Identification of potential forest wildfire risk in Angeles National Forest by random forest and logistic regression

Abstract

Frequent wildfires have a growing negative impact on Los Angeles. Understanding the driving factors of wildfire disasters positively impacts predicting and preventing potential wildfire risk. From June to September every year, wildfire disasters frequently occur in the United States. Each wildfire event is associated with human or climatic factors. This paper compares the wildfire prediction ability of two different methods: Logistic Regression (LR) and Random Forest (RF), to determine the main structural factors that explain the possibility of fire in Los Angeles National Forest Park. Natural environmental factors and human social activities are considered essential predictors of fire. The prime natural drivers are natural resource location, climatic conditions, and topographic factors. With the rapid development of society, human activity has become an essential driving factor of wildfire. All samples were randomly divided into training samples, test samples, and validation sets. The results show that the RF model in the machine learning method is more scientific and effective in wildfire risk prediction. The RF model (area under the curve is 89%) provides higher prediction accuracy than the LR model (area under the curve is 81%) and ranks the importance of the factors. The results of the RF model show the spatial distribution of the high-risk regions for wildfire, which allows for a more efficient allocation of fire resources in the park.