Developing a Probabilistic Tsunami Hazard Assessment Framework for Pacific Sources: USGS Powell Center Meeting Summary Open Access

The overall goal of the U.S. Geological Survey (USGS) Powell Center Working Group (WG) called Tsunami Source Standardization for Hazards Mitigation in the United States is to develop and implement a transparent, scientifically based methodology for characterizing hypothetical but realistic sources of tsunamis that pose a potential hazard to U.S. populations, commerce, and infrastructure. The results of this process will provide federal, state, and local authorities with probabilities of tsunami occurrence for improved mitigation and planning related to potential hazards from tsunamis. There have been efforts at a national scale elsewhere (e.g., in Australia, Davies and Griffin, 2018, and New Zealand, Power et al., 2017). This is the first time a WG has been organized to bring together such an extensive knowledge base to help in developing national-scale tsunami hazards analyses with an impact upon U.S. states and territories. The purpose of this specific WG meeting series was to leverage expert knowledge to form a consensus regarding the characteristics and parameters deemed most important for developing and analyzing tsunami sources in specific geographic regions. The process by which consensus is reached for respective regions varies due to differences in the quantity and quality of information available but includes parameter selection and weighting, iterative refinement, and implementation to derive outcomes that meet the needs of a broad constituency. Consistently defined tsunami sources will provide states and territories with a measure of consistency across shared source regions and an increased understanding of tsunami hazards to U.S. coastal communities. The March 2023 WG meeting that is the subject of this paper focused on tsunami sources throughout the Pacific Ocean Basin, except for those from the Alaska–Aleutian and Cascadia megathrusts, which had their own meetings devoted to those regions.

Probabilistic tsunami hazard analysis (PTHA) is a process used to develop a framework to inform tsunami modeling and tsunami hazard and risk analyses. This process uses a logic tree approach to best characterize the full spectrum of possible tsunamigenic behaviors and to quantify the likelihood that the tsunamigenic processes involved may strongly influence tsunami hazards (Geist and Lynett, 2014). The logic tree is intended to capture the wide range of variation in the processes that contribute to tsunami hazards. The logic tree includes a branch for each process, and tree branches are assigned weights that represent the relative importance of that process (the likelihood of that process to occur) as typically judged by a community of subject matter experts (SMEs), who are all members of the WG. Some processes are included even if they have a low likelihood of occurring so that later users of these data can understand the entire range of possibilities that were considered during PTHA. An important aspect of PTHA is that the process allows for the inclusion and quantification of epistemic and aleatory uncertainties (Grezio et al., 2017). Epistemic uncertainty refers to our knowledge of the physical processes and systems. Aleatory uncertainty refers to our ability to properly characterize a given phenomenon or phenomena.

The National Tsunami Hazard Mitigation Program (NTHMP) is the primary organization that supports tsunami hazard programs in U.S. states and territories. Established following the 1992 Cape Mendocino earthquake and tsunami, the NTHMP developed fundamental components of tsunami hazard mapping in the United States (e.g., tsunami model software benchmarking) and is the principal source of funding for the preparation of tsunami hazard maps. The 2024–2029 NTHMP Strategic Plan outlines support for PTHA under strategy 1.1.2: “Develop probabilistic tsunami hazard maps where sources are community-models and consistent on a national scale, especially for use in updating ASCE/Building Code ‘Tsunami Design Zone’ maps and in state, regional, and national risk assessments so results will be consistent and comparable to other hazards” (NTHMP, 2024). ASCE refers to the American Society of Civil Engineers, the organization that develops proposed engineering standards that are typically adopted by lead agencies as part of the building code. ASCE first incorporated tsunamis into its building code standards in ASCE 7-16 (ASCE, 2017).

