Machine learning algorithm-based forest fire risk prediction in Yuanling, Hunan Province

In order to accurately assess the risk level of forest fires, help forest patrol, optimize resource layout, and improve fire prevention efficiency, Yuanling County was taken as the research object. Based on the data of terrain, fuel, meteorological, and human activity factors, a machine learning algorithm was used to build a forest fire occurrence prediction model, which has certain reference significance for forest fire prevention.

With comprehensive consideration of terrain, fuel, meteorology, and human activities, 11 driving factors were extracted in the study area, including elevation, slope, slope direction, normalized vegetation index, vegetation type, precipitation, air temperature, wind speed, distance from road, distance from residential area, and distance from water system, and the driving factors of forest fire were evaluated. The historical fire point data of the study area is obtained based on the MODIS fire product. The forest fire occurrence prediction model was constructed by a machine learning algorithm, and the prediction accuracy of the model was comprehensively evaluated by using the confusion matrix evaluation index and the ROC curve. The forest fire occurrence prediction model was constructed by a machine learning algorithm, and the prediction accuracy of the model was comprehensively evaluated by using the confusion matrix evaluation index and the ROC curve.

The distance from the road and the distance from the residential area are two driving factors with the largest weight, and other driving factors also affect the occurrence of forest fires. The ROC curves of three models showed that the random forest model had good accuracy, with the accuracy reaching 78.15% and the area under the curve 0.85, while the logistic regression prediction model had the accuracy reaching 74.81% and the area under the curve 0.81. The accuracy of the SVM prediction model is 70.74%, and the area under the curve is 0.79.

The random forest model shows better prediction ability than the logistic regression model and support vector machine model. The areas with high and extremely high risk of forest fire accounted for 26.62% in the study area. The forest fire risk level map is helpful for the relevant departments to take relevant preventive measures and effectively protect the safety of forest resources.