Assessing wildfire susceptibility and spatial patterns in diverse forest ecosystems across China: An integrated geospatial analysis

Understanding the spatial and temporal distribution characteristics of fires, their driving factors and accurately predicting fire occurrences are essential for effective forest management. Therefore, it is essential to identify and predict areas susceptible to fires, particularly in a country like China, where environmental and social conditions have undergone significant changes. In this study, we analyzed the spatial patterns of fires in distinct forest ecosystems across China. By incorporating RS and GIS technologies, and machine learning methodologies, we examined the factors influencing fires and developed a susceptibility model for different forest ecosystems. To generate fire susceptibility maps, we employed three machine learning models to establish connections between fire occurrences data and 17 predictor variables including climate, topography, vegetation, and human disturbances, namely artificial neural network, random forest, and the extreme gradient boosting models. The results showed that the fire points in different forest ecosystems showed a significant clustering distribution in space, and the driving factors of fire were different. We observed satisfactory performance across all the fire prediction models employed. Specially, extreme gradient boosting model exhibited superior performance with an AUC = 0.82-0.95; accuracy = 0.79-0.87; recall = 0.78-0.89; and F-Measure = 0.78-0.86. Forest fires in Heilongjiang Province are mainly caused by vegetation factors, while in Sichuan, human factors are the primary cause of fire incidents. Topographical factors play a crucial role in influencing the occurrence of forest fires in Shanxi and Fujian. Climate factors play a crucial role in Guangdong and Yunnan. The temporal and spatial patterns of fires in various ecosystems could be analyzed in combination with forest fire factors, providing important scientific information for regional forest fire early warning and monitoring.