The 2004 December 26 Indian Ocean tsunami severely hit Sri Lanka. Although it was not in the direct path of the initial tsunami waves, the western coast was struck by diffracted waves that caused much damage. The numerical model GEOWAVE is used to compute tsunami generation, propagation and inundation from the earthquake source to the Sri Lankan coast. A nested grid system is constructed to increase the resolution until Galle Bay, on the southwestern coast, where a 20 m-grid is used. The six nested topobathymetric grids are interpolated from ETOPO2 and high resolution data, at sea as onshore. Simulation results are compared with tsunami height data from National Oceanic and Atmospheric Administration (NOAA; US) and Geological Survey & Mines Bureau (GSMB; Sri Lanka). When the grid resolution increases, the discrepancy between the model and the data remains, on average, good, whereas its spread increases. We then conclude that the order of magnitude of the tsunami height is consistent from the 180 m-resolution grid, but the spatial imprecision is too high to locally predict reliable water heights. Nevertheless, the comparison between computed time-series of sea surface elevation at the Colombo tide station and tide-gauge data shows a very good agreement as both amplitude, and arrival time of the first wave are well reproduced. When focusing onshore, the modelled inundation limit is compared with the limit measured in the field. With its a priori setup, computed inundation spreads much farther behind the field limit. We then integrate more accurate nearshore conditions into the model. Non-linear shallow water equations are chosen instead of fully non-linear Boussinesq equations; the bottom friction on land is increased to a much higher value than at sea; the buildings cover and the low tide conditions are taken into account in the DEM. The resulting high resolution simulation agrees better with field data, even if discrepancies are still locally observed in places of DEM imprecision and in a river valley. This simulation, however, demonstrates that taking into account nearshore and onshore features may significantly improve tsunami impact assessment.
Tsunamis, Indian Ocean
The 2004 December 26 Sumatra—Andaman earthquake was one of the most disastrous recorded earthquakes, due, in part, to its large magnitude, Mw between 9.1 and 9.3 (Ammon et al. 2005; Park et al. 2005; Stein & Okal 2005). The earthquake generated a tsunami, which propagated through the Indian Ocean and caused extreme inundation and destruction along the surroundings coasts (Chanson 2005). The resulting death toll is estimated to be between 200 000 and 300 000. The most devastated country was Indonesia, as it lies near the subduction zone where the earthquake occurred, but the tsunami also severely struck Sri Lanka, where more than 30 000 people were killed (Liu et al. 2005; Goff et al. 2006; Inoue et al. 2007). The tsunami hit the island because it was on the main propagation axis, and then diffracted around it, so that the southwestern zone was as greatly damaged as the eastern zone.
Since 2005, numerous studies have attempted to model the tsunami propagation: some focused on the seismic source to discriminate between inverted slip distributions (Pietrzak et al. 2007) or to directly invert the slip distribution (Grilli et al. 2007; Piatanesi & Lorito 2007; Sladen & Hébert 2008); others attempted to model the tsunami impact on local targets like Thailand or the Mascarene Islands (Ioualalen et al. 2007; Hébert et al. 2007); and one study examined the propagation of the tsunami over the entire world and the spatial transport of energy in all oceans (Kowalik et al. 2007).
The potential influence of nearshore propagation effects on local tsunami height at the shoreline and onshore is well known. Most tsunami models, however, only operate on rough bathymetric data (Dao & Tkalich 2007; Geist et al. 2007; Ioualalen et al. 2007). The difficulty is to obtain high resolution data and to handle the whole extent of tsunami propagation, for example, with an irregular mesh or a nested system (Pietrzak et al. 2007; Hébert et al. 2007). Without such accurate data, numerical modelling may only compute tsunami amplitude at sea and cannot appraise the tsunami impact on the coasts (Geist et al. 2007).
