Forest fires on the rise: Mapping high-risk zones and socio-economic exposure in İzmir, Turkey

In Turkey, forest fires pose a serious threat to the environment since they persistently destroy infrastructure, lives and forest ecosystems. The General Directorate of Forestry's official statistics indicate a rising trend in the frequency of wildfires, highlighting the need for better risk assessment and sustainable land management techniques. Adequate identification of areas that are fire-prone via a reliable modelling framework is therefore important for mitigation planning and resource distribution. Based on historical forest fire record coupled with other anthropogenic and environmental conditioning factors, this research performs a detailed assessment of the İzmir region's susceptibility to forest fires. The evaluated variables include climatic variables (wind speed, precipitation, temperature), topographic factors (altitude, slope, aspect, curvature, topographic wetness index (TWI), terrain ruggedness index (TRI), topographic position index (TPI)), land features (land use/land cover (LULC), geology, tree cover density, forest types), and closiness variables (distance to settlements, roads, and rivers). The influence of each parameter was evaluated through feature importance analysis to determine the key factors driving wildfire susceptibility. Fire susceptibility modelling was carried out employing three different machine learning algorithms, which are eXtreme Gradient Boosting (XGBoost), Decision Tree (DT), and Support Vector Machine (SVM). Model performance was carefully analyzed and compared to determine which one exhibits the highest predictive framework accuracy. The one with the best outcome was finally selected to produce a high-resolution susceptibility map and to forecast future wildfires under amended temperature and precipitation conditions. Feature importance analysis indicated that geology, temperature, aspect, and precipitation are the most important factors influencing current wildfire susceptibility patterns in the study zone. The socio-economic exposure was also evaluated by intersecting areas of high fire probability predicted by the XGBoost model with spatial data on vulnerable assets. This combined analysis provides practical insights to support regional wildfire management and mitigation strategies. Nonetheless, the generated susceptibility maps present an effective decision-support tool for authorities, enabling the identification of priority zones for intervention, strengthening early prevention strategies, and enhancing the spatial distribution of monitoring and fire suppression efforts. This research generates a scientific framework for well-informed decision-making and aimed interventions, amplifying disaster risk mitigation and long-term environmental sustainability.