGF location
dislocation is directly used as the initial sea-surface elevation is (82 E, 6.8 N) off the east coast of Sri Lanka. The fault
of (7), because the water wave celerity is much smaller than the parameters used in the tsunami source and other details of the
velocity of elastic seismic waves [9]. comparison can be found in [11].
It should be noted that the RGF has to be modified under a
B. RGF in Tsunami Propagation
spherical coordinate system. The GF is forced by elevated water
Tsunami propagation in the deep ocean, including the numer- mass at the source, and hence is proportional to the area of the
ical GF and its reciprocal, is computed with a linear governing source grid. For spherical coordinate system, the receiver grid
equation such as the following linear shallow-water equations: and the source grid have unequal grid area, and, hence, the GF is
not exactly reciprocal unless the RGF is multiplied by the area
ratio of the source grid to the receiver grid. For example, in the
case of Chilean tsunami shown in Section III-A, the tsunami
source is near 36 S and the vulnerable cities in Taiwan are near
24 N. The area ratio of the source grid to the receiver grid is
(6) the ratio of (36 )/ (24 ), which is approximately 0.9, and,
hence, the RGF in Chile area is multiplied by 90%.
where is the free-surface elevation, and are fluxes along The tsunami forecast by the RGF is as quick as the traditional
the - and -directions, is time, is gravity acceleration, and GF, but the computation requirement is significantly reduced: In
is the water depth taken from the real bathymetry. The initial the traditional GF approach, many computation cases, as many
condition for a GF is given by assuming zero initial velocity and as the number of grid points, have to be evaluated to obtain
zero free-surface elevation everywhere except for the source the numerical GF of each grid point . That means hundreds of
grid point , where the velocity is still zero but the free-sur- thousands of traditional GFs are needed for a huge tsunami. On
face elevation is unity. The evolution of the initial condition the other hand, for the RGF approach, only one RGF is required
with time is then calculated by a modified leapfrog finite differ- for a receiver grid point or a vulnerable city. A great deal of
ence scheme in the validated Cornell multigrid coupled tsunami computation time and computer storage is saved if the RGF is
model (COMCOT) [13], [14]. The numerical response of all used.
ocean points in the computational domain is the desired GF. For example, a 25-h RGF that covers the whole Pacific basin
Once the GF is computed for every point in the source zone, with 1-min resolution in time and 1/30 resolution in space re-
the response of any point in a tsunami incidence can be obtained quires 130 GB to store in the netcdf file format. If the same cov-
simply by multiplying the GF and the initial vertical deforma- erage and resolution are applied to the traditional GF, millions
tion , then superposing over the grid points of the whole of hard disc drives have to be used to store these GFs, and it is
source zone [3], [5]. For example, let be the location of the impossible. Although the disk space for GF approach can be as
source point and a location near the vulnerable city (receiver). small as the RGF approach by recording only the GF informa-
The tsunami height can be forecast simply by superposing the tion related to the vulnerable city, the information not provided
product of initial surface elevation of point in the rupture by the RGF is usually stored for application in other locations,
fault zone and the precomputed GF associated with by and the total disk space is still much larger than the RGF ap-
proach.
(7) Note that, in practice, we do not need a 25-h RGF even though
the numerical fundamental solution is simulated for 25 h. Be-
40 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 40, NO. 1, JANUARY 2015

Fig. 2. Comparison of the RGF method (dashed line) and the direct COMCOT tsunami simulation for the time series of offshore wave profile in Indian Ocean
Tsunami.

