- Email: [email protected]
Tsunami hazard and early warning system in South China Sea
Journal of Asian Earth Sciences 36 (2009) 2–12
Contents lists available at ScienceDirect
Journal of Asian Earth Sciences journal homepage: www.elsevier.com/locate/jaes
Tsunami hazard and early warning system in South China Sea Philip L.-F. Liu a,b,*, Xiaoming Wang a, Andrew J. Salisbury a a b
School of Civil and Environmental Engineering, Cornell University, Hollister Hall, Hollister, Ithaca, NY 14853, USA Institute of Hydrological and Oceanic Sciences, Jhongli, Taiwan
a r t i c l e
i n f o
Article history: Received 23 June 2008 Received in revised form 15 December 2008 Accepted 26 December 2008
a b s t r a c t In this paper, we discuss the potential tsunami hazard in the South China Sea region. We focus our discussions on the characteristics of tsunamis generated from earthquakes along the Manila subduction zone. A procedure is presented for establishing a tsunami early warning system in the region. A scenario case is used to demonstrate the feasibility of the early warning system. Ó 2009 Elsevier Ltd. All rights reserved.
Keywords: Tsunami Early warning system Subduction zone Inverse problem Numerical simulation South China Sea
1. Introduction Tsunami is one of the most devastating natural coastal disasters. Most of large tsunamis are generated by submarine earthquakes occurring in subduction zones. Tsunamis can also be triggered by volcano eruptions and large landslides. Although the skill for predicting earthquake is still in its infancy, a tsunami warning system is still possible for a distant tsunami if the tsunami can be detected in the open-ocean. Information on the arrival time and the height of leading tsunami wave in areas far away from the source region can be predicted with some confidence (e.g., Wei et al., 2003). Such a tsunami warning system is now operational in the Pacific Ocean and has proven its effectiveness and validity for several recent tsunami events. If a similar early warning system had been available in the Indian Ocean, the 2004 Indian Ocean tsunami would have caused much less damage in property and loss in human lives in Sri Lanka, India, Maldives and other coastal regions. In the South China Sea (SCS) region, the Manila subduction zone has been identified as a high hazardous tsunamigenic earthquake source region. No earthquake larger than Mw = 7.6 has been recorded in the past 100 years in this region, suggesting a high probability for larger earthquakes in the future. And most alarmingly, there is no operational tsunami warning system in place in this region. If a tsunamigenic earthquake were to occur in this region in * Corresponding author. Address: School of Civil and Environmental Engineering, Cornell University, Hollister Hall, Hollister, Ithaca, NY 14853, USA. Tel.: +1 607 255 5090. E-mail address: [email protected] (P.L.-F. Liu). 1367-9120/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.jseaes.2008.12.010
the near future, a tragedy with the magnitude similar to the 2004 Indian Ocean tsunami could repeat itself. In this paper, potential tsunami hazard in SCS region is studied and a procedure for establishing a tsunami early warning system is presented (Satake, 1987; Titov et al., 2001). This system will be capable of releasing early warning information, including both tsunami arrival times and wave heights, to the surrounding countries for earthquakes along the Manila subduction zone. Using this information as input, inundation forecasts will also be possible by applying tsunami runup models in coastal regions of interest. 2. Hazardous tsunamigenic zones in South China Sea In the 2006 USGS tsunami source workshop (Kirby et al., 2006), three subduction zones, the Manila subduction zone, Ryukyu subduction zone and N. Sulawesi subduction zone, were identified as having high potentials to generate hazardous tsunamis (Fig. 1). Preliminary numerical studies have indicated that tsunamis generated from the Ryukyu subduction zone and N. Sulawesi subduction zone will not directly affect countries surrounding the South China Sea. Tsunamis generated from the Ryukyu subduction zone mostly propagate into the Pacific Ocean due to the strike angle of the potential thrust faults in this region. On the other hand, tsunamis generated from the N. Sulawesi subduction zone are most likely trapped inside the Celebes Sea. Neither of these subduction zones will have significant impacts on the surrounding countries of the South China Sea. Therefore, in this paper only the tsunamis generated from the Manila subduction zone will be investigated.
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
3
Wells and Coppersmith (1994) studied 421 historical earthquake events and 244 of these earthquakes with relatively reliable fault parameters were statistically analyzed. Empirical relationships among moment magnitude (Mw), rupture length, rupture width and dislocation were established. The calculated width of each fault plane with Wells and Coppersmith’s relationships is given in Table 2. In this table, ‘‘SRL” denotes surface rupture length and ‘‘RLD” subsurface rupture length. It is clear that the width of each fault plane segment varies slightly. However, for simplicity, we choose a single width value, 35 km, for all fault plane segments in the present study. The fault parameters for these hypothetical fault planes are summarized in Table 3. If we assume an earthquake with magnitude Mw = 8.0, with the fault parameters listed in Table 3, the amount of slip motions can be computed using the following equations:
Mo ¼ lDLW 2 Mw ¼ log10 M o 10:7 3 Fig. 1. Subduction zones near/in South China Sea (source: Kirby et al., 2006 USGS workshop).
2.1. Source region parameters along Manila Trench The Manila subduction zone, or so-called Manila Trench, is where the Eurasian Plate subducts under the Philippine Sea Plate at a speed of 70 mm/year (Lin, 2000). Manila Trench, starting from the northern tip of Palawan, Philippine, evolves to the north along the Western edge of Luzon, Philippine and ends in Taiwan, with a total length about 1000 km. Earthquake records show that the largest earthquake in this subduction zone in the past 100 years is about Mw = 7.5 (the 1999 Chi–Chi earthquake was 7.6 and the 1934 earthquake offshore from the northern Luzon was 7.5), making this region the most tsunami-hazardous in the future (Lee, personal communication). Along Manila Trench, six fault segments have been identified in USGS tsunami sources workshop 2006 (Kirby et al., 2006) as shown by the red crosses in Fig. 2.1 The epicenters are slightly different from those recommended by Kirby et al. (2006) which are shown by yellow crosses in Fig. 2 (right panel). The same dip angles, as indicated by Kirby et al., are adopted for each fault plane. The rake angle is assumed 90° for all the fault planes, which will make a maximum contribution to the seafloor deformation. The focal depth, defined as the distance from the seafloor to top edge of a fault plane, is assumed to be 15 km, which is commonly observed in several major earthquake events in this region in the past. The fault parameters for each fault segment are summarized in Table 1. The task for determining the width of a fault plane is non-trivial. The true fault plane width can only be estimated after an earthquake. Thus, the width can only be inferred from past earthquakes and/or from other empirical studies. Using the database of National Earthquake Information Center (NEIC), several historical events are used to estimate the width of fault plane. On December 11, 1999 an Mw = 7.3 earthquake occurred. Its epicenter is marked by a yellow cross in Fig. 3. Seven small aftershocks are indicated by white dots. By drawing a rectangular region enclosing the main shock and aftershocks, the size of the source area for this event is outlined and the width of the fault plane for this event is estimated as 34 km (width of fault plane = measured width in Fig. 3 divided by cos(dip angle) = 30 km/cos(28°)). Alternatively, the size of the ruptured area during an earthquake event may also be determined from empirical formulae. 1 For interpretation of color in Figs. 2, 3, and 9 the reader is referred to the web version of this article.
