Machine learning characterization of fuel temperature and moisture dynamics in wildland-urban interface for urban...

With accelerating urbanization and intensifying climate change, wildfire risks are increasing worldwide, making the wildland-urban interface (WUI) a critical high-risk zone for fire occurrence and environmental management. This study employs data-driven machine learning approaches-including Long Short-Term Memory (LSTM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)-to characterize dynamic fuel temperature and moisture across WUI areas in Beijing, using meteorological observations from eleven fire risk monitoring stations. Model performance was evaluated using Mean Squared Error (MSE), Mean Absolute Error (MAE), and the coefficient of determination (R²), and compared with a conventional Fire Weather Index (FWI)-based baseline model. Results indicate that LSTM generally achieved lower errors and higher R² values than RF and XGBoost, demonstrating enhanced capability in capturing temporal variability in WUI microclimate conditions. No significant linear relationship was observed between nearest WUI distance and model performance, suggesting that localized environmental heterogeneity exerts greater influence than simple spatial proximity. The proposed framework supports improved fuel condition forecasting and climate-resilient wildfire risk management in rapidly urbanizing WUI regions.