Future ground-level summer ozone concentrations in Europe: the importance of ozone as predictor in a statistical downscaling approach

Statistical downscaling models are used to estimate historical and future ozone concentrations in Europe for the summer months from April to September. The basis is formed by over 700 measuring stations from the hourly Air Quality eReporting ozone pollution data from the European Environment Agency. Daily maximum 8-hr running means (MDA8) as well as daily maximum 1-hr values (MDA1) of ozone are the target variables. Meteorological (ERA5) as well as ozone (CAMS) reanalysis data serve as predictors in the perfect prognosis (PP) approaches. A station-specific, individualized predictor screening guarantees site-specific optimums. The predictor selection is performed using regularization with varying shrinkage. Multiple Linear Regression (MLR) is used to model the relationship between all selected predictors and MDA8/ MDA1. An ensemble of seven CMIP6 Earth system models is used to estimate future ozone concentrations. Projections are calculated for the years 2041-2060 and 2081-2100 under two different future scenarios (SSP2-4.5 and SSP3-7.0). The CMIP6 data is bias corrected with a univariate quantile delta mapping method, before being used in the statistical models.

With respect to predictor selection, a sensitivity study is conducted, testing different sets of predictors to examine their influence on future ozone concentrations. Predictor sets with and without ozone, with only thermal and radiative predictors, and additionally thermo-dynamical or circulation-dynamical information are analyzed, in addition to the site-specific optimums described above.

The projection results of the different predictor settings highlight the importance of ozone as predictor. In all predictor sets with included ozone, the results of the two scenarios SSP2-4.5 and SSP3-7.0 differ in their sign depending on the scenario used. The SSP2 scenario, called "Middle of the Road", leads to decreasing ozone concentrations in Europe, while the more pessimistic SSP3 scenario results in partly strong increases of harmful ground-level ozone concentrations. In contrast, all predictor sets that do not take ozone into account show consistently positive change signals.

On the one hand, our results point to the need to include information on emission changes in statistical assessments in order to obtain a realistic picture of future ozone development. Furthermore, our research underscores the need to further reduce air pollution in Europe to better protect human health from direct emissions such as NO_x and indirect pollutants such as ozone.