ð1Þ ð2Þ
where l ¼ 3:0 1010 N=M is the rigidity of earth mantle, D is the amount of slip motion (slip) and L is the length of the fault plane and W is the width of the fault plane, Mo is the scalar moment of an earthquake and Mw is the moment magnitude of an earthquake. The calculated slip motion on each fault plane segment is shown in the last column in Table 3. With all the fault parameters shown in Table 3, the seafloor deformation over each fault segment can be computed via Okada’s elastic fault model (Okada, 1985). 3. Simulation of tsunamis generated along Manila Trench In this section, we analyze the tsunami arrival time and amplitude distribution using the hypothetical earthquake of magnitude Mw = 8.0 with the fault plane parameters as listed in Table 3. The simulated results will help to identify the correlation between most affected coastal areas with the fractured fault plane segment. The arrival time analysis will also provide important information in selecting the locations of tsunami deep-ocean sensors. In simulating tsunami propagations, we first assume that the initial sea surface profile mimics the final seafloor deformation after the earthquake. This is reasonable since the duration of an earthquake is usually very short and the size of the rupture area is very large compared to the water depth. Therefore, there is not enough time for the water above the seafloor deformation to drain out. Tsunami propagations are simulated with a validated numerical model – COMCOT, which uses an explicit Leap-Frog finite difference scheme to solve the Shallow Water Equations (Liu et al., 1994, 1995, 1998). In this study, uniform 2-min grids using bathymetry data from ETOPO2, are implemented for all the simulations. The numerical domain ranges from 99°E to 133°E in longitude and 1°S–33°N in latitude with a grid dimension 1021 1021 (see Fig. 4). A vertical wall boundary is assumed along shorelines, where water depth is 5.0 m. 3.1. Arrival time and tsunami amplitude distribution In Fig. 5 the contours of arrival time are plotted for different tsunami scenario corresponding to rupturing different fault plane segment. In this paper, the arrival time is defined as the instant when the water surface is elevated more than 1 cm above the mean sea level due to the arrival of the leading tsunami wave. In regions where tsunami waves are relatively high, this criterion works very well. However, in regions far from the source area and where tsunami wave amplitude is very small (close to 1 cm), especially where bathymetry is complicated as well (with islands, submarine
4
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 2. Suggested fault plane segments along Manila Trench (Kirby et al., 2006).
Table 2 Rupture width estimated by Wells and Coppersmith’s relationships (1994).
Table 1 Fault parameters suggested by Kirby et al. (2006). Fault
Lon (deg.)
Lat (deg.)
Length (km)
Strike (deg.)
Dip (deg.)
Rake (deg.)
Segment
Length (km)
Mw (SRL)
Mw (RLD)
Width (SRL) (km)
Width (RLD) (km)
E1 E2 E3 E4 E5 E6
120.5 119.8 119.3 119.2 119.6 120.5
20.2 18.7 17.0 15.1 13.7 12.9
160 180 240 170 140 100
10 35 359 3 320 293
10 20 28 30 22 26
90 90 90 90 90 90
E1 E2 E3 E4 E5 E6
160 180 240 170 140 100
7.69 7.75 7.90 7.72 7.62 7.44
7.77 7.85 8.04 7.81 7.69 7.47
34.9 36.9 42.6 35.9 32.7 27.6
37.6 40.6 48.6 39.1 34.9 28.4
Fig. 3. Estimated fault plane size for the December 11, 1999 event. The black rectangular is outlined by enclosing the main shock and all its aftershocks.
features, etc.), the criterion becomes unreliable. Nevertheless, this simple definition, for most of part, serves its purpose in the present study.
Numerical results show that when the fault segment 1 is ruptured, which is closest to Taiwan, the leading tsunami wave will reach the southern part of Taiwan within about 20 min. The arrival time plots for all five different scenarios, corresponding to five different rupture segments, indicate that the generated tsunamis will arrive at coastal regions of Southeast China (Fujian province, Guangdong province, Hong Kong, Macao and Hainan Island) in 2–3 h. The coastal communities in Vietnam will also be affected in about 2 h, while tsunamis will strike Malaysia in 3 h. This suggests that one or two well placed sensors in the South China Sea can provide useful information for an effective early warning system for most of the region. The tsunami amplitude distribution plots indicate that for earthquakes occurring within the fault segment planes 1 and 2, the southern part of Taiwan will be greatly affected. For earthquakes occurring in other segments, the tsunami impacts on Taiwan will be relatively small. The plots also show that a major part of tsunami energy generated by earthquakes within fault plane segments 1–4 will travel to the west and the northwest. As a result, the coastal regions of Southeast China will be threatened by these tsunami sources. For coastal areas along Vietnamese coast, fault plane segments 4 and 5 are the most hazardous tsunami sources. On the other hand, the coastal regions of Malaysia are only significantly affected by tsunamis generated from fault plane segment 5. The hazardous tsunami source regions for surrounding countries in the South China Sea are summarized in Table 4.
5
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12 Table 3 Finalized hypothetical fault plane characterizations along Manila Trench. Fault
Lon (deg.)
Lat (deg.)
Length (km)
Width (km)
Strike (deg.)
Dip (deg.)
Rake (deg.)
Slip (m)
E1 E2 E3 E4 E5 E6
120.5 119.8 119.3 119.2 119.6 120.5
20.2 18.7 17.0 15.1 13.7 12.9
160 180 240 170 140 100
35 35 35 35 35 35
10 35 359 3 320 293
10 20 28 30 22 26
90 90 90 90 90 90
6.68 5.94 4.45 6.29 7.63 10.69
Note: the focal depth is chosen as 15.0 km for all fault plane segments.
Fig. 4. Numerical domain of tsunami hazard study.
4. Development of a tsunami early warning and forecasting model in South China Sea 4.1. General procedures The establishment of a tsunami early warning and forecasting system involves the following stages:
Tsunami detection. When a tsunami is detected by one or more sensors in deepocean, the time history of free surface measurements for the leading wave are processed for the inverse calculations. Data assimilation and inverse calculations.
Deployment of tsunami sensors. The locations of tsunami sensors in deep-ocean need to be determined based on the understanding on source region characteristics, tsunami arrival times and the coastal areas to be protected.
Based on the measurements at deep-ocean sensors, inverse calculations are performed to determine the seafloor displacements in the source region, i.e., a set of weighting factors associated with unit sources is determined. Tsunami forecasting and warning.
