Dynamical and statistical downscaling of seasonal temperature forecasts in Europe: Added value for user applications

This work describes the results of a comprehensive intercomparison experiment of dynamical and statistical downscaling methods performed in the framework of the SPECS (http://www.specs-fp7.eu) and EUPORIAS (http://www.euporias.eu) projects for seasonal forecasting over Europe, a region which exhibits low-to-moderate seasonal forecast skill. We considered a 15-member hindcast provided by the ECEARTH global model (similar to ECMWF System 4, but using bias corrected SST) for the period 1991-2012. In particular, we focus on summer mean temperature and evaluate the added value of downscaling for representation of the local climatology (mean values and extremes), improvement of model skill and performance in particular heatwave episodes. Whereas the suitability of dynamical downscaling for reducing the orographic biases of the global model depends on the region and model considered, statistical downscaling can systematically reduce errors in different order moments, from the mean to the extremes (as represented by the 95th percentile here). However, both dynamical and statistical methods lead to similar skill patterns with about the same overall performance as the global model, which shows higher values in south-eastern Europe. Therefore, no relevant added value is found in terms of model skill improvement. Finally, when focusing on the heatwaves of 2003, 2006, 2010 and 2012, the limitations of the global model to detect these hot episodes are inherited by all dynamical and statistical downscaling methods so no added value is neither found in this aspect. This work provides, to our knowledge, the largest and most comprehensive intercomparison of statistical and dynamical downscaling for seasonal forecasting over Europe.This study was supported by the SPECS and EUPORIAS projects, funded by the European Commission through the Seventh Framework Programme for Research under grant agreements 308378 and 308291, respectively. We are also grateful to the E-OBS dataset from the EU-FP6 project ENSEMBLES and the data providers in the ECA&D project. One of the authors (EvM) wants to thank Michael Kolax from SMHI (Norrköping, Sweden) for making available the full EC-EARTH hindcast ensemble for dynamical downscaling at KNMI. Finally, for the WRF simulations, the authors acknowledge the access provided to the Altamira Supercomputer at the Institute of Physics of Cantabria (IFCA-CSIC), member of the Spanish Supercomputing Network (http://grid.ifca.es/wiki/Supercomputing)