Taking into account model inadequacies is a key challenge in climate forecasting. As part of the FP7-SPECS project, we examine how stochastic perturbations of atmospheric model dynamics impact seasonal forecast quality of the CNRM coupled model. The method described in this paper helps derive model error statistics as well as improve key aspects of our forecasting system such as systematic errors over the North Atlantic mid-latitudes.