Unravelling Seismic-Tsunami Complexity in Northern Chile Through Stochastic Scenario Clustering

Resumen

Accurate tsunami hazard assessment requires the consideration of large ensembles of earthquake scenarios to represent the source variability and associated uncertainties adequately. However, simulating hundreds or thousands of tsunami scenarios is computationally demanding, limiting its applicability in operational and planning contexts. This study presents a stochastic clustering and classification framework aimed at reducing and organizing tsunami generating earthquake scenarios while preserving their physical and geological representativeness. A total of 257 synthetic megathrust earthquake scenarios with magnitude Mw 9.0 were generated for the northern Chilean subduction margin (19°S–25°S) and propagated through numerical tsunami simulations to estimate inundation-related impact metrics for the coastal city of Mejillones. Multivariate K-Means clustering was applied to group scenarios with similar tsunamigenic behavior based on key variables such as rupture depth, maximum slip, inundation height, inundated area, and distance to the site. Subsequently, interpretable decision tree models (CART) were used to identify the dominant variables and thresholds controlling cluster differentiation. Results indicate that tsunami impact variability in Mejillones is primarily governed by rupture depth, tectonic domain, and distance to the site, with shallow ruptures located within the megathrust Domain B producing the highest inundation levels. The proposed approach enables a substantial reduction of scenarios without loss of physical consistency, providing an efficient and robust framework for tsunami hazard assessment and risk-informed coastal planning along the northern Chilean margin.