Observations, Effects and Real Time Assessment of the March 11, 2011 Tohoku-oki Tsunami in New Zealand

The great Tohoku-oki earthquake of March 11, 2011 generated a devastating tsunami in the near field as well as substantial far-field effects throughout the Pacific Ocean. In New Zealand, the tsunami was widely observed and instrumentally recorded on an extensive array of coastal tidal gauges and supplemented by current velocity data from two sites. While the tsunami's first arrival was on the morning of March 12 in New Zealand, the strongest effects occurred throughout that afternoon and into the following day. Tsunami effects consisted primarily of rapid changes in water level and associated strong currents that affected numerous bays, harbors, tidal inlets and marine facilities, particularly on the northern and eastern shores of the North Island. The tsunami caused moderate damage and significant overland flooding at one location. The tsunami signal was clearly evident on tide gauge recordings for well over 2 days, clearly illustrating the extended duration of far field tsunami hazards. Real time analysis and modelling of the tsunami through the night of March 11, as the tsunami crossed the Pacific, was used as a basis for escalating the predicted threat level for the northern region of New Zealand. A comparison to recorded data following the tsunami shows that these real time prediction models were accurate despite the coarse near-shore bathymetry used in the assessment, suggesting the efficacy of such techniques for future events from far-field sources.     

Tsunami Hazard Posed to New Zealand by the Kermadec and Southern New Hebrides Subduction Margins: An Assessment Based on Plate Boundary Kinematics, Interseismic Coupling, and Historical Seismicity

We assess the tsunami hazard posed to New Zealand by the Kermadec and southern New Hebrides subduction margins. Neither of these subduction zones has produced tsunami large enough to cause significant damage in New Zealand over the past 150?years of well-recorded history. However, as this time frame is short compared to the recurrence interval for major tsunamigenic earthquakes on many of the Earth’s subduction zones, it should not be assumed that what has been observed so far is representative of the long term. For each of these two subduction zones we present plate kinematic and fault-locking results from block modelling of earthquake slip vector data and GPS velocities. The results are used to estimate the current rates of strain accumulation on the plate interfaces where large tsunamigenic earthquakes typically occur. We also review data on the larger historical earthquakes that have occurred on these margins, as well as the Global CMT catalogue of events since 1976. Using this information we have developed a set of scenarios for large earthquakes which have been used as initial conditions for the COMCOT tsunami code to estimate the subsequent tsunami propagation in the southwest Pacific, and from these the potential impact on New Zealand has been evaluated. Our results demonstrate that there is a significant threat posed to the Northland and Coromandel regions of New Zealand should a large earthquake (M _w ?8.5) occur on the southern or middle regions of the Kermadec Trench, and that a similarly large earthquake on the southern New Hebrides Trench has the potential to strongly impact on the far northern parts of New Zealand close to the southern end of the submarine Three Kings Ridge. We propose logic trees for the magnitude–frequency parameters of large earthquakes originating on each trench, which are intended to form the basis for future probabilistic studies.     

A Probabilistic Tsunami Hazard Study of the Auckland Region, Part I: Propagation Modelling and Tsunami Hazard Assessment at the Shoreline

Regional source tsunamis represent a potentially devastating threat to coastal communities in New Zealand, yet are infrequent events for which little historical information is available. It is therefore essential to develop robust methods for quantitatively estimating the hazards posed, so that effective mitigation measures can be implemented. We develop a probabilistic model for the tsunami hazard posed to the Auckland region of New Zealand from the Kermadec Trench and the southern New Hebrides Trench subduction zones. An innovative feature of our model is the systematic analysis of uncertainty regarding the magnitude-frequency distribution of earthquakes in the source regions. The methodology is first used to estimate the tsunami hazard at the coastline, and then used to produce a set of scenarios that can be applied to produce probabilistic maps of tsunami inundation for the study region; the production of these maps is described in part II. We find that the 2,500 year return period regional source tsunami hazard for the densely populated east coast of Auckland is dominated by events originating in the Kermadec Trench, while the equivalent hazard to the sparsely populated west coast is approximately equally due to events on the Kermadec Trench and the southern New Hebrides Trench.     

