Simulation of analysis error contributions arising from uncertainties in the parametrisation of deep convection

Abstract

AbstractIn order to simulate analysis error contributions resulting from observation errors and model errors, a possible methodology is to run a data assimilation ensemble that includes observation perturbations and a multiphysics approach. This method is examined here formally and experimentally, in order to investigate the effects of deep convection parametrisation uncertainties on analysis errors. A linearised expansion of initial condition errors is first employed to decompose them into contributions arising from observation errors and model errors. This is compared formally with a similar expansion for analysis differences resulting from the use of two different model versions, as in a multiphysics approach. Experimental results with the Action de Recherche Petite Echelle Grande Echelle global model are then examined. Observation error contributions are simulated by analysis differences between an unperturbed experiment and another experiment that includes observation perturbations that are drawn from specified observation error covariances. This is compared with analysis differences resulting from the use of two different parametrisation schemes of deep convection. The results indicate that the contribution of convection uncertainties is relatively large in the Tropics, with similar amplitudes to the contribution of observation errors. Simulated model errors are also found to be pronounced in the boundary layer and in convective areas where the analysis constraint by observations is relatively small, due to lower observation density.