A spatial explicit method combining HYSPLIT and random forest for predicting the next-day pollen index at urban scale

Abstract Airborne pollen is a significant cause of allergies leading to various discomforting symptoms. Many cities worldwide, including Beijing, face airborne pollen challenges. Accurate forecasting of airborne pollen levels can provide critical risk alerts for residents. However, existing studies at urban scale often overlook the spatial variability of pollen levels, focusing predominantly on pollen from individual plant sources or using station-based forecasts, which limits their applicability. To address these shortcomings with available data, this study integrates the Hybrid Single-Particle Lagrangian Integrated Trajectory model with Random Forest to develop a novel method for predicting the spatial distribution of next-day pollen indexes derived from dominant tree species. Using Beijing as a case study, the proposed method demonstrates promising prediction accuracy, with R 2 of 0.67 for the city-averaged index. An analysis of pollen index distributions during Beijing's spring pollen season reveals that urban areas consistently exhibit higher predicted pollen indexes for the included tree species compared to the citywide average throughout the season. This trend suggests an elevated allergy risk in urban regions from these specific pollen types, likely due to pollen accumulation in semi-enclosed terrains influencing pollen transport and deposition. Furthermore, the study highlights that eliminating the included dominant tree pollen sources in city centers is insufficient to achieve significant reductions in the predicted urban pollen index. This research presents a new approach for generating spatially resolved pollen index forecasts, offering enhanced spatial detail compared to traditional station-based method, and enhanced spatial detail based on available operational data and dominant vegetation surveys, offering valuable insights for allergy prevention strategies.