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

Area of Science:

• Environmental science
• Aerobiology
• Computational modeling

Background:

• Accurate airborne pollen forecasts are crucial for managing respiratory allergies like allergic rhinitis and asthma.
• Long-term pollen predictions can also aid biodiversity conservation and ecosystem health.
• Existing forecasting methods often lack generalization across varied climatic regions.

Purpose of the Study:

• To develop and validate an ensemble forecasting model for airborne grass (Poaceae) pollen concentrations.
• To assess the model's performance across three climatically distinct European cities.
• To identify key meteorological and aerobiological predictors for pollen forecasting.

Main Methods:

• An ensemble model was created using seven machine learning families (including NNETAR, ARIMA, XGBoost, Random Forest).
• Pollen data (2018-2024) from Hirst-type traps and meteorological data were utilized.
• Model weights were optimized based on predictive performance, with 2024 data used for validation.

Main Results:

• The ensemble model achieved high predictive accuracy (R²: 0.66 Augsburg, 0.62 Córdoba, 0.84 Thessaloniki).
• Lagged pollen concentrations and previous-day temperature were identified as significant predictors.
• Incorporating data from an automatic pollen monitor enhanced predictive performance (R² = 0.89 in Augsburg).

Conclusions:

• Ensemble-based pollen forecasting effectively generalizes across diverse bioclimatic regions.
• The model is sensitive to local ecological and climatic factors, enabling improved allergy risk management.
• This framework supports the development of advanced real-time forecasting systems and offers insights into vegetation dynamics.