fore the first tsunami wave arrives, the RGF is simply zero and symbols are the same as (6). The shear stress is calculated by
does not need to be recorded. On the other hand, because of the assuming Manning’s equals 0.02.
uncertainty in the determination of the tsunami source, only the The coast of Taiwan can be affected by tsunami waves gener-
first few hours of the forecast are enough to provide useful infor- ated at any point in the whole Pacific basin. The Pacific basin has
mation for tsunami hazard mitigation. Hence, it is not necessary many possible tsunami sources, including the famous “ring of
to keep the whole simulation results. Instead, only the first 5 h fire.” The combinations for all possible earthquake magnitudes,
after the arrival of the first tsunami wave are recorded as the hypocenter locations, and fault angles are too many to be com-
RGF and a single RGF needs approximately 28 GB to store. pletely included if the inundation simulation is to be executed
beforehand. To save computer time for the inundation simula-
C. Nearshore Tsunami Inundation Map tion which is very time consuming, the inundation is not based
A detailed nearshore simulation is very time consuming. Al- on the direct numerical simulation for all moments of magnitude
though Titov et al. [4] proposed that real-time inundation simu- with all the possible hypocenters. Instead, the inundation map is
lation can be achieved by parallel computation, it is not feasible obtained through a detailed nearshore simulation with an inci-
for most government agencies to maintain and operate such a dent wave of different wave heights. Since the incident tsunami
system continuously. Thus, precomputed inundation maps are waveform is not known, normal incidence of a bell-shaped wave
a common practice in tsunami hazard mitigation. Usually the is assumed and the inundation is precomputed. The time scale is
tsunami wave is simulated from the beginning of a tsunamigenic set to be 20 min to be close to that of a real tsunami. As will be
earthquake to the inundation in the vulnerable city. It should discussed in Section IV-C, a comparison between two inunda-
be noted that, for coastal waters, the linear equations (6) are tion simulations of an earthquake scenario and the bell-shaped
not applicable because of the high nonlinearity and the bottom wave case shows that a 20-min bell-shaped wave provides a rea-
friction in wave propagation. Hence, the following nonlinear sonable estimation of the inundation.
shallow-water equations are used: For an incident wave with offshore wave height in between
two simulated bell-shaped waves, the inundation map is ob-
tained by interpolating the distance to the shoreline. These inun-
dation maps are then transformed into Google Earth kmz format.
As shown in Fig. 3, hospitals, drug stores, schools, grocery
stores, gas stations, and restaurants are shown in the inundation
(9) map. Thus, useful resources and the location of vulnerable facil-
ities can be demonstrated together via this geographic informa-
where is the total water depth, and is the bottom tion system so that people in charge can easily make decisions
shear stress with subscript representing its direction. All other to mitigate the hazard.
CHEN et al.: FORECAST SYSTEM FOR OFFSHORE WATER SURFACE ELEVATION WITH INUNDATION MAP INTEGRATED FOR TSUNAMI EARLY WARNING 41

Fig. 3. Inundation map in Google Earth format. The inundated region of red color has water depth higher than 3 m, while the yellow and blue regions represent
water depth higher than 2 and 1 m, respectively.

Fig. 4. Integrated forecast system on a GUI.

D. Forecast Procedures and GUI on the regression relation and the elastic half-space dislocation
model explained in Section II-A, the sea-bottom dislocation is
The three forecast modules discussed above are integrated by calculated and drawn, as shown in the upper middle of the inter-
a GUI of MATLAB so that their operation can be more conve- face in Fig. 4, which is also the initial sea-surface elevation
nient and user friendly. As shown in Fig. 4, the earthquake pa- of tsunami used in (7) and (8). Every vulnerable city has its pre-
rameters such as the moment magnitude, the longitude, latitude, computed RGF for every source point in the ocean. Multiplying
and depth of the hypocenter, as well as the three fault angles of the initial elevation with the RGF and summing together over
strike, dip, and slip obtained from the centroid moment tensor the whole source region, the offshore free-surface elevation for
(CMT) solution, are shown in the “fault parameter” zone in the the vulnerable city is promptly obtained in the upper right of
upper left of the interface. These parameters can be given manu- the interface. The choice of the vulnerable city is provided in
ally, or provided automatically through the Internet. Then, based the lower left of the interface in Fig. 4.
42 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 40, NO. 1, JANUARY 2015

