Natural Hazards Research Summit 2022: Uncertainty Quantification for Global Geospatial Liquefaction Models

Geospatial natural hazard models are simple empirical models that take advantage of broadly available geospatial proxies that capture location-specific information on geologic, geomorphologic, and hydrological characteristics and have predictive power for natural hazard predictions. Such models are useful for regional hazard assessments in both post-disaster response and pre-disaster planning phases. Due to the intrinsic variability of natural hazards, model simplification, and measurement errors, the accuracy and uncertainty of these geospatial models vary from region to region and event to event. There is a strong research need to quantify the uncertainty of geospatial natural hazard models to guide future data collection and model updating. We present a framework that addresses this problem by quantifying three types of uncertainties in geospatial natural hazard models: parametric uncertainty, model bias, and input uncertainty of geospatial proxies. We employ forward uncertainty propagation to illustrate how different types of uncertainties affect the outputs of geospatial natural hazard models. The proposed uncertainty quantification framework provides a measure of uncertainty on model predictions and can be applied to any logistic regression models and other geospatial modeling problems.