Abstract:

In the context of global climate change,the forest fire prevention situation on Hainan Island is becoming increasingly severe,and it is urgently needed to analyze the driving factors of tropical forest fires on Hainan Island and build a strong predictive model that is applicable.Utilizing historical forest fire data compiled by the forestry department from ground surveys and MOD14A fire detection,a comprehensive dataset was established for Hainan Island.This dataset was combined with climate,vegetation,topography,and human activity data to construct a predictive model using the random forest methodology.1)The average monthly temperature is the most influential factor on forest fire risk in Hainan Province,followed by the average monthly precipitation.2)Comparative model analysis shows the random forest model,with an AUC value of 1,outperforms the geographically weighted logistic regression model,which has an AUC value of 0.88,indicating that the random forest model is more suitable for predicting the probability of tropical forest fires on Hainan Island than the geographically weighted logistic regression model.3)The spatial distribution of forest fire risk on Hainan Island mainly occurs in the west.This study believes that the random forest model is more applicable than the geographically weighted logistic regression model in building a predictive model for tropical forest fire risk.

Key words: Hainan Province, forest fires, generalized linear model, random forest model, probability of occurrence prediction