Enhancing Fire Susceptibility Mapping in Semnan Province: Integrating Machine Learning and Geospatial Analysis

Aims: This study assesses the impacts of natural and human factors on fire occurrences, identifies key contributors to fire susceptibility maps, and employs machine learning algorithms (MLAs) to enhance the spatiotemporal patterns of fire susceptibility maps. Materials & Methods: Data were collected from ۱۱۰ fire locations and ۱۱۰ non-fire points spanning from ۲۰۰۱ to ۲۰۲۲ at annual scale. Various auxiliary variables, including climate data, terrain features, Normalized Difference Vegetation Index (NDVI), and distance to roads, were analyzed to model fire susceptibility. The study employed multiple MLAs, including Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Decision Trees (GBDT), to generate the fire susceptibility maps. Findings: About ۷۰% of fires occurred within ۲ km of roads, indicating significant human influence. Grasslands had the highest fire rates, with over ۲۵% of fires from ۲۰۰۱-۲۰۲۲ due to flammable fuels. The RF and mean models identified ۰.۴% and ۱.۳۱% of the area as very high susceptibility (۳۸,۸۰۰ km² and ۱۲,۶۰۰ km²), while the GBDT and SVM models identified ۲.۴۲% and ۱.۸۶% (۲۳۴,۷۰۰ km² and ۱۸۰,۰۰۰ km²). The very high susceptibility class, though small in percentage, covers large areas. Conclusion: This research highlights the importance of integrating environmental and human factors for predicting fire events in arid regions and developing comprehensive fire susceptibility maps, critical for protecting vulnerable ecosystems. These outcomes provide valuable tools for fire management and mitigation strategies within vulnerable ecosystems. Moreover, developing targeted fire management strategies focused on high-risk areas, such as juniper and broadleaf forests must be a priority.

کلیدواژه ها:

Auxiliary variables ، Climatic indicators ، Juniper Forest ، Distance to Road ، Fire Susceptibility