Construction and application of a high-resolution pollen numerical model system based on phenology and XGBoost

The allergenic characteristics of anemophilous pollen differ among taxa, and its atmospheric concentration plays a critical role in influencing allergy risks and health assessments. Based on phenology and artificial intelligence methods, the study constructed a high-precision pollen numerical model system using 15-year pollen monitoring data of Beijing in spring. The model system considered the phenological phases and multiple plant taxa, and based on thermal accumulation, optimized the start and end dates of pollen release for Cupressaceae and Salicaceae (Stage I) and Pinaceae (Stage II). A normalized pollen emission potential model was constructed by integrating dual-threshold temperature accumulation with phenological probability. XGBoost was used to simulate seasonal pollen integral (SPIn), enabling spatiotemporal modeling of spring pollen emission. To represent pollen transport and dispersion in the atmosphere, a new pollen module was incorporated into the WRF-Chem framework, coupling pollen emission with meteorological adjustment factors such as wind speed, precipitation, and humidity, thus forming the WRF-Chem-Pollen system for high-resolution, hourly pollen simulations. The model demonstrated strong performance, with R exceeding 0.60 in 60 % of the years, confirming its spatiotemporal reliability. Further analysis revealed a compound influence of meteorological factors on SPIn, characterized by a "early-season promotion and late-season suppression" effect within distinct time windows. Stage I pollen dominated spring loads, while Stage II contributions were more stable. Overall, this study establishes a comprehensive framework for modeling spring pollen and elucidates the meteorological drivers of pollen release. The findings provide scientific support for pollen exposure assessment, health forecasting, and urban ecological management.