WG members presented and evaluated existing peer-reviewed literature comprising the state of knowledge for tsunami sources surrounding the Pacific, other than the Alaska–Aleutian and Cascadia active margins. (The list of all WG members can be found in Supplemental Material at https://www.aegweb.org/e-eg-supplements.) Core group member (core member of the WG) Baoning Wu recorded notes each day and, at the beginning of each subsequent day, presented a summary of the discussions from the previous day. The core team members are all listed as co-authors of this paper. The various WG principal investigators (PIs) presented a proposed framework for the logic tree prepared for each of the tsunami sources (e.g., Thio, 2019, Figure C-2). Proposed logic tree branches include floating vs. segmented rupture, earthquake size, earthquake event rate, and earthquake fault slip distribution. Earthquake size includes various parameters that are split across multiple logic tree branches (e.g., rupture length and width, slip distribution, event rate, and magnitude). Because earthquake size, to a first order, is the principal factor contributing to tsunami size (Abe, 1979; Okal and Synolakis, 2004), WG members provided feedback on the design and content of the logic tree that focused primarily on the floating vs. segmented rupture and earthquake event rates branches (see the next section).

Uncertainty in earthquake size is accounted for by creating two alternative logic tree branches for floating and segmented fault rupture models. Floating rupture slip extents are imposed along a fault geometry with slip constrained by a magnitude distribution such as the Gutenberg–Richter (GR) relations, maximum magnitude, or characteristic earthquake models (Thio, 2019). These earthquakes are integrated randomly over a distribution of magnitudes and locations. Because of the random nature of these factors that control earthquakes, this approach may ignore pertinent geophysical and/or geological observations and data. In other words, floating ruptures are not constrained by geophysical and/or geological observations and data (e.g., a geological feature cutting across a tsunami source that may serve to control the size of an earthquake on that source; a floating rupture would ignore that geological feature).

A segmented fault slip model uses dimensions of fault segments to constrain earthquake magnitudes and can include multiple-segment ruptures. Segments are defined by the SMEs based on previous earthquakes, tectonic features, seafloor topography, and other geophysical characteristics, such as estimates of fault coupling, thickness of sediment overlying the subducting plate, or changes in plate age across fracture zones in the subducting plate, all of which may control possible segmentation (Ruff and Kanamori, 1980). Each of these types of geological information can increase the certainty of the potential for a segment boundary to serve as a boundary to earthquake rupture. Persistent boundaries in the model may artificially reduce the hazard near those boundaries, and this approach could potentially underestimate contributions to the tsunami hazards from smaller earthquakes. The conversations during the meeting focused on characteristics (listed above in this paragraph) of the subduction zones that help to identify possible segment boundaries and evaluation of the probability of a rupture going through a segment boundary (generating a multi-segment earthquake rupture).

Logic tree branches that handle source rate characterization compute the recurrence rate of earthquakes as a function of magnitude and location. Approaches include the use of plate rate, paleoseismicity, and seismicity catalog models, as discussed below. These models are built on data derived from plate rate information, paleoseismic studies that define recurrence and magnitude rates, and regression of seismicity catalogs, as briefly discussed for each approach in the next three paragraphs.

Plate rate models are stable away from microplates, and the use of plate rate models requires the determination of seismic coupling ratios (Wang and Dixon, 2004). The coupling ratio is the proportion of the plate convergence rate that is accumulated as tectonic strain, as opposed to contributing to aseismic slip (Lindsey et al., 2021). Tectonic strain is released during slip on earthquake faults, and aseismic slip is fault slip that happens between earthquake events. Event rates are computed from effective convergence rates and magnitude distributions and are considered to be the best option when other information is unavailable. The best plate rate model must capture all pertinent facets of the geological and tectonic setting (as these pertain to tsunamigenesis).

A second approach includes the use of paleoseismic models, in which earthquake event recurrence is based on identification of past historic and prehistoric geologic events. One challenge for this approach is to determine the earthquake size for these events that are often represented by a single point in time and space. Another challenge is to distinguish single events from multiple events when age uncertainty overlaps from adjacent paleoseismology locations. The paleoseismic model approach can be used to validate results from the other two approaches and is the preferred method where the prehistorical record is sufficiently long.

