Comparing Jupiter and Saturn: Similar Envelopes and Different Core Structures Revealed by Machine Learning

Jupiter and Saturn provide key constraints on the internal structure of giant planets through precise gravity measurements obtained by the Juno and Cassini missions. These data are complemented by additional observations, including in situ atmospheric measurements for Jupiter and ring seismology for Saturn. Despite these advances, inferring their internal structures from the available constraints remains a challenging problem.

In this work, we present a unified approach that combines accurate interior modeling with a machine-learning surrogate, enabling efficient exploration of a broad range of planetary structures under consistent assumptions for both gas giants. The models include a self-consistent treatment of zonal flows, linking the observed atmospheric dynamics to the deep interior through their gravitational signatures.

From large model ensembles, a small number of characteristic interior structures emerge for each planet. In Saturn, uncertainty in atmospheric helium abundance permits a wide subset of these models, but the inclusion of ring seismology constraints reduces the solutions to a single preferred structure. For Jupiter, atmospheric composition and the temperature at 1 bar provide comparable constraints. Overall, the combined observational constraints reduce the solutions to one characteristic structure for each planet, which share similar underlying architectures but differ in the structures that best match the data. We find that both planets favor low-metallicity envelopes, with Jupiter having a smaller compact core than Saturn.

These results demonstrate the power of combining machine learning with physically consistent modeling to efficiently explore complex parameter spaces and highlight the importance of integrating diverse observations to characterize and compare giant-planet interiors.