Probabilistic Tsunami Hazard Analysis: High Performance Computing for Massive Scale Inundation Simulations

Abstract

Probabilistic Tsunami Hazard Analysis (PTHA) quantifies the probability of exceeding a specified inundation intensity at a given location within a given time interval. PTHA provides scientific guidance for tsunami risk analysis and risk management, including coastal planning and early warning. Explicit computation of site-specific PTHA, with an adequate discretization of source scenarios combined with high-resolution numerical inundation modelling, has been out of reach with existing models and computing capabilities, with tens to hundreds of thousands of moderately intensive numerical simulations being required for exhaustive uncertainty quantification. In recent years, more efficient GPU-based High-Performance Computing (HPC) facilities, together with efficient GPU-optimized shallow water type models for simulating tsunami inundation, have now made local long-term hazard assessment feasible. A workflow has been developed with three main stages: 1) Site-specific source selection and discretization, 2) Efficient numerical inundation simulation for each scenario using the GPU-based Tsunami-HySEA numerical tsunami propagation and inundation model using a system of nested topo-bathymetric grids, and 3) Hazard aggregation. We apply this site-specific PTHA workflow here to Catania, Sicily, for tsunamigenic earthquake sources in the Mediterranean. We illustrate the workflows of the PTHA as implemented for High-Performance Computing applications, including preliminary simulations carried out on intermediate scale GPU clusters. We show how the local hazard analysis conducted here produces a more fine-grained assessment than is possible with a regional assessment. However, the new local PTHA indicates somewhat lower probabilities of exceedance for higher maximum inundation heights than the available regional PTHA. The local hazard analysis takes into account small-scale tsunami inundation features and non-linearity which the regional-scale assessment does not incorporate. However, the deterministic inundation simulations neglect some uncertainties stemming from the simplified source treatment and tsunami modelling that are embedded in the regional stochastic approach to inundation height estimation. Further research is needed to quantify the uncertainty associated with numerical inundation modelling and to properly propagate it onto the hazard results, to fully exploit the potential of site-specific hazard assessment based on massive simulations.

The Making of the NEAM Tsunami Hazard Model 2018 (NEAMTHM18)

“…Similar approaches are being followed in New Zealand (MCDEM, 2016) and the United States (American Society of Civil Engineers, 2017), based on the local PTHA. The region-wide NEAMTHM18 is also being taken as a reference for higher-resolution sitespecific PTHAs (Gibbons et al, 2020) and applications dealing with critical infrastructures at risk (e.g., Argyroudis et al, 2020). The development of standardized PTHA products (hazard curves, hazard and probability maps, exhaustive and transparent documentation, web tools for dissemination and analysis) is the first step to include tsunamis in multi-hazard risk assessment and mitigation.…”

Tsunami risk management for crustal earthquakes and non-seismic sources in Italy

“…The increasing number and quality of probabilistic hazard and risk analyses show that there are a growing capability and intent to study and characterize complex sources such as the atypical ones, especially the seismic ones. Recent hazard studies on seismically generated tsunamis have already started including non-megathrust events [11,31,52,53], and many efforts are also ongoing to improve the computational efficiency [66,67,69] and to extend toward non-seismic sources [11,17,32,98,191]. These hazard studies may provide important information also for tsunami warning, such as (i) the description of the full variability of the source mechanisms (e.g., [31]); (ii) the prioritization of sources based on the capability of generating tsunamis and the long-term frequency of tsunamigenic events (e.g., based on hazard disaggregation; [31,381]); (iii) the development of databases of pre-defined sources and related tsunami propagation scenarios that may be used in real-time tsunami warning (e.g.…”

“…Non-linear shallow water models are most frequently used for this purpose (e.g., the review [64]); Boussinesq model applications are used as well, despite that they are more prone to instabilities (see discussion in [220]). Despite the relatively high accuracy of these modelling techniques, large uncertainty still exists (e.g., [54,63,69,92,127,221]), and the potential impact of uncertainty in initial/boundary conditions or of different modelling techniques still have to be fully understood.…”

“…In comparison, large subduction zone earthquake behaviour is indeed more predictable, as it is largely dominated by thrust events on the plate interface. To model this variability a very large set of scenarios is needed (e.g., > 10 7 , considering only seismic sources in a small basin like the Mediterranean, see [31]) and thus their integration into tsunami hazard and warning depends on the development of specific methodological and computational strategies [31,54,[66][67][68][69].…”

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

“…For this purpose, a large number of inundation scenarios are needed to quantify the epistemic uncertainty and bias caused by simplifications introduced through approximate methods. A local PTHA application using more than 40,000 earthquake sources (Gibbons et al, 2020) is only a start.…”