An efficient decoupled 3-D axial anisotropic resistivity inversion for magnetotelluric data with OpenMP parallelization
Issued Date
2025-12-01
Resource Type
ISSN
13438832
eISSN
18805981
Scopus ID
2-s2.0-85218453389
Journal Title
Earth, Planets and Space
Volume
77
Issue
1
Rights Holder(s)
SCOPUS
Bibliographic Citation
Earth, Planets and Space Vol.77 No.1 (2025)
Suggested Citation
Thongyoy W., Siripunvaraporn W., Amatayakul P., Rung-Arunwan T., Baba K. An efficient decoupled 3-D axial anisotropic resistivity inversion for magnetotelluric data with OpenMP parallelization. Earth, Planets and Space Vol.77 No.1 (2025). doi:10.1186/s40623-024-02113-5 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/105502
Title
An efficient decoupled 3-D axial anisotropic resistivity inversion for magnetotelluric data with OpenMP parallelization
Corresponding Author(s)
Other Contributor(s)
Abstract
We have developed two novel axial anisotropic inversion codes for magnetotelluric (MT) data: a full axial inversion and a decoupled axial inversion. Both codes are based on the data space Gauss–Newton inversion method used in the WSINV3DMT code, enhanced with OpenMP parallelization and Intel MKL PARDISO direct solver for improved computational efficiency. The full axial inversion searches for three axial anisotropic resistivity elements ρxx, ρyy, and ρzz. Based on our prior influence studies, we identified contrasting influence patterns: ρxx has a strong influence on Zxy and Zyy, whereas ρyy has a strong influence on Zyx and Zxx, while ρzz has negligible impact on all responses. We, therefore, proposed a decoupled axial inversion. This novel technique splits the full axial inversion into two independent modes: ρxx- and ρyy-modes. The ρxx-mode requires just the Zxy and Zyy responses to recover ρxx, while the ρyy-mode requires the other two responses. By splitting the inversion, each mode requires half the data parameters and reduces model parameters to one-third compared to the full axial inversion. The substantial parameter reduction is the key factor leading to significantly faster processing times and lower memory requirements. Both codes were validated with the synthetic model. Utilizing OpenMP and direct solvers, the full axial inversion processed 256 MT stations in 65 min per iteration or a total of 195 min for three iterations to converge. The decoupled inversion achieved significantly faster processing, requiring just 32 min per iteration or a total of 96 min to finish the inversion. In addition, the decoupled inversion requires just a fraction of the memory used in the full axial inversion. This makes it practical to operate even on the standard personal computers of current technology. We further applied the decoupled inversion to real MT data acquired in the Northwestern Pacific Ocean. The resulting 3-D inverted ρxx and ρyy models support the existence of the anisotropy occurring at 60–120 km depth beneath the ocean floor, in agreement with previous studies.