Based on a Novel Hierarchical Convolutional Neural Network for Forest Fire Spread Research

Forest fires rank among the most destructive natural hazards, leading to extensive harm across ecosystems, economies, and human settlements. Their recurring nature necessitates precise modeling of spread dynamics and path forecasting to support rapid emergency interventions. This work introduces FSS-CNN, a novel hierarchical convolutional neural network designed to simulate the spread of forest fires. A core feature of the model is its hierarchical state-condition architecture, which effectively captures advanced features from diverse input sources. Additionally, the study formulates a fire spread spatiotemporal distribution field (FSSDF) to represent the dynamic propagation process, allowing for end-to-end learning and global optimization. The framework's performance is assessed using a case study of the wildfire in Xichang, Sichuan Province, with simulation outputs compared to Sentinel-2 satellite observations. The close correspondence between predicted and actual burned area distributions demonstrates the model's high predictive accuracy.