A Source Clustering Approach for Efficient Inundation Modeling and Regional Scale Probabilistic Tsunami Hazard Assessment

Abstract

For coastal regions on the margin of a subduction zone, near-field megathrust earthquakes are the source of the most extreme tsunami hazards, and are important to handle properly as one aspect of any Probabilistic Tsunami Hazard Assessment. Typically, great variability in inundation depth at any point is possible due to the extreme variation in extent and pattern of slip over the fault surface. In this context, we present an approach to estimating inundation depth probabilities (in the form of hazard curves at a set of coastal locations) that consists of two components. The first component uses a Karhunen-Loève expansion to express the probability density function (PDF) for all possible events, with PDF parameters that are geophysically reasonable for the Cascadia Subduction Zone. It is then easy and computationally cheap to generate a large N number of samples from this PDF; doing so and performing a full tsunami inundation simulation for each provides a brute force approach to estimating probabilities of inundation. However, to obtain reasonable results, particularly for extreme flooding due to rare events, N would have to be so large as to make the tsunami simulations prohibitively expensive. The second component tackles this difficulty by using importance sampling techniques to adequately sample the tails of the distribution and properly re-weight the probability assigned to the resulting realizations, and by grouping the realizations into a small number of clusters that we believe will give similar inundation patterns in the region of interest. In this approach, only one fine-grid tsunami simulation need be computed from a representative member of each cluster. We discuss clustering based on proxy quantities that are cheap to compute over a large number of realizations, but that can identify a smaller number of clusters of realizations that will have similar inundation depths. The fine-grid simulations for each cluster representative can also be used to develop an improved strategy, in which these are combined with cheaper coarse-grid simulations of other members of the cluster. We illustrate the methodology by considering two coastal locations: Crescent City, CA and Westport, WA.

Probabilistic Tsunami Hazard and Risk Analysis: A Review of Research Gaps

“…However, the fastest HPC simulation workflows (e.g., de la Asunción et al, 2013;Oishi et al, 2015;Macías et al, 2017;Musa et al, 2018) still require typically 10-60 min to simulate tsunami inundation at a scale of tens of meters, rendering them unsuitable for extensive PTRA studies with up to millions of scenarios (Basili et al, 2021). To overcome this "challenge of scales", modeling approximations are presently necessary for PTHA feasibility and can either involve 1) largely reducing the number of inundation scenarios (e.g., González et al, 2009;Lorito et al, 2015;Volpe et al, 2019;Williamson et al, 2020), 2) use of approximate models or statistics such as amplification factors (e.g., Løvholt et al, 2012;Kriebel et al, 2017;Gailler et al, 2018;Glimsdal et al, 2019), or 3) machine learning-based tsunami emulators (e.g., Sarri et al, 2012;Salmanidou et al, 2017;Giles et al, 2020).…”

From offshore to onshore probabilistic tsunami hazard assessment via efficient Monte-Carlo sampling

“…Synthetic catalogues are often inefficient for tsunami inundation hazard calculations (Sepúlveda et al, 2019;Williamson et al, 2020). If seismicity follows a Gutenberg-Richter like distribution then they tend to be dominated by smaller magnitudes, while larger magnitude scenarios are resolved less well despite often being more relevant in tsunami hazard applications.…”

“…To improve on the efficiency of synthetic catalogues, PTHAs often employ stratified-sampling by earthquake magnitude (e.g. De Risi & Goda, 2017;Williamson et al, 2020;Basili et al, 2021;Zamora et al, 2021). The Monte-Carlo accuracy of this approach is analysed below, and techniques are presented to estimate the errors both before and after high-resolution tsunami simulation.…”

“…Develop efficient methods to compute Q(e) using high-performance-computing and/or approximation methods (e.g. de la Asunción et al, 2013;de Baar & Roberts, 2017;Guillas et al, 2018;Arce Acuña & Aoki, 2018;Williamson et al, 2020;Fukutani et al, 2021;Giles et al, 2021);…”

“…Our approach builds upon stratified-sampling, in which scenarios are binned by magnitude and a fixed number are randomly sampled from each bin (e.g. De Risi & Goda, 2017; Williamson et al, 2020;Basili et al, 2021;Zamora et al, 2021). This standard approach is improved by leveraging the offshore PTHA to automatically focus sampling on scenarios with larger waves near the site of interest, and on magnitude-bins that contribute most to Monte-Carlo errors.…”