Introducing geoMCMC: A General-Purpose Bayesian Inference Framework for Planetary Interiors and its Application to Ganymede and the Juice mission

Abstract

In planetary geophysics, the translation of geodetic and geophysical measurements into interior properties represents a complex inverse problem (see e.g. [1-2]). This challenge is inherently characterized by non-uniqueness, mathematically correlated parameters, and the necessity for rigorous uncertainty quantification. With upcoming flagship missions—such as ESA’s Jupiter Icy Moons Explorer (Juice) [3] and NASA’s Europa Clipper [4]—preparing to deliver high-precision datasets from the Jovian system, the planetary science community requires advanced computational frameworks to interpret these observations. Historically, MCMC Bayesian inversions for constraining planetary interiors have relied on ad-hoc workflows, hard-coded parameterizations, and target-specific libraries (see e.g. [1-2]). Such an approach can be limiting, as it restricts cross-disciplinary reproducibility and scalability and a systematic comparison of analyses between different target bodies.

To overcome these limitations, we present geoMCMC, an open-source Python library developed to provide a unified, consistent, and modular computational framework for planetary interior analysis. By standardizing the inversion workflow, this library aims to enhance model sharing, cross-disciplinary collaboration, and methodological reproducibility across the planetary science community. Originally conceptualized to investigate the subsurface oceans of icy satellites [5], geoMCMC has evolved into a generalized, high-performance computational framework designed for the Bayesian inversion of both rocky and icy planetary bodies.

At its core, the software relies on a plugin-based architecture orchestrated by a central inference problem pattern. As illustrated in Fig. 1, this architecture explicitly decouples statistical inference from physical modeling through a five-pillar design: the Body (managing planetary layer states and geometry), the Parameter Set (linking statistical priors to interior attributes), the Observation Set (handling geophysical forward models), the Constraint Set (enforcing physical assumptions), and the Sampler (the Bayesian inference engine). This modularity ensures that state definitions, geometry, model constraints, and sampling algorithms function as independent, interchangeable components. For instance, users can seamlessly swap constraint modules, i.e. the functional blocks responsible for enforcing fundamental a priori physical assumptions, such as mass balance, hydrostatic equilibrium, or advanced thermodynamic models. In fact, these modules automatically resolve interdependent properties to guarantee physical consistency before any forward model is evaluated. Furthermore, the architecture is designed for high-performance parallel processing, allowing computationally intensive physical models to be offloaded to a high-efficiency compiled C++ kernel.