Improving the Ensemble Representation of Model Uncertainty for Coupled Land–Atmosphere Data Assimilation

This presentation reviews progress towards developing a coupled land/atmosphere data assimilation system for NWP, focusing on the necessary advances in estimating land model background uncertainty. This coupled land/atmosphere DA uses NOAA's GSI Hybrid EnVar’s EnKF to assimilate land observations, including estimating the land model background uncertainty directly from the GSI atmospheric ensemble. The land surface (and near-surface) spread in the GSI ensemble, and in most other atmospheric ensemble systems, is under-dispersive. Initial tests showed that directly applying Gaussian perturbations to the land states, as is done in many offline land EnKF schemes, produces ensembles with biased soil moisture and surface fluxes. Instead, we apply a stochastic perturbation of physics tendencies (SPPT) scheme to perturb the soil moisture and temperature tendencies, which produces relatively unbiased ensembles. Also, typical offline land EnKF schemes force all ensemble members with the same atmospheric realization, while in the GSI ensemble each land member interacts with a unique atmospheric realization, giving more realistic ‘errors of the day’. Based on independent estimates of model uncertainty, compared to the current GSI, or to a typical offline land EnKF, the GSI with land SPPT produces ensembles that better represent model uncertainty in soil moisture, and screen-level temperature and specific humidity.