Data-driven machine learning for multi-hazard fragility surfaces in seismic resilience analysis

Abstract

Offshore earthquakes and subsequent tsunamis pose significant risks to many coastal populations worldwide. This paper introduces a data-driven machine learning model that synthesizes accurate 3D earthquake–tsunami fragility surfaces from randomly selected 2D fragility curves. The integration of physics-based simulations enhances the model's reliability for these specific hazards, making it a valuable tool for multi-hazard analysis in earthquake–tsunami contexts. Additionally, by shifting 2D fragility curves to represent retrofitted structural systems, the model can generate earthquake–tsunami fragility surfaces for community-level mitigation studies. While the model is demonstrated for earthquake–tsunami scenarios, its methodology architecture has the potential to contribute to other multi-hazard situations for the initial conditions in multi-hazard community resilience analysis.