.png){kind=icon}
.png){kind=icon}
Return
Efficient Transient Electromagnetic Inversion With Automatic Differentiation
.png){kind=icon}
PRE
AI
DOI：10.1109/TGRS.2026.3665936.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The transient electromagnetic (TEM) method is recognized as a highly efficient geophysical prospecting technique for rapid acquisition of subsurface resistivity distribution. The conventional gradient-based iterative inversion for TEM usually adopts finite difference (FD) to compute the Jacobian matrix to minimize the loss function and update the model parameters. However, when inverting large datasets for complex geological models, FD-based Jacobian computations not only could compromise gradient accuracy but also incur substantial computational costs. To address the demand for rapid inversion of field TEM data, we propose ADABFM—a novel TEM inversion method that integrates automatic differentiation (AD) with adaptive Born forward mapping (ABFM). Leveraging the gradient-calculation capability of AD, ADABFM can use essentially error-free gradients to update model parameters and obtain highly accurate resistivity models. Furthermore, ADABFM embeds system response characteristics directly into the forward modeling kernel functions. This design eliminates the complexities associated with explicit system response modeling and enhances the robustness of inversion. To validate the new method, we first conduct systematic tests with synthetic datasets. Then, we apply ADABFM to two distinct field datasets: 1) the open-source tTEM20AAR dataset collected in the Aare Valley, Switzerland and 2) a dataset acquired by our proprietary TEM system for the Quaternary overburden in Urumqi, China. For the tTEM20AAR dataset, the results obtained by ADABFM are essentially consistent with those obtained by the conventional FD-based iterative method with spatial constraints, while revealing more detailed subsurface features. For the Urumqi dataset, the inversion results also demonstrate the satisfactory consistency with the direct borehole measurements. Benefiting from the improvement in computational efficiency, ADABFM enables the rapid inversion from multiple initial models, which can effectively alleviate the local minima issue inherent in iterative inversion. The ADABFM method proposed in this study should provide an accurate, efficient, and robust solution for rapid TEM inversion.
Keywords:
Adaptive Born forward mapping (ABFM)
automatic differentiation (AD)
gradient computation
transient electromagnetic (TEM) inversion
Journal
IF:
8.6
Papers:
2.1W
Citations:
10.7W
Organization
View more organizations