Construction of unit sources. The entire subduction zone needs to be divided into small segments; each segment is treated as ‘‘unit” source for tsunami generation.
Once the weighting factors are obtained, tsunami amplitudes and arrival times can be calculated everywhere by summing up the pre-calculated results from each unit source weighted by the weighting factor.
Pre-tsunami calculations.
4.2. A tsunami early warning system in the South China Sea
Tsunami wave fields generated from each unit source are calculated and the free surface elevation and velocity everywhere inside the domain are stored into a database.
In this section, a prototype tsunami early warning system is developed for the South China Sea region.
6
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 5. Contours of arrival time and tsunami amplitude distribution (red line: 30-min contour; pink line: 1.0-h contour; blue line: 2.0-h contour; green line: 3.0-h contour; yellow line: 4.0-h contour; and gray color scale is in meter for wave amplitude). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12 Table 4 Hazardous tsunami source regions for different coastal area. Countries/regions
Philippine Southeast China Taiwan Vietnam Malaysia
Hazardous tsunami sources Fault segments
Latitude range
All segments 1, 2, 3 and 4 1, 2 and 3 3 and 4 5
12.7°N–21.2°N 14.2°N–21.2°N 15.9°N–21.2°N 14.2°N–18.0°N 13.1°N–14.2°N
4.2.1. Locations of tsunami sensors Basically, information needed for tsunami forecasting comes from two origins: field measurements and numerical analysis. To have accurate measurements and make them useful for tsunami forecasting in a timely manner, the locations for tsunami sensors need to be considered carefully. Also, because of the costs, the number of sensors will be very limited. Thus, the locations chosen should allow sensors detecting tsunamis from as many unit sources as possible and still leaving enough time for forecasting. In the South China Sea, at least two tsunami sensors are needed to detect the south and north portions of the Manila Trench. In case of instrument failure, an additional sensor is recommended. Several factors have to be considered when deciding the sensor locations. Firstly, sensors need to be deployed in an area with relatively flat seafloor so as to reduce the local disturbances from complex bathymetry, coastal reflection and the trembling of source region. This implies that sensors should be some distance away from the source region. By examining the bathymetry map of South China Sea, two areas generally satisfy these requirements: southwest of Luzon and northwest of Luzon, where the sea floor is flat and water depth is around 3.0–4.0 km. Secondly, tsunami forecasting requires that sensors are capable of detecting tsunamis as early as possible such that sufficient time is available for forecasting. This implies that the location of these sensors should be close enough to the source regions. Furthermore, to detect tsunamis as early as possible, the sensors should be deployed on the paths where tsunamis travel fastest (the fastest portion of the arrival time contours). By examining the arrival
7
time contours, three areas can be identified: the west of fault plane segment 1, the west of interface between fault plane segments 2 and 3, and the southwest of the interface between fault plane segments 4 and 5. Thus three tsunami sensors are recommended to be deployed in areas corresponding to these observations. From the analysis presented in the previous section, tsunamis generated from fault plane segments 1 and 2 will arrive at Taiwan in 20–30 min. Considering the period of the leading tsunami wave is in the order of magnitude of 30 min, there is not enough time for utilizing the early warning system via the procedure described above since measurements of the leading wave are required. On the other hand, if the fault plane segment 3 ruptures, the generated tsunami will strike Taiwan in 1.0 h. Therefore, the tsunami sensor should be deployed no more than 10 min away from the source region, which is equivalent to about 120 km. This distance will allow more than 1.0 h warning time for all other coastal areas in China, Vietnam, and Malaysia. By considering the above requirements, we suggest three possible locations, which are outlined by red circles in Fig. 6: B1 (119.40°E, 20.10°N), B2 (118.15°E, 18.40°N) and B3 (117.60°E, 13.50°N). 4.2.2. Construction of unit sources The entire fault plane in the Manila subduction zone is divided into small ‘‘units”. Each unit contains a fault plane of 70 km long and 35 km wide, which becomes the unit source for tsunami precalculations (Figs. 7 and 8). The strike angles are determined from the trench orientation; the dip angles are estimated from Kirby et al. (2006 USGS tsunami source workshop). The width of fault plane is inferred from the aftershock analysis of the past earthquake events as discussed in the previous section. 4.2.3. Pre-tsunami calculations COMCOT is used to calculate tsunami wave fields generated by each of the 39 constructed unit sources and the results are stored in a database. The wave fields from each unit source are denoted as Gi(x, y, t), where G represents free surface elevation (or velocity) at location (x, y) at time t after the main shock; the subscript i presents the ith unit source.
Fig. 6. Suggested locations of tsunami sensors.
8
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 7. ‘‘Unit” source along Manila Trench (red dots denote the epicenter of each unit source). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 8. Fault plane segments.
4.2.4. Data assimilation and inverse calculations During an earthquake event, a tsunami is first detected at sensor j (i.e., either sensor B1, B2 or B3), the time history of sea surface elevation measurements at this sensor will be made available for
inverse calculations. Assume that the sensor is located at (xo, yo), the measurements at this sensor can be written as
Zðxo ; yo ; tk Þ;
k ¼ 1; Nt
ð3Þ
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 9. The seafloor deformation and the time history of measurements at sensor B3.
Table 5 Estimation of the rupture length. Earthquakes
Moment magnitude
Estimated empirically (km)
Seismic analysis (km)
1986/11/14 Hualien 1999/09/21 Chi–Chi 2003/05/21 Algeria 2005/03/28 Sumatra 2004/12/26 Sumatra 2006/11/15 Kuril Islands
6.5 7.6 6.8 8.6 9.3 8.3
22 120 35.5 570 1690 360
25 100 36.5 400 1400 400
where Nt represents the total number of measurements within a given time duration, usually last through the period of the leading wave. Once the complete form of leading wave is obtained, the inverse calculations start immediately to determine the weighting factors for each unit source by solving
min kAc bk2 c
ð4Þ
Fig. 11. Time history of sea surface elevations measured at sensor B3.
Fig. 10. Unit source numbering and cities for comparisons.
9
10
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
subject to the non-negative constraint with
8 G1 ðxo ; yo ; t1 Þ > > < A¼ > > : G1 ðxo ; yo ; t Nt Þ 9 8 Zðxo ; yo ; t1 Þ > > > > = < .. b¼ . > > > > ; : Zðxo ; yo ; t Nt Þ
9 GNs ðxo ; yo ; t1 Þ > > = ; > > ; GNs ðxo ; yo ; t Nt Þ
.. .
in which Ns denotes the total number of unit sources included in the inverse calculation. Since the above set of equations is over-determined (the number of measurements larger than the number of unit sources), the solution is unique. If the seafloor deformation due to unit source i is represented by Fi(x, y), the resultant seafloor deformation by the earthquake can be determined as
9 8 c > = < 1> .. c¼ ; . > > ; : c Ns ð5Þ
Fðx; yÞ ¼
Ns X
ci F i ðx; yÞ
i¼1
Fig. 12. Comparisons between the predicted and measured time histories near coastal cities.