In this study, we extend the global model of Grilli et al. (2007), with focus on the impact on southwestern Sri Lanka. We built a simulation chain to increase the resolution of the tsunami calculation in this area. Along the process, computed tsunami heights on Sri Lanka shore are compared with several sets of observations. The last step of the simulation chain is a focus on the Bay of Galle. There, model results are compared with field data of inundation limits. At this stage, we run several calculations with different options and parameters in order to refine the model so that it agrees, as well as possible, with the data.
2 Numerical Modelling
2.1 Tsunami model
In this study, we used the tsunami generation, propagation and inundation model GEOWAVE (Watts et al. 2003). The initial tsunami (surface water deformation and velocity) is computed from the vertical displacements of seafloor produced by an elastic dislocation simulating the earthquake (Okada 1985). Then, propagation and inundation of tsunami waves are computed with FUNWAVE, a model based on fully non-linear Boussinesq equations, accounting for frequency dispersion (Wei et al. 1995). The methodology of GEOWAVE is described in Grilli et al. (2007).
2.2 Nesting system
The aim is to compute tsunami waves until they reach the Sri Lankan coast. In a first stage, the most important thing is to describe the tsunami wavelength well, which is related to the source characteristics. For an earthquake-generated tsunami, the wavelength generally reaches a few hundred kilometres if the concerned fault is long. For the 2004 Sumatra earthquake, the wavelength is estimated, from the first wave measured by the Jason-1 satellite altimeter, as around 500–600 km (Gower 2005). The spatial resolution of the computational grid must be chosen according to this wavelength. In our case, a grid resolution of a few kilometres is enough to model tsunami propagation across the deep part of the Indian Ocean. However, when the tsunami comes near the coast, its wavelength decreases due to the decreasing depth, so that the resolution of the computational grid must increase. As GEOWAVE can solve tsunami propagation only on a regular square grid, we had to implement a system of nested grids. Considering the layout of the affected zone, a one-way nested scheme appears to be appropriate. Such systems are commonly used in tsunami modelling with non-linear shallow water equations (Koshimura & Mofjeld 2001; Titov et al. 2005), but this approach was not previously applied to Boussinesq-type models. In such a one-way nested grid scheme, wave propagation is first computed on the entire domain of the coarse grid. Information (velocity and water height) are extracted from the coarse grid simulation at the boundaries of the included fine grid and then used for the calculation on the fine grid. Information are not returned to the coarse grid from the fine one. Indeed, in the case of the December 2004 tsunami, the waves propagate in one preferential direction, they enter the Sri Lanka region and then exit without coming back again. The trapped waves are fully computed in the finer grids. It appears that some waves are reflected by the Maldives and then come back towards Sri Lanka, but the simulation shows that they have a negligible amplitude compared with direct and trapped waves (see Fig. 3). In practice, a first complete run is performed on grid of δx0-uniform spacing, during which water height and velocities are recorded at the limits of a second domain included within the first. Then, these conditions are interpolated on the limits of second domain, according to the chosen nested grid resolution δx1 < δx0 and used to compute through a new run more accurate tsunami propagation on this finer grid.
Shaded relief map of sea surface height 4 hr after the earthquake (S0). White arrows indicate the first wave reflected from Sri Lanka, whereas black ones mark the wave reflected from the Maldives.
Shaded relief map of sea surface height 4 hr after the earthquake (S0). White arrows indicate the first wave reflected from Sri Lanka, whereas black ones mark the wave reflected from the Maldives.
In our simulations, we chose to use a nested ratio of 3, so that δx1 = δx0/3. As we intend to reach a local scale around the Bay of Galle in Sri Lanka, the nesting process is iterated as needed. The resolution of all grids composing the nesting system are reported in Table 1. The first three simulations comprise the whole of Sri Lanka (Fig. 1). The last three are centred on the southwestern part of the island and then on the Bay of Galle.
Location of the six nested grids used in our simulation chain (coordinates are in Mercator projection).
Location of the six nested grids used in our simulation chain (coordinates are in Mercator projection).