Forecasting tsunamis in Poverty Bay,New Zealand,with deep-ocean gauges

The response/transfer function of a coastal site to a remote open-ocean point is introduced, with the intent to directly convert open-ocean measurements into the wave time history at the site. We show that the tsunami wave at the site can be predicted as the wave is measured in the open ocean as far as 1,000+ km away from the site, with a straightforward computation which can be performed almost instantaneously. The suggested formalism is demonstrated for the purpose of tsunami forecasting in Poverty Bay, in the Gisborne region of New Zealand. Directional sensitivity of the site response due to different conditions for the excitation of the shelf and the bay’s normal modes is investigated and used to explain tsunami observations. The suggested response function formalism is validated with available records of the 2010 Chilean tsunami at Gisborne tide gauge and at the nearby deep-ocean assessment and reporting of tsunamis (DART) station 54401. The suggested technique is also demonstrated by hindcasting the 2011 Tohoku tsunami and 2012 Haida Gwaii tsunami at Monterey Bay, CA, using an offshore record of each tsunami at DART station 46411.

     

Estimation of Tsunami Hazard in New Zealand due to South American Earthquakes

We develop a probabilistic model for estimating the tsunami hazard along the coast of New Zealand due to plate-interface earthquakes along the South American subduction zone. To do this we develop statistical and physical models for several stages in the process of tsunami generation and propagation, and develop a method for combining these models to produce hazard estimates using a Monte-Carlo technique. This process is largely analogous to that used for seismic hazard modelling, but is distinguished from it by the use of a physical model to represent the tsunami propagation, as opposed to the use of empirical attenuation models for probabilistic seismic hazard analysis.     

The July 15, 2009 Fiordland,New Zealand Tsunami: Real-Time Assessment

On 15 July 2009, a Mw 7.8 earthquake occurred off the New Zealand coast, which by serendipitous coincidence occurred while the International Tsunami Symposium was in session in Novosibirsk, Russia. The earthquake generated a tsunami that propagated across the Tasman Sea and was detected in New Zealand, Australia and as far away as the US West coast. Small boats close to the epicenter were placed in jeopardy, but no significant damage was observed despite a measured run-up height of 2.3 m in one of the Sounds in close proximity to the source (Wilson in GNS Science Report 46:62 2009). Peak-to-trough tsunami heights of 55 cm were measured at Southport, Tasmania and a height of 1 m was measured in Jackson Bay, New Zealand. The International Tsunami Symposium provided an ideal venue for illustration of the value of immediate real-time assessment and provided an opportunity to further validate the real time forecasting capabilities with the scientific community in attendance. A number of agencies with responsibility for tsunami forecast and/or warning, such as the NOAA Center for Tsunami Research, the Pacific Tsunami Warning Center, GNS Science in New Zealand, the Australian Bureau of Meteorology and the European Commission Joint Research Centre were all represented at the meeting and were able to demonstrate the use of state of the art numerical models to assess the tsunami potential and provide warning as appropriate.     

