Canada’s Wildfire Susceptibility Assessment Using Statistical Data-Driven Models

Abstract

Wildfire Susceptibility Assessment (WSA) is one of the critical approaches to wildfire risk management. In this study, we employed a hybrid approach by integrating two distinct statistical models, namely Frequency Ratio (FR), Weight of Evidence (WoE), with Shannon Entropy (SE) (i.e., FR-SE and WoE-SE) for WSA. To meet the aim, 18538 historical wildfire data were collected and separated into two sections for model development and validation. Next, 13 wildfire-influencing parameters, including slope degree, aspect, topographic wetness index, elevation, evapotranspiration, land use/land cover, normalized differences vegetation index, distance from the lake, precipitation, distance from the rivers, distance from the roads, soil moisture, and mean annual maximum temperature were prepared and feed the models. Finally, model performance were evaluated using the validation data set and receiver operating characteristic (ROC) curve technique. Findings shows that the integration of models has improved the modeling performance, as WOE-SE model has the highest performance (96.5%), followed by WoE (96.3%), SE-RF (95.9%) and RF (95.2%) model respectively. Result of SE model showed that mean annual maximum temperature has the highest impact on the wildfire occurrence across Canada, while topographic wetness index is the lowest effective parameter.

Keywords: Wildfire, statistical models, Canada, Shannon Entropy, Frequency ratio, Weight of Evidence.