Downscaling of monthly PM10 indices at different sites in Bavaria (Germany) based on circulation type classifications

Atmospheric circulation affects local concentrations of particulate matter with an aerodynamic diameter of 10μm or less (PM10) in different ways: Via the determination of local meteorological conditions favoring or suppressing the formation and the accumulation of PM10, and through its control on short–and long–range transport of particles and precursors. The quantitative assessment of the connections between the large–scale atmospheric circulation and local PM10 is relevant not only for the understanding of observed variations in PM10 concentrations. It is even more important for estimating the potential effects of projected future changes in large–scale atmospheric circulation on PM10. In this contribution, daily atmospheric circulation types (CTs), resulting from variants of three different classification methods, and their monthly occurrence frequencies have been utilized in three different downscaling approaches for estimating monthly indices of PM10 for the period 1980–2010 at 16 locations in Bavaria (Germany). All variants of approaches have been evaluated via a leave–one–out cross validation procedure in order to attain reliable performance ratings to detect the most suitable downscaling approaches. Results indicate that the highest performance of downscaling approaches is achieved in winter when the best performing models explain on average roughly 50% of the observed PM10 variance. From this it can be concluded that classification–based approaches are generally suitable for the downscaling of PM10, particularly during winter when PM10 concentrations in Bavaria reach maximum values. As preferable settings of the downscaling approaches, the usage of rather small spatial domains and a relatively high number of classes for circulation type classification and furthermore the utilization of multiple linear regression analyses or random forest analyses for relating CTs to PM10 have been ascertained. These findings provide the basis for further enhancements of the classification–based downscaling of monthly PM10 that will be realized in successive investigations.