ð6Þ
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
11
Fig. 12 (continued)
Fig. 13. Comparison between earthquake-generated seafloor deformation and synthetic seafloor deformation.
and the moment magnitude of the earthquake can be derived from Eq. (2). 4.2.5. Tsunami forecasting With weighting factor, ci, obtained for each unit source, the tsunami amplitude at any location can be forecasted as
Z f ðx; y; tÞ ¼
Ns X
ci Gi ðx; y; tÞ
ð7Þ
i¼1
4.2.6. An example of tsunami forecasting To test the forecasting model, a hypothetical earthquake with magnitude Mw = 8.0 is assumed to occur over the fault segment 4 rupturing the entire segment. Fault parameters are the same as those used in tsunami-hazardous analysis for fault segment 4 in the previous section. The epicenter of the earthquake is located at (119.2°E, 15.1°N). Time histories of sea surface elevations are recorded at the sensors B1, B2, B3 and other pre-selected tidal gage locations for inverse calculations and validating the predictions. Since B3 is the closest sensor to the source region, its record for the free surface elevations is used for inversely calculating the weighting factors
for unit sources. Thirty-six minutes after the earthquake, the wave form of the leading tsunami wave has been collected at B3. The free surface elevation records from t = 6 to 36 min contain 31 measurements (records between blue lines in Fig. 9). However, the number of unit sources to be included in the reverse calculations should be considered carefully. It is neither necessary nor practical to include every unit source, since the total number of unit sources cannot be larger than the number of measurements. In addition, for the selected range of measurements, the waves generated by unit source far away from sensor B3 and the epicenter would have not arrived at B3. Consequently, the weighting factors for these far away unit sources are less accurate. Therefore, we must limit the number of unit sources and only select those unit sources within a reasonable range of the epicenter in the inverse calculations. This range is determined from the empirical relationship between the rupture length and moment magnitude of an earthquake derived by Wells and Coppersmith (1994), which has the form
L ¼ 10
Mw 4:49 1:49
ð8Þ
where Mw is the moment magnitude of an earthquake and L is the subsurface rupture length by the earthquake (e.g., length of the
12
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
fault plane). For several historical earthquakes, the estimated rupture with Eq. (8) and the length from seismic wave analysis are given in Table 5. In general, Wells and Coppersmith’s empirical estimates are very close to those from seismic wave analysis. For larger earthquakes, the empirical relationship slightly overestimates the rupture length. In our inverse calculations, we adopt the empirical rupture length, which is 300 km, to determine how many unit sources are included in the calculation. If the epicenter of any unit sources falls into a circular region, centered at the epicenter of the earthquake (119.2°E, 15.1°N) with a radius 150.0 km, it will be included in the reverse calculation. In this example, the unit sources included in the calculations are 4, 5, 6, 7, 18, 20, 21 and 22 (see Fig. 10), covering an area of 280 km by 105 km, which is much larger than the source region (e.g., fault segment 4). And the weighting factors for these unit sources are 0.1600, 0.1597, 0.4739, 0.2609, 0.0143, 0.0000 and 0.1103, respectively. The numbering of unit sources is given in Fig. 11 (left panel). And these weighting factors are used to obtain the predictions at the nearshore of cities shown in Fig. 10 (right panel). The predicted and measured time histories are also compared at the nearshore of cities around the rim of the South China Sea (Fig. 12). The predictions match well with the measurements. Additionally, the seafloor deformation reconstructed from the weighting factors is also compared with that of the earthquake (Fig. 13). Reasonable agreement is shown. The moment magnitude calculated from the inverse solution is Mw = 8.04, which is very close to the actual earthquake magnitude Mw = 8.0 we used for the tsunami generation. Finally, our early warning system allows 30 min leading time for Taiwan, 2.5 h for China, 1 h for Vietnam, and 2.0 h for Malaysia for necessary evacuation. In general, we conclude that the proposed tsunami early warning system gives a very satisfactory performance. 5. Concluding remarks In this paper, we have demonstrated that potential for a strong future earthquake along the Manila subduction zone is real. Using
a numerical model, we have also shown that most of surrounding countries in the South China Sea will be affected by the tsunamis generated by the future earthquake. The simulated results show that the leading tsunami waves will reach most of coastal communities in China, Vietnam, Malaysia, and Taiwan within 1–2 h. We have proposed a procedure to establish a tsunami early warning system for the South China Sea region. A hypothetical earthquake is used to demonstrate the effectiveness of the early warning and forecasting system. Acknowledgement The research presented in this paper has been supported by National Science Foundation through grants to Cornell University. References Kirby, S. et al., 2006. In: USGS Tsunami Sources Workshop (2006): Great Earthquake Tsunami Sources: Empiricism and Beyond, April 21–22. Lee, W., 2006. Personal communication. Lin, C.-H., 2000. Thermal modeling of continental subduction and exhumation constrained by heat flow and seismicity in Taiwan. Tectonophysics 324, 189– 201. Liu, P.L.-F., Cho, Y.-S., Yoon, S.B., Seo, S.N., 1994. Numerical simulations of the 1960 Chilean tsunami propagation and inundation at Hilo, Hawaii. In: El-Sabh, M.I. (Ed.), Recent Development in Tsunami Research. Kluwer Academic Publishers, pp. 99–115. Liu, P.L.-F., Cho, Y.-S., Briggs, M.J., Synolakis, C.E., Kanoglu, U., 1995. Run-up of solitary waves on a circular island. J. Fluid Mech. 302, 259–285. Liu, P.L.-F., Woo, S.-B., Cho, Y.-S., 1998. Computer Programs for Tsunami Propagation and Inundation. Technical Report, Cornell University. Okada, M., 1985. Surface deformation due to shear and tensile faults in a half-space. Bull. Seismol. Soc. Am. 75 (4), 1135–1154. Satake, K., 1987. Inversion of tsunami waveforms for the estimation of a fault heterogeneity: method and numerical experiments. J. Phys. Earth 35, 241–254. Titov, V.V., Mofjeld, H.O., Gonzalez, F.I., Newman, J.C., 2001. Offshore forecasting of Alaskan tsunamis in Hawaii. In: Hebenstreit, G.T. (Ed.), Tsunami Research at the End of a Critical Decade. Kluwer Academic Publishers, Birmingham, England, Netherlands, pp. 75–90. Wei, Y., Cheung, K.F., Curtis, G.D., McCreery, C.S., 2003. Inverse algorithm for tsunami forecasts. J. Waterway, Port, Coastal Ocean Eng. 129 (2), 60–69. Wells, D.L., Coppersmith, K.J., 1994. New empirical relationships among Magnitude, rupture length, rupture width, rupture area and surface displacement. Bull. Seismol. Soc. Am. 84 (4), 974–1002.