Based on the time series of the sea-surface elevation in the


upper right of the interface, the max offshore wave height is also
calculated. This wave height is used to obtain the inundation
map by interpolating among the detailed simulations of various
incident wave heights, as described in Section II-C.
On the top of the “fault parameter” zone in the upper left
there are three icons. The “configure” icon is the setup of the
three websites of seismic information which will be used when
the “real-time” option in the lower left is activated. Along the
whole coastline of Taiwan there are quite a few towns; it is not
possible to include every vulnerable town onto the same screen,
and, hence, all the coastal towns are divided into a few groups
and the switch among these groups is controlled via the “area”
icon. The third icon “report” on the upper left of the interface
screen is to summarize the tsunami incident.
Besides the above functions, a choice of the time window ex-
tension for the time series in the upper right is given in the lower
left. Finally, the inundation map in the Google Earth kmz format
is shown on the Google Earth platform after the “kmz” icon is Fig. 5. Map of Taiwan that shows the position of wave stations mentioned in
this study.
chosen. A complete demonstration of the interface system on a
desktop personal computer (PC) is recorded as a video file in avi
format, and is included as an auxiliary material of this paper. angles, and 8.9 moment magnitude according to the U.S. Geo-
logical Survey website. Later, the moment magnitude and focal
III. APPLICATION TO REAL TSUNAMI EVENTS depth were modified to 9.0 and 10 km, respectively, and the
In the development of the present system, two tsunami events three fault angles and the epicenter were also modified to 29
occurred that affected the coast of Taiwan, including the 2010 (strike), 77 (dip), 95 (slip), and (38.486 N, 142.597 E) [18].
Chilean Tsunami and the 2011 Great Eastern Japan (Tohoku) The length of the fault plane was more than 600 km according
Tsunami. No damages were reported in Taiwan, but the wave to the regression (1). Note that the length of the subduction zone
record at stations shown in the map of Fig. 5 provides precious east of Honshu was about 500 km [19]. That is, for an Mw 9.0
in situ data for the validation of the present GUI system. earthquake, the fault is much shorter than the average case, and,
hence, the length and the width estimated from the regression
A. The 2010 Chilean Tsunami formula are not quite correct. Still the fault dimensions and the
The Chilean Earthquake of moment magnitude 8.8 occurred dislocation at the sea bottom all have order-of-magnitude ac-
on February 27, 2010 UTC at (72.47 W, 35.77 S) with 30-km curacy and the comparison with the in situ tsunami waveforms
focal depth, 19 strike, 30 dip, and 104 slip angles [17]. The shows that the tsunami forecast gives reasonable wave height.
forecasted tsunami wave has approximately 1-h period with a At some stations such as Suao and Taipei Harbors, the wave
leading small hump followed by a larger depression wave at height is correct although the predicted wave length is larger
wave stations shown in the map of Fig. 5. The in situ record is re- than the field measurement. The reason why the forecast system
trieved from the 1-Hz surface wave monitoring station shown in gives an order-of-magnitude accuracy for tsunami wave height
Fig. 6(a) that measures 2048 successive points ( 34 min) every is because the area of the rupture zone and, hence, the vertical
hour. Because of the 26-min gap in each hour, crests and troughs dislocation given by (1) and (5) are reasonable. In Kaohsiung,
of the tsunami wave are not recorded at most observatories ex- the forecast system underestimates the wave height, while the
cept for Keelung in northern Taiwan where the first trough that forecast in Keelung and Hualien overestimate the wave height.
has the highest amplitude can be identified. The residual mode, The comparisons of observed and predicted tsunami waveforms
after filtering out shorter waves with the implicit modal func- in Suao, Keelung, and Hualien are given in Fig. 7. Overall,
tion of Hilbert Huang transform, is the most significant long the forecast system gives an order-of-magnitude accuracy for
wave and can represent the tsunami wave. Its comparison with tsunami wave height at all wave stations shown in the map of
the tsunami forecast is shown in Fig. 6(b). The measured wave Fig. 5; thus, this system provides a preliminary estimation which
height is close to the tsunami wave calculated by the RGF fore- is quick and reliable.
cast system, and the arrival time lags behind the forecast only
by 9 min. Thus, the sea-surface elevation recorded in Keelung C. Simulation Time Used by the Present Integrated System
during the 2010 Chilean Tsunami validates the applicability of A feasible methodology to quickly evaluate a tsunami hazard
the forecast system. must be both accurate and quick so that a tsunami early warning
can be broadcast in time and the chance of false-alarm can be
B. The 2011 Great Eastern Japan Tsunami minimized. The accuracy is tested in Sections III-A and III-B,
The devastating Great Eastern Japan (Tohoku) Tsunami on while the simulation time is discussed in this section.
March 11, 2011, took place at 5:16 UTC with epicenter (37.32 Using the present forecast system, the central processing
N, 141.77 E and 24-km depth, 28 strike, 78 dip, 102 slip unit (CPU) time increases only slightly with domain area: for
CHEN et al.: FORECAST SYSTEM FOR OFFSHORE WATER SURFACE ELEVATION WITH INUNDATION MAP INTEGRATED FOR TSUNAMI EARLY WARNING 43

