A Bayesian approach to postfire debris-flow probability modeling for southern California, United States

Background Intense rainfall on steep, burned landscapes can initiate postfire debris flows (PFDFs). Watershed-scale estimates of PFDF probability in response to a given rainfall intensity are beneficial for assessing PFDF hazards, including prioritizing areas for hazard mitigation. Aims We utilize a database of PFDF observations from southern California, US, to develop a logistic regression model for PFDF probability. Methods The model structure, including input parameters related to terrain steepness, burn severity and rainfall, is based on an existing model. However, we adopt a Bayesian approach to model development, resulting in calibrated posterior distributions over all parameters. Key results The model performs well when applied to a southern California test dataset and can be used to estimate watershed-specific rainfall intensity-duration thresholds for PFDFs. We demonstrate how integration of conditional PFDF probabilities over the full marginal distribution of peak 15 min rainfall intensity can quantify climatologically-driven spatial variations in PFDF probability during the first postfire year. Conclusions A Bayesian approach to PFDF probability modeling results in a framework that performs well when tested on another southern California dataset and provides estimates of predictive uncertainty. Implications PFDF hazard assessments can benefit from modeling approaches that propagate uncertainties from rainfall data through the model structure to output probabilities.