Predicting Wildfires Triggered by Anthropogenic Burning: A Spatial Framework Integrating AI Models

Anthropogenic ignition sources—such as agricultural residue burning, land clearing, and open waste disposal—are increasingly contributing to wildfire occurrence, particularly in areas undergoing rapid land use change and urban expansion. In South Korea, over 40% of recent wildfires are linked to such activities, with a growing concentration along the urban–rural interface. This study presents a spatially explicit wildfire risk assessment framework that integrates Generalized Additive Models (GAM), Random Forest (RF), and Geographically Weighted Regression (GWR) to identify key environmental and anthropogenic drivers of wildfire ignition. Using a national wildfire database (2001–2025), combined with high-resolution climatic, land-use, and demographic data, the study establishes operational risk thresholds and generates predictive risk maps. Among the models, RF exhibited the highest predictive performance (AUC = 0.82), effectively classifying high- and low-risk areas. GAM revealed complex nonlinear relationships—fire probability increased sharply below 15% relative humidity, peaked at wind speeds around 15 m/s, and was highest at intermediate NDVI values (0.2–0.4). Interestingly, the highest wildfire risks were observed not in remote interiors but along peri-urban zones near Seoul, Incheon, and Daejeon, where land fragmentation, recreational activities, and ignition sources converge. GWR further identified spatial heterogeneity in predictor effects, emphasizing localized risk dynamics. High-risk areas (probability ≥ 0.7) comprised 16.96% of the national territory, especially in eastern Gangwon-do, northern Gyeonggi-do, and parts of Chungcheong and Jeolla provinces. This integrated modeling framework supports region-specific wildfire prevention, land-use planning, and climate-adaptive fire management, and offers a transferable approach for other fire-prone regions.