2.3 Source parameters
The source parameters are taken from Grilli et al. (2007), who adjusted a five-segment fault model to satellite track records and some tide gauge data around the most devastasted part of the Indian Ocean coasts. We also use the same time sequence spanning 1200 s, which could be debated as it largely exceeds measured rupture duration (Ammon et al. 2005; Ni et al. 2005; Park et al. 2005). Grilli et al. (2007) defend their choice by referring to rising time effects, low shear-wave speed in the accretionary prism and since other tsunami modelling also assumed reduced rupture speed (e.g. Fujii & Satake 2007).
The output of each calculation consists of time-series of gridded sea surface elevation (SSE), maximum SSE on the domain (zmax) and, eventually, time-series of SSE at selected points taken as numerical gauges. The time-series results are important as they help to constrain the arrival time of the tsunami at the shore, and they also enable tsunami period estimation.
3 Data Used In This Study
3.1 Bathymetric and topographic data
Bathymetric grids are derived from several data sets of various resolutions.
The first grid S0 is derived from the ETOPO2 data set (U.S. Department of Commerce & Atmospheric Administration 2006). ETOPO2 is made from several data sources, mainly, satellite altimetry and has a 2′× 2′ spatial resolution.
A loose data set of sounding points on the complete Sri Lanka margin was provided by the DEOCOM project (Delimitation of the Outer Edge of the Continental Margin of Sri Lanka under the UNCLOS; Fig. 2). Nearshore detailed bathymetric profiles were provided by the Coastal Conservation Department (CCD), but only a few kilometres along north of Galle. Around Galle Bay and Weligama Bay, fine resolution data were provided by the National Hydrographic Office (NHO) of Sri Lanka (Garcin et al. 2007). These data have been checked and corrected before computing topobathymetric grids at various resolutions up to 20 m (Garcin et al. 2007, 2008).
Topography is needed to model inundation and tsunami reflection when the tsunami hits the coast. It, thus, mainly plays a role in high-resolution simulations, that is S4 and S5. We used a fine topographical data set on a 3–4-km wide coastal strip, where the main part is below 10 m elevation. A 20 m specific DEM was interpolated from the Survey Department of Sri Lanka data (elevation points and contour lines) and covers the coastal strip from Beruwala to Weligama Bay (thus covering the extent of S4 extent; Garcin et al. 2007). Another DEM made by the United Nations University (UNU) from a 1 : 5000 map digitizing and kinematic GPS data covers the Galle urban area with a 5 m resolution. Both these DEMs helped to build S4 and S5 topographical grids through subsampling. Elsewhere, SRTM elevation data with a 3″× 3″ resolution are used but have almost no influence on the tsunami calculation. Because of their limited extent, fine topographical data are integrated only in S3 and subsequent grids.
We use the Mercator projection to build a nesting system of six rectangular uniform grids from topobathymetric data with spatial resolutions from 4860 to 20 m (Table 1).
3.2 Tsunami observation data: tsunami height and inundation
As in the NOAA data set, what we call here ‘tsunami height’ may refer to the maximum elevation reached by the water either at sea (in case of tide gauges) or onshore (when talking about run-up).
Tsunami height data sets from several sources were used in this study. We first retrieved NOAA data (http://www.ngdc.noaa.gov/hazard/tsu.shtml), which is well spatially distributed data for the entire global domain of calculation. NOAA data for Sri Lanka are used to particularly check S0 to S2 simulations. When focusing on southern Sri Lanka (S2 and S3), we use GSMB data collected in the field a few days after the tsunami disaster (2004 December 30; Siriwardana et al. 2005).
 The megathrust earthquake that struck near Indonesia on 26 December 2004 at 0h58′53″ UTC (+7h for Thailand local time) was likely the 3rd largest earthquake ever recorded [Stein and Okal, 2005]. From its epicenter, located 80 km west of the coast of northern Sumatra (at approximately 95°51′W, 3°25′N), the earthquake proceeded approximately northward, rupturing 1200–1300 km of the Andaman-Sunda trench in about 8–10 min [Ammon et al., 2005; Lay et al., 2005] (Figure 1). Liberating enormous energy, corresponding to a Mw ≃ 9.3 moment magnitude, the earthquake triggered a tsunami that was one of the most devastating natural disasters ever witnessed in modern history, causing more than 292,000 fatalities in 12 countries bordering the Indian Ocean basin (T. Kawata et al., The December 26, 2004 earthquake tsunami disaster of Indian Ocean. Research Group on The December 26, 2004 Earthquake Tsunami Disaster of Indian Ocean, 2006, http://www.drs.dpri.kyoto-u.ac.jp/sumatra/index-e.html#casualty) (hereinafter referred to as Kawata et al., online report, 2006). The largest tsunami runups, over 30 m, occurred south of Banda Aceh, Sumatra, whose shore is closest to the epicenter, only about 10 minutes away in terms of tsunami propagation time (Figure 1). This area suffered the majority of fatalities (almost 230,000 dead or missing) and the most intense and widespread destruction during the 12/26/04 event (T. Kawata et al., Comprehensive analysis of the damage and its impact on coastal zones by the 2004 Indian Ocean tsunami disaster, 2005, Disaster Prevention Research Institute, http://www.tsunami.civil.tohoku.ac.jp/sumatra2004/report.htm, 2005) (hereinafter referred to as Kawata et al., online report, 2005). The next most heavily impacted area was the coast of Thailand, although it is located on the other side of Sumatra, not in direct line of the epicenter. It took the tsunami 1h45′ to 2h to reach this location [Tsuji et al., 2006]. Thousands of fatalities occurred in Thailand even though, on this east side of the fault, the first tsunami wave to arrive was a large depression wave that caused a significant withdrawal of the ocean at many locations, a crucial sign of tsunami arrival that often was not correctly read.
 All of the six Thai provinces that border the Andaman coast (Ranong, Phang Nga (Khao Lak area), Phuket, Krabi, Trang, and Satun; Figures 2–5) have exposed coastlines that were severely damaged by the tsunami. Among these, the province of Phang Nga suffered the most fatalities, accounting for 71% of the 8,500 people reported dead or missing in Thailand [Bagai et al., 2005; Kawata et al., online report, 2006] and widespread coastal destruction. Throughout this province, most of the fishing villages and their associated ecological environment were completely destroyed; many cultural landmarks suffered partial or total destruction. The largest tsunami runups (11 to 14 m) and destruction in Phang Nga province were observed near Khao Lak [Tsuji et al., 2006] (see also A. Siripongse, Investigation and risk evaluation on tsunami disaster and suggestions on monitoring and prevention of tsunami, in A First Report Under the Project: Investigation for Reclamation of Natural Resources and Environment by Chulalongkorn University, submitted to the Ministry of Natural Resources and Environment, Thailand, 2005) (hereinafter referred to as Siripongse, submitted manuscript, 2005) on a 20 km stretch of shoreline that includes several popular beaches and resorts (Bahn Khao Lak, Nang Thong, Bang Niang, Pa Ka Rang, and Pak Tawib; from south to north in Figure 3, in the Khao Lak area). Damage to tourist resorts, residential areas, and commercial buildings was widespread. A number of pictures and personal video recordings made in this area show that, after the initial ocean withdrawal, a large bore appeared, maybe reaching up to 8 m in height, and propagated as an almost straight line front approaching the Khao Lak beach and causing large runup. The second most impacted area in Thailand was the island of Phi Phi, which is located in Krabi province, 80 km east of the southern tip of Phuket (Figure 5; 98.8°E, 7.8°N). Phi Phi island suffered 15% of the fatalities reported in Thailand, when up to 6 m waves submerged a highly populated, narrow and low-lying sand isthmus (∼100–1,000 m wide and 2–2.5 m elevation), connecting two mountainous headlands between Tonsai bay (south coast of Phi Phi island) and Lohdalum bay (north of Phi Phi island). Eyewitnesses reported that waves hit the sand isthmus from both bays, first from the north side of the island, and a few minutes later from the south side; this was confirmed by personal pictures and video recordings (SEATOS, Sumatra earthquake and tsunami offshore survey, Cruise Report, 2005, http://www.oce.uri.edu/seatos/report.html) (hereinafter referred to as SEATOS, online report, 2005). Finally, Phuket Island was the third region of Thailand to be severely impacted by the tsunami, although it was much less heavily devastated than the Khao Lak area, and only locally, suffering 5% of the total fatalities in Thailand. A 5.5- to 6-m-high wave hit the western coast of the island, causing large runups (up to 10 m; Figure 3) and major damage, particularly at Kamala and Patong beaches. This resulted in 9% of the fatalities suffered in Thailand, with Kamala beach experiencing the most significant loss of life on the island. Destruction was widespread in Patong Beach, where not a single property escaped damage and eyewitnesses reported at least a 2-m-high surge that lasted for well over an hour, following the initial withdrawal.
 To better understand the large runups and destruction observed in coastal Thailand, and in view of the likelihood of similar future events occurring in the region (large earthquakes with Mw = 7.8–9.0 have occurred in 1797, 1833, 1861, 1881, 1907 and 1941 along this plate boundary [Lay et al., 2005]), in this study, we perform detailed numerical simulations of tsunami runup and impact along the coast of Thailand for the 12/26/04 event. In earlier work [Grilli et al., 2007], using a state-of-the-art Boussinesq model of tsunami generation, propagation, and runup, we had iteratively calibrated and validated a tsunami source for this event by comparing tsunami predictions with observations made at tide gauges in the Indian Ocean and the Andaman Sea, and JASON-1 satellite altimeter data measured in deep water. Here further model simulations are performed with a much finer regional grid defined over a smaller geographic area, using highly resolved bathymetric and topographic data in coastal Thailand. Specifically, the objectives of this study are to simulate: (1) runups over the whole Andaman coast of Thailand, where most post-tsunami field observations were made [Tsuji et al., 2006; Choi et al., 2006; Kawata et al., online report, 2005; Siripongse, submitted manuscript, 2005]; and (2) the sequence of events, at locations where these are available from eyewitness reports [e.g., Papadopoulos et al., 2006]. We will show that our simulation is robust, in the sense that it explains most of the observed features of the tsunami along the Andaman Coast of Thailand, without these having been used to calibrate the tsunami source. Once these objectives are reached, we will use our validated synoptic predictions of tsunami impact in Thailand to globally analyze the event, including in areas where no observations were made. We will thus assess which areas may be safe or most likely vulnerable to future tsunamis in the region.
2. Overview of the Sumatra Fault Tectonics
 The relative motion between the Indian and Sunda Plates is on the order of 4 cm per year in direction 20°N while, between the Australian and Sunda plates, it is on the order of 5 cm per year in direction 8°N [Socquet et al., 2006] (Figure 1). The 26 December 2004 Mw ≃ 9.3 megathrust earthquake [Stein and Okal, 2005] was a consequence of strain accumulated in the Indian/Sunda junction, some of which had not experienced a large earthquake for the past 150 years or so. Recent large events in the region include Mw ∼ 8.4 in 1797, Mw ∼ 9 in 1833, and Mw ∼ 8.5 in 1861, for the Australian/Sunda boundary, and weaker Mw ∼ 7.9 events for the Indian/Sunda boundary in 1881 and 1941 [Lay et al., 2005]. This unbalanced partition of past earthquake magnitudes and recurrence times between the two plate boundaries indicates that larger strains had accumulated in the Indian/Sunda boundary prior to the 26 December 2004 event, and explains both the epicenter location at the junction between the subducting Indian and Australian plates and the overriding Eurasian plate (Burma and Sunda subplates) and the northward rupture propagation, where most of the aftershocks were recorded along a ∼1300 km arc of the Andaman trench [Lay et al., 2005]. The 28 March 2005 Mw = 8.7 event was a second large megathrust earthquake that occurred farther south, liberating additional strain on another stretch of the Australian/Sunda boundary and generating a small tsunami, locally causing a 4 m runup near the Islands of Nias. Finally, more recently, on 17 July 2006, a Mw = 7.7 earthquake occurred off southwest Java, liberating some more strain even further south along the same plate boundary and causing a devastating tsunami along 150 km of Java's coastline.
