The increasing incidence of large wildfires in fire-prone regions, such as California, presents significant challenges to the insurance industry in accurately pricing and managing wildfire risks. Here we introduce a machine learning-based fire modeling framework encompassing three major process-based components: fire ignition, fire spread, and fire sampling, designed for interannual predictive wildfire risk assessment. Within each component, we have developed machine learning (ML) and deep learning (DL) models, including Extreme Gradient Boosting (XGBoost), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) Networks, to simulate fire processes and burn probability at 1-km resolution. Retrospective evaluations indicate promising modeling performance over high-risk regions, with a high recall score (0.91-0.95), a moderate precision score (0.01-0.11), and a balanced F-1 score (0.02-0.19) for historical large wildfires in California during 2020-2022. This ML-based fire modeling framework has been integrated into insurance business practices, providing essential insurance products at a more affordable price. Such initiative aims to serve as a safety net for homeowners confronting the evolving threats of climate change.

Machine Learning,Fire Modeling,Risk Assessment,Insurance Industry