A Data-Driven Assessment of Machine Learning Models for Climate-Sensitive Wildfire Susceptibility in Alberta

The accelerating intensification of wildfire activity across Western Canada underscores the need for adaptive, climate-responsive risk assessment frameworks. Alberta has experienced increasingly severe fire seasons driven by rising temperatures, prolonged drought conditions, altered wind regimes, and vegetation stress linked to climate variability. This study proposes a hybrid geospatial–artificial intelligence framework designed to enhance wildfire susceptibility assessment under evolving climatic conditions. Thirteen environmental and anthropogenic conditioning variables, including topographic attributes, vegetation indices, hydro-climatic indicators, and proximity-based human activity factors, were integrated within a Geographic Information System (GIS) environment. To address uncertainty in multi-criteria decision-making, a fuzzy-based weighting scheme combined with the Analytic Hierarchy Process was implemented to generate a climate-informed susceptibility surface.

Building upon this spatial foundation, multiple machine learning classifiers, including Support Vector Machine, k-Nearest Neighbor, and Extreme Learning Machine, were trained using historical wildfire records (1991–2023). Model performance was evaluated using cross-validation on accuracy, precision, recall, F1 Score, and Receiver Operating Characteristic–Area Under the Curve (ROC–AUC) metrics. The results indicate that high- and very-high-susceptibility zones dominate substantial portions of the study region, reflecting increasing exposure to climate-driven ignition conditions. Among the tested algorithms, the Extreme Learning Machine demonstrated superior predictive capability, achieving an overall accuracy of 96% and a ROC–AUC exceeding 0.98. The proposed framework advances wildfire hazard modeling by integrating fuzzy uncertainty treatment with data-driven predictive analytics, offering a scalable decision-support tool for climate adaptation planning, emergency preparedness, and sustainable land management strategies