Large-scale 3D Inversion of Airborne Potential Field Data

Abstract

3D inversion of gravity and/or magnetic data is a difficult problem for a number of reasons. Firstly, airborne surveys are characterized by very large data volumes. Second, modelling large-scale surveys is computationally challenging. Finally, potential field data are finite and noisy; their inversion is an ill-posed problem, meaning that regularization must be introduced to recover the most geologically plausible solutions from the infinite number of numerically equivalent solutions. In this paper, we discuss these difficulties. We show that the key to practical large-scale 3D inverse problems is with a moving footprint. We exploit the fact that a potential field’s footprint is significantly smaller than the area of an airborne survey. We implement this in a re-weighted regularized conjugate gradient method for minimizing the objective functional, and have incorporated a wide variety of regularization options. We have developed our 3D potential field inversion in fully parallelized software such that we avoid storage of the sensitivity matrix. As such, we are effectively unlimited in the size of inversion we can perform. We demonstrate this with a case study for the 3D inversion of 1700 line km of both FALCON® and TMI data over Timmins to an earth model with over 128 million cells.