A third approach includes the regression of seismicity catalogs (e.g., GR), which represent a small window of time for earthquake fault systems. One challenge with this method is to distinguish megathrust subduction zone interface earthquakes from earthquakes within subducting plates (i.e., slabs). It is possible that GR relations are inappropriate for megathrust earthquakes, because seismicity rates may underpredict large-magnitude earthquake recurrence, as is found to be the case for the Cascadia subduction zone. The reason for this potential for underprediction is based on several factors (Lay et al., 1982). The Cascadia megathrust lacks any recorded subduction zone earthquakes, so the earthquake catalog is dominated by smaller-magnitude events (Stein and Wysession, 2003; Goldfinger et al., 2012). Therefore, the catalog does not include the probability of large-magnitude events. In addition, the paleoseismic evidence for earthquakes in Cascadia shows that they occur less frequently than GR predictions based on earthquakes of smaller magnitude (Atwater et al., 2005). However, in places where there is a record of historical tsunamigenic earthquakes, like along the Nankai subduction zone offshore of eastern Japan, the use of GR scaling can be compared with the historical record. For Nankai, the historical record matched well with GR-based regression (Baba et al., 2022).

WG members used an online independent polling tool to collect their opinions for various factors. The key part of the logic tree that requires WG member input is the segmentation of the megathrust faults. For each plate margin, WG members considered hypothetical segment boundaries. WG member Serge Lallemand provided digital maps on which the WG members placed segment boundaries for discussion via videoconferencing software.

At the beginning of the meeting, core group member and PI Stephanie Ross reviewed the process that the weeklong meeting would follow and outlined the agenda for the first day as follows: (1) present the results from previous meetings, (2) discuss end-user needs as viewed from states and territories, (3) hear topical presentations relevant to PTHA, and (4) preview a hypothetical logic tree to be considered at the meeting. Ross summarized how the past meetings focused on tsunami sources in the Alaska–Aleutian subduction zone; the U.S. Gulf Coast, U.S. East Coast, and U.S. Caribbean Territories; and the Cascadia subduction zone. Core group member Hong Kie Thio used the results of the Alaska–Aleutian tsunami source meeting to discuss the components of the PTHA process as applied during that meeting. Thio’s presentation helped WG members understand the scope and goals of the Pacific tsunami sources meeting. Core group member Jason Patton then presented a summary of the Cascadia tsunami source meeting, including a review of its draft logic tree, also to help WG members prepare for the Pacific tsunami sources meeting. In addition, Patton discussed the ongoing post-meeting efforts to complete the Cascadia logic tree, which includes an updated megathrust fault geometry based on recent seismic reflection data acquisitions from the CASIE21 multi-channel seismic survey conducted in 2021 (Carbotte et al., 2023).

Core group member and PI Rick Wilson presented a summary of end-user needs to develop products based on the PTHA process. These end-user needs were based on information from the NTHMP state and territory representatives. PTHA is applied in California in several ways. First, it allows the California Building Standards Commission to update the Tsunami Design Zones for use in the California Building Code (ASCE, 2017). PTHA also provides the basis for the California Geological Survey (CGS) to develop project-level tsunami hazard zones that will help the respective lead agencies mitigate tsunami hazards through evacuation or building design. The results from PTHA let the Federal Emergency Management Agency (FEMA) update its flood hazard risk analyses via the Risk Mapping, Assessment and Planning process (FEMA, 2024c). The PTHA results provide a basis for the California Governor’s Office of Emergency Services and the CGS to update evacuation maps, assist communities with planning for construction of vertical evacuation structures that provide protection from a tsunami, and update FEMA Hazus risk modeling (FEMA, 2024a). There are several communities in California, Oregon, and Washington (e.g., Crescent City, CA; Seaside, OR; and Ocean Shores, WA) where the time it takes to get out of the tsunami hazard zone is too short to successfully evacuate, and vertical evacuation structures remain the main way to mitigate tsunami hazards in these locations. Tsunami modeling based on PTHA provides the basis for the engineering design for these vertical evacuation towers. FEMA’s Natural Hazards Risk Assessment Program manages the Hazus risk analysis program, which specifically includes hazards from tsunamis (FEMA, 2024b). Hazus software combines hazard information (e.g., probabilistic tsunami model results) with population and exposure information to calculate the potential impact of tsunamis on casualties and infrastructure damage.