Fig. 6. (a) Location of the wave monitoring station. (b) Retrieved tsunami wave (red) and the tsunami forecast of the integrated system (blue) in Keelung
during the 2010 Chilean Tsunami.

a trans-basin tsunami in the Pacific Ocean, it takes 75 s for undation modeling. The former takes 4.5 h for 28 h of propaga-
a desktop PC of Microsoft Windows 7 operating system with tion simulation and the latter needs 12.5 h (simulating time) for
4-GB RAM and an Intel Core 2 Duo E8500 with two 3.16-GHz 40-min inundation simulation on an HP ProLiant ML350p Gen
CPU. 8 workstation of Linux operating system with 32-GB RAM and
On the other hand, using a direct simulation with the shallow- 2 Hexa-Core Intel Xeon 2.0-GHz CPUs.
water equations (6) and (9), the CPU time increases sharply The above comparison suggests that the present forecast
with the domain area. For the same trans-basin tsunami in the system can save much time for tsunami wave propagation over
Pacific Ocean, there are 20 million (3451 by 5761, 2-min res- a large domain and the much more time-consuming inundation
olution) coarse-grid points for propagation modeling and four simulation in the nearshore area. The application of the present
million (1911 by 2001, 5-m resolution) fine-grid points for in- system is proven to be quick and economical.
44 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 40, NO. 1, JANUARY 2015

IV. DISCUSSION TABLE I


DIMENSIONS OF THE GREAT EASTERN JAPAN TSUNAMI AND THE
CORRESPONDING CURVE COLORS IN FIG. 7
A. Accuracy and Robustness of the Present System

In an earthquake, most of the source parameters cannot be


directly measured and have to be calculated based on available
seismometer measurements. There are some uncertainties, and
it takes time in determining these fault parameters.
Still the forecast in previous tsunami events in Taiwan has
reasonable accuracy. Even for the Great Eastern Japan Tsunami
where the width of the fault plane is significantly larger than the
regression formula of Wells and Coppersmith, while the length is
significantly smaller according to (1) and [19], the wave height
forecast is accurate to an order of magnitude. As discussed in
Section II-B, the tsunami wave height is accurate because the area
of the rupture zone and hence the vertical dislocation given by the
regression formula both have reasonable accuracy.
As the present source model gives fault dimensions different
from the real value, it is reasonable to ask if the difference in
waveform and wave height is due to the source model. Hence,
the influence of the fault dimensions is discussed in the following.
A few combinations for the determination of the fault width and
length are shown in Table I. The first method is exactly (1) with
fault length SRL and width RW. From the second to
the fifth method, another regression formula of the rupture zone
area (RA)

