Toward the integration of seismic analysis and tsunami model for rapid inundation forecasting system

Aditya Riadi Gusman
Earthquake Research Institute, the University of Tokyo

1. INTRODUCTION

Great earthquakes in the subduction zones generated large tsunamis that devastated coastal areas. Huge tsunamis such as the 2004 Indian Ocean or the 2011 Tohoku tsunamis caused fatal damages in areas more than 1 km away from the coastlines. Unfortunately, existing tsunami early warning systems in the world do not have the capability to forecast tsunami inundation in the near field.

We developed a methodology for Near-field Tsunami Inundation Forecasting (NearTIF) that can produce high-resolution tsunami inundation forecast maps in around 3 minutes after the source of the tsunami has been determined (Gusman et al., 2014). This algorithm requires database of pre-computed tsunami waveforms and tsunami inundations.

The algorithm has been tested in a retrospective forecast of the 2011 Tohoku tsunami along the Sanriku Coast, Japan. Tsunami inundation maps were forecasted with earthquake source model that was estimated from the W phase of seismic data. We found that the resulting tsunami inundations were similar to the observed tsunami run-up heights.

We carried out tsunami evacuation drill in Kushiro city Hokkaido that used results from our algorithm to evaluate the effectiveness of tsunami inundation forecast maps.

2. NEAR-FIELD TSUNAMI INUNDATION FORECASTING (NearTIF)

Tsunami inundations from different scenarios can be similar to each other as long as the tsunami waveforms at near-shore points are also similar. Tsunami waveforms at near-shore point deeper than 50 m from low-resolution and high-resolution simulations are similar. Base on these assumptions, we developed a methodology for high-resolution tsunami inundation forecasting. This method required pre-computed tsunami waveforms and tsunami inundation database.

When an earthquake happens, tsunami waveforms at near-shore points can be simulated on low-resolution grid by solving the linear shallow water equations. This kind of simulation does not take long time and can be done using regular computer. A scenario in the database than produced tsunami waveform that is similar to the simulated one can be found by RMS analysis. Finally the corresponding tsunami inundation map from that best-fit scenario is selected as tsunami inundation forecast map. This process takes only approximately 3 minutes on a regular computer.

3. FAULT MODEL FOR THE 2011 TOHOKU TSUNAMI

W phase is a long period phase (200 to 1,000 s) that arrives before S phase, and can be used for rapid and robust determination of great earthquake’s source parameters with sufficient accuracy for tsunami early warning purposes (Kanamori and Rivera, 2008). About 5 minutes after the 2011 Tohoku earthquake Japanese F-net stations recorded seismic data that is enough to accurately estimate the seismic moment by W phase inversion. The estimated scalar moment for the 2011 Tohoku earthquake from the 5 min W phase data is 3.69 × 1022 Nm (Mw 9.0). The moment tensor solution was used to make a fault model for the 2011 Tohoku earthquake. The length (L) and width (W) of the fault model using scaling relation of Hanks and Bakun (2002) and simple relation of L = 2 × W are 246 km and 123 km, respectively (Gusman and Tanioka, 2013).

4. RETROSPECTIVE TSUNAMI INUNDATION FORECAST MAPS FOR THE 2011 TOHOKU TSUNAMI

The fault model based on the 5-min W phase solution was used as an input for NearTIF to produce tsunami inundation maps in 15 locations along the Sanriku Coast. The retrospective tsunami inundation forecast from this fault model can explain the observed limit of tsunami inundation and inland tsunami heights very well in all locations (Gusman et al., 2014). For example, tsunami inundation maps for Rikuzentakata, Japan produced by the NearTIF algorithm is shown in Fig. 1. The forecasted limit of inundation and inland tsunami heights are very similar with the observations (Fig. 1).

Fig. 1 Tsunami inundation forecast in Rikuzentakata produced by the NearTIF algorithm from the fault model of 5-min W phase solution. Green bars represent the measured tsunami heights, magenta dots represent the forecasted tsunami heights, blue lines represent the actual limit of inundation, and black dots are the measured positions. K value indicates the relative size of the observed and simulated tsunami heights, and κ (kappa) value indicates the precision of the simulated tsunami heights (Gusman et al., 2014).

Fig. 1 Tsunami inundation forecast in Rikuzentakata produced by the NearTIF algorithm from the fault model of 5-min W phase solution. Green bars represent the measured tsunami heights, magenta dots represent the forecasted tsunami heights, blue lines represent the actual limit of inundation, and black dots are the measured positions. K value indicates the relative size of the observed and simulated tsunami heights, and κ (kappa) value indicates the precision of the simulated tsunami heights (Gusman et al., 2014).

The tsunami inundation map produced by the NearTIF algorithm is comparable with that from numerical forward modeling. However, the numerical forward modeling required approximately 2 hours of CPU time to produce tsunami inundation map for a single location (Rikuzentakata). The required time is much longer if we simulate more locations. Based on performance test of NearTIF vs. numerical forward modeling in our previous study, the NearTIF algorithm is 800 times faster than numerical forward modeling to produce tsunami inundation maps in 15 locations (Gusman et al., 2014).

5. TSUNAMI EVACUATION DRILL IN KUSHIRO CITY, HOKKAIDO

From historical and pre-historical tsunamigenic earthquake studies (Hatori, 1984; Hirata et al., 2003; Satake et al., 2005; Tanioka et al., 2007; Ioki, 2013), Kushiro city has been identified as a tsunami prone area. Kushiro city office has provided to the public a tsunami hazard map that can increase the effectiveness of evacuation plans for the community.

We evaluated the effectiveness of the NearTIF algorithm in the real world by carrying out a tsunami evacuation drill in Kushiro city, Hokkaido, Japan, involving the residents. The drill started by an announcement of tsunami warning to evacuate the residents to the nearest evacuation building. About 10 minutes after the announcement, the participants used tablet computers to see the tsunami inundation forecast map that was uploaded to the Internet. It was easy for the participants to see their current location on the tsunami inundation forecast map because the tablet computer was equipped with GPS device. The participants found that the use of the tsunami inundation forecast map produced by NearTIF was effective to make a better decision with high confidence during the tsunami evacuation drill.

6. CONCLUSIONS

Tsunami inundation forecast on high-resolution topography can help to make a decision for evacuation during a tsunami event. Tsunami inundation can be simulated accurately by solving the nonlinear shallow water wave equations. However, high-resolution tsunami simulation is numerically expensive. To resolve this challenge, we developed a methodology for Near-field Tsunami Inundation Forecasting (NearTIF) that is equipped with a database of pre-computed tsunami waveforms and pre-computed tsunami inundation. The tsunami inundation forecasted by the NearTIF algorithm is similar to the result from numerical forward simulation and thus are a fast alternative to the slower numerical simulation.

W phase data can give a reliable earthquake magnitude estimate (Kanamori, 2008; Duputel et al., 2011; Gusman and Tanioka, 2013; Benavente and Cummins, 2013). Reliable centroid moment tensor solutions of the 2011 Tohoku earthquake can be estimated using 5 min of W phase data recorded at Japanese F-net stations (Gusman and Tanioka 2013; Gusman et al., 2014).

We evaluated the effectiveness of the NearTIF algorithm in the real world by carrying out a tsunami evacuation in Kushiro city, Hokkaido, Japan. The participants found that the use of the tsunami inundation forecast map was effective to make a better decision with high confidence during the tsunami evacuation process.
The NearTIF algorithm is recommended to be use as part of the reconstruction policy by local authorities to improve the evacuation efficiency particularly in tsunami prone areas. We recommend the use of the NearTIF method in developing future tsunami forecasting systems with a capability of providing tsunami inundation forecast maps for locations near the tsunami source area.

