| The aim is to obtain probabilistic information on climate change signals from ensembles of global and regional climate models. Since single-integration climate models only provide one possible realisation of climate variability, ensembles are an common approach to address the uncertainty in climate modelling. Recently, several methods, mostly from the field of numerical weather prediction, have been developed (SKD, AKD, NGR, BMA). The method presented here is related to ensemble kernel dressing and is extended to a multivariate approach in order to include (spatio-)temporal dependence. Furthermore, we discuss the general problem of how to estimate model-specific weights, especially when dealing with a-priori distinguishable as well as a-priori indistinguishable simulations. The method is applied to ensembles of coupled general circulation models, e.g. ECHAM5, as well as regional climate simulations over Europe using the CLM model. |
|