The plethora of available General Circulation Model (GCM) data allows an inclusion of model evaluation techniques to weigh the contribution of GCMs for the use in a statistically downscaled ensemble. We present the development of a pipeline in which GCM data is evaluated based on different criteria, downscaled using an analog technique and the performance impact of the evaluation is validated via the SPARTACUS dataset for Austria.
For the evaluation of GCMs we employ multiple metrics, which are calculated for all GCMs as well as reanalysis data. For the latter we use JRA55 and ERA5 reanalyses, which are used as historical reference. The inclusion of two reanalysis data sets allows the estimation of reanalysis uncertainty contained in the employed evaluation methodology. A principal component based Northern Atlantic Oszillation Index is calculated from which various performance measures based on the time coefficients and empirical orthogonal functions are determined. Furthermore the Central European Zonal Index is used as an additional performance metric. Time-independent quantities are then calculated for all performance measures to reflect the non-time-synchronous characteristic of GCMs. Finally, different weighting schemes are determined based on the performance metrics and additional factors such as model independence, yielding weighting coefficients for each model. The GCMs are downscaled using an analog approach with a random choice analog selection based on the ten best analogs for each given day. Predictands to be downscaled are the near-surface temperature and precipitation totals with a daily resolution. The downscaled ensemble is then compared to the SPARTACUS dataset and weighting schemes are evaluated by minimizing the error of the ensemble mean over a historical period.