3. The 26 December 2004 Earthquake and Tsunami Events
 Before witnessing this event, scientists analyzed tsunamis generated by small-scale seismic ruptures as instantaneously triggered. This was a fairly good approximation because, for small rupture propagation times, the delay between tsunami time of triggering by coseismic bottom motion, and actual fault rupture was generally negligible as compared to travel time to the nearest coasts (e.g., the 16 November 1999 Vanuatu earthquake and tsunami [Ioualalen et al., 2006]).
 The large size of the ruptured area of the 26 December 2004 event, however, raised many questions regarding the relationships between rupture speed and tsunami modes and timing of triggering by coseismic bottom motion. As far as past large-scale earthquakes and derived tsunamis, none were sufficiently well observed (through seismic and hydrographic networks) to initiate a comprehensive study of these relationships. The 26 December 2004 event is a milestone in this respect, because of its widespread observation with a sufficiently comprehensive and dense network to initiate such studies.
3.1. Summary of Earthquake Mechanism
 The earthquake occurred at 0h58′53″ UTC off the northern coast of Sumatra, Indonesia, at 95°51′, 3°25′ (Figure 1). The earthquake was measured in great detail over the Indian Ocean basin, using seismographs and GPS stations. Seismic inversion models [Ammon et al., 2005; Bilham et al., 2005; Lay et al., 2005] indicate that, for about 500 s, the rupture propagated approximately northward from the epicenter, along 1,200–1,300 km of the Andaman-Sunda trench (with an average rupture speed of 2.5–3 km/s), causing up to ∼6 m of bottom subsidence and ∼10 m of uplift over a region 100–150 km wide across the subduction area. According to Bilham , up to 10 m uplift and subsidence were generated by the earthquake elastic rebound, offshore of Banda Aceh (northern tip of Sumatra). Seismic inversion and GPS records further indicate that fault slip was not homogeneous along the ruptured area varying between 15 and 25 m, with a gradual decrease northward from the epicenter [Vigny et al., 2005]. (See Grilli et al.  for a more detailed overview of rupture and bottom processes.)
3.2. Tsunami Observations
 Many real time observations of the tsunami were made in the Indian Ocean, perhaps so extensively for the first time owing to recent progress in observational techniques. Thus data are available from many tide gauges [Merrifield et al., 2005; Nagarajan et al., 2006] (also Royal Thai Navy, http://www.navy.mi.th/hydro/tsunami.htm, 2005) (hereinafter referred to as Royal Thai Navy, online report, 2005), a few satellite altimeters [e.g., Gower, 2005; Smith et al., 2005], and a satellite Multi-angle Imaging Spectro-Radiometer (MISR) [e.g., Garay and Diner, 2007]. The very large extent of the ruptured area and large associated tsunami that was generated also contributed to their easier detection over a large domain. Beside these instrument records, numerous post-tsunami field surveys were made over the whole Indian Ocean basin [e.g., Tsuji et al., 2006; Choi et al., 2006; Kawata et al., online report, 2005; Siripongse, submitted manuscript, 2005]. This large amount of nonseismic data has helped better characterize the earthquake through constraints provided by the associated tsunami, such as arrival time of successive waves at tide gauges and along satellite transects.
 Thus, in our earlier work, we used many hydrographic data sets, including amplitude, timing, periodicity and sequence, of tsunami waves measured by various instruments, to iteratively develop and calibrate parameters of a multisegment coseismic tsunami source for the 26 December 2004 event [Grilli et al., 2007] (Figure 1). The two main data sets used in this calibration are detailed below.