RA

is used to obtain RA (67 608 km ). Since area is the product of


length and width, the fault zone width in the second method is B. Fault Parameters Without Sufficient Earthquake
determined by the area divided by the SRL. The third method is Information
similar but the SRL is replaced by the subsurface rupture length
(RLD). The fourth method uses 500-km length and 200-km width Once an earthquake occurs, the epicenter location and focal
according to [19]. For the fifth method, the fault length is fixed as depth are usually broadcast with the earthquake news, but the
500 km following [19] and the width is RA divided by 500 km. determination of the fault angles takes time. In the integrated
Although the regression formulas of [6] shown in (1)–(3) are system, the fault parameters are obtained through the U.S. Ge-
widely used, their applicability for earthquakes of high moment ological Survey or the F-net Broadband Seismograph Network
magnitude is a problem as the earthquake data used for regres- of the National Research Institute for Earth Science and Dis-
sion are not very large. Thus, the most recent regression for- aster Prevention, Japan. The former usually requires one hour,
mulas [20] are also used for comparison where Mw up to 8.9 while the latter provides the fault parameter in about 15 min. The
is used for regression. This method is shown in the last row of earthquake parameters may change after their first announce-
Table I. The three panels in Fig. 7 show the robustness of the ment.
system as the forecasts are almost identical. However, our experience in the recent tsunami events sug-
The wave height of all the predicted tsunami is close to the in gests that the system can provide the order-of-magnitude accu-
situ measurement; however, its wavelength is longer than the real racy for the sea-surface elevation even during the Great Eastern
tsunami, suggesting there are some detailed structures not pre- Japan Tsunami where the dimension of the ruptured zone is sig-
dictable by the source model. As indicated in [21], in this incident, nificantly different from other magnitude 9.0 earthquakes.
theinversionofthesedetailedslippatternsneedsnear-sourcedata. If there is no update information on the fault angles, the
It should be noted that tsunami hindcast that takes months to present forecast system cannot operate automatically, and
complete is not the object of this study. The integrated system these parameters have to be filled out manually. Following
is established to forecast tsunami in a few minutes and its ap- the experience of Japan in developing inundation maps, the
plication on both the 2010 Chilean tsunami and the 2011 Great 45 dip angle and the 90 slip angle can be used for the safe
Eastern Japan Tsunami shows order-of-accuracy accuracy. The side [22]. The strike angle can be assumed to be along the
inundation map is produced based on the offshore wave height trench in the neighboring area. Thus, the first estimation of the
and, hence, is not affected by the over-prediction of the wave- tsunami hazard is available, which is very useful, especially for
length if a bigger tsunami comes. Overall, this system is robust a near-field tsunami that arrives at a coastal city before more
even with the uncertainties in the generating mechanism. detailed earthquake information is available.
CHEN et al.: FORECAST SYSTEM FOR OFFSHORE WATER SURFACE ELEVATION WITH INUNDATION MAP INTEGRATED FOR TSUNAMI EARLY WARNING 45

Fig. 7. Observed tsunami wave (black curve) and the tsunami predicted by the six methods (represented by red, blue, green, yellow, magenta, and aqua curves)
in Table I in (a) Keelung, (b) Suao, and (c) Hualien during the 2011 Great Eastern Japan Tsunami.

C. Inundation Map by Bell-Shaped Wave the determination of the inundation map is not straightforward.
As there are countless possible waveforms, including the In the present stage, a hump of 20-min time scale is used as
combination of both the wave period and the wave amplitude, the incident tsunami waveform because the amount of water
46 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 40, NO. 1, JANUARY 2015

Fig. 8. Area inundated in an earthquake scenario (blue) and the area inundated by a bell-shaped wave (red).

over the mean sea level is larger than most waveforms, and the D. Future Work
20-min period can represent most tsunami incidents. The present interface integrates three modules and is easy to
According to [23], the extreme runup height and other wave apply in a tsunami event. To make the best use of this interface,
characteristics on a beach are only weakly dependent on the more information can be integrated and the interface can serve
shape of incident waves; this implies that the effect of wave as a platform for tsunami hazard mitigation.
period is negligible if it is similar to that of a tsunami. This result For example, the measured tsunami can be used to improve
also justifies our approach in employing one specific waveform the forecast, such as the system described in [4]. Earthquake
to represent the incidence of various tsunami waveforms. information can also be added separately besides the CMT so-
Since the incident tsunami waveform is not known before- lution provided by the U.S. Geological Survey and the F-net
hand and the inundation simulation is very time consuming, it Broadband Seismograph Network, so that there will be at least
would be better to choose a waveform that induces the worst some useful data before the CMT solution is available. Another
condition. To show that bell-shaped waves can generate a hazard example could be a comprehensive information system that aids
map of the worst condition, an inundation simulation based on the people in charge of hazard mitigation, rescue, and recovery
tsunamis generated by a scenario of a magnitude 8.3 subma- to make better decisions.
rine earthquake off the coast of Keelung in northern Taiwan is
given below. The epicenter is assumed at (122.0 E, 26.0 N) REFERENCES
with 10-km focal depth, 105 strike, 70 dip, and 90 slip an- [1] M.-K. Hsu and C.-T. Lee, “Tsunamis around Taiwan,” (in Chinese)
Acta Oceanographica Taiwanica, vol. 35, no. 1, pp. 1–16, 1996.
gles, so that the generated tsunami will propagate normally to
[2] D. Reymond, E. A. Okal, H. Hébert, and M. Bourdet, “Rapid forecast
Keelung city as the bell-shaped wave aims to represent the most of tsunami wave heights from a database of pre-computed simulations,
hazardous situation. and application during the 2011 Tohoku tsunami in French Polynesia,”
Geophys. Res. Lett., vol. 39, 2012, DOI:10.1029/2012GL051640,
Based on (1) and (5) in Section II-A, the rupture length, width,
L11603.
and dislocation are 247.7 km, 62 km, and 7.6 m, respectively. [3] Y. Wei, K. F. Cheung, G. D. Curtis, and C. S. McCreery, “Inverse
Numerical tsunami simulation shows that the offshore wave algorithm for tsunami forecast,” J. Waterway Port Coast. Ocean. Eng.,
vol. 129, no. 2, pp. 60–69, 2003.
height is 6 m, and the inundation is compared with another sim-
[4] V. V. Titov, F. I. González, M. C. Eble, H. O. Mofjeld, J. C. Newman,
ulation of the bell-shaped wave of the same height and 20-min and A. J. Venturato, “Real-time tsunami forecasting: Challenges and
time scale. solutions,” Nature Hazards, vol. 35, pp. 41–58, 2005.
[5] A. Sánchez and K. F. Cheung, “Tsunami forecast using an adaptive in-
The inundated areas are given in Fig. 8 where the blue and red
verse algorithm for the Peru-Chile source region,” Geophys. Res. Lett.,
lines enclose the area in the earthquake scenario and the bell- vol. 34, 2007, DOI: 10.1029/2007GL030158, L13605.
shaped wave case, respectively. The results show that the inun- [6] D. L. Wells and K. J. Coppersmith, “New empirical relationships
among magnitude, rupture length, rupture width, rupture area and
dated areas are quite similar for both cases, but the bell-shaped
surface displacement,” Bull. Seismol. Soc. Amer., vol. 84, no. 4, pp.
wave inundates a wider range: Of the 90 721 land grid points in 974–1002, 1994.
the simulation domain, 8934 grids have inundation deeper than [7] T. C. Hanks and H. Kanamori, “Moment magnitude scale,” J. Geophy.
Res., vol. 84, no. B5, pp. 2348–2350, 1979.
0.5 m in the submarine earthquake scenario, while the number
[8] L. Mansinha and D. E. Smylie, “The displacement fields of inclined
of inundated grids due to a bell-shaped wave is 9772. This com- faults,” Bull. Seismol. Soc. Amer., vol. 61, no. 5, pp. 1433–1440, 1971.
parison shows that the 20-min bell-shape wave provides a rea- [9] A. Ben-Menahem and M. Rosenman, “Amplitude patterns of tsunami
waves from submarine earthquakes,” J. Geophys. Res., vol. 77, pp.
sonable estimation of the inundation.
3097–3128, 1972.
CHEN et al.: FORECAST SYSTEM FOR OFFSHORE WATER SURFACE ELEVATION WITH INUNDATION MAP INTEGRATED FOR TSUNAMI EARLY WARNING 47