REFERENCES
Benavente, R., and P.R. Cummins (2013), Simple and reliable finite fault solutions for large earthquakes using the W-phase: The Maule (Mw = 8.8) and Tohoku (Mw = 9.0) earthquakes, Geophys. Res. Lett., 40, 3591–3595, doi:10.1002/grl.50648.
Duputel, Z., L. Rivera, H. Kanamori, G. P. Hayes, B. Hirsorn, and S. Weinstei (2011), Real-time W phase inversion during the 2011 off the Pacific coast of Tohoku Earthquake, Earth Planets Space, 63(7), 535–539, doi: 10.5047/eps.2011.05.032.
Gusman, A.R. and Y. Tanioka (2013), W phase inversion and tsunami inundation modeling for tsunami early warning: Case study for the 2011 Tohoku event, Pure Appl. Geophys., doi:10.1007/s00024-013-0680-z.
Gusman, A.R, Y. Tanioka, B. MacInnes, and H. Tsushima (2014), A methodology for near-field tsunami inundation forecasting: Application to the 2011 Tohoku tsunami, J. Geophys. Res. doi: 10.1002/2014JB010958.
Hanks, T.C., and W.H. Bakun (2002), A bilinear source-scaling model for M-log A observations of continental earthquakes, Bull. Seism. Soc. Am., 92 (5), 1841-1846.
Hatori, T. (1984), Source area of the east Hokkaido tsunami generated in April, 1843, Bull. Earthq. Res. Inst. Univ. Tokyo, 59, 423- 431, (in Japanese with English abstract).
Hirata, K, E. Geist, K. Satake, Y. Tanioka, and S. Yamaki (2003), Slip distribution of the 1952 Tokachi-Oki earthquake (M8.1) along the Kurile Trench deduced from tsunami waveform inversion, J. Geophys. Res., 108, 2196, doi:10.1029/2002JB001976.
Ioki, K. (2013), Source process of great earthquakes along the Kurile trench estimated from tsunami waveforms and tsunami deposit data, Doctoral thesis, Graduate School of Science, Hokkaido University.
Kanamori, H., and L. Rivera (2008), Source inversion of W phase: Speeding up seismic tsunami warning, Geophys. J. Int., 175, 222–238, doi:10.1111/j.1365-246X.2008.03887.x.
Satake, K., S. Nanayama, S. Yamaki, Y. Tanioka, and K. Hirata (2005), Variability along tsunami source in the 17th-21st centuries along the southern Kurile trench, Tsunamis: Case Studies and Recent Developments, edited by K. Satake, pp. 157-170, Springer.
Tanioka, Y., K. Satake, and K. Hirata (2007), Recurrence of Recent Large Earthquakes Along the Southernmost Kurile-Kamchatka Subduction Zone, Geophysical monograph, 172, 145-152.

A methodology for near-field tsunami inundation forecasting and its application to the 2011 Tohoku tsunami

Aditya Riadi Gusman
Institute of Seismology and Volcanology, Hokkaido University
Kita 10 Nishi 8 Kitaku, Sapporo, Hokkaido, Japan

Yuichiro Tanioka
Institute of Seismology and Volcanology, Hokkaido University
Kita 10 Nishi 8 Kitaku, Sapporo, Hokkaido, Japan

Japan Geoscience Union Meeting, Yokohama, Japan, 28 April – 2 May 2014. Website

Abstract

We develop a new methodology for near-field tsunami inundation forecasting (NearTIF). This method required site-specific pre-computed tsunami inundation and pre-computed tsunami waveform database. Information about tsunami source of an event is required as an input for the method to work. By this method, we will not attempt to obtain a reliable earthquake source model for an event. Instead, any available information about tsunami source such as earthquake’s moment magnitude, earthquake fault model, or tsunami source model will be used. After information about the tsunami source is obtained, tsunami waveforms at near-shore points can be simulated in real-time during an event. Simulating tsunami waveforms by solving the linear shallow water equation on low-resolution bathymetric data does not take long time, therefore it is suitable to be used in real-time. By using root mean square analysis, a scenario that gives the most similar tsunami waveforms in the database is selected as the best-fit site-specific scenario. Then the corresponding pre-computed tsunami inundation of the best scenario is selected as the tsunami inundation forecast.

The pre-computed tsunami database is built from thrust earthquake scenarios of simple rectangular fault models with moment magnitude ranged from Mw 8.0 to 9.0. We arrange a total of 56 reference points along the subduction zone off the east coast of Honshu, Japan as the center top of the fault planes. The points are grouped into four depth categories of shallowest, upper intermediate, lower intermediate, and deepest plate interface. The earthquake scenarios for each depth category have moment magnitude range of Mw 8.0 to 9.0, Mw 8.0 to 8.9, Mw 8.0 to 8.8, and Mw 8.0 to 8.7, respectively, from the shallowest to the deepest plate interface, making a total of 532 scenarios.

Sites are chosen based on their coastal geomorphology (i.e. bay, lagoon, isthmus) or location of coastal community. Virtual observation points at which tsunami waveforms is computed are placed strategically near-shore, around a bay at depth of deeper than 30 or 50 m depending on the bathymetry.

