Advances in pollen forecast quality across CAMS regional models

The allergenic pollen forecast has become a standard part of the daily air quality production chain of the regional CAMS (Copernicus Atmosphere Monitoring Service) models. The first operational European pollen forecasts started in 2013 based on the developments within MACC (Monitoring Atmospheric Composition and Climate, the predecessor of CAMS) and were released during its Interim Implementation phase (MACC-II). The four-day birch predictions were computed by seven regional models (CHIMERE, EMEP, EURAD-IM, LOTOS-EUROS, MATCH, MOCAGE, SILAM) and aggregated into a multi-model ENSEMBLE. The establishment of European pollen forecasts and their evaluation were supported by cooperation with the European Aeroallergen Network (EAN), which serves as the main provider of pollen observations in Europe. Over the years, the CAMS model cluster was expanded with four new independent models DEHM, GEMAQ, MINNI, MONARCH and the list of pollen forecasts was gradually extended with olive, alder, grass, ragweed, and mugwort species. Currently, the CAMS European pollen forecast is delivered by eleven state-of-the-art chemical transport models and their ENSEMBLE for six pollen species. Only recently has the systematic evaluation of CAMS pollen predictions developed into regular reports published at the end of the season.

Present work will demonstrate the progress in the model’s performance across 11 CAMS regional models and their ENSEMBLE, based on the evaluation of daily timeseries from first-day hourly model forecasts. The forecast accuracy will be assessed in terms of the model mean bias, temporal correlation coefficient, root mean square error, and shifts in the pollen season start and end of the aerobiological season.

The analysis shows that there is no ultimate ”best” model. Depending on the type of pollen and the evaluation score, the different models appear to be good. In 2024 the ENSEMBLE scores were among the best, sometimes outperforming all individual models. The median ENSEMBLE was, in most cases, capable of disregarding the outliers, still providing good forecasts. Exception was the alder forecast, where the majority of models were very low, and the resulting underestimation penetrated through the median, leading to a strongly low-biased ENSEMBLE. Practically all models showed some deviations from the main ENSEMBLE for individual pollen species.