The European Pollen Reanalysis for alder, birch, and olive airborne pollen was released at the end of 2023, with only technical validation of the procedure presented in the paper accompanying the dataset. The current presentation provides a first glance on the quality of the reanalysis and highlights its strengths and areas of improvement. The analysis is performed at a few individual stations in different parts of Europe: Finland, Latvia, Lithuania, Belgium, and Spain.
The reanalysis was built using only assimilation of the total seasonal pollen production, i.e. the phenological models and their parameters were left completely intact. This allowed an almost-complete separation of the absolute values of the predicted time series and their temporal behavior. The first ones were corrected during the assimilation, the second ones were practically the same in both the initial SILAM run and the final reanalysis.
SPIn time series at individual stations.
The Seasonal Pollen Integral at individual stations is the most-affected parameter by the assimilation: with exception of the long-range transport, SPIn is directly proportional to the regional pollen production. Not surprisingly, the reanalysis demonstrated a strong improvement of this parameter practically at all stations and all tree genera. The most significant improvement was for birch pollen, except for Southern Europe, where local pollen production is very small, and the season is almost entirely decided by the long-range transport.
Absolute bias and RMSE.
The reanalysis is generally unbiased by its design. However, predictions in some individual years and at specific stations could manifest noticeable deviations, offset in other years. As a result, RMSE was mostly decided by the temporal correlation: low-correlating stations showed higher RMSE.
Temporal correlation at individual stations.
The intra-seasonal concentration evolution was not affected by the assimilation (except for the changes in the long-range transport), so the model skills remained practically the same for the initial and final runs. The most-challenging genera was alder, which SILAM considers as a single taxon. Such simplification was acceptable in regions with only one dominant species or with close flowering times but led to significant errors if several species flowered at different times in the same region. The birch timeseries are less sensitive to this feature because different taxa almost always have close flowering periods.
Season start and end.
These are the parameters practically not affected by the assimilation. Therefore, the key features of the operational SILAM version have been manifested in all three genera. The model was reproducing the season start with a comparatively high accuracy (majority of the considered stations showed just a few days of an error), whereas the season end was quite significantly delayed.
