Representing Model Uncertainty in the ECMWF Convection Scheme

Model uncertainty is currently represented in the ECMWF model, the Integrated Forecasting System (IFS), using two stochastic schemes. The Stochastically Perturbed Parametrisation Tendencies (SPPT) scheme represents random errors associated with model uncertainty due to the model's physical parametrisation schemes, and so uses multiplicative noise to perturb the parametrised tendencies about the average value that a deterministic scheme represents (Palmer et al., 2009). In contrast, the Stochastic Kinetic Energy Backscatter scheme (SKEB) represents a physical process absent from the model: the streamfunction is randomly perturbed to represent upscale kinetic energy transfer (Berner et al., 2009). We are interested specifically in the representation of uncertainty in the convection parametrisation scheme. SPPT is used to represent uncertainty in the other physics schemes (clouds, radiation etc.). This representation is tested in areas where the effect of convection is small, and shown to be reliable. We compare several representations of uncertainty in the convection scheme. As a benchmark, we consider the operational SPPT scheme, which uses the same random number to perturb each physics scheme. Firstly, we considered the effect of a perturbed parameter ensemble. The values of four parameters in the convection scheme were estimated using the Ensemble Prediction and Parameter Estimation System (Järvinen et al, 2011) by colleagues at ECMWF. Each of the 50 ensemble members in the ensemble prediction system was assigned a different set of these parameters, drawn to sample the joint posterior distribution. The perturbed parameter representation improves on the deterministic convection scheme, but is less skilful than the standard SPPT scheme. A stochastically perturbed parameter scheme, in which the parameter perturbations are modulated stochastically in time and space, was also tested. This scheme is a physically motivated way of including stochastic physics into an ensemble system, and performs similarly to the fixed perturbed parameter scheme. Finally, we considered an independent SPPT scheme, which uses a different random number field for each of the five physics schemes. This resulted in a large increase of spread in areas of significant convection, correcting the under-dispersive ensemble in these regions. In regions where the effects of convection are small and the ensembles are well calibrated, this scheme has little effect, and the ensemble forecasts for these regions remain reliable. Further testing of this scheme at operational resolution is currently underway. J. Berner et al. A spectral stochastic kinetic energy backscatter scheme and its impact on flow dependent predictability in the ECMWF ensemble prediction system. Journal of the Atmospheric Sciences, 66(3):603-626, 2009. H. Järvinen et al. Ensemble prediction and parameter estimation system: the concept. Q. J.Royal Meteorol. Soc., 138(663), 281-288, 2012. T. N. Palmer et al. Stochastic parametrization and model uncertainty. Technical Report 598, European Centre for Medium-RangeWeather Forecasts, 2009.