Enhancing Mass Density Estimation of Physical Models in the Thermosphere Using Data Assimilation

Mass density is an important variable for the near-Earth space environment and is essential forapplications such as precise orbit determination, satellite lifetime prediction, and re-entry forecasting.At present, the temporal and spatial coverage of mass density observations in the upper atmosphereis heterogeneous. For instance, mass density can be derived from accelerometers on-board ofsatellites. Although these densities have a temporal resolution of a few seconds, few satellites areequipped with accelerometers sufficiently precise enough for this purpose. Additionally, the massdensity can be derived from any tracked satellite based on its orbital decay. However, this approachhas a temporal resolution of several hours to days. Consequently, one relies on models to obtain themass density at any given time and location. Semi-empirical models are constructed by fitting a largedatabase of observations to mathematical functions. Given position, time, and solar activityrepresented by proxy indices, they return the average mass density for these inputs. They performbest for frequently observed conditions, and less so when evaluated for less frequently observedconditions such as geomagnetic storms. In contrast to empirical models, physical models forward aninitial state of the atmosphere based on first principles. They are limited by discretization, externalforcings, unaccounted subgrid processes, assumptions and simplifications. Consequently, they do notprovide better mass density estimates than empirical models. Both kind of models currently do notprovide sufficiently accurate mass density estimates. To enhance mass density simulations of physicalmodels we combine them with observations using data assimilation with the goal of outperformingempirical models in the thermosphere. We used the Parallel Data Assimilation Framework (PDAF) todevelop a version of the NCAR HAO Thermosphere Ionosphere Electrodynamics General CirculationModel (TIE-GCM) capable of sequential data assimilation. It uses an efficient ensemble-based Kalmanfilter. Data located on satellite trajectories or regular grids (e.g., obtained from empirical models) canbe assimilated. We use a two-week long test period, including a strong geomagnetic storm todemonstrate and test our assimilation strategies. First, we assimialte accelerometer-derived massdensities along the orbits of CHAMP and GRACE individually and simultaneously. As a localizationscheme is employed to avoid non-physical updates caused by spurious long-range correlations withinthe ensemble, only regions in the vicinity of the satellites are updated. Thus, without co-estimatingmodel dynamics this approach cannot achieve global enhancements. Secondly, to achieve a globalupdate of the mass density, we have developed a two-step approach. We calibrate a semi-empiricalmodel with CHAMP accelerometer-derived densities, evaluate it on a global grid, and assimilate it.Independent GRACE accelerometer derived densities are used for validation. The median difference ofthe accelerometer derived densities to the assimilation experiment is smaller than to the empiricalmodel. Thirdly, we want to exploit the coupling in the MIT and assimilate electron number densities toenhance mass density estimation. To this end, we investigate the assimilation of global 3D electrondensities from a semi-empirical model developed at TUM-DGFI that employs a B-spline representationand constraint optimization to guarantee realistic electron density profiles.