[10] H. G. Loomis, “Tsunami prediction using the reciprocal property of Guan-Yu Chen (M’12) was born in Taiwan in 1966.
Green’s functions,” Mar. Geodesy, vol. 2, no. 1, pp. 27–39, 1979. He received the M.S. degree in civil engineering from
[11] G.-Y. Chen, C.-H. Lin, and C.-C. Liu, “Quick evaluation of the National Taiwan University, Taipei City, Taiwan,
runup height and inundation area for early warning of tsunami,” in 1989 and the Ph.D. degree in physical oceanog-
J. Earthquake Tsunami, vol. 1250005, pp. 1–23, 2012, DOI: raphy from the University of Michigan, Ann Arbor,
10.1142/S1793431112500054. MI, USA, in 1998.
[12] G. F. Carrier, T. T. Wu, and H. Yeh, “Tsunami runup and drawdown Currently, he is an Associate Professor with the
on a plane beach,” J. Fluid Mech., vol. 475, pp. 79–99, 2003. Department of Oceanography, National Sun Yat-sen
[13] P. L.-F. Liu, S.-B. Woo, and Y.-S. Cho, Computer Programs for University, Kaohsiung, Taiwan. His research inter-
Tsunami Propagation and Inundation. Ithaca, NY, USA: Cornell ests include internal waves, infragravity waves, and
Univ., 1998. tsunami.
[14] X. M. Wang and P. L.-F. Liu, “An analysis of 2004 Sumatra earthquake Dr. Chen is a member of the American Geophysical Union.
fault plane mechanisms and Indian Ocean tsunami,” J. Hydraulic Res.,
vol. 44, no. 2, pp. 147–154, 2006.
[15] Z. Xu, “The all-source Green’s function and its applications to
tsunami,” Sci. Tsunami Hazards, vol. 26, no. 1, pp. 59–69, 2007. Chin-Chu Liu was born in Taiwan in 1979. He
[16] G.-Y. Chen and C.-C. Liu, “Evaluating the location of tsunami sensors: received the M.S. degree in physical oceanography
methodology and application to the northeast coast of Taiwan,” Terr. from the National Sun Yat-sen University, Kaoh-
Atmos. Ocean. Sci., vol. 20, no. 4, , pp. 563–571, 2009, DOI: 10.3319/ siung, Taiwan, in 2005.
TAO.2008.08.04.01(T). Currently, he is a Research Assistant with the
[17] U.S. Geological Survey, Nov. 25, 2013 [Online]. Available: http:// Department of Oceanography, National Sun Yat-sen
earthquake.usgs.gov/earthquakes/eqinthenews/2010/us2010tfan/ University. His research interests include numerical
neic_tfan_cmt.php simulation of internal waves and tsunami.
[18] U.S. Geological Survey, Jul. 25, 2012 [Online]. Available: http://
earthquake.usgs.gov/earthquakes/eqinthenews/2011/usc0001xgp/
neic_c0001xgp_cmt.php
[19] E. Kalkan and V. Sevilgen, “March 11, 2011 M9.0 Tohoku, Japan
Earthquake,” 2012 [Online]. Available: http://nsmp.wr.usgs.gov/
ekalkan/Tohoku/ index.html
[20] Y.-T. Yen and K.-F. Ma, “Source-scaling relationship for M 4.6–8.9 Cheng-Chung Yao was born in Taiwan in 1963. He
earthquakes, specifically for earthquakes in the collision zone of received the M.S. degree in surveying engineering
Taiwan,” Bull. Seismol. Soc. Amer., vol. 101, no. 2, pp. 464–481, from the National Cheng Kung University, Tainan
2012, DOI: 10.1785/0120100046. City, Taiwan, in 1989.
[21] F. Romano, A. Piatanesi, S. Lorito, N. D’Agostino, K. Hirata, S. Atzori, Currently, he is a Lecturer with the Department of
Y. Yamazaki, and M. Cocco, “Clues from joint inversion of tsunami Civil Engineering, Chienkuo Technology University,
and geodetic data of the 2011 Tohoku-oki earthquake,” Sci. Rep. 2, vol. Changhua, Taiwan. His research interests include re-
385, 2012, DOI: 10.1038/srep00385. mote sensing and geographic information system.
[22] T. Yamashita and R. Sato, “Generation of tsunami by a fault model,”
J. Phys. Earth, vol. 22, pp. 415–440, 1974.
[23] I. Didenkulova, E. Pelinovsky, and T. Soomere, “Runup characteris-
tics of symmetrical solitary tsunami waves of ‘unknown’ shapes,” Pure
Appl. Geophys., vol. 165, pp. 1–16, 2008.

You might also like