Fire is one of the most important causes of forest degradation, especially in semiarid forest ecosystems. The increase in annual fire occurrence and the complexity of environmental factors affecting fire occurrence in the Zagros vegetation zone have increased the importance of modeling factors affecting fire occurrence in this region. Therefore, forecasting fire-prone areas and practical factors can help forest managers to prevent destructive fires. This study aims to modulate the fire-sensitive areas using machine learning methods, including support vector machine (SVM), random forest (RF), and generalized linear model (GLM). Materials & Methods:Fire-effective factors were categorized into four classes (physiographic, biological, climatic, and anthropogenic factors) and 16 raster-based variables. The fire susceptibility maps were validated using the area under the curve (AUC) values extracted from the receiver operating characteristics (ROC) curve. In addition, the RF model was used to determine the relative importance of each variable. Findings: Results showed that fires happened in the middle elevation (300-2000m), lower slopes (<20%), and in the west and southwest slope aspects. More fires were also in agricultural and residential areas. The validation of fire susceptibility maps showed that the RF model (AUC=0.911) has higher accuracy than the SVM (AUC=0.864) and GLM models (AUC=0.824). Based on the RF model, high and very high-risk had low areas (9.48 and 5.97%, respectively). The most effective factors on fire occurrence were anthropogenic (distance from residential, distance from agricultural lands, and distance from roads) and climatic factors (relative air humidity, wind speed, and slope aspect), and the least important factors were distance from rivers and slope aspect. Conclusion: Given the role of anthropogenic factors in the occurrence of fires, it is suggested that nature-based education be increased and people’s dependence on these forest ecosystems be reduced. Given the lack of sufficient information on fires and the importance of research on forest fires, it is recommended that a database of past and ongoing fires in the forests of the study area using remote sensing and geographic information systems and a history of fires in these areas be prepared to evaluate fire occurrence models in future research.