Validation and Verification of Tsunami Numerical Models

In the aftermath of the 26 December, 2004 tsunami, several quantitative predictions of inundation for historic events were presented at international meetings differing substantially from the corresponding well-established paleotsunami measurements. These significant differences attracted press attention, reducing the credibility of all inundation modeling efforts. Without exception, the predictions were made using models that had not been benchmarked. Since an increasing number of nations are now developing tsunami mitigation plans, it is essential that all numerical models used in emergency planning be subjected to validation—the process of ensuring that the model accurately solves the parent equations of motion—and verification—the process of ensuring that the model represents geophysical reality. Here, we discuss analytical, laboratory, and field benchmark tests with which tsunami numerical models can be validated and verified. This is a continuous process; even proven models must be subjected to additional testing as new knowledge and data are acquired. To date, only a few existing numerical models have met current standards, and these models remain the only choice for use for real-world forecasts, whether short-term or long-term. Short-term forecasts involve data assimilation to improve forecast system robustness and this requires additional benchmarks, also discussed here. This painstaking process may appear onerous, but it is the only defensible methodology when human lives are at stake. Model standards and procedures as described here have been adopted for implementation in the U.S. tsunami forecasting system under development by the National Oceanic and Atmospheric Administration, they are being adopted by the Nuclear Regulatory Commission of the U.S. and by the appropriate subcommittees of the Intergovernmental Oceanographic Commission of UNESCO.     

A Probabilistic Tsunami Hazard Study of the Auckland Region, Part II: Inundation Modelling and Hazard Assessment

Regional source tsunamis pose a potentially devastating hazard to communities and infrastructure on the New Zealand coast. But major events are very uncommon. This dichotomy of infrequent but potentially devastating hazards makes realistic assessment of the risk challenging. Here, we describe a method to determine a probabilistic assessment of the tsunami hazard by regional source tsunamis with an “Average Recurrence Interval” of 2,500-years. The method is applied to the east Auckland region of New Zealand. From an assessment of potential regional tsunamigenic events over 100,000 years, the inundation of the Auckland region from the worst 100 events is modelled using a hydrodynamic model and probabilistic inundation depths on a 2,500-year time scale were determined. Tidal effects on the potential inundation were included by coupling the predicted wave heights with the probability density function of tidal heights at the inundation site. Results show that the more exposed northern section of the east coast and outer islands in the Hauraki Gulf face the greatest hazard from regional tsunamis in the Auckland region. Incorporating tidal effects into predictions of inundation reduced the predicted hazard compared to modelling all the tsunamis arriving at high tide giving a more accurate hazard assessment on the specified time scale. This study presents the first probabilistic analysis of dynamic modelling of tsunami inundation for the New Zealand coast and as such provides the most comprehensive assessment of tsunami inundation of the Auckland region from regional source tsunamis available to date.     

Forecasting tsunamis in Poverty Bay, New Zealand, with deep-ocean gauges

The response/transfer function of a coastal site to a remote open-ocean point is introduced, with the intent to directly convert open-ocean measurements into the wave time history at the site. We show that the tsunami wave at the site can be predicted as the wave is measured in the open ocean as far as 1,000+ km away from the site, with a straightforward computation which can be performed almost instantaneously. The suggested formalism is demonstrated for the purpose of tsunami forecasting in Poverty Bay, in the Gisborne region of New Zealand. Directional sensitivity of the site response due to different conditions for the excitation of the shelf and the bay’s normal modes is investigated and used to explain tsunami observations. The suggested response function formalism is validated with available records of the 2010 Chilean tsunami at Gisborne tide gauge and at the nearby deep-ocean assessment and reporting of tsunamis (DART) station 54401. The suggested technique is also demonstrated by hindcasting the 2011 Tohoku tsunami and 2012 Haida Gwaii tsunami at Monterey Bay, CA, using an offshore record of each tsunami at DART station 46411.     

