Quantifying Model Misrepresentation in Geophysical Inversion for Critical Mineral Exploration

Geophysical inversion often leans on simplified geological models due to a lack of detailed geological information in greenfield critical mineral exploration. While the actual orebodies are more complex in geometry, the simplified model representation can introduce large uncertainties, thereby adversely affecting decisions for field development. We introduce a stochastic quantification and treatment of the model misrepresentation errors when inverting geophysical data. To test whether misrepresentation errors exist, we provide a generalized hypothesis testing approach. We start with the assumption that our current geometry model accurately represents the subsurface (null hypothesis). We then use geophysical data to test this assumption. If the null hypothesis is rejected, the geometry is known to be false. We employ a Bayes factor to quantify where in space the model error is most significant. The latter will be used as information to reparametrize the assumed geometry, to reduce the misrepresentation errors. Our approach also allows accounting for data error and model error jointly, by using a pushforward formulation of the inverse problem. We provide a sequential Monte Carlo (SMC) sampling algorithm to solve the pushforward formulation, enabling practical computation of the Bayes factor. Both the synthetic and real field studies showed the Bayes factor can locate where the model is misrepresented. The real field application also shows that, through mitigating the misrepresentation errors, we reduced the uncertainty in key decision-making parameters concerning the field development, including the prospecting conductors’ volume and depth. We further provide open-source code to facilitate practical applications.