Following these informative introductory presentations, Baoning Wu discussed information about the implications of dynamic rupture models for tsunami hazards (e.g., Wu et al., 2019). WG member Allison Shumway presented an overview of the USGS National Seismic Hazard Model (NSHM), including details about segmentation, applying event rates to this segmentation, and magnitude scaling relations, because PTHA is intended to be consistent with this NSHM (Petersen et al., 2023). WG member James Goff addressed how tsunami source information can be inferred from field studies, emphasizing the need for the development of a global tsunami database for prehistoric and historical tsunamis (Goff et al., 2012). Core group member Hong Kie Thio discussed PTHA modeling and how it might be applied to the Pacific sources (Thio, 2019). WG member Gareth Davies presented a global-scale PTHA, the importance of developing sensitivity studies for that PTHA, how logic trees are used to develop earthquake magnitude–frequency distributions, and ways to obtain onshore inundation modeling results (Grezio et al., 2017). WG member William Power discussed the tsunami hazard analysis for New Zealand and how a PTHA was developed. This New Zealand PTHA included tsunami source characterization, magnitude–frequency distribution, tsunami modeling with COMCOT software (Wang and Liu, 2006), and hazard calculation (Power et al., 2017). Finally, core group member and PI Marie Eble presented an introduction of the tsunami sources chosen to be evaluated at this meeting, based on a tsunami sensitivity analysis for Pacific Ocean Basin subduction zones. These margins were initially organized as follows: (1) Kamchatka/Kuril, (2) Nankai/Ryukyu, (3) Izu–Bonin, (4) Mariana, (5) Philippine, (6) New Britain/Manus/Kilinailau, (7) Solomon/Vanuatu, (8) Tonga/Kermadec, (9) Ecuador/Peru, and (10) Chile (Figure 1). Core group member Pat Lynett wrapped up the day by showing how weight percentages would be compiled for each logic tree branch using online polling.

The meeting continued with a series of global overview presentations followed by talks that focused on specific margins. WG member Emile Okal presented a general discussion about subduction earthquakes and tsunamigenesis and noted important examples of tsunami earthquakes, irregular fault fragmentation, and other sources, such as outer-rise earthquakes. WG member Shannon Graham discussed a circum-Pacific evaluation of plate motions and megathrust fault coupling ratios (Graham et al., 2021). The Graham et al. (2021) coupling ratios are based on a global-scale block model that incorporates plate motion rates that impart strain accumulation within tectonic blocks and along block boundaries (the plate boundary faults). This modeling generates estimates of fault slip deficit (how much of the relative plate motions are stored along faults, in units of millimeter per year). The coupling fraction is calculated by dividing the slip deficit by the plate motion rate. Graham et al. (2021) used these model results to generate potential rupture scenarios for subduction zone systems and compared them with previous global earthquake forecasts (e.g., Plescia and Hayes, 2020). WG member Kirstie Haynie presented an overview of the USGS Slab 2.0 model subduction zone geometries, the methods used to prepare these models, and how the models will be updated in the future (Hayes et al., 2018). Data used to create Slab 2.0 include active seismic source data, the global centroid moment tensor catalog, earthquake locations, bathymetry, and seismic receiver functions.

