Parameterizing Model Uncertainty due to Unresolved Dynamics in Ensemble Forecasts (Invited Presentation)

Model uncertainty associated with unresolved dynamics should be represented in ensembles forecasts. Typicaly this is done either through the use of ad-hoc inflation techniques (in ensemble data assimilation), or through the use of stochastic parameterizations in the forecast model used to generate the background ensemble. In the current NOAA operational EnKF system, a stochastic kinetic energy backscatter scheme is used in the ensemble forecast step. This scheme, originally developed by Judith Berner and Glenn Shutts, uses a dissipation estimate to inject spatially and temporally correlated random wind perturbations in regions where energy is being dissipated to model the missing backscatter of energy from unresolved and heavily damped scales. An alternative approach to backscatter has recently been proposed by V. Resseguier and colleagues, which models the uncertainty due to unresolved dynamics in terms of location uncertainty in advection. Random spatially and temporally correlated perturbations are added to the wind field used to compute the transport terms in the model dynamics. No explicit dissipation estimate is needed, and the advection by the random component of the wind field acts as a multiplicative stochastic forcing. In this talk I will compare the impact of this new randomized transport scheme in ensemble forecasts relative to the more familiar stochastic energy backscatter approach. Results with both a simple quasi-geostrophic turbulence model and a the new FV3-based version of the NOAA operational global forecast model will be presented.