 The first data set consists of digital tide gauge or point surface elevation records. Most of these are tide gauges that are part of the Global Sea Level Observing system (GLOSS) network, monitored by the Joint Technical Commission for Oceanography and Marine Meteorology (JCOMM). Tide gauges that were used in the source calibration are located in Hannimaadhoo, Male and Gan (Maldives), Colombo (Sri Lanka), Diego Garcia (British Territory) and Cocos Island (Australia). UHSLC provides digital tide residuals, which can be directly compared with the simulated time series. A discussion of the tsunami signal detected by the tide gauges, including arrival times and sequences of tsunami waves, was given by Merrifield et al. . Additional tide gauges operated in Thailand were used, particularly that at Taphao-Noi (Royal Thai Navy, online report, 2005). Finally, a depth sounding record made a mile off Nai Harn Bay near the southwestern end of Phuket Island, onboard the yacht Mercator, in 12 m of water, was used that showed the arrival of three main waves over a duration of 35′.
 The second data set is the sea level anomaly detected by JASON-1's satellite altimeter, which happened to cut across the evolving tsunami wave pattern in a north-south direction approximately 2 hours after the earthquake, during cycle 109 of pass 129 [Gower, 2005; Smith et al., 2005]. Grilli et al.  calculated the sea level anomaly over a diagonal transect in the Indian Ocean by subtracting measurements made during the earlier cycle 108 from cycle 109; they corrected for the travel speed of the satellite in their comparison with model results. Phenomena other than the tsunami may affect sea surface anomaly and cause errors, such as the internal and wind-forced variability of the ocean but, at relatively low latitudes such as here, the dominant timescales derived from basin-wide eddies are much larger than the period between two satellite cycles (around 10 days). Still, the obtained signal was noisy, maybe because of the relatively small Bay of Bengal basin, which may locally yield higher variability; thus the discrepancy between two cycles can be on the order of 20%, with or without a tsunami signal. Nevertheless, considering its magnitude (up to 1.20 m from peak to trough), the tsunami signal can be clearly identified in the records.
 A third data set, used here but not in the tsunami source calibration, consists in the runup values measured during post-tsunami field surveys made along the Andaman coast of Thailand [Tsuji et al., 2006; Choi et al., 2006; Siripongse, submitted manuscript, 2005]. Mostly densely populated areas, however, were surveyed, such as resort beaches in Khao Lak, Phuket, and Phi Phi island. Again it is one purpose of this work, through model simulations, to provide a synoptic and complete picture of tsunami impact in Thailand, including at locations where measurements were not made, and try to identify regions vulnerable to future tsunamis, independent of the density of the population. Such information would help in future regional development plans that might be considered.
 Grilli et al.  performed model simulations using a 1′ × 1′ grid (and 1.2 s time step), in a computational domain covering the entire Bay of Bengal, the Andaman Sea, and part of the Southern Indian Ocean (from 72° to 102°E in longitude and from 13°S to 23.5°N in latitude). They simulated runups only at key locations (Banda Aceh in Indonesia, Khao Lak in Thailand), where the tsunami was most destructive, and favorably compared these with observations. In the present work, we use a finer 0.25′ grid (with a 0.5 s time step), starting west of the northern tip of Sumatra and covering the Andaman sea up to the northern coast of Thailand, i.e., from 91° to 101°E in longitude and from 3.6°N to 12°N in latitude (Figure 5).
4. Tsunami Simulations
 Numerical simulations of tsunami coastal impact require three components: (1) a source, reflecting the known geology and seismology of the event; (2) ocean bathymetry and coastal topography, and (3) a tsunami propagation and runup model, representing the relevant physics.
 Here we simulate tsunami propagation and inundation with FUNWAVE, a Boussinesq water wave model developed at the University of Delaware [Wei and Kirby, 1995; Wei et al., 1995