WG members Marco Pagani and Kendra Johnson presented the Global Earthquake Model (GEM) and discussed the components of their model, the methods used to prepare their model, and their approach to make further improvements in the future (Pagani et al., 2020). GEM is a non-profit, public–private partnership that leads a global collaborative effort to develop tools and resources for earthquake hazard and risk assessment. GEM develops probabilistic seismic hazard analyses for subduction zones that are based on subduction zone fault geometry, along-strike segmentation, a classification of seismicity (so that earthquakes are associated with the correct source: interface, crustal, intraslab, volcanic, or unclassified), the source occurrence rates, and epistemic uncertainties. WG member Gareth Davies presented a summary of the 2018 Australian PTHA and made suggestions about preparation of PTHA models (Davies and Griffin, 2018). Davies and Griffin (2018) incorporate fault geometry from Slab 1.0 and Slab 2.0, apply three types of earthquake slip to those geometries (fixed-area uniform slip, variable-area uniform slip, and heterogeneous slip), and use various scaling relations for the constraints on earthquake size (e.g., scaling relations and GR magnitude distribution; Strasser et al., 2010). WG member Brian Atwater discussed methods used to conduct paleotsunami investigations and to apply those methods in the field. Atwater emphasized that the taphonomy of deposits can be a confounding factor that limits the preservation of tsunami deposits in the geological record (i.e., crab bioturbation). WG member James Goff discussed a hypothetical Pacific paleotsunami database and provided some transdisciplinary examples of collaboration (e.g., between archeologists and anthropologists; Goff et al., 2012, 2022).

WG member Matías Carvajal introduced specific margins by discussing an overview of the seismotectonic setting in the subduction zone along the west coast of South America. Carvajal showed several examples of subdivisions of the margin into segments, both large and small (Philibosian and Meltzner, 2020; Molina et al., 2021). Carvajal concluded his presentation with an overview of the relations between earthquake and tsunami size for the Chilean margin (Carvajal et al., 2017a, 2017b) and emphasized the relations between earthquake size and far-field tsunami size. Tsunami size is often controlled by the down-dip slip distribution (a larger slip in the up-dip region leads to a larger tsunami; Carvajal et al., 2017a). WG member Emile Okal discussed the subduction zones along the Peru, Ecuador, and Central America margins. WG members Bill Fry, William Power, and Laura Wallace discussed the subduction zones along the Hikurangi, Kermadec, Tonga, Vanuatu, Solomon, New Britain, Manus, and Kilinailau margins. Some regions have a rich history and prehistory of earthquakes and tsunamis (like the Hikurangi margin), whereas other regions do not (so PTHA inputs are more heavily reliant upon modeling methods, like for the Kermadec and Tonga subduction zones).

WG member Yuichiro Tanioka discussed paleotsunamis, earthquake slip models, and fault coupling along Japanese subduction zones, including the Kuril, Japan, Nankai, Ryukyu, and Izu–Bonin margins. The Japan and Nankai margins have a rich record of historical events that can be used to inform PTHA. WG member Harold Tobin presented evidence for splay faults along subduction zones focusing on the Nankai margin (e.g., Bassett et al., 2022). Tobin used the 1944 and 1946 earthquakes in this region to show how the eastern half of the Nankai margin (1946 earthquake) has a well-defined splay fault but the western margin (1944 earthquake) does not (Bassett et al., 2022). WG member Breanyn MacInnes discussed modern tsunamis and paleotsunami evidence for the Kuril/Kamchatka subduction zone (MacInnes et al., 2016; Pinegina et al., 2018). Evidence for paleotsunamis along the Kurils extends to 7,500 years ago, and MacInnes used the correlation of these deposits, along with mapped fracture zones along the margin, to suggest possible ways to segment the margin and to suggest that there may be changes in earthquake frequency with time. The central Kurils have a higher rate of tsunamis (recurrence around 200–300 years), and the southern Kurils and Hokkaido have a lower rate of tsunamis (recurrence around 400–800 years). WG member Serge Lallemand summarized the elements that could control tsunamigenesis in the Mariana and Philippine subduction zones, including an overview of the utility of the Submap database (Lallemand et al., 2018). The Submap database includes a summary of subduction zones using about 200 parameters (including structural, slab, kinematic, and seismogenic zone parameters) and uses about 260 margin normal transects to explore these parameters. Lallemand emphasized how structures in the subducting plate (e.g., seamounts, ridges, and fracture zones), which constitute forms of roughness in the subducting plate, can control interplate coupling and seismic rupture. In short, a rough seafloor can lead to low coupling, and a smooth seafloor can lead to high coupling (Ruff and Kanamori, 1980, 1983). There is also a SubQuake database with rupture information from 169 subduction zone earthquakes since 1900 (van Rijsingen et al., 2018). WG member William Power continued his discussion of the subduction tectonics along the Manus, New Britain, and Solomon Islands margins. WG member Shannon Graham presented a brief overview of the Mexico subduction zone. Core group members Hong Kie Thio and Jason Patton brought the discussion back to the logic tree topic in the context of segment boundaries that had been proposed over the previous days.

