Enhancing Thermospheric Density Modeling Using Bias Factor Analysis

Accurate thermospheric density modeling is critical for predicting atmospheric drag and satellite orbit evolution. Empirical models often show biases under different solar and geomagnetic conditions. In this study, we developed TT-NRL, a hybrid temporal convolu tional network(TCN)-Transformer framework designed to calibrate the NRLMSISE-00 using satellite orbit data. The model is trained on CHAMP accelerometer-derived densities combined with space weather and positional parameters, using the density ratio between observations and model outputs as the target. Evaluations across annual, monthly, and daily scales demonstrate consistent error reduc-tions compared with NRLMSISE-00 during both quiet and storm-time conditions. TT-NRL provides balanced improvements in both short and long-term scenarios, establishing it as a reliable framework for enhancing empirical thermospheric density models. 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining. Al training, and similar technologies.