Contents lists available at ScienceDirect
Journal of Asian Earth Sciences journal homepage: www.elsevier.com/locate/jaes
Tsunami hazard and early warning system in South China Sea Philip L.-F. Liu a,b,*, Xiaoming Wang a, Andrew J. Salisbury a a b
School of Civil and Environmental Engineering, Cornell University, Hollister Hall, Hollister, Ithaca, NY 14853, USA Institute of Hydrological and Oceanic Sciences, Jhongli, Taiwan
a r t i c l e
i n f o
Article history: Received 23 June 2008 Received in revised form 15 December 2008 Accepted 26 December 2008
a b s t r a c t In this paper, we discuss the potential tsunami hazard in the South China Sea region. We focus our discussions on the characteristics of tsunamis generated from earthquakes along the Manila subduction zone. A procedure is presented for establishing a tsunami early warning system in the region. A scenario case is used to demonstrate the feasibility of the early warning system. Ó 2009 Elsevier Ltd. All rights reserved.
Keywords: Tsunami Early warning system Subduction zone Inverse problem Numerical simulation South China Sea
1. Introduction Tsunami is one of the most devastating natural coastal disasters. Most of large tsunamis are generated by submarine earthquakes occurring in subduction zones. Tsunamis can also be triggered by volcano eruptions and large landslides. Although the skill for predicting earthquake is still in its infancy, a tsunami warning system is still possible for a distant tsunami if the tsunami can be detected in the open-ocean. Information on the arrival time and the height of leading tsunami wave in areas far away from the source region can be predicted with some confidence (e.g., Wei et al., 2003). Such a tsunami warning system is now operational in the Pacific Ocean and has proven its effectiveness and validity for several recent tsunami events. If a similar early warning system had been available in the Indian Ocean, the 2004 Indian Ocean tsunami would have caused much less damage in property and loss in human lives in Sri Lanka, India, Maldives and other coastal regions. In the South China Sea (SCS) region, the Manila subduction zone has been identified as a high hazardous tsunamigenic earthquake source region. No earthquake larger than Mw = 7.6 has been recorded in the past 100 years in this region, suggesting a high probability for larger earthquakes in the future. And most alarmingly, there is no operational tsunami warning system in place in this region. If a tsunamigenic earthquake were to occur in this region in * Corresponding author. Address: School of Civil and Environmental Engineering, Cornell University, Hollister Hall, Hollister, Ithaca, NY 14853, USA. Tel.: +1 607 255 5090. E-mail address: [email protected] (P.L.-F. Liu). 1367-9120/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.jseaes.2008.12.010
the near future, a tragedy with the magnitude similar to the 2004 Indian Ocean tsunami could repeat itself. In this paper, potential tsunami hazard in SCS region is studied and a procedure for establishing a tsunami early warning system is presented (Satake, 1987; Titov et al., 2001). This system will be capable of releasing early warning information, including both tsunami arrival times and wave heights, to the surrounding countries for earthquakes along the Manila subduction zone. Using this information as input, inundation forecasts will also be possible by applying tsunami runup models in coastal regions of interest. 2. Hazardous tsunamigenic zones in South China Sea In the 2006 USGS tsunami source workshop (Kirby et al., 2006), three subduction zones, the Manila subduction zone, Ryukyu subduction zone and N. Sulawesi subduction zone, were identified as having high potentials to generate hazardous tsunamis (Fig. 1). Preliminary numerical studies have indicated that tsunamis generated from the Ryukyu subduction zone and N. Sulawesi subduction zone will not directly affect countries surrounding the South China Sea. Tsunamis generated from the Ryukyu subduction zone mostly propagate into the Pacific Ocean due to the strike angle of the potential thrust faults in this region. On the other hand, tsunamis generated from the N. Sulawesi subduction zone are most likely trapped inside the Celebes Sea. Neither of these subduction zones will have significant impacts on the surrounding countries of the South China Sea. Therefore, in this paper only the tsunamis generated from the Manila subduction zone will be investigated.
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
3
Wells and Coppersmith (1994) studied 421 historical earthquake events and 244 of these earthquakes with relatively reliable fault parameters were statistically analyzed. Empirical relationships among moment magnitude (Mw), rupture length, rupture width and dislocation were established. The calculated width of each fault plane with Wells and Coppersmith’s relationships is given in Table 2. In this table, ‘‘SRL” denotes surface rupture length and ‘‘RLD” subsurface rupture length. It is clear that the width of each fault plane segment varies slightly. However, for simplicity, we choose a single width value, 35 km, for all fault plane segments in the present study. The fault parameters for these hypothetical fault planes are summarized in Table 3. If we assume an earthquake with magnitude Mw = 8.0, with the fault parameters listed in Table 3, the amount of slip motions can be computed using the following equations:
Mo ¼ lDLW 2 Mw ¼ log10 M o 10:7 3 Fig. 1. Subduction zones near/in South China Sea (source: Kirby et al., 2006 USGS workshop).
2.1. Source region parameters along Manila Trench The Manila subduction zone, or so-called Manila Trench, is where the Eurasian Plate subducts under the Philippine Sea Plate at a speed of 70 mm/year (Lin, 2000). Manila Trench, starting from the northern tip of Palawan, Philippine, evolves to the north along the Western edge of Luzon, Philippine and ends in Taiwan, with a total length about 1000 km. Earthquake records show that the largest earthquake in this subduction zone in the past 100 years is about Mw = 7.5 (the 1999 Chi–Chi earthquake was 7.6 and the 1934 earthquake offshore from the northern Luzon was 7.5), making this region the most tsunami-hazardous in the future (Lee, personal communication). Along Manila Trench, six fault segments have been identified in USGS tsunami sources workshop 2006 (Kirby et al., 2006) as shown by the red crosses in Fig. 2.1 The epicenters are slightly different from those recommended by Kirby et al. (2006) which are shown by yellow crosses in Fig. 2 (right panel). The same dip angles, as indicated by Kirby et al., are adopted for each fault plane. The rake angle is assumed 90° for all the fault planes, which will make a maximum contribution to the seafloor deformation. The focal depth, defined as the distance from the seafloor to top edge of a fault plane, is assumed to be 15 km, which is commonly observed in several major earthquake events in this region in the past. The fault parameters for each fault segment are summarized in Table 1. The task for determining the width of a fault plane is non-trivial. The true fault plane width can only be estimated after an earthquake. Thus, the width can only be inferred from past earthquakes and/or from other empirical studies. Using the database of National Earthquake Information Center (NEIC), several historical events are used to estimate the width of fault plane. On December 11, 1999 an Mw = 7.3 earthquake occurred. Its epicenter is marked by a yellow cross in Fig. 3. Seven small aftershocks are indicated by white dots. By drawing a rectangular region enclosing the main shock and aftershocks, the size of the source area for this event is outlined and the width of the fault plane for this event is estimated as 34 km (width of fault plane = measured width in Fig. 3 divided by cos(dip angle) = 30 km/cos(28°)). Alternatively, the size of the ruptured area during an earthquake event may also be determined from empirical formulae. 1 For interpretation of color in Figs. 2, 3, and 9 the reader is referred to the web version of this article.
