Can a physics-based fire behavior model predict burn severity and post-fire debris flow hazard?
Post-wildfire flooding and debris flows pose a threat to ecosystems and infrastructure. As the frequency and size of severe wildfires increase, there is a growing need for predictive debris flow hazard modeling in unburned but fire-prone landscapes. However, no crosswalks have been established between physics-based models of fire behavior and the remotely sensed burn severity inputs needed for debris flow modeling. Here, we explore this connection by creating a large ensemble of fire behavior simulations using QUIC-Fire – a fast-running three-dimensional fire model – in basins across five major-disaster wildfires in California and Washington. We then conducted generalized linear mixed modeling (GLMM) and Random Forest modeling to identify QUIC-Fire outputs that best predicted remotely sensed burn severity. We found that the selected linear predictors exhibited weak and unintuitive relationships with soil burn severity and differenced Normalized Burn Ratio. Moreover, model evaluations of GLMMs and Random Forests showed they were poorly predictive of both metrics of remotely sensed burn severity. This suggests that there are barriers to integrating QUIC-Fire into a wildfire-hazards modeling framework, which could include mismatches in scale between fire models and remotely sensing, or simplifications of weather data and fire initiation. Further research is needed to establish connections between the fire physics outputs of QUIC-Fire and post-fire effects to vegetation and soils that are essential for evaluating debris flow hazard.