An ensemble-forecasting model for airborne grass pollen at three climatically distinct sites

Precise airborne pollen forecasting is essential for mitigating exposure risks in individuals with pollen-related respiratory diseases such as allergic rhinitis and asthma and for supporting timely public health warning. Moreover, long-term accurate pollen forecasts could also support biodiversity conservation, ecosystem functions, and public-health protection. We developed an ensemble forecasting model for airborne grass (Poaceae) pollen concentrations in three climatically distinct European cities: Augsburg (Germany, transitional temperate-continental), Córdoba (Spain, dry Mediterranean), and Thessaloniki (Greece, humid Mediterranean). Pollen data (2018-2024) from Hirst-type volumetric traps were combined with meteorological parameters (temperature, humidity, precipitation). The 2024 pollen data were used for validation. Of 61 candidates, seven representative model families (Regularized Linear Regression, Extreme Gradient Boosting, Neural Network Autoregression [NNETAR], Random Forest, Support Vector Regression, Prophet-XGBoost hybrid, and Autoregressive Integrated Moving Average [ARIMA]) were selected for the ensemble. Model weights were assigned according to predictive performance. The ensemble achieved R² values of 0.66 in Augsburg, 0.62 in Córdoba and 0.84 in Thessaloniki, with NNETAR and/or ARIMA contributing most strongly during the pollen season. Lagged pollen concentrations and previous-day temperature emerged as key predictors. When incorporating data from an automatic pollen monitor (BAA500, Helmut Hund GmbH) in Augsburg, the model achieved higher predictive performance (R² = 0.89). Our findings demonstrate that ensemble-based pollen forecasting may generalize across contrasting bioclimatic regions, while remaining sensitive to local ecological and climatic controls. This framework provides a foundation for more powerful (real-time) forecasting systems aimed primarily at improving daily allergy risk management, while potentially offering complementary insights into longer-term vegetation dynamics under climate variability.