Abstract: A forest fire prediction model was established in the forest areas of Hunan Province, to provide a theoretical basis for the prevention, control and management of forest fires. Extracted forest fire hotspot monitoring satellite data from 2016 to 2020, and used ArcGIS 10.7 to establish random points that formed 1:1 data with fire points, making the data conform to binomial distribution. Spatial information was extracted from 24 types of driving factors such as daily average maximum temperature, daily average precipitation, sunshine duration, and altitude, and GWR 4.0 was used to predict and analyze the driving factors that affect the forest fires. A forest fire prediction model for the region was built, using AIC, BIC, ROC curves, and AUC for model fitting testing, and conduct fire risk zoning for the region. The research results indicate that: nine variables, including "daily average wind speed" "daily average pressure" "sunshine time" "daily average temperature" "daily average relative humidity" "altitude", and "normalized index value", are the driving factors; The GWLR model has a good fitting effect, with an AUC value of 0.966 and an optimal classification threshold of 0.810; In the fire risk zoning, medium to high fire risk areas are mainly concentrated in the central and southern regions of Hunan Province, and this area should further strengthen the publicity and education of forest fire prevention, establish and improve the control of outdoor fire pipes, and strengthen the construction of infrastructure in key fire risk areas.

Key words: forest fire prediction, geographically weighted logistic regression, forest fire driving factors, fire hazard area division