Middle Atmosphere Operational Data Assimilation with the Use of Ensembles

David Kuhl* and Karl Hoppel, Douglas R. Allen, John McCormack, Jun Ma, Steve Eckermann, Nancy Baker, Elizabeth Satterfield, Sergey Frolov and Craig H. Bishop
NRL-DC, NRL-DC, NRL-DC, NRL-DC, Computational Physics Inc, NRL-DC, NRL-Monterey, NRL-Monterey, University Corp. for Atmos. Research and NRL-Monterey

Proper modeling of the middle atmosphere (stratosphere and mesosphere) is extremely important for seasonal forecasting (e.g., deep troposphere-stratosphere teleconnections critical to Arctic prediction) as well as for providing proper lower boundary conditions for driving space weather models. The upper stratosphere and mesosphere (~50-100km) exhibit fast decorrelation times due to energetic divergent resolved gravity waves and short radiative relaxation times, in sharp contrast to the lower stratosphere where motion is dominated by “balanced” quasi-nondivergent vortical dynamics and long radiative relaxation times. The large-scale dynamics of the mesosphere, which are dominated by a rich spectrum of solar tides and planetary waves propagating upwards from the troposphere and stratosphere, are impacted locally by physical processes peculiar to this region, such as non-orographic gravity-wave drag, shortwave oxygen and ozone heating, exothermic chemical heating and infrared radiative cooling under conditions of nonlocal thermodynamic equilibrium.

Over the past several years, a scientific version of the U.S. Navy’s operational model NAVGEM (Navy Global Environmental Model) has been extended to cover the middle atmosphere region through updates to the physics and chemistry of the model. We will present results on extending the operational data assimilation into the middle atmosphere. Four systems are being tested and evaluated, 1) the current operational 4D-Var, 2) the soon to be operational hybrid 4D-Var, 3) an Ensemble-Var, and 4) an experimental Local Ensemble Tangent Linear Model (LETLM) 4D-Var system. Two specific aspects of the current operational 4D-Var algorithm appear to limit accuracy in the middle atmosphere: firstly, the background forecast error covariance is poorly specified for the middle atmosphere and secondly, 4D-Var involves a linear approximation to the forecast model to propagate perturbations to the background forecast, known as a tangent-linear model (TLM). The three new DA systems to a lesser or greater degree address these limitations and we will evaluate their effectiveness.

*email: david.kuhl@nrl.navy.mil
*Preference: Oral