Evaluating the performance of species distribution models Biomod2 and MaxEnt in forest grassland fire prediction: an example from Sichuan Province, China

Accurate identification of forest and grassland fire-prone areas is essential for effective fire management and ecosystem protection. This study aimed to evaluate the performance of two species distribution models, MaxEnt and Biomod2, in predicting forest grassland fire risks in Sichuan Province, and to accurately identify regions with high fire risk. Using fire occurrence data from 2011 to 2020 and relevant environmental variables, both models were applied to generate fire risk maps and identify key influencing factors. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and the true skill statistics (TSS). Results showed that the Biomod2 model outperformed MaxEnt, with the ensemble model EMwmean achieving the highest accuracy (AUC = 0.93, TSS = 0.71). Mean temperature and precipitation were the most influential variables, with human activity also playing a significant role. High-fire-risk areas were concentrated in southwestern Sichuan, the western populated zones, and the northern Sichuan Basin. We recommended that the MaxEnt model be considered when both model prediction accuracies are adequate for subsequent applications, or when only local fire point data are available and the global distribution must be predicted. Otherwise, the Biomod2 model is to be preferred.