Objective This paper aims to select and validate suitable forest fire prediction models for the study area, identify key driving factors of fire occurrence, and map fire risk zoning, then providing scientific basis and decision support for forest fire prevention and management.

Method Using historical fire data from 1981 to 2020 and integrating multi-source data (meteorological conditions, topography, vegetation, human activities, and socio-economic factors), the applicability of four machine learning methods in predicting forest fires in the Daxing’an Mountains of Inner Mongolia of northern China was compared. Based on the significant factors influencing fire occurrence, maps of fire occurrence probability and fire risk zoning were generated.

Result (1) The boosted regression tree model (BRT) showed an area under the curve (AUC) value of 0.967, and the random forest model (RF) achieved an AUC of 0.947, both demonstrating excellent predictive performance. The predictive accuracy of the Logistic regression model (LR) and the Gompit regression model (GR) was slightly lower than former two models, but still met the basic predictive requirements for the study area, with AUC values of 0.852 and 0.851, respectively. (2) Meteorological factors, such as diurnal temperature range and daily minimum relative humidity, were the dominant factors influencing forest fires in the Daxing’an Mountains of Inner Mongolia. Elevation also ranked high in the relative importance of driving factors. Human activities and socio-economic factors, such as distance to roads, distance to fire lookout towers, and per capita GDP, also had some influence on fire occurrence. (3) Large areas of medium to high fire risk were present in the eastern and southeastern parts of the Daxing’an Mountains of Inner Mongolia, while the northern China-Russia border and the southwestern China-Mongolia border also exhibited elevated fire risk. Factors such as average temperature and average surface temperature during fire prevention period in autumn of previous year influenced forest fire occurrences in the following year.

Conclusion Among the four models compared, the BRT was identified as the most suitable one for predicting forest fire occurrence in the Daxing’an Mountains of Inner Mongolia. Meteorological factors and elevation significantly influence fire occurrence, while human activities and socio-economic factors also have a certain impact on the occurrence of fires. The high and medium fire risk areas are primarily concentrated in the eastern and southeastern parts of the study area, with some fire risks presenting in the northern and southwestern regions.