From Jupiter to Saturn: Characterizing Interior Structures with Machine Learning

Understanding the interiors of both Jupiter and Saturn is essential for building a consistent picture of giant planet formation and evolution. While the two planets share many similarities, each provides unique observational windows into its internal structure: Jupiter through atmospheric abundances measured by the Galileo entry probe and the Juno mission, and Saturn through oscillation modes detected via ring seismology. In both cases, high-precision gravity measurements, by Juno for Jupiter and Cassini for Saturn, offer strong constraints on interior models. However, despite their accuracy, these measurements cannot uniquely determine the internal structure, given the complexity and variability of possible structural configurations.

To address this, we develop a unified modeling framework that combines NeuralCMS, a deep neural network trained on interior models computed with the concentric Maclaurin spheroid (CMS) method, with a self-consistent wind model. This approach enables efficient exploration of a wide parameter space of Jupiter interior models without relying on prior assumptions. Using clustering analysis on the multidimensional model space, we identify four key classes of interior structures, characterized by differences in core configuration and envelope properties. We also show that Jupiter’s structure can be effectively described using only two key parameters, significantly reducing the complexity of the problem.

We then extend this approach to model Saturn’s interior, enabling a systematic and meaningful comparison between the two planets within a shared framework. The comparative analysis provides a broader perspective on the diversity of giant planet interiors and the processes that shape them. This work demonstrates the value of unified, data-driven modeling approaches in advancing our understanding of giant planet interiors across the Solar System.