Overview
This research combines physical models of wildfire spread with learned corrections. The goal is to forecast fireline geometry and arrival time while examining how physical constraints can support useful predictions under changing conditions.
Approach
I develop fireline forecasts using differentiable anisotropic Eikonal propagation, and build FireBench and WRF-SFIRE simulation and evaluation pipelines. Comparisons consider physical and neural baselines, with fireline geometry and arrival-time errors as evaluation targets.
Results & Current Status
Current work covers simulation and evaluation pipelines, fireline forecasting, and investigation of fire-intensity prediction and uncertainty modelling using fuel-consumption and energy-budget constraints. Research is ongoing; no final performance claim is made here.
