Conveners
GPU-Accelerated Digital Twins for Tsunami Hazards: Integrating Non-Hydrostatic Multilayer HySEA Model for Landslide Generated Tsunamis and gPCE Surrogates
- Manuel J. Castro Díaz (University of Málaga)
Description
Probabilistic Tsunami Hazard Assessment (PTHA) for landslide-generated events is often constrained by the prohibitive computational cost of simulating thousands of potential scenarios. To address this, we present a "Digital Twin" framework that integrates high-fidelity GPU-accelerated physical modeling with efficient statistical surrogate models.
The physical foundation is the Non-Hydrostatic Multilayer HySEA model. It employs a depth-averaged formulation capturing essential dispersive effects and vertical velocity profiles. Using very few vertical layers, the model realistically simulates the complex interaction between granular mass movements—modeled via Savage-Hutter type rheology, and the water column. The implementation is fully GPU-accelerated, significantly reducing simulation time.
To enable rapid uncertainty quantification, we employ generalized Polynomial Chaos Expansions (gPCE) to construct surrogate models. These surrogates approximate tsunami metrics (elevation, velocity, and arrival time) as continuous functions of uncertain parameters such as volume, location, and rheology. By using Delayed Gauss-Patterson (DGP) sparse grids, we achieve model convergence with a small set of simulations per zone.
We validate this framework in Mayotte (France), where a recent seismo-volcanic crisis has raised concerns regarding submarine slope stability. The resulting surrogates produce probabilistic hazard maps and density functions in less than 2 seconds, providing a critical tool for faster-than-real-time forecasting and risk mitigation.
Joint work with: Cléa Denamiel (Ruđer Bošković Institute/IACRK, Croatia), Alexis Marboeuf, Anne Mangeney, Anne Le Friant, Antoine Lucas (University Paris Cité, IPGP, France), Marc Peruzzetto (BRGM, France), and Enrique Fernández-Nieto (Universidad de Sevilla, Spain).
