Rift Related Structure Near the Keweenaw Peninsula from a Novel Trans-dimensional Bayesian Inversion of Gravity and Structural Data

Abstract:

Gravity surveys provide great insight into continental rift-related structures, from their deep crust signature to economically vital ore bodies concentrated by rift magmatism. Yet, absent other geophysical or geological constraints, solutions to the gravity problem are non-unique due to tradeoff in volume, density, and depth. Attempts to overcome this limitation commonly require simplifying assumptions of 2D geometry, user-directed searches, and model regularization, which makes it difficult to fully characterize uncertainty and tradeoff in density models. To address these shortcomings, we develop a trans-dimensional Bayesian method for inferring 3D multiscale density structure from gravity and structural data and apply it to structures related to the Keweenawan Rift.

We model subsurface density anomalies as tessellations of constant-density Voronoi polyhedra and compute the resulting gravity anomaly and its gradient. Based on the misfit with observed gravity and structural priors, the model is then iteratively updated utilizing the Metropolis-Hastings algorithm, yielding an ensemble of models. The number, densities, and locations of Voronoi nuclei are allowed to vary within the algorithm. We analyze the resulting ensemble for posterior distributions of depth, volume, and density contrast of anomalies, from which we can make inferences about the magmatism produced by rifting.

We present an application of our method to data compiled from the USGS to examine a large anomaly near the Porcupine Mountains. The igneous units produced by the rifting at the triple junction are near to the surface in this region, allowing for strong constraints on absolute density. Mining activity also provides a surfeit of prior information from structural and borehole data. We also present tests investigating the effect of structural priors and multi-altitude gravity observations on reducing model uncertainty.