Instrument-Aware Closed-Loop Network for High-Resolution Electrical Resistivity Tomography

Abstract

Electrical resistivity tomography (ERT) inversion is fundamentally constrained by the nonlinear and ill-posed nature of the problem, as well as the intrinsic limitations of measurement instruments. Conventional inversion methods, relying primarily on smoothness constraints, often oversmooth subsurface structures and blur anomaly boundaries, especially under instrument-imposed data precision limits. To address these challenges, this study proposes an instrument-aware structure-aware closed-loop ERT network (SACL-ERTNet), which integrates a differentiable forward subnetwork with an inversion subnetwork to form a closed-loop architecture. By enforcing consistency between predicted and measured apparent-resistivity responses, the network incorporates observation-domain constraints that are directly related to the instrument’s measurement characteristics, reducing solution ambiguity while improving reconstruction fidelity. Additionally, an edge-supervision branch models structural boundaries, coupling global resistivity reconstruction with local edge-aware learning to enhance boundary delineation. Synthetic experiments show improved resistivity- and boundary-reconstruction metrics, while the field application demonstrates clearer structural continuity and favorable observation-domain consistency among the evaluated methods.