Impact of a Stochastic Parameterization of Cumulus Convection Using Cellular Automata in a Meso-Scale Ensemble Prediction System.
Thursday, 18 December 2014: 8:15 AM
A common way of addressing forecast uncertainty in Numerical Weather Prediction (NWP) models is to use ensemble prediction. The idea behind ensemble prediction
is to simulate the sensitivity of the forecast to the initial and boundary conditions, as well as model construction error, such as sub-grid physical parameterizations. Existing methods used in order to account for such model consturction uncertainty include; multi-model ensembles, adding random perturbations to the tendencies produced by the parameterizations, or perturbing parameters within the parameterizations. Although such methods are successful in a probabilistic sense, individual ensemble members can be degraded in a deterministic sense by adding random non-physical perturbations. Furthermore, different ensemble members can have different bias (and skill) since they are based on separate models and/or parameters. Another way of accounting for model uncertainty (due to sub-grid variability) is to introduce random variability in the convection parameterization itself. Here we will present the impact of the stochastic deep convection parameterization using cellular automata described in Bengtsson et. al. 2013, as implemented in the high resolution meso-scale ensemble prediction system HarmonEPS. The questions we would like to answer are; can we improve the forecast skill both in a deterministic and probabilistic sense using the stochastic convection scheme? Can the stochastic parameterization in terms of the spread/skill relationship compete with the multi-model approach? Furthermore, the stochastic parameterization proposed in Bengtsson et. al. 2013 addresses lateral communication between model grid-boxes by using a cellular automaton. It was demonstrated that the scheme in a deterministic model is capable of contributing to the organization of convective squall-lines and meso-scale convective systems. We study if and how uncertainties with origin on the sub-grid scale transfer to the larger atmospheric scales through such convective organization. This is done by studying the ensemble spread of the kinetic energy spectra at various forecast lead-times. Bengtsson, L., Steinheimer, M., Bechtold, P and Geleyn, J-F (2013): A Stochastic Parameterization for Deep Convection Using Cellular Automata. QJRMS, 139, 1533-1543.