Evaluating data source effects on forest fire prediction models in Inner Mongolia

Background Frequent intense fires, influenced by both natural and social factors, present significant forest management challenges. Predicting forest fire occurrence is essential for early warning and effective response. Aims Quantify discrepancies between human records and satellite fire data, and assess the predictive accuracy of six models. Methods This study analyzes fire data from 2001 to 2022, incorporating manual records and satellite remote sensing. Utilizing six models, it examines variations across different administrative regions and fire prevention periods, exploring the relative significance of the optimal predictive model and its driving factors. Key results The fundamental relationship between key variables and fire probability was established, revealing a significant spatiotemporal differentiation pattern in the variables related to fire occurrence in Inner Mongolia. Conclusions The Boosted Regression Tree model was identified as optimal for predicting forest fires in Inner Mongolia. Meteorological factors are the primary variables related to fire occurrence, followed by topographical and socioeconomic factors. Frequency and probability of fire occurrence gradually decrease from east to west, most fires being concentrated in the northeastern and central regions. Implications Investigating spatiotemporal patterns of fire occurrence and accurately predicting forest fire probability at different scales can provide a scientific basis for the rational allocation of regional fire prevention resources.