ð1Þ ð2Þ
where l ¼ 3:0 1010 N=M is the rigidity of earth mantle, D is the amount of slip motion (slip) and L is the length of the fault plane and W is the width of the fault plane, Mo is the scalar moment of an earthquake and Mw is the moment magnitude of an earthquake. The calculated slip motion on each fault plane segment is shown in the last column in Table 3. With all the fault parameters shown in Table 3, the seafloor deformation over each fault segment can be computed via Okada’s elastic fault model (Okada, 1985). 3. Simulation of tsunamis generated along Manila Trench In this section, we analyze the tsunami arrival time and amplitude distribution using the hypothetical earthquake of magnitude Mw = 8.0 with the fault plane parameters as listed in Table 3. The simulated results will help to identify the correlation between most affected coastal areas with the fractured fault plane segment. The arrival time analysis will also provide important information in selecting the locations of tsunami deep-ocean sensors. In simulating tsunami propagations, we first assume that the initial sea surface profile mimics the final seafloor deformation after the earthquake. This is reasonable since the duration of an earthquake is usually very short and the size of the rupture area is very large compared to the water depth. Therefore, there is not enough time for the water above the seafloor deformation to drain out. Tsunami propagations are simulated with a validated numerical model – COMCOT, which uses an explicit Leap-Frog finite difference scheme to solve the Shallow Water Equations (Liu et al., 1994, 1995, 1998). In this study, uniform 2-min grids using bathymetry data from ETOPO2, are implemented for all the simulations. The numerical domain ranges from 99°E to 133°E in longitude and 1°S–33°N in latitude with a grid dimension 1021 1021 (see Fig. 4). A vertical wall boundary is assumed along shorelines, where water depth is 5.0 m. 3.1. Arrival time and tsunami amplitude distribution In Fig. 5 the contours of arrival time are plotted for different tsunami scenario corresponding to rupturing different fault plane segment. In this paper, the arrival time is defined as the instant when the water surface is elevated more than 1 cm above the mean sea level due to the arrival of the leading tsunami wave. In regions where tsunami waves are relatively high, this criterion works very well. However, in regions far from the source area and where tsunami wave amplitude is very small (close to 1 cm), especially where bathymetry is complicated as well (with islands, submarine
4
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 2. Suggested fault plane segments along Manila Trench (Kirby et al., 2006).
Table 2 Rupture width estimated by Wells and Coppersmith’s relationships (1994).
Table 1 Fault parameters suggested by Kirby et al. (2006). Fault
Lon (deg.)
Lat (deg.)
Length (km)
Strike (deg.)
Dip (deg.)
Rake (deg.)
Segment
Length (km)
Mw (SRL)
Mw (RLD)
Width (SRL) (km)
Width (RLD) (km)
E1 E2 E3 E4 E5 E6
120.5 119.8 119.3 119.2 119.6 120.5
20.2 18.7 17.0 15.1 13.7 12.9
160 180 240 170 140 100
10 35 359 3 320 293
10 20 28 30 22 26
90 90 90 90 90 90
E1 E2 E3 E4 E5 E6
160 180 240 170 140 100
7.69 7.75 7.90 7.72 7.62 7.44
7.77 7.85 8.04 7.81 7.69 7.47
34.9 36.9 42.6 35.9 32.7 27.6
37.6 40.6 48.6 39.1 34.9 28.4
Fig. 3. Estimated fault plane size for the December 11, 1999 event. The black rectangular is outlined by enclosing the main shock and all its aftershocks.
features, etc.), the criterion becomes unreliable. Nevertheless, this simple definition, for most of part, serves its purpose in the present study.
Numerical results show that when the fault segment 1 is ruptured, which is closest to Taiwan, the leading tsunami wave will reach the southern part of Taiwan within about 20 min. The arrival time plots for all five different scenarios, corresponding to five different rupture segments, indicate that the generated tsunamis will arrive at coastal regions of Southeast China (Fujian province, Guangdong province, Hong Kong, Macao and Hainan Island) in 2–3 h. The coastal communities in Vietnam will also be affected in about 2 h, while tsunamis will strike Malaysia in 3 h. This suggests that one or two well placed sensors in the South China Sea can provide useful information for an effective early warning system for most of the region. The tsunami amplitude distribution plots indicate that for earthquakes occurring within the fault segment planes 1 and 2, the southern part of Taiwan will be greatly affected. For earthquakes occurring in other segments, the tsunami impacts on Taiwan will be relatively small. The plots also show that a major part of tsunami energy generated by earthquakes within fault plane segments 1–4 will travel to the west and the northwest. As a result, the coastal regions of Southeast China will be threatened by these tsunami sources. For coastal areas along Vietnamese coast, fault plane segments 4 and 5 are the most hazardous tsunami sources. On the other hand, the coastal regions of Malaysia are only significantly affected by tsunamis generated from fault plane segment 5. The hazardous tsunami source regions for surrounding countries in the South China Sea are summarized in Table 4.
5
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12 Table 3 Finalized hypothetical fault plane characterizations along Manila Trench. Fault
Lon (deg.)
Lat (deg.)
Length (km)
Width (km)
Strike (deg.)
Dip (deg.)
Rake (deg.)
Slip (m)
E1 E2 E3 E4 E5 E6
120.5 119.8 119.3 119.2 119.6 120.5
20.2 18.7 17.0 15.1 13.7 12.9
160 180 240 170 140 100
35 35 35 35 35 35
10 35 359 3 320 293
10 20 28 30 22 26
90 90 90 90 90 90
6.68 5.94 4.45 6.29 7.63 10.69
Note: the focal depth is chosen as 15.0 km for all fault plane segments.
Fig. 4. Numerical domain of tsunami hazard study.
4. Development of a tsunami early warning and forecasting model in South China Sea 4.1. General procedures The establishment of a tsunami early warning and forecasting system involves the following stages:
Tsunami detection. When a tsunami is detected by one or more sensors in deepocean, the time history of free surface measurements for the leading wave are processed for the inverse calculations. Data assimilation and inverse calculations.
Deployment of tsunami sensors. The locations of tsunami sensors in deep-ocean need to be determined based on the understanding on source region characteristics, tsunami arrival times and the coastal areas to be protected.
Based on the measurements at deep-ocean sensors, inverse calculations are performed to determine the seafloor displacements in the source region, i.e., a set of weighting factors associated with unit sources is determined. Tsunami forecasting and warning.
Construction of unit sources. The entire subduction zone needs to be divided into small segments; each segment is treated as ‘‘unit” source for tsunami generation.
