Spatial modeling the probability of wildfire occurrence using frequency ratio and weight- of-evidence models

In recent years, many attempts have been made to manage and control wildfires. Modeling and mapping wildfire probability across fire-prone landscapes is one of the most important measures. In the present study, the capability of frequency ratio and weight-of-evidence models for predicting the probability of wildfires occurrence in the Kohgiluyeh and Boyer-Ahmad province were investigated. The modeling process and prediction of future fires were based on an analysis of the relationship between 271 historical fires occurred during the 2002-2014 period and 10 predictor variables including slope degree, aspect, altitude, temperature, wind speed, land use, NDVI, and proximity to rivers, roads, and human settlement. During the modeling process, the significance of each variable class on wildfire occurrence was quantified. The model results were used to produce distribution maps of wildfire probability. The results of the evaluation and comparison of the models, which were carried out using the receiver operating characteristic method, success rate, prediction rate, and Wilcoxon test showed that the weight-of-evidence model with success and prediction rates of 0.886 and 0.821 performed better than the frequency ratio model in both training and validation datasets. Overall, the results revealed that approximately 30% of the forests and rangelands of the province fall within the high and very high probability to wildfire occurrence, which requires prudent management measures to mitigate the risk of fire.