Deep Learning-based Forest Fire Prediction Model Research in the Daxing'anling Mountains,Inner Mongolia
[Objective]To predict forest fires in the Daxing'anling Mountains of Inner Mongolia and provide important support for forest fire prevention.[Method]Based on the Daxing'anling Mountains of Inner Mon-golia as the research object,combined with MCD64 A1 monthly fire point products,terrain,climate and other data,the forest fire potential impact factor data set was constructed,and the convolutional neural network,random forest and support vector machine models were used respectively to predict and visual-ize the probability of forest fires in the study area.The developed models were evaluated and the spatial distribution characteristics of forest fires were analyzed.[Results]The main driving factors of forest fire in the Daxing'anling Mountains were altitude,average temperature,total precipitation and the distance from water area in order of importance.The AUC value of CNN,RF and SVM was 0.838,0.794 and 0.788,re-spectively,and the accuracy of CNN was the highest.CNN can effectively divide areas with high and low forest fire susceptibility,which is conducive to dividing forest fire warning areas.[Conclusion]The CNN model is more suitable for predicting the probability of forest fires in the Daxing'anling Mountains than RF and SVM models.The spatial distribution of forest fire risk in the Greater Khingan Mountains is obviously regional,mainly occurring in the southeast region.
万方数据
维普
