Stochastic-Dynamic Parameterizations in Numerical Weather Prediction

Abstract
Although state-of-the-art models used for numerical weather prediciton run at very high resolution, they still exhibit model error. One known source of model error are the effects of unresolved subgrid-processes onto the resolved flow. While conventional parameterizations represent the effects of the unresolved scales deterministically as function of the local state, stochastic parameterization allow for nonlocal subgrid-scale fluctuations.

This talk will give a brief overview on the stochastic parameterization efforts at ECMWF in the context of probabilistic weather foreasting. The first is aimed at injecting a fraction of dissipated energy into the resolved flow ("stochastic backscatter")and uses a cellular automaton as nonlocal pattern-generator. The second tries to improve the representation of tropical variability by introducing stochastic cloud-clusters on the supergrid scale.

Back to agenda