Modeling of the 2011 Japan Tsunami: Lessons for Near-Field Forecast

During the devastating 11 March 2011 Japanese tsunami, data from two tsunami detectors were used to determine the tsunami source within 1.5 h of earthquake origin time. For the first time, multiple near-field tsunami measurements of the 2011 Japanese tsunami were used to demonstrate the accuracy of the National Oceanic and Atmospheric Administration (NOAA) real-time flooding forecast system in the far field. To test the accuracy of the same forecast system in the near field, a total of 11 numerical models with grids telescoped to 2 arcsec (~60 m) were developed to hindcast the propagation and coastal inundation of the 2011 Japanese tsunami along the entire east coastline of Japan. Using the NOAA tsunami source computed in near real-time, the model results of tsunami propagation are validated with tsunami time series measured at different water depths offshore and near shore along Japan’s coastline. The computed tsunami runup height and spatial distribution are highly consistent with post-tsunami survey data collected along the Japanese coastline. The computed inundation penetration also agrees well with survey data, giving a modeling accuracy of 85.5 % for the inundation areas along 800 km of coastline between Ibaraki Prefecture (north of Kashima) and Aomori Prefecture (south of Rokkasho). The inundation model results highlighted the variability of tsunami impact in response to different offshore bathymetry and flooded terrain. Comparison of tsunami sources inferred from different indirect methods shows the crucial importance of deep-ocean tsunami measurements for real-time tsunami forecasts. The agreement between model results and observations along Japan’s coastline demonstrate the ability and potential of NOAA’s methodology for real-time near-field tsunami flooding forecasts. An accurate tsunami flooding forecast within 30 min may now be possible using the NOAA forecast methodology with carefully placed tsunameters and large-scale high-resolution inundation models with powerful computing capabilities.     

Comparison of the 2010 Chile and 2011 Japan Tsunamis in the Far Field

In this study we analyze water level data from coastal tide gauges and deep-ocean tsunameters to explore the far-field characteristics of two major trans-Pacific tsunamis, the 2010 Chile and the 2011 Japan (Tohoku-oki) events. We focused our attention on data recorded in California (14 stations) and New Zealand (31 stations) as well as on tsunameters situated along the tsunami path and proximal to the study sites. Our analysis considers statistical analyses of the time series to determine arrival times of the tsunami as well as the timing of the largest waves and the highest absolute sea levels. Fourier and wavelet analysis were used to describe the spectral content of the tsunami signal. These characteristics were then compared between the two events to highlight similarities and differences between the signals as a function of the receiving environment and the tsunami source. This study provides a comprehensive analysis of far-field tsunami characteristics in the Pacific Ocean, which has not experienced a major tsunami in nearly 50 years. As such, it systematically describes the tsunami response characteristics of modern maritime infrastructure in New Zealand and California and will be of value for future tsunami hazard assessments in both countries.     

Obtaining natural oscillatory modes of bays and harbors via Empirical Orthogonal Function analysis of tsunami wave fields

To a tsunami wave, bays and harbors represent oscillatory systems, whose resonance (normal) modes determine the response to tsunami and consequently the potential hazard. The direct way to obtain the resonance modes of a water reservoir is by solving the boundary problem for the eigenfunctions of the linearized shallow-water wave equation. The principal difficulty of posing such a problem for a basin coupled to an ocean is specifying the boundary between the two. The technique developed in this work allows the normal modes of a semi-enclosed water body to be obtained without a-priori restricting the resonator area. The technique utilizes complex Empirical Orthogonal Function analysis of modeled tsunami wave fields. On the examples of Poverty Bay in New Zealand and Monterey Bay in California (United States), we demonstrate that the normal modes can be identified and isolated using the EOFs of a data set comprised of the concatenated time-series collected from different tsunami scenarios in a basin. The analysis of the modeled tsunami wave fields for the normal modes can also answer the question of how likely and under which conditions the different modes are exited, due to feasible natural events.     