We test the algorithm to hindcast tsunami inundation along the Sanriku coast that was generated by the 2011 Tohoku earthquake. To produce accurate tsunami inundation map, accurate information about tsunami source is required. We used source models for the 2011 Tohoku earthquake previously estimated from GPS, W phase, or offshore tsunami waveform data. These source models could be available before tsunami hits the shore. The forecasting algorithm is capable of providing a tsunami inundation map that is similar to that obtained by numerical forward modeling, but with remarkably faster speed. Using a regular laptop computer, the time required to forecast tsunami inundation in coastal sites from the Sendai Plain to Miyako City is approximately 3 min after information about the tsunami source is obtained. We found that the tsunami inundation forecasts from the GPS (5 min), W phase (5 min and 10 min) fault models, and tsunami source model (35 min) are reliable for tsunami early warning purposes and considerably similar to the observation. This method can be used to develop a future tsunami forecasting systems with a capability of providing tsunami inundation forecasts in the near field locations.

Keywords: near-field tsunami inundation forecast, pre-computed tsunami database, tsunami early warning.

Comparative evaluation of tsunami-GPS and teleseismic body wave inversion methods for the 2014 Iquique, Chile, earthquake

Aditya Riadi Gusman, #Satoko Murotani, Kenji Satake, Mohammad Heidarzadeh, Shingo Watada (ERI, Univ. Tokyo), Endra Gunawan (Nagoya Univ.), and Bernd Schurr (GFZ)

Seismological Society of Japan Meeting, Niigata, Japan, 24-26 November 2014.

1. Introduction
Tsunami waveforms, land based GPS data and teleseismic body waves respectively provide good estimate on spatial slip distribution of submarine earthquake, static slip distribution beneath land, and the precise timing of slip history. Here we compare the slip distributions of the April 1, 2014 Iquique earthquake (Mw 8.2: USGS) from teleseismic inversion and a joint inversion of tsunami waveforms and GPS data.

2. Data and methodology
We used tsunami waveforms recorded at DART buoys across the Pacific Ocean and near-field tide gauges, and co-seismic displacements recorded by northern Chile GPS networks in the joint multi-window inversion to estimate the spatial and temporal slip distribution. In constructing the tsunami Green’s function we consider the effects of the elastic loading of the seafloor by the tsunami, the density variation of the compressible seawater, and the geopotential variation due to the movement of water mass on the linear long-waves (Watada et al., 2014: JGR), which gives the ability to use far-field tsunami waveforms in the inversion.
For the teleseismic waveforms, we inverted 54 P-wave vertical components and 1 SH-wave horizontal component using the method of Kikuchi and Kanamori (1991: BSSA) based on the same fault geometry of the tsunami and GPS data joint inversion.
Rupture durations of 45 s and 44 s on each subfault are assumed for the joint inversion and teleseismic inversion, respectively. To evaluate the sensitivity of rupture velocity to the estimation of slip distribution, we used different rupture velocities.

3. Estimated rupture processes
The teleseismic inversion with different rupture velocities yielded similar moment rate functions (Fig.b) but their spatial slip distributions are different. On the contrary, the joint inversion gives a stable spatial slip distribution (Fig.a) for different rupture velocities. Tsunami waveforms do not have the resolution in determining precise rupture history; therefore, the maximum rupture duration has to be guided by the moment rate function estimated from teleseismic inversion. When the same duration is assumed, the total seismic moment estimated by the teleseismic inversion (2.33 × 1021 Nm, Mw 8.2) is larger than that from the joint inversion (1.20 × 1021 Nm, Mw 8.0).
Untitled
Fig. a) Slip distribution of the 2014 Iquique earthquake estimated from tsunami waveform and GPS data. b) Moment rate functions from teleseismic body waves using different rupture velocities. c) Moment rate functions from tsunami waveforms and GPS data using different rupture velocities.

