TIEGCM_ROPE_PART_II_Density_Modeling

This repository focuses on time series modeling of reduced-order thermospheric states derived from a Convolutional Orthogonal Autoencoder (COAE). The COAE compresses high-dimensional TIE-GCM outputs into a compact 10-dimensional latent space, enabling efficient and scalable forecasting of thermospheric density. We implement and compare multiple sequence modeling approaches—including DMD, SINDy, LSTM, GRU, and Transformer—to learn the temporal evolution of these latent coefficients. To improve forecast accuracy and robustness, especially during geomagnetic storms, we incorporate storm-tuned models and various ensemble strategies, including equal weighting and meta-learning. The repository includes modular utilities for sequence generation, model evaluation, and forecasting, supporting reproducible research in thermospheric modeling and space weather applications.