Earth’s upper atmosphere—including the ionosphere and thermosphere—is influenced by both weather patterns from below and space weather from above. These layers are shaped by complex interactions between different types of winds, waves, and electric currents. Sudden stratospheric warmings, geomagnetic storms, and tropospheric convection can all trigger large changes in the upper atmosphere, but understanding how these changes ripple across different variables remains a challenge.

In this work, we apply a new machine learning approach to learn how one atmospheric variable can be used to predict another. For example, we use data from the WACCM-X climate model to train our system to estimate meridional (north-south) winds and vertical motion based on zonal (east-west) wind patterns. This method captures complex, nonlinear relationships that traditional models often miss.

Our model performs well during dynamic events, revealing how disturbances from the lower or upper atmosphere affect winds and circulation patterns at different altitudes. This approach can help researchers explore how energy and momentum move through the whole atmosphere and complements both observational data and physics-based simulations. Ultimately, it offers a new tool for improving our understanding of atmospheric coupling and space weather impacts.