Continental-scale probabilistic resistivity imaging of Australia using deep learning: Implications for geology, groundwater, and critical minerals

Abstract

Electrical resistivity is a fundamental physical property that provides powerful constraints on subsurface structure, lithology, fluid distribution, and mineralization. We present a continental-scale electrical resistivity model of Australia's shallow crust with quantified uncertainty, extending to depths of up to ~650 m. The model is derived from probabilistic inversion of more than 26.6 million airborne electromagnetic (AEM) soundings acquired along 352,600 line-kilometers from the AusAEM program. We develop a deep learning-based probabilistic inversion framework using invertible neural networks (INNs). To accommodate differences among AEM surveys, we trained 11 networks using a common set of synthetic resistivity models and the same training strategy, enabling consistent integration of inversion results across Australia. The entire inversion was completed in only 4.37 GPU hours, overcoming long-standing computational barriers to continental-scale probabilistic geophysical imaging. The resulting resistivity model exhibits clear spatial patterns that closely align with Australia's major geological provinces, sedimentary basins, aquifer and mineralizing systems. We further define the depth of investigation using posterior resistivity distributions to support uncertainty-informed geological interpretation. The resistivity model together with quantified uncertainty provides new physical constraints on regolith thickness, basin architecture, paleochannel and mineralized systems across Australia, serving as a foundation for interpreting continental-scale electrical structure and its geological, hydrological, and mineral system implications.