Background
More than 25% of adults in Europe suffer from pollinosis, although variability across countries might be quite large. In Belgium, at least ~10% of people develop allergies due to birch pollen. Thus, many people may benefit from the birch pollen forecasting system established for the Belgian territory in 2023 based on the SILAM model (System for Integrated modeLling of Atmospheric composition). The key question, however, is which uncertainty can be expected when modelling and forecasting airborne pollen levels?
Materials & Methods
The uncertainty in modelling airborne birch pollen levels near the surface using SILAM is quantified based on a Monte-Carlo error approach for the season of 2018 in Belgium using varying major model input data and ECMWF ERA5 meteorological data. The relative Coefficient of Variation (CV%) is used as a measure for uncertainty. The studied key model input datasets that drive the birch pollen model are:
- Map with the amount (areal fraction) and location of birch trees on a native 0.1° x 0.1° grid.
- Map with the start and end of the birch pollen season on a native 1° x 1° grid.
- Ripening temperature of birch catkins.
For each input dataset, 100 randomly sampled data layers were prepared for running SILAM 100 times. For the maps, in each 1° by 1° block containing 100 grid cells (0.1° x 0.1° native grid), we randomly redistributed the birch pollen emission sources 100 times. From the resulting 100 SILAM model runs, 100 spatial-temporal datasets on surface birch pollen levels were produced, and their variation was summarized by the CV%.
Results
From the analysis, we find that the uncertainty in the amount and locations of birch pollen emission sources in SILAM on resulting modelled airborne birch pollen levels near the surface in Belgium is substantially high, with CV% values ranging between ~15% and ~35%. The parameters indicating the start and end of the season, however, are at least equally important. CV% values up to 50% are found in the southeastern parts of Belgium. By adding up all the model input uncertainties, including the impact of the catkins-ripening temperature, we obtain CV% values of 50% and more. If we assume an accumulated error of 20-40% from all meteorological data, a CV% value near 60% can be expected. These error values in modelled pollen levels are in the same order of magnitude as the reported errors in monitored pollen levels, based on the reference Hirst method.
Conclusions
The uncertainty in airborne birch pollen levels near the surface modelled using SILAM and quantified as the CV% is more than 50%. It is the same order of magnitude as the reported errors from observed pollen counts using Hirst-type devices at monitoring stations.
