Empirical Orthogonal Analysis-Based Modeling of Thermospheric Neutral Density

Thermospheric density (TD) variations have a significant impact on orbit determination, low-Earth satellites, and prediction. As a result, the modeling of TD variability is very important for space weather forecasting, satellite communication, and navigation. In this study, we constructed an empirical model of the TD variability at 450 km altitude using the empirical orthogonal function (EOF) analysis and nonlinear least-squares fitting method based on the TD observations obtained from the Swarm-C satellite in the period 2014–2020. The model can describe the TD variability as a function of latitude, longitude, universal time, solar and geomagnetic activities, and day of year. To validate the model, the EOF model outputs were compared to the TD observations obtained from the JB2008 and MSIS2.0 models, and its error was substantially smaller than that of the JB2008 and MSIS2.0 models. The RMSE between the TD observations and EOF models is 0.084. TD observations and JB2008 model outputs have an RMSE of 0.106, while TD observations and MSIS2.0 model outputs have an RMSE of 0.18. The model and observations differ at weak solar activity due to solar radiation pressure and aerodynamic drag modeling inaccuracies. For future work, we will apply machine learning techniques to multi-satellite observations of TD and investigate the dependence of thermospheric system on various physical forcing such as Sym-H and AE.