Answers are grounded in this paper and cite their sources.
Computational study demonstrates accurate 2D subsurface resistivity imaging across complex terrain using automatic differentiation, indicating a flexible approach that eliminates manual gradient math.
This study presents a two-dimensional magnetotelluric (MT) inversion framework based on automatic differentiation (AD) and evaluates its applicability for models with complex topography. The forward problem is formulated in the frequency domain and solved using the finite difference method (FDM) with apparent resistivity and phase responses calculated for both transverse electric (TE) and transverse magnetic (TM) modes. The forward-modelling workflow is implemented in a PyTorch-based differentiable programming framework, allowing the gradients of the discrete objective function with respect to the conductivity to be obtained through backpropagation. In this framework, the objective function, model parameterization, and optimizer type can be trivially modified without the need to rederive or re-implement the gradient of the objective function or the optimization procedure, as in traditional inversions. Synthetic experiments for model with and without topography show that even with the addition of 5% Gaussian noise, the proposed method can accurately recover the structure for the synthetic models, and exhibits excellent resistance to noise and good adaptability to complex topographic conditions. The method is further validated using the COPROD2 data collected from Canada and recovers the main conductive structural features identified in previous studies. The main purpose of this study is to construct an inversion framework based on automatic differentiation, and verify that this method can reduce manual work in the gradient derivation.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: