P-617 - An interpretable machine learning technique to identify the key meteorological factors influencing the incidence of Fusarium head blight in Korea
Seoul National University Seoul, Seoul-t'ukpyolsi, Republic of Korea
Abstract Text: Fusarium head blight (FHB), mainly caused by Fusarium asiaticum, poses a serious threat to wheat production in Korea, leading to yield losses, economic damage, and health risks from mycotoxin exposure in humans and animals. However, environmental conditions favoring FHB at specific growth stages remain poorly understood. This study applied interpretable machine learning (ML) to identify key meteorological factors influencing FHB incidence in Korea, using nationwide survey data from 2015 to 2021. Two ML models—random forest and boosted regression trees—were selected for their robustness to multicollinearity and efficiency with relatively small datasets. Three critical environmental thresholds for FHB were identified: 75% relative humidity (RH) during heading, and a combination of 75% RH and ≥60 mm precipitation during flowering. Exceeding two or more of these thresholds substantially increased disease incidence compared to exceeding only one. These findings highlight the potential of interpretable ML to uncover stage-specific environmental drivers of FHB and improve disease forecasting in wheat and other crops.
