中国全土のアレルゲン性花粉リスクを高解像度マッピングするアンサンブル機械学習手法

Airborne pollen is a key environmental allergen affecting millions across China. As pollen levels and allergy prevalence continue to rise under rapid urbanization and climate change, developing spatially explicit, long-term pollen datasets becomes increasingly important for public health and ecological risk assessment. In this study, we developed a novel ensemble machine learning framework integrating random forest and gradient boosting models to estimate daily tree and herbaceous pollen concentrations across mainland China from 2011 to 2023. Models were trained using daily pollen data from 27 monitoring sites during 2019-2024 and a rich set of predictors, including meteorological, vegetation, land use, and spatiotemporal variables. By applying the trained models to historical environmental datasets, we reconstructed nationwide daily pollen concentrations for 2011-2023 to extend the temporal coverage beyond the observational record. The models achieved high accuracy, with R² values of 0.90 (tree) and 0.89 (herbaceous), and root mean square errors of 0.58 and 0.49, respectively. Tree pollen peaked in early spring in eastern, northeastern, central, and southwestern regions, while herbaceous pollen peaked in late summer in northern and northwestern areas. Seasonal timing, temperature, and vegetation indices were key drivers, with short-term lagged temperature (0-7 days) strongly influencing predictions. This study provides the first nationwide, long-term, daily pollen dataset for China derived from observation-based modeling and historical reconstruction, serving as an important resource for ecological research and public health applications. The established modeling framework offers a robust foundation for pollen exposure assessment, allergy forecasting, and climate-responsive risk management of aeroallergens under changing environmental conditions.