Ensemble forecasting and data assimilation with stochastic physics

Jeff Whitaker


All data assimilation systems that use ensemble forecasts to estimate background-error covariances require some method for representing missing sources of uncertainty, such as model error. The current NCEP operational hybrid 3d-ensemble Var system uses inflation (both multiplicative and additive). In this talk I will discuss the possibility of replacing the additive component with a stochastic representation of model uncertainty included in the GFS model itself. Several different schemes, including vorticity confinement, stochastic kinetic energy backscatter, perturbed boundary-layer humidity and stochastically-perturbed physics tendencies have been tested, both in the context of the 3d-ensemble Var data assimilation at reduced resolution, and in medium-range ensemble forecasting at the current operational (T254L64) resolution.