From Model Validation to Epistemic Assurance in Disaster Simulations

Abstract

Disaster simulation models are increasingly used to support decisions about future hazards, climate adaptation, and post-disaster recovery, yet many of the socio-economic processes they represent cannot be directly observed or fully validated. Instead, modelers rely on proxies, inferred relationships, and simplified mechanisms to represent latent processes, creating the possibility that defensible representations produce different decision-relevant conclusions despite similar agreement with historical observations. We argue that confidence in a model implementation and confidence in the conclusions it supports are distinct epistemic questions. We propose an epistemic assurance framework for disaster simulation models that distinguishes proxy, structural, and outcome validity as complementary properties of the model while introducing epistemic robustness as a property of decision-relevant conclusions. The framework first establishes whether a model provides a defensible basis for inference and then evaluates whether important conclusions depend on particular representational choices. Together, these complementary assessments provide a structured basis for qualifying the scope of the model-based conclusions that can be responsibly defended and used to support decisions. Illustrative examples demonstrate how alternative representations can reveal fragile conclusions even when competing models exhibit similar agreement with historical observations. More broadly, the framework complements existing approaches to model validation and uncertainty analysis by reframing the evaluation of disaster simulations from asking whether a model has been validated to asking which conclusions can be responsibly supported, under what assumptions, and for which intended purposes.