Wildfire risk assessment based on Light Gradient Boosting Machine model

Abstract

With the expansion of the power grid and the limitation of the geographical environment, some areas have to adopt the way of crossing the forest area to arrange the transmission lines. Some forest areas are sparsely populated and the vegetation is lush. Once a mountain fire occurs, it is easy to spread to the vicinity of the transmission corridor, resulting in the failure of transmission line tripping and reclosing. In order to effectively predict wildfires, this paper proposes a wildfire risk assessment model based on LightGBM. Combining vegetation factors, meteorological factors, terrain factors, and human factors, the moderately correlated fire point characteristics were screened out based on correlation analysis, and a wildfire risk assessment model was constructed. After that, the fire point products of NPP and MODIS are used as the validation data of the model, and the acracy of the model is predicted by the accuracy, precision, recall, F1-Score and AUC values. A comprehensive evaluation showed that the accuracy of the model was 0.86 and the AUC value was 0.83. The results showed that the model could effectively predict wildfire risk.