Forest Fire Risk Prediction Model Integrating Spatiotemporal Information

Forest fires pose a significant threat to forest ecosystems and human life, making scientifically based prediction and early warning systems crucial for their control and prevention. This study conducts a comprehensive analysis of historical forest fire data from the Southwest China region from 2001 to 2018 and combines multiple driving factors, including topography, meteorology, vegetation conditions, and fire weather indices. Machine learning models, including random forests, support vector machines, and deep neural networks were applied, particularly developing a new neural network model integrating spatiotemporal information, named spatiotemporal weighted back propagation (STW_BP), to predict the risk of forest fires. The results show that the introduction of spatiotemporal features significantly enhances the model’s predictive capabilities; the STW_BP model outperforms other models in terms of accuracy, recall, and area under the curve values. It effectively captures the spatiotemporal distribution characteristics of fire risk and applies them to fire risk prediction, hence providing a scientific basis for fire prevention and resource allocation. This study not only advances our understanding of the patterns of forest fire occurrence but also further validates the importance and application value of spatiotemporal information in fire risk analysis.