Core group member Pat Lynett led WG members in a discussion about constructing the logic tree branches for the Pacific tsunami sources, emphasizing that each margin is different and that logic trees for each margin would differ based on these differences. The WG decided to organize the margins in the following manner (Figure 1):

1. Chile/Colombia/Ecuador/Peru

2. Middle America

3. Tonga/Kermadec/Hikurangi

4. Solomon/Vanuatu

5. New Britain/Manus/Kilinailau

6. Mariana/East Philippine/Izu–Bonin

7. Japan/Nankai/Ryukyu

8. Kamchatka/Kuril

WG member Raphaël Paris led a 2-hour session on volcano tsunami sources that included the WG members from the main meeting plus many additional volcanic source WG members. The goal of this session was to explore the different tsunami source mechanisms associated with volcanic eruptions, focusing on what we know, what we do not know, and what is needed to fill the gaps. Paris listed the eight volcanogenic mechanisms: subaerial landslide, submarine landslide, underwater explosion, caldera collapse, column collapse, pyroclastic flow, volcano–tectonic earthquake, and atmospheric forcing. Lynett pointed out that a major challenge for incorporating volcanic sources in PTHA is the source of uncertainty in our ability to link physical mechanisms with tsunami size (epistemic uncertainty). This volcano tsunami discussion inspired the volcanologists and the WG to collaborate in holding an additional meeting to more fully address volcanic sources.

The WG members completed the segmentation and weight voting for the remaining subduction zone margins on the final day of the meeting. As a follow-up to the weeklong meeting, some WG members joined a 2-hour online session to discuss segmentation of the Manus, Kilinailau, New Britain, Solomon, and Vanuatu margins. A second online follow-up activity was undertaken to establish segmentation boundaries for the subduction zone forming the Middle America trench.

The core group, comprising the co-authors of this paper, continues working to complete the logic tree so that it can be incorporated into PTHA. This logic tree will be combined with the logic trees developed for the Alaska-Aleutian, Cascadia, crustal fault sources, and the Gulf of Mexico, Caribbean, and U.S. East Coast logic trees. The next steps will be to develop tsunami wave models for each source based on the logic trees. Results from the wave models will be used to generate equally spaced offshore points along the coast with tsunami hazard curves. These hazard curves will contain tsunami wave amplitudes for a range of exceedance probabilities. States and territories will then be able to use these hazard curves to select deterministic tsunami sources that meet probabilities of exceedance for any given return period. The first main use for these data will be in tsunami modeling at the 2,475-year return period (2 percent probability of exceedance in 50 years), which will form the basis for the ASCE 7-28 update that will be the Tsunami Design Zone incorporated into the building code for essential and critical facilities. ASCE 7-28, which is updating ASCE 7-22, is scheduled to be released in 2028 and to be adopted into the building code and by lead agencies in the years following 2028 (ASCE, 2022).

Since the 2023 meeting, a comprehensive publication on PTHA has been released (Goda et al., 2024), which outlines recent advances in the research and practice of PTHA. The PTHA process could continue to be improved (e.g., for seismic sources in relation to low number of historical events, fault parameterization, dynamics of fault mechanics, and empirical scaling relations) as outlined by Behrens et al. (2021). In addition, the future of PTHA may come in the form of probabilistic inundation modeling, which requires the application of modeling workflows customized for graphics processing unit–based high-performance computing methods (Gibbons et al., 2020; Davies et al., 2022).

The authors thank the USGS John Wesley Powell Center for Analysis and Synthesis (https://www.usgs.gov/centers/john-wesley-powell-center-for-analysis-and-synthesis) for supporting this project, especially Jill Baron, Demi Jasmine Bingham, Ian Steinig, and Leah Colasuonno. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.