Summary Gravity inversion is an essential technique for recovering subsurface density variations. Constructing stratigraphic models from gravitational observations, however, remains challenging because gravity data provide limited vertical resolution and the inverse problem is strongly non-unique. To address these limitations, we develop a Bayesian geometry-based gravity inversion framework aimed specifically for stratigraphic reconstruction. Stratigraphic interfaces are parameterized using a Fourier series, which provides a compact set of variables, enables a controllable representation of layer geometry, and improves the vertical resolution of the recovered density model. Uncertainty in the inferred stratigraphy is quantified with Stein variational inference, yielding an ensemble approximation to the posterior distribution. The resulting posterior models reveal the confidence level and spatial variability of stratigraphic layers. The proposed method has been validated using two toy examples to illustrate the main concepts. Two synthetic lunar basin models, representing alternative interpretive hypotheses for upper-crustal layering, are further designed to evaluate algorithm performance. Finally, application to satellite gravity observations from the Orientale Basin demonstrates that the proposed framework can recover the basin’s large-scale tectonic structure.
Zhang et al. (Fri,) studied this question.