Ecological Informatics (May 2025)
Ensemble-based forecasting of wildfire potentials using relative index in Gangwon Province, South Korea
Abstract
Wildfire indices have been widely used to assess wildfire potential under varying climate conditions. However, their region-specific applicability remains limited due to inherent incompatibilities among various indices. This study proposes post-processing procedures for wildfire forecasting by applying statistical index-merging methods to enhance the utility of conventional wildfire indices in forested regions of South Korea. Accordingly, 126 wildfire cases from 2014 to 2023 are analyzed, and the performance of conventional indices is assessed individually and in combination using three merging methods: simple averaging, variance-covariance (VC), and triple collocation (TC). The forecasting capabilities of individual and merged indices are evaluated using hit/miss metrics, specifically the probability of detection and false alarm ratio. The results reveal that ensemble merging techniques can partially enhance the forecasting performance of conventional indices that are otherwise suboptimal for the target region. The forecasting performance of both individual and merged indices is higher in forested areas, highlighting vegetation as a significant factor in wildfires. Notably, the forecasting performance of the VC method, which incorporates inter-index correlations, is superior to that of TC, which does not account for these correlations. This robust performance is evident regardless of wildfire size or severity. Furthermore, the findings underscore the critical role of climate conditions in wildfire detection, highlighting the need to address the effects of climate change. Consequently, the application of statistical index-merging methods enhances wildfire forecasting accuracy in Gangwon Province, South Korea, surpassing that of conventional indices alone, despite their global reliability.