The Tsunami of 26 December, 2004: Numerical Modeling and Energy Considerations

A numerical model for the global tsunamis computation constructed by Kowalik et al. (2005), is applied to the tsunami of 26 December, 2004 in the World Ocean from 80°S to 69°N with spatial resolution of one minute. Because the computational domain includes close to 200 million grid points, a parallel version of the code was developed and run on a Cray X1 supercomputer. An energy flux function is used to investigate energy transfer from the tsunami source to the Atlantic and Pacific Oceans. Although the first energy input into the Pacific Ocean was the primary (direct) wave, reflections from the Sri Lankan and eastern shores of Maldives were a larger source. The tsunami traveled from Indonesia, around New Zealand, and into the Pacific Ocean by various routes. The direct path through the deep ocean to North America carried miniscule energy, while the stronger signal traveled a considerably longer distance via South Pacific ridges as these bathymetric features amplified the energy flux vectors. Travel times for these amplified energy fluxes are much longer than the arrival of the first wave. These large fluxes are organized in the wave-like form when propagating between Australia and Antarctica. The sources for the larger fluxes are multiple reflections from the Seychelles, Maldives and a slower direct signal from the Bay of Bengal. The energy flux into the Atlantic Ocean shows a different pattern since the energy is pumped into this domain through the directional properties of the source function. The energy flow into the Pacific Ocean is approximately 75% of the total flow to the Atlantic Ocean. In many locations along the Pacific and Atlantic coasts, the first arriving signal, or forerunner, has lower amplitude than the main signal which often is much delayed. Understanding this temporal distribution is important for an application to tsunami warning and prediction.     

Evaluation of the 15 July 2009 Fiordland,New Zealand Tsunami in the Source Region

We undertake detailed near-field numerical modelling of the tsunami generated by the 15 July 2009 earthquake (Mw 7.8) in Fiordland, New Zealand. High resolution bathymetry and topography data at Breaksea and Dusky Sounds, and Chalky and Preservation Inlets are derived mostly from digitised New Zealand nautical charts, Shuttle Radar Topographic Mission (SRTM) 3 arc-second data, and General Bathymetric Chart of the Ocean (GEBCO) 30 s data. A combination of continuous and campaign Global Positioning System (GPS), satellite radar (ALOS/PALSAR InSAR images) and seismology data are used to constrain the seafloor deformation for the initial tsunami condition. This source model, derived independently of DART observations, provides an excellent fit to observed tsunami elevations recorded by DART buoy 55015. The model results in the near field show maximum tsunami elevations in the range 0.5–2.0 m inside the sounds and inlets with maximum flow speeds of 3.0 m/s. Along the open coast, maximum tsunami elevations reach 2.0 m. The high flow speeds through the inlets may change the inlet stratifications and water mass inside the sounds. Media reports and field reconnaissance data show some tsunami evidence at Cormorant Cove, Duck and Goose Coves, and Passage Point.     

The July 1998 Papua New Guinea Earthquake: Mechanism and Quantification of Unusual Tsunami Generation

— The unusual tsunami generated by the July 17, 1998 Papua New Guinea earthquake was investigated on the basis of various geophysical observations, including seismological data, tsunami waveform records, and on-land and submarine surveys. The tsunami source models were constructed for seismological high-angle and low-angle faults, splay fault, and submarine slumps. Far-field and near-field tsunamis computed from these models were compared with the recorded waveforms in and around Japan and the measured heights along the coast around Sissano Lagoon, respectively. In order to reproduce the far-field tsunami waveforms, small sources such as splay fault or submarine slump alone were not enough, and a seismological fault model was required. Relocated aftershock distribution and observed coastal subsidence were preferable for the low-angle fault, but the low-angle fault alone could not reproduce the large near-field tsunamis. The low-angle fault with additional source, possibly a submarine slump, is the most likely source of the 1998 tsunami, although other possibilities cannot be excluded. Computations from different source models showed that the far-field tsunami amplitudes are proportional to the displaced water volume at the source, and the comparison with the observed tsunami amplitudes indicated that the displaced water volume at the 1998 tsunami source was ～0.6 km³. The near-filed tsunami heights, on the other hand, are determined by the potential energy of displaced water, and the comparison with the observed heights showed that the potential energy was ～2 × 10¹² J.     

