|
چکیده
|
Fire is one of the most pervasive factors destroying forest ecosystems, with ecological, economic, and social consequences. Therefore, analyzing changes in the situation of the region in terms of fire occurrence danger and also the parameters affecting it will be very useful in the field of fire risk management and control approaches. This is especially true in semi-arid oak forests. The current study was conducted in several stages. First, thirteen parameters, including topographical, climatic, biological, anthropogenic, and soil surface moisture as effective factors in forest fire risk, were assessed and modulated using machine learning methods. Raster-based maps for these criteria were generated using integrated geographic information systems (GIS) and remote sensing. Subsequently, fire risk maps were developed using historical fire occurrence data. Model’s accuracy assessment revealed that the Random Forest (RF) model outperformed both Generalized Linear Models (GLM) and Support Vector Machine (SVM) models in fire risk detection. According to the RF model results, 33.95 and 18.84% of the studied areas were classified as low and high fire risk classes, respectively. The investigation of the factors affecting the occurrence of fire showed that anthropogenic factors (distance from residential areas, distance from agricultural lands), climatic factors (temperature, wind speed, relative humidity), and topographical factors (elevation) played a more important role in places with a history of fire. Therefore, to mitigate fire frequency and associated damages, it is essential to address the causes and motivations of fire ignition while minimizing fire-prone conditions through preventive measures.
|