Fire risk mapping using machine learning method and remote sensing in the Mediterranean region

Abstract

Forest fires are a notable phenomenon that causes substantial destruction to natural resources. To control or prevent forest fires, modeling fire risk levels is essential. Using fire occurrence data, 36 different variables, and the maximum entropy (MaxEnt) model, a forest fire risk map for the Mediterranean region was generated in this study. The fire risk map was developed using four primary variables: topography, climate, forest structure, and human interference. Low, moderate, high, and extreme classes covered 35.6 %, 34.2 %, 17.0 %, and 13.2 % of the forested areas, according to the forest fire risk map based on MaxEnt. According to the MaxEnt model, the variable that most affected the forest fire risk was tree species, which accounted for approximately 44 % of the fire events. The accuracy of the model was evaluated using the receiver operating characteristic (ROC) analysis, and the area under the curve (AUC) was calculated to be 0.810. In order to assess the efficiency of the model, the burned areas caused by the mega-fires in Manavgat and Gündoğmuş in 2021 were determined using Landsat satellite images and the difference normalized burn ratio (dNBR) index. The fire risk map generated by the MaxEnt model was overlaid with the forest fire burned area maps for Manavgat and Gündoğmuş. On the fire risk map, approximately 90 percent and 98 percent of the forest fire-burned areas in Manavgat and Gündoğmuş, respectively, fell into the high and extreme risk categories. Based on these findings, it was determined that the created fire risk map is highly accurate and compatible with the fire events.