Once the weighting factors are obtained, tsunami amplitudes and arrival times can be calculated everywhere by summing up the pre-calculated results from each unit source weighted by the weighting factor.
Pre-tsunami calculations.
4.2. A tsunami early warning system in the South China Sea
Tsunami wave fields generated from each unit source are calculated and the free surface elevation and velocity everywhere inside the domain are stored into a database.
In this section, a prototype tsunami early warning system is developed for the South China Sea region.
6
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 5. Contours of arrival time and tsunami amplitude distribution (red line: 30-min contour; pink line: 1.0-h contour; blue line: 2.0-h contour; green line: 3.0-h contour; yellow line: 4.0-h contour; and gray color scale is in meter for wave amplitude). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12 Table 4 Hazardous tsunami source regions for different coastal area. Countries/regions
Philippine Southeast China Taiwan Vietnam Malaysia
Hazardous tsunami sources Fault segments
Latitude range
All segments 1, 2, 3 and 4 1, 2 and 3 3 and 4 5
12.7°N–21.2°N 14.2°N–21.2°N 15.9°N–21.2°N 14.2°N–18.0°N 13.1°N–14.2°N
4.2.1. Locations of tsunami sensors Basically, information needed for tsunami forecasting comes from two origins: field measurements and numerical analysis. To have accurate measurements and make them useful for tsunami forecasting in a timely manner, the locations for tsunami sensors need to be considered carefully. Also, because of the costs, the number of sensors will be very limited. Thus, the locations chosen should allow sensors detecting tsunamis from as many unit sources as possible and still leaving enough time for forecasting. In the South China Sea, at least two tsunami sensors are needed to detect the south and north portions of the Manila Trench. In case of instrument failure, an additional sensor is recommended. Several factors have to be considered when deciding the sensor locations. Firstly, sensors need to be deployed in an area with relatively flat seafloor so as to reduce the local disturbances from complex bathymetry, coastal reflection and the trembling of source region. This implies that sensors should be some distance away from the source region. By examining the bathymetry map of South China Sea, two areas generally satisfy these requirements: southwest of Luzon and northwest of Luzon, where the sea floor is flat and water depth is around 3.0–4.0 km. Secondly, tsunami forecasting requires that sensors are capable of detecting tsunamis as early as possible such that sufficient time is available for forecasting. This implies that the location of these sensors should be close enough to the source regions. Furthermore, to detect tsunamis as early as possible, the sensors should be deployed on the paths where tsunamis travel fastest (the fastest portion of the arrival time contours). By examining the arrival
7
time contours, three areas can be identified: the west of fault plane segment 1, the west of interface between fault plane segments 2 and 3, and the southwest of the interface between fault plane segments 4 and 5. Thus three tsunami sensors are recommended to be deployed in areas corresponding to these observations. From the analysis presented in the previous section, tsunamis generated from fault plane segments 1 and 2 will arrive at Taiwan in 20–30 min. Considering the period of the leading tsunami wave is in the order of magnitude of 30 min, there is not enough time for utilizing the early warning system via the procedure described above since measurements of the leading wave are required. On the other hand, if the fault plane segment 3 ruptures, the generated tsunami will strike Taiwan in 1.0 h. Therefore, the tsunami sensor should be deployed no more than 10 min away from the source region, which is equivalent to about 120 km. This distance will allow more than 1.0 h warning time for all other coastal areas in China, Vietnam, and Malaysia. By considering the above requirements, we suggest three possible locations, which are outlined by red circles in Fig. 6: B1 (119.40°E, 20.10°N), B2 (118.15°E, 18.40°N) and B3 (117.60°E, 13.50°N). 4.2.2. Construction of unit sources The entire fault plane in the Manila subduction zone is divided into small ‘‘units”. Each unit contains a fault plane of 70 km long and 35 km wide, which becomes the unit source for tsunami precalculations (Figs. 7 and 8). The strike angles are determined from the trench orientation; the dip angles are estimated from Kirby et al. (2006 USGS tsunami source workshop). The width of fault plane is inferred from the aftershock analysis of the past earthquake events as discussed in the previous section. 4.2.3. Pre-tsunami calculations COMCOT is used to calculate tsunami wave fields generated by each of the 39 constructed unit sources and the results are stored in a database. The wave fields from each unit source are denoted as Gi(x, y, t), where G represents free surface elevation (or velocity) at location (x, y) at time t after the main shock; the subscript i presents the ith unit source.
Fig. 6. Suggested locations of tsunami sensors.
8
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 7. ‘‘Unit” source along Manila Trench (red dots denote the epicenter of each unit source). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 8. Fault plane segments.
4.2.4. Data assimilation and inverse calculations During an earthquake event, a tsunami is first detected at sensor j (i.e., either sensor B1, B2 or B3), the time history of sea surface elevation measurements at this sensor will be made available for
inverse calculations. Assume that the sensor is located at (xo, yo), the measurements at this sensor can be written as
Zðxo ; yo ; tk Þ;
k ¼ 1; Nt
ð3Þ
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
Fig. 9. The seafloor deformation and the time history of measurements at sensor B3.
Table 5 Estimation of the rupture length. Earthquakes
Moment magnitude
Estimated empirically (km)
Seismic analysis (km)
1986/11/14 Hualien 1999/09/21 Chi–Chi 2003/05/21 Algeria 2005/03/28 Sumatra 2004/12/26 Sumatra 2006/11/15 Kuril Islands
6.5 7.6 6.8 8.6 9.3 8.3
22 120 35.5 570 1690 360
25 100 36.5 400 1400 400
where Nt represents the total number of measurements within a given time duration, usually last through the period of the leading wave. Once the complete form of leading wave is obtained, the inverse calculations start immediately to determine the weighting factors for each unit source by solving
min kAc bk2 c
ð4Þ
Fig. 11. Time history of sea surface elevations measured at sensor B3.
Fig. 10. Unit source numbering and cities for comparisons.
9
10
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
subject to the non-negative constraint with
8 G1 ðxo ; yo ; t1 Þ > > < A¼ > > : G1 ðxo ; yo ; t Nt Þ 9 8 Zðxo ; yo ; t1 Þ > > > > = < .. b¼ . > > > > ; : Zðxo ; yo ; t Nt Þ
9 GNs ðxo ; yo ; t1 Þ > > = ; > > ; GNs ðxo ; yo ; t Nt Þ
.. .
in which Ns denotes the total number of unit sources included in the inverse calculation. Since the above set of equations is over-determined (the number of measurements larger than the number of unit sources), the solution is unique. If the seafloor deformation due to unit source i is represented by Fi(x, y), the resultant seafloor deformation by the earthquake can be determined as
9 8 c > = < 1> .. c¼ ; . > > ; : c Ns ð5Þ
Fðx; yÞ ¼
Ns X
ci F i ðx; yÞ
i¼1
Fig. 12. Comparisons between the predicted and measured time histories near coastal cities.