海底地震有限断层破裂模型对近场海啸数值预报的影响

快速准确的海啸源模型是近场海啸精确预警的关键.尽管目前还没有办法直接对其进行正演定量计算,但是可以通过多源地震、海啸观测数据进行反演或联合反演推算.不同的海啸源可能导致不同的预警结论,了解不同类型海啸源适用性、评估海啸源特征差异对近场海啸的影响,无论对于海啸预警还是海啸模拟研究尤为重要.本文评估分析了6种不同同震断层模型对2011年3月11日日本东北地震海啸近场数值预报的影响,重点对比分析了有限断层模型与均一滑动场模型对近场海啸产生、传播、淹没特征的影响及各自的误差.研究表明:近场海啸波能量分布主要取决于海啸源分布特征,特别是走向角的差异对海啸能量分布影响较大;有限断层模型对海啸灾害最为严重的39°N以南沿岸地区的最大海啸爬坡高度明显优于均一滑动场模型结果;综合对比DART浮标、GPS浮标及近岸潮位站共32个站次的海啸波幅序列结果发现有限断层模型整体平均绝对/相对误差比均一滑动场模型平均误差要低,其中Fujii海啸源的平均绝对/相对误差最小,分别是0.56m和26.71%.UCSB海啸源的平均绝对/相对误差次之.3个均一滑动场模型中USGSCMT海啸源模拟精度最高.相对于深海、浅海观测站,有限断层模型比均一滑动场模型对近岸观测站计算精度更高.海啸源误差具有显著的方向性,可能与反演所采用的波形数据的代表性有关;谱分析结果表明Fujii海啸源对在12至60min主频波谱的模拟要优于UCSB海啸源.海啸源中很难真实反映海底地震破裂过程,然而通过联合反演海啸波形数据推算海啸源的方法可以快速确定海啸源,并且最大限度的降低地震破裂过程与海啸产生的不确定性带来的误差.     

Seismic vulnerability scenarios of Unreinforced Masonry churches in New Zealand

A vulnerability analysis of c.300 unreinforced Masonry churches in New Zealand is presented. The analysis uses a recently developed vulnerability index method (Cattari et al. in Proceedings of the New Zealand Society for Earthquake Engineering NZSEE 2015 conference, Rotorua, New Zealand, 2015a; b; SECED 2015 conference: earthquake risk and engineering towards a Resilient World, Cambridge; Goded et al. in Vulnerability analysis of unreinforced masonry churches (EQC 14/660)—final report, 2016; Lagomarsino et al. in Bull Earthq Eng, 2018), specifically designed for New Zealand churches, based on a widely tested approach for European historical buildings. It consists of a macroseismic approach where the seismic hazard is defined by the intensity and correlated to post seismic damage. The many differences in typologies of New Zealand and European churches, with very simple architectural designs and a majority of one nave churches in New Zealand, justified the need to develop a method specifically created for this country. A statistical analysis of the churches damaged during the 2010–2011 Canterbury earthquake sequence was previously carried out to develop the vulnerability index modifiers for New Zealand churches. This new method has been applied to generate seismic scenarios for each church, based on the most likely seismic event for 500 years return period, using the latest version of New Zealand’s National Seismic Hazard Model. Results show that highly vulnerable churches (e.g. stone churches and/or with a weak structural design) tend to produce higher expected damage even if the intensity level is lower than for less vulnerable churches in areas with slightly higher seismicity. The results of this paper provide a preliminary tool to identify buildings requiring in depth structural analyses. This paper is considered as a first step towards a vulnerability analysis of all the historical buildings in the country, in order to preserve New Zealand’s cultural and historical heritage.     

