Answers are grounded in this paper and cite their sources.
SUMMARY The direct current (DC) resistivity method is widely used for near-surface investigations due to its ability to characterize subsurface resistivity structures. Machine learning (ML) techniques have shown potential for achieving high-precision resistivity inversion by modeling complex nonlinear relationships. Nevertheless, their application to three-dimensional (3-D) resistivity problems is often constrained by the need for large training datasets and the high computational cost associated with network training. In practice, conventional deterministic inversion is the dominant approach for 3-D resistivity imaging due to its computational efficiency and robustness. However, the regularization term in deterministic inversion tends to smooth anomaly features, leading to reduced inversion resolution. To address these limitations, this study proposes a novel inversion framework that integrates conventional deterministic inversion with ML. First, a deterministic inversion method, the limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) algorithm, is applied to obtain a preliminary resistivity model. The density-based spatial clustering of applications with noise (DBSCAN) is then employed to identify the spatial extents of anomaly zones. By this means, high-quality training resistivity models that closely resemble the target subsurface resistivity structures are generated by varying resistivity values within the identified anomaly zones. Subsequently, a broad learning (BL) network is employed to establish the nonlinear mapping between apparent resistivity data and subsurface resistivity. Synthetic tests on Single-, Multiple-, Nested-, and Coupled-Model scenarios using only 1 500 high-quality training samples show that the proposed method reduces the mean absolute percentage error (MAPE) by 23–37 per cent and increases the structural similarity index measure (SSIM) by 122–171 per cent relative to conventional L-BFGS inversion. Field experiments further demonstrate reliable recovery of high- and low-resistivity anomalies. These results demonstrate that embedding deterministic inversion characteristics into the training samples improves inversion accuracy while substantially reducing reliance on large training datasets, providing an efficient approach for 3-D resistivity inversion with potential applications in near-surface surveys.
No takes yet. Share an insight, caveat, or question.
Tao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: