A Robust Three-Dimensional Inversion Scheme for Controlled-Source Audio-Frequency Magnetotelluric Data

Abstract Controlled-source audio-frequency magnetotelluric (CSAMT) methods are widely employed in geothermal, mineral, and groundwater exploration due to their capability to image subsurface resistivity distributions. The application of three-dimensional (3D) inversion techniques to process CSAMT field data and recover 3D resistivity models has become increasingly common. However, traditional 3D inversion schemes often exhibit low computational efficiency, poor convergence, and limited accuracy in large-scale applications, particularly when Dirichlet boundary conditions are used. To address these challenges, a robust inversion scheme is evaluated, which integrates a 3D nonlinear conjugate gradient (NLCG) framework with four key strategies—(ⅰ) short-offset acquisition, (ⅱ) a perfectly matched layer (PML)-based forward modelling approach, (ⅲ) a cooling strategy for the regularization factor, and (ⅳ) confinement of the inversion domain to high-sensitivity regions—to enhance model accuracy, numerical stability, and computational efficiency. Synthetic tests demonstrate that the proposed scheme achieves faster convergence and more reliable resistivity recovery, with improved resistivity values and anomaly boundaries compared to conventional Dirichlet-based methods. A field case study in the Yanqing geothermal area further illustrates its practical advantages, producing resistivity images that better align with borehole and geologic information, with significantly improved recovery of deep targets and reduced data misfit.