In Japan, Japanese oak wilt has caused widespread and long-term damage, prompting growing attention toward remote sensing technologies for its detection. This study proposes a method for detecting Japanese oak wilt damage through time-series analysis of Sentinel-2 satellite imagery. We analyzed the phenological changes of vegetation indices (NDGI, NDVI, and NWI) in northern Awaji Island, Japan, during the severe 2020 outbreak, using a Random Forest classifier trained with phenology-based features. Detection performance varied with the label definition, and it was maximized when a lower dead tree area threshold (1%) was used for training labels, suggesting that low within-pixel damage signals are informative. The proposed time-series approach substantially reduced false positives relative to a single-date analysis, thereby improving precision. These findings highlight the high potential of our approach for wide-area screening of oak wilt damage.
Takeda et al. (Thu,) studied this question.