Spatial wildfire risk assessment for Oahu, Hawaii, using swarm-intelligence-optimized CatBoost and the analytic network process

초록

Wildfires in the Hawaiian Islands are an increasing concern due to climate change and the spread of invasive species. This study thus assessed the wildfire risk on Oahu, Hawaii. Wildfire susceptibility was modeled using the categorical boosting (CatBoost) algorithm. To improve model performance, its hyperparameters were optimized through a comparative analysis of two swarm-intelligence (SI) metaheuristics: the pathfinder algorithm (PFA) and the sparrow search algorithm (SSA). These SI approaches automatically explored the hyperparameter space and maximized the prediction accuracy for the validation set, resulting in two optimized models (PFA-CatBoost and SSA-CatBoost). Socioeconomic vulnerability was then quantified using the analytic network process (ANP), based on eight socioeconomic and demographic indicators that were weighted, fuzzified, and integrated into composite index. The wildfire susceptibility results showed that the hybrid PFA-CatBoost and SSA-CatBoost models outperformed the baseline CatBoost model, with SSA-CatBoost achieving the best performance, reaching an accuracy of 0.984 on the testing set. The ANP analysis identified annual household income, population density, and education level as the primary socioeconomic indicators associated with wildfire vulnerability. According to the resulting wildfire risk map, 11.7% of Oahu fell within the high or very-high risk categories. The high-risk areas were primarily concentrated in densely populated corridors such as Makaha, Waianae, Maili, Nanakuli, Ewa Gentry, Waipahu, and the Honolulu-Waialae-Kahala corridor. Overall, the proposed framework provides a robust approach to spatial wildfire risk assessment, supporting risk-informed decision-making for wildfire mitigation, land-use planning, and emergency management on Oahu Island.

키워드