New Publication: Implicit Neural Representation for 2D Magnetotelluric Inversion
New research by Yanyi Wang, Xavier Garcia, Eric Attias, Boyao Zhang, Fansheng Xiong, Jianxin Liu, and Zhenwei Guo is now available online in Computers & Geosciences. Building on the work presented at EGU 2026, the study develops a physics-constrained implicit neural representation (INR) approach for two-dimensional magnetotelluric (MT) inversion.
The method represents subsurface resistivity as a continuous neural field and compares its predictions with observed MT data using a Maxwell-equation-based forward model. This allows inversion without labelled training datasets, while the network itself provides implicit regularization.
Tests on synthetic models and the COPROD2 field dataset demonstrate the method's ability to recover complex resistivity structures. In the layered synthetic example, it improves model accuracy and structural similarity relative to MARE2DEM while achieving a comparable data fit.
Well done, Yanyi and team!
Read the paper: "Implicit neural representation for 2D magnetotelluric inversion: A physics-constrained deep learning approach."
Computers & Geosciences, 218, 106277. Published online 29 September 2026; January 2027 issue.
