Neural Network based calibration of atmospheric density models

Atmospheric density is the most important factor for accurate estimation of the drag force exerted on spacecraft at Low Earth orbits. Empirical models provide the most accurate estimation of the density currently available, although they still suffer from estimation errors. This work presents a novel approach based on Neural Networks for reducing the error in the density estimated by empirical models, along the orbit of a spacecraft. The Neural Networks take as inputs the density estimated by DTM-2013, NRLMSISE-00 and JB2008, three of the latest empirical atmospheric models available. Density estimated from the accelerometers of the CHAMP and GRACE missions are used as targets for the training, validation and testing of the Neural Networks. In addition, this work studies the use of the spacecraft's average speed as an input to the Neural Networks. The test results indicate that the Neural Networks produce density estimates with less error than the density from the three empirical models studied.