Details
| Originalsprache | Englisch |
|---|---|
| Aufsatznummer | 103985 |
| Fachzeitschrift | Journal of applied geophysics |
| Jahrgang | 175 |
| Publikationsstatus | Veröffentlicht - Apr. 2020 |
| Extern publiziert | Ja |
Abstract
Due to multi-exponential decay-time properties of the subsurface volume units or layers, magnetic resonance sounding (MRS) relaxation data exhibit a multi-exponential behavior. MRS inverse problem in a multi-exponential modeling framework brings about a very large size of the parameter space which is computationally costly. In this paper, a fast and memory efficient inversion algorithm to retrieve the aquifer properties in terms of water content and relaxation time is presented. The original nonsymmetric linearized forward matrix is projected onto a Krylov subspace with smaller dimension using an iterative Golub-Kahan-Lanczos bidiagonalization (GKL) method. Because of ill-conditioning of the projected linearized forward matrix a regularized damped least squares equation is applied at each step of the GKL factorization method to extract the best possible approximation of the partial water content. Numerical experiments based on synthetic and field data demonstrate that the proposed inversion method provides a good estimation of the water content and relaxation time compared to the standard algorithm with computationally more efficient functionality.
ASJC Scopus Sachgebiete
- Erdkunde und Planetologie (insg.)
- Geophysik
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
in: Journal of applied geophysics, Jahrgang 175, 103985, 04.2020.
Publikation: Beitrag in Fachzeitschrift › Artikel › Forschung › Peer-Review
}
TY - JOUR
T1 - A fast multi-exponential inversion of magnetic resonance sounding using iterative Lanczos bidiagonalization algorithm
AU - Fallahsafari, Mahdi
AU - Ghanati, Reza
AU - Hafizi, Mohammad Kazem
AU - Müller-Petke, Mike
N1 - Funding information: We would like to thank Leibniz Institute for Applied Geophysics for providing us with the field data. The authors are grateful to the Institute of Geophysics, University of Tehran (UT) for all its support. The authors would like to acknowledge the financial support from UT for this research under grant number 6201010/1/07 . M.K. Hafizi acknowledges UT for the opportunity provided to him for sabbatical leave at Memorial University of Newfoundland. The authors are grateful to the editor and two ananymous reviewers for the insightful and useful comments and suggestions. We would like to thank Leibniz Institute for Applied Geophysics for providing us with the field data. The authors are grateful to the Institute of Geophysics, University of Tehran (UT) for all its support. The authors would like to acknowledge the financial support from UT for this research under grant number 6201010/1/07. M.K. Hafizi acknowledges UT for the opportunity provided to him for sabbatical leave at Memorial University of Newfoundland. The authors are grateful to the editor and two ananymous reviewers for the insightful and useful comments and suggestions.
PY - 2020/4
Y1 - 2020/4
N2 - Due to multi-exponential decay-time properties of the subsurface volume units or layers, magnetic resonance sounding (MRS) relaxation data exhibit a multi-exponential behavior. MRS inverse problem in a multi-exponential modeling framework brings about a very large size of the parameter space which is computationally costly. In this paper, a fast and memory efficient inversion algorithm to retrieve the aquifer properties in terms of water content and relaxation time is presented. The original nonsymmetric linearized forward matrix is projected onto a Krylov subspace with smaller dimension using an iterative Golub-Kahan-Lanczos bidiagonalization (GKL) method. Because of ill-conditioning of the projected linearized forward matrix a regularized damped least squares equation is applied at each step of the GKL factorization method to extract the best possible approximation of the partial water content. Numerical experiments based on synthetic and field data demonstrate that the proposed inversion method provides a good estimation of the water content and relaxation time compared to the standard algorithm with computationally more efficient functionality.
AB - Due to multi-exponential decay-time properties of the subsurface volume units or layers, magnetic resonance sounding (MRS) relaxation data exhibit a multi-exponential behavior. MRS inverse problem in a multi-exponential modeling framework brings about a very large size of the parameter space which is computationally costly. In this paper, a fast and memory efficient inversion algorithm to retrieve the aquifer properties in terms of water content and relaxation time is presented. The original nonsymmetric linearized forward matrix is projected onto a Krylov subspace with smaller dimension using an iterative Golub-Kahan-Lanczos bidiagonalization (GKL) method. Because of ill-conditioning of the projected linearized forward matrix a regularized damped least squares equation is applied at each step of the GKL factorization method to extract the best possible approximation of the partial water content. Numerical experiments based on synthetic and field data demonstrate that the proposed inversion method provides a good estimation of the water content and relaxation time compared to the standard algorithm with computationally more efficient functionality.
KW - Inverse problem
KW - Krylov subspace
KW - Lanczos bidiagonalization
KW - Magnetic resonance sounding
KW - Multi-exponential
UR - http://www.scopus.com/inward/record.url?scp=85081953019&partnerID=8YFLogxK
U2 - 10.1016/j.jappgeo.2020.103985
DO - 10.1016/j.jappgeo.2020.103985
M3 - Article
AN - SCOPUS:85081953019
VL - 175
JO - Journal of applied geophysics
JF - Journal of applied geophysics
SN - 0926-9851
M1 - 103985
ER -