Rupture process of the 2014 Iquique Earthquake estimated from tsunami waveform and GPS data

Aditya Riadi Gusman1, Kenji Satake1, Satoko Murotani1, Mohammad Heidarzadeh1, Endra Gunawan2, and Shingo Watada1

1) Earthquake Research Institute, University of Tokyo
2) Nagoya University

American Geophysical Union Fall Meeting, San Francisco, 15-19 December 2014. AbstractAGU

Abstract

A great earthquake (Mw 8.2) occurred on April 1, 2014 at 23:46:46 off the coast of Iquique, Chile (USGS). The earthquake generated a tsunami that was recorded at four DART buoy and seven tide gauge stations. The tsunami amplitudes recorded at two closest tide gauge stations from the epicenter (Pisagua and Iquique) are both 187 cm. A GPS station in Iquique located about 98 km SE of the epicenter recorded the co-seismic displacement caused by the earthquake. The recorded horizontal and vertical displacements are -28.90(± 0.16) cm E, 3.90(± 0.12) cm N, and -3.80(± 0.60) cm Z. We use tsunami waveforms and co-seismic displacement data in a joint inversion to estimate the slip distribution of the 2014 Iquique earthquake. We apply a multiple time window inversion to show the spatial and temporal slip distribution. The fault geometry that we use is based on the SLAB1.0 model and aftershock depths. The fault geometry resembles the curvature of SLAB1.0 but with shallower depth; strike angle of 347°, rake angle of 90°, and sub-fault size of 20 km × 20 km are used. By assuming the rigidity of 4 × 1010 N/m2, the seismic moment for the 2014 Iquique earthquake calculated from the estimated slip distribution is 1.04 × 1021 Nm, which is equivalent to Mw 8.0. The dimension of major slip region of the slip distribution is 80 km × 40 km, and the maximum slip amount is 7.06 m. The major slip region is located down dip of the hypocenter and the rupture propagates from the epicenter to the SE direction. Instead of an instantaneous rupture process, the use of a more realistic rupture velocity (2.0, 2.8, or 3.0 km/s) can better explain the tsunami data. The selection among these three realistic rupture velocities does not strongly affect the estimated slip distribution. We also estimate the slip distribution using three other fault geometries, two of which have the same curvature as our final model but with different depths, and the third has a single dip angle. Our final slip distribution gives simulated tsunami waveforms and calculated co-seismic displacements that are slightly closer to the observations compared to those that used other fault geometries. Deeper fault geometries give larger total seismic moment, and the precise slip amount on each sub-fault is strongly dependent on the assumed fault geometry.

A methodology for near-field tsunami inundation forecasting

Yuichiro Taniokaand Aditya Riadi Gusman2

1) Institute of Seismology and Volcanology, Hokkaido University
2) Earthquake Research Institute, University of Tokyo

American Geophysical Union Fall Meeting, San Francisco, 15-19 December 2014. AbstractPoster PDFAGU

Abstract

Here we describe a new methodology for near-field tsunami inundation forecasting. We designed an algorithm that can produce high-resolution tsunami inundation maps of near-field sites before the actual tsunami hits the shore. This algorithm relies on a database of precomputed tsunami waveforms at several near-shore points and precomputed tsunami inundation maps from various earthquake fault model scenarios. By using numerical forward model, it takes several hours to simulate tsunami inundation in each site from each fault model. After information about a tsunami source is estimated, tsunami waveforms at near-shore points can be simulated in real-time. A scenario that gives the most similar tsunami waveforms is selected as the site-specific best scenario and the tsunami inundation from that scenario is selected as the tsunami inundation forecast. To test the algorithm, tsunami inundation along the Sanriku coast is forecasted by using source models for the 2011 Tohoku earthquake estimated from GPS, W phase, or offshore tsunami waveform data. The forecasting algorithm is capable of providing a tsunami inundation forecast that is similar to that obtained by numerical forward modeling, but with remarkably smaller CPU time. The time required to forecast tsunami inundation in 15 coastal sites from the Sendai Plain to Miyako City is approximately 3 minutes after information about the tsunami source is obtained. We found that the tsunami inundation forecasts from the 5-min GPS, 10-min W phase fault models, and 35-min tsunami source model are all reliable for tsunami early warning purposes and quantitatively match the observations well, although the latter model gives tsunami forecasts with highest overall accuracy. We evaluated the effectiveness of this algorithm in the real world by carrying out a tsunami evacuation drill in Kushiro City, Hokkaido, Japan, involving the city residents. The participants found that the use of the tsunami inundation forecast map produced by NearTIF was effective in helping them make better decisions with high confidence during the tsunami evacuation drill. This method can be useful in developing future tsunami forecasting systems with a capability of providing tsunami inundation forecasts for locations near the tsunami source area.