Advancing airborne pollen mapping with integrated ground and satellite observations

Pollen in the atmosphere can affect the Earth's climate by impacting cloud formation and can harm human health by triggering respiratory diseases such as allergic rhinitis (hay fever) and asthma. However, the current understanding of atmospheric pollen is limited by sparse observational data. To address this gap, we integrate spaceborne aerosol optical properties with meteorological and phenological data in a machine learning framework to estimate atmospheric pollen concentrations and composition across the continental United States. The model is based on a random forest algorithm trained using multi-year observations from 40 airborne pollen monitoring stations. Model performance is evaluated through 5-fold cross-validation using three metrics designed to assess mean concentration (Symmetric Mean Absolute Percentage Error, SMAPE), peak intensity (Normalized Average Range difference, NAR), and interannual variability (Normalized Average Interquartile Range difference, NAIQR). The model performs best in regions with dense observational coverage, particularly the northeastern and southeastern U.S.; for example, the average SMAPE for Poaceae in these regions is 0.40, compared to 0.56 in other regions. Our results also suggest that satellite-observed aerosol optical properties significantly improve model performance. Compared to models without optical inputs, our integrated approach achieves up to a 130% improvement in NAR and NAIQR for key taxa such as Ambrosia, Pinaceae, Populus, Ulmus, and Betula. This novel approach provides a consistent and comprehensive data-driven mapping of pollen concentrations, with the potential to extend predictive capability into regions lacking historical observations. Our results demonstrate that model performance improves largely when even limited local data are available for pre-training. This work advances our understanding of pollen distributions in data-sparse regions and provides a valuable tool for climate, ecosystem, and public health research.