ð6Þ
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
11
Fig. 12 (continued)
Fig. 13. Comparison between earthquake-generated seafloor deformation and synthetic seafloor deformation.
and the moment magnitude of the earthquake can be derived from Eq. (2). 4.2.5. Tsunami forecasting With weighting factor, ci, obtained for each unit source, the tsunami amplitude at any location can be forecasted as
Z f ðx; y; tÞ ¼
Ns X
ci Gi ðx; y; tÞ
ð7Þ
i¼1
4.2.6. An example of tsunami forecasting To test the forecasting model, a hypothetical earthquake with magnitude Mw = 8.0 is assumed to occur over the fault segment 4 rupturing the entire segment. Fault parameters are the same as those used in tsunami-hazardous analysis for fault segment 4 in the previous section. The epicenter of the earthquake is located at (119.2°E, 15.1°N). Time histories of sea surface elevations are recorded at the sensors B1, B2, B3 and other pre-selected tidal gage locations for inverse calculations and validating the predictions. Since B3 is the closest sensor to the source region, its record for the free surface elevations is used for inversely calculating the weighting factors
for unit sources. Thirty-six minutes after the earthquake, the wave form of the leading tsunami wave has been collected at B3. The free surface elevation records from t = 6 to 36 min contain 31 measurements (records between blue lines in Fig. 9). However, the number of unit sources to be included in the reverse calculations should be considered carefully. It is neither necessary nor practical to include every unit source, since the total number of unit sources cannot be larger than the number of measurements. In addition, for the selected range of measurements, the waves generated by unit source far away from sensor B3 and the epicenter would have not arrived at B3. Consequently, the weighting factors for these far away unit sources are less accurate. Therefore, we must limit the number of unit sources and only select those unit sources within a reasonable range of the epicenter in the inverse calculations. This range is determined from the empirical relationship between the rupture length and moment magnitude of an earthquake derived by Wells and Coppersmith (1994), which has the form
L ¼ 10
Mw 4:49 1:49
ð8Þ
where Mw is the moment magnitude of an earthquake and L is the subsurface rupture length by the earthquake (e.g., length of the
12
P.L.-F. Liu et al. / Journal of Asian Earth Sciences 36 (2009) 2–12
fault plane). For several historical earthquakes, the estimated rupture with Eq. (8) and the length from seismic wave analysis are given in Table 5. In general, Wells and Coppersmith’s empirical estimates are very close to those from seismic wave analysis. For larger earthquakes, the empirical relationship slightly overestimates the rupture length. In our inverse calculations, we adopt the empirical rupture length, which is 300 km, to determine how many unit sources are included in the calculation. If the epicenter of any unit sources falls into a circular region, centered at the epicenter of the earthquake (119.2°E, 15.1°N) with a radius 150.0 km, it will be included in the reverse calculation. In this example, the unit sources included in the calculations are 4, 5, 6, 7, 18, 20, 21 and 22 (see Fig. 10), covering an area of 280 km by 105 km, which is much larger than the source region (e.g., fault segment 4). And the weighting factors for these unit sources are 0.1600, 0.1597, 0.4739, 0.2609, 0.0143, 0.0000 and 0.1103, respectively. The numbering of unit sources is given in Fig. 11 (left panel). And these weighting factors are used to obtain the predictions at the nearshore of cities shown in Fig. 10 (right panel). The predicted and measured time histories are also compared at the nearshore of cities around the rim of the South China Sea (Fig. 12). The predictions match well with the measurements. Additionally, the seafloor deformation reconstructed from the weighting factors is also compared with that of the earthquake (Fig. 13). Reasonable agreement is shown. The moment magnitude calculated from the inverse solution is Mw = 8.04, which is very close to the actual earthquake magnitude Mw = 8.0 we used for the tsunami generation. Finally, our early warning system allows 30 min leading time for Taiwan, 2.5 h for China, 1 h for Vietnam, and 2.0 h for Malaysia for necessary evacuation. In general, we conclude that the proposed tsunami early warning system gives a very satisfactory performance. 5. Concluding remarks In this paper, we have demonstrated that potential for a strong future earthquake along the Manila subduction zone is real. Using
a numerical model, we have also shown that most of surrounding countries in the South China Sea will be affected by the tsunamis generated by the future earthquake. The simulated results show that the leading tsunami waves will reach most of coastal communities in China, Vietnam, Malaysia, and Taiwan within 1–2 h. We have proposed a procedure to establish a tsunami early warning system for the South China Sea region. A hypothetical earthquake is used to demonstrate the effectiveness of the early warning and forecasting system. Acknowledgement The research presented in this paper has been supported by National Science Foundation through grants to Cornell University. References Kirby, S. et al., 2006. In: USGS Tsunami Sources Workshop (2006): Great Earthquake Tsunami Sources: Empiricism and Beyond, April 21–22. Lee, W., 2006. Personal communication. Lin, C.-H., 2000. Thermal modeling of continental subduction and exhumation constrained by heat flow and seismicity in Taiwan. Tectonophysics 324, 189– 201. Liu, P.L.-F., Cho, Y.-S., Yoon, S.B., Seo, S.N., 1994. Numerical simulations of the 1960 Chilean tsunami propagation and inundation at Hilo, Hawaii. In: El-Sabh, M.I. (Ed.), Recent Development in Tsunami Research. Kluwer Academic Publishers, pp. 99–115. Liu, P.L.-F., Cho, Y.-S., Briggs, M.J., Synolakis, C.E., Kanoglu, U., 1995. Run-up of solitary waves on a circular island. J. Fluid Mech. 302, 259–285. Liu, P.L.-F., Woo, S.-B., Cho, Y.-S., 1998. Computer Programs for Tsunami Propagation and Inundation. Technical Report, Cornell University. Okada, M., 1985. Surface deformation due to shear and tensile faults in a half-space. Bull. Seismol. Soc. Am. 75 (4), 1135–1154. Satake, K., 1987. Inversion of tsunami waveforms for the estimation of a fault heterogeneity: method and numerical experiments. J. Phys. Earth 35, 241–254. Titov, V.V., Mofjeld, H.O., Gonzalez, F.I., Newman, J.C., 2001. Offshore forecasting of Alaskan tsunamis in Hawaii. In: Hebenstreit, G.T. (Ed.), Tsunami Research at the End of a Critical Decade. Kluwer Academic Publishers, Birmingham, England, Netherlands, pp. 75–90. Wei, Y., Cheung, K.F., Curtis, G.D., McCreery, C.S., 2003. Inverse algorithm for tsunami forecasts. J. Waterway, Port, Coastal Ocean Eng. 129 (2), 60–69. Wells, D.L., Coppersmith, K.J., 1994. New empirical relationships among Magnitude, rupture length, rupture width, rupture area and surface displacement. Bull. Seismol. Soc. Am. 84 (4), 974–1002.