The aim of our project is to develop regional ensemble techniques to improve decadal climate predictions for Europe by means of dynamical downscaling and statistical post-processing. To achieve this goal the project will follow a four step strategy.
Development of a regional decadal prediction strategy: In a first step, a strategy for regional decadal predictions will be developed based on existing experience with regional multi-member ensembles. Results from in house PhD studies as well as from several projects like ENSEMBLES, Clavier and Klimzug will be assessed and synthesized in a strategy for REDCLIP.
Downscaling of decadal predictions: In step two, selected global baseline system decadal hindcasts will be dynamically downscaled using the regional climate model REMO. At least two ensemble members of each hindcast period (defined by the global simulations with ECHAM6/MPIOM in Module D (Synthesis)) will be used and directly downscaled to 0.22° horizontal resolution. The choice of the members will be coordinated within Module C to achieve a sufficient ensemble size for the regional decadal predictions. A detailed analysis in cooperation with partners in Module C but also with Module E (Validation) of the downscaled predictions will be done to assess the predictive skill of the regional hindcasts. It is planned to carry out a workshop with several partners from Module C and E to coordinate the analysis and skill assessment.
Generation of regional ensembles: A regional ensemble based on a stochastic physics approach and domain shifting will be set up and tested. Several regional ensembles of selected global hindcast members based on the strategy developed in step one will be generated (preferable in 0.44° resolution to be able to run as many members as possible). A detailed skill analysis of the ensembles will follow.
Statistical post-processing – recalibration: Statistical post-processing techniques such as the climate conserving recalibration (CCR) by Weigel et al. (2009) used in seasonalforecasting will be transferred to decadal climate predictions. Hence, the downscaling of the baseline decadal predictions already performed in step two will be completed. This is to get a more confident statistical basis available for the application of the statistical postprocessing. The recalibration method will then be applied to the regional and global baseline hindcasts. A comparison of the skill between recalibrated regional and global hindcasts for Europe will be evaluated.