Prediction and susceptibility mapping of forest fires using machine learning and remote sensing methods

In Türkiye, forest fires occur during the summer months. This is especially the case in the Mediterranean and Aegean regions. Climate change, humanitarian reasons and climatic characteristics of the region cause forest fires, which destroy the forest resources where they occur and cause economic and various environmental problems. For this reason, it is of great importance to detect forest fires in advance and to analyze the areas with the potential of fire occurrence. In this study, remote sensing and machine learning techniques have been used to detect and predict forest fires in Turkiye. Temporal forest fire data has been obtained from the FIRMS dataset, and a new dataset has been created by obtaining various climatic characteristics of Antalya province from the ERA-5, NASA GLDAS, and NASA SRTM databases. The new data set has been evaluated using Ada Boost, Decision Tree, Gradient Boosting Machine (GBM), K-Nearest Neighbors (KNN), Logistic Regression, Support Vector Machine (SVM) and Artificial Neural Network (ANN) algorithms and the results have been classified using Jenks Natural Breaks classification method to create forest fire susceptibility maps of the region. According to the experimental results, the most successful algorithm has been determined to be Ada Boost (sensitivity=0.93, specificity=0.95, accuracy=0.94, kappa=0.88, AUC=0.99). These models can be used to create monthly and seasonal forest fire susceptibility maps for the Mediterranean and Aegean regions of Türkiye, so that necessary preventive measures can be taken before forest fires occur.