Enhancing Numerical Simulation of Mass Density in Earth's Upper Atmosphere using Data Assimilation

The atmosphere's mass density is variable in space and time and directly proportional to atmospheric drag, which decelerates all objects in the atmosphere. Thus, the mass density should be specified with high accuracy and precision for applications depending on atmospheric drag acceleration, such as precise orbit determination, satellite lifetime assessment, and satellite re-entry prediction. The lower a satellite's orbit, the larger the atmospheric drag. Thus, atmospheric drag is especially of concern for low-Earth orbiting satellites. The mass density is not directly observed along satellite orbits but is simulated by physics-based numerical or empirical models. Numerical models providing the mass density suffer from simplifications, assumptions, discretization, uncertain parameters, idealized external forcings with limited temporal resolution, and unrealistic boundary conditions. Thus, the mass density predictions of numerical models show significant differences compared to other models and observations. Data assimilation is the combination of observations and models, taking into account their uncertainties. Several studies demonstrated that data assimilation enhances the prediction skills of numerical atmosphere models. However, data assimilation experiments require significant computational resources and typically cover only periods lasting a few days. In addition, the uncertainty of the model forecasts is tailored to the specific conditions of the assimilation experiment and is not transferable to other periods. Moreover, spurious correlations in the model covariances often require localization, which limits the improvements of the models to the vicinity of the sparse observations. Here, I implement a new assimilation system for the Thermosphere Ionosphere Electrodynamics General Circulation Model using the Parallel Data Assimilation Framework to address those limitations. Time-variable perturbations of the model inputs allow a realistic representation of the model's uncertainty. They reduce spurious ...