Wildfire probability assessment and analysis based on multi-source data in Guangxi Province, China

Abstract

Wildfires are key environmental variables in global climate monitoring systems that significantly impact ecosystems, human life, and property. This study assessed the probability of wildfires in Guangxi Province using multi-source data and the Evidence Belief Function (EBF) model. A stratified random sampling method was applied to evaluate the accuracy of the three wildfire remote sensing products (MCD64A1, Fire_CCI51, and MCD14ML). Among these, MCD64A1 was identified as the most reliable dataset for mapping wildfires in the region. By dividing the year into three wildfire occurrence periods (A, B, and C) and combining 14 wildfire factors from 2003 to 2018, the EBF model was used to map and analyze the spatial distribution of wildfire risk. The relative importance of each wildfire factor was evaluated using a Random Forest (RF) model. The validation results confirmed that the AUC values of the success rate of the EBF model were all greater than 0.7, indicating that the model performed well in wildfire risk assessments. The derived wildfire probability map showed strong spatial consistency between the assessed probability levels and observed wildfire events, with high-risk areas in the northwest, central, and northeast regions of Guangxi. Furthermore, relative humidity was identified as the most significant factor influencing wildfire probability, while land surface temperature (LST), wind speed, and a human influence indicator, the Human Footprint (HPF), also had a significant effect on wildfire probability. These findings highlight the reliability of the EBF and RF models as tools for prediction evaluation and provide insights into effective resource allocation and wildfire prevention strategies in the region.