Explainable multivariable damage modeling for water-related hazards

Quantifying physical vulnerability is central for enhancing the resilience of communities and infrastructures to natural hazards. This task, however, remains challenging because damage processes are inherently multivariable, emerging from the interaction among hazard intensity, exposure characteristics and system-specific resistance. Addressing such complexity requires modeling frameworks capable of capturing nonlinear dependencies among these factors while maintaining robustness and interpretability. This contribution explores two strategies for tackling this challenge in the context of water-related hazards: synthetic, physically based models and data-driven, machine-learning approaches. Emphasis is placed on key model requirements such as transparency, robustness, transferability and the explicit representation of uncertainty. Within the first approach, we present and discuss the advantages of flood damage modeling tools that leverage physically based probabilistic frameworks, explicitly linking damage to its explanatory variables, such as the INSYDE model for residential buildings and contents, as well as AGRIDE-c for crops. The second approach, relying on post-event data, is presented here with a focus on the 2011 Great East Japan tsunami, supported by the availability of extensive and detailed datasets on building and road damage as well as human fatalities. Integration with explainable AI methods, such as feature importance ranking and individual conditional expectation plots, provides insights into the complex interactions among hazard, exposure and vulnerability factors, thereby enhancing interpretability and predictive reliability of developed models. Together, the two different approaches then demonstrate how both physically interpretable and data-driven models can improve the understanding and quantification of vulnerability across different hazard contexts. This integrated perspective can contribute to advancing transparent, transferable, and evidence-based frameworks for risk assessment and disaster resilience.