Small bodies global gravity inversion via the level-set method

We propose an approach to infer large-scale heterogeneities within a small celestial body from measurements of its gravitational potential, provided for instance by spacecraft radio -tracking. The non -uniqueness of the gravity inversion is here mitigated by limiting the solutions to piecewise-constant density distributions, thus composed of multiple regions of uniform density (mass anomalies) dispersed in a background medium. The boundary of each anomaly is defined implicitly as the 0 -level surface of a scalar field (called the level -set function), so that by modifying this field the shape and location of the anomaly are varied. The gravitational potential associated with a density distribution is here computed via a line -integral polyhedron method, yielding the coefficients of its spherical harmonics expansion. The density distribution is then adjusted via an iterative least -squares approach with Tikhonov regularization, estimating at every iteration corrections to the level -set function, the density contrast of each anomaly, and the background density, in order to minimize the residuals between the predicted gravity coefficients and those measured. Given the non -convexity of the problem and the lack of prior knowledge assumed (save for the shape of the body), the estimation process is repeated for several random initial distributions, and the resulting solutions are clustered based on global properties independent of the input measurements. This provides families of candidate interior models in agreement with the data, and the spread of the local density values across each family is used to assess the uncertainties associated with the estimated distributions. We present multiple synthetic tests with increasingly more realistic settings (in terms of gravity resolution and precision, and of shape, size and distribution of the internal heterogeneities), showing that our method is generally able to retrieve a ground -truth mass distribution even with noisy data. For further validation, we present an application of the method to real data, namely the Bennu gravity coefficients measured by the OSIRIS-REx team.