Reduced Order Probabilistic Emulation for Physics‐Based Thermosphere Models

The geospace environment is volatile and highly driven. Space weather has effects on Earth's magnetosphere that cause a dynamic enigmatic response in the thermosphere, particularly evolution of neutral mass density. Many models exist use space drivers to produce density response, but these are typically computationally expensive or inaccurate for certain conditions. In this work aims employ probabilistic machine learning (ML) method create an efficient surrogate Thermosphere Ionosphere Electrodynamics General Circulation Model (TIE-GCM), physics-based thermosphere model. Our leverages principal component analysis reduce dimensionality TIE-GCM recurrent neural networks model behavior much quicker than numerical newly developed reduced order emulator (ROPE) uses Long-Short Term Memory perform time-series forecasting state provide distributions future We show across available data, ROPE similar error previous linear approaches while improving storm-time modeling. also conduct satellite propagation study significant November 2003 storm which shows can capture position resulting from with < 5 km bias. Simultaneously, point estimates result biases 7 - 18 km.