Numerical Simulation of the 2011 Tohoku Tsunami Based on a New Transient FEM Co-seismic Source: Comparison to Far- and Near-Field Observations

In this work, we simulate the 2011 M9 Tohoku-Oki tsunami using new coseismic tsunami sources based on inverting onshore and offshore geodetic data, using 3D Finite Element Models (FEM). Such FEMs simulate elastic dislocations along the plate boundary interface separating the stiff subducting Pacific Plate from the relatively weak forearc and volcanic arc of the overriding Eurasian plate. Due in part to the simulated weak forearc materials, such sources produce significant shallow slip (several tens of meters) along the updip portion of the rupture near the trench. To assess the accuracy of the new approach, we compare observations and numerical simulations of the tsunami's far- and near-field coastal impact for: (i) one of the standard seismic inversion sources (UCSB; Shao et al. 2011); and (ii) the new FEM sources. Specifically, results of numerical simulations for both sources, performed using the fully nonlinear and dispersive Boussinesq wave model FUNWAVE-TVD, are compared to DART buoy, GPS tide gauge, and inundation/runup measurements. We use a series of nested model grids with varying resolution (down to 250 m nearshore) and size, and assess effects on model results of the latter and of model physics (such as when including dispersion or not). We also assess the effects of triggering the tsunami sources in the propagation model: (i) either at once as a hot start, or with the spatiotemporal sequence derived from seismic inversion; and (ii) as a specified surface elevation or as a more realistic time and space-varying bottom boundary condition (in the latter case, we compute the initial tsunami generation up to 300 s using the non-hydrostatic model NHWAVE). Although additional refinements are expected in the near future, results based on the current FEM sources better explain long wave near-field observations at DART and GPS buoys near Japan, and measured tsunami inundation, while they simulate observations at distant DART buoys as well or better than the UCSB source. None of the sources, however, are able to explain the largest runup and inundation measured between 39.5° and 40.25°N, which could be due to insufficient model resolution in this region (Sanriku/Ria) of complex bathymetry/topography, and/or to additional tsunami generation mechanisms not represented in the coseismic sources (e.g., splay faults, submarine mass failure). This will be the object of future work.     

Tsunami hazard in the shorelines of Khark island (Persian Gulf), Iran

Iran is located in one of the seismically active regions of the world. Due to the high probability of earthquakes throughout the country and the potential for tsunami inundation along the coasts and offshore, comprehensive studies on the interaction of these natural phenomena are necessary. In this study, the most conservative scenarios are determined for possible earthquakes within the Khark zone (Persian Gulf) based on experimental relations between the fault length, magnitude and displacement, which are parameters for determining tsunamigenic sources. Subsequently, the maximum height of tsunami waves are calculated based on the specifications of the seismic source and its distance from the shore as well as the coastal slope. A zoning map of tsunami hazard is finally presented.     

Tsunami Probability in the Caribbean Region

We calculated tsunami runup probability (in excess of 0.5 m) at coastal sites throughout the Caribbean region. We applied a Poissonian probability model because of the variety of uncorrelated tsunami sources in the region. Coastlines were discretized into 20 km by 20 km cells, and the mean tsunami runup rate was determined for each cell. The remarkable ~500-year empirical record compiled by O’Loughlin and Lander (2003) was used to calculate an empirical tsunami probability map, the first of three constructed for this study. However, it is unclear whether the 500-year record is complete, so we conducted a seismic moment-balance exercise using a finite-element model of the Caribbean-North American plate boundaries and the earthquake catalog, and found that moment could be balanced if the seismic coupling coefficient is c = 0.32. Modeled moment release was therefore used to generate synthetic earthquake sequences to calculate 50 tsunami runup scenarios for 500-year periods. We made a second probability map from numerically-calculated runup rates in each cell. Differences between the first two probability maps based on empirical and numerical-modeled rates suggest that each captured different aspects of tsunami generation; the empirical model may be deficient in primary plate-boundary events, whereas numerical model rates lack backarc fault and landslide sources. We thus prepared a third probability map using Bayesian likelihood functions derived from the empirical and numerical rate models and their attendant uncertainty to weight a range of rates at each 20 km by 20 km coastal cell. Our best-estimate map gives a range of 30-year runup probability from 0–30% regionally.