A challenge in Earth and space sciences is the capability to continuously monitor the key parameters of space systems that frequently experience complex, multiscale dynamics. The ionosphere-thermosphere (I-T) system is a region where the direct measurement of essential quantities such as electric fields, neutral winds, and temperatures remain largely inaccessible. Here, we present a generalizable inverse modeling framework that bridges sparse observations with physics-based simulation to recover these hidden state variables. By leveraging high-resolution, data-driven reconstructions of electron density in combination with the Thermosphere-Ionosphere-Electrodynamics General Circulation Model (TIEGCM), our approach enables end-to-end inference of latent I-T parameters without relying on explicit forward models. Through supervised neural networks trained on large-scale simulation data, the system autonomously learns the intrinsic couplings among I-T variables, enabling fast, robust, and physically informed parameter estimation. This fusion of physical modeling and machine learning transcends traditional retrieval methods, offering a scalable pathway toward real-time geospace monitoring and improved representation of upper-atmosphere dynamics in Earth system models.
Chen et al. (Fri,) studied this question.