Title | Preliminary development of a pseudo-3D MT inversion using deep learning and its application to the Mount Meager geothermal area, British Columbia |
| |
Author | Liu, X ; Craven, J
A; Tschirhart, V ; Hanneson, C; Unsworth, M; Grasby, S E |
Source | Geoconvention 2022, abstracts; 2022 p. 1-5 Open
Access |
Links | Online - En ligne
|
Image |  |
Year | 2022 |
Alt Series | Natural Resources Canada, Contribution Series 20210702 |
Publisher | Geoconvention Partnership |
Meeting | GeoConvention 2022; Calgary, AB; CA; June 20-22, 2022 |
Document | Web site |
Lang. | English |
Media | paper; on-line; digital |
File format | pdf |
Province | British Columbia |
NTS | 92J/05; 92J/06; 92J/11; 92J/12; 92J/13; 92J/14 |
Area | Mount Meager |
Lat/Long WENS | -123.7442 -123.2817 50.7517 50.4839 |
Subjects | geophysics; Science and Technology; Nature and Environment; energy resources; geothermal energy; geophysical surveys; magnetotelluric surveys; geophysical interpretations; magnetotelluric
interpretations; resistivity; modelling; volcanism; volcanoes; magmas; Mount Meager Volcanic Complex; Artificial intelligence; Methodology |
Illustrations | flow diagrams; 3-D models; plots; location maps; geophysical profiles |
Program | Energy Geoscience Program Coordination |
Released | 2022 06 01 |
Abstract | (Summary) The inversion of magnetotelluric (MT) data is the process used to determine the subsurface resistivity structure from a set of surface observations. The computation time for running
3D MT inversions is very long, and even running with high performance computing (HPC). In recent years, an alternative technique utilizing artificial neural networks has emerged to determine subsurface physical properties from surface geophysical
data, which improved inversion efficiency. Building on the previous research, we propose a pseudo-3D inversion algorithm within a convolutional neural network (CNN) framework to generate a sparse subsurface resistivity distribution model from MT
response in less time than conventional geophysical inversion. The sensitivity of the MT response to a conductor at different locations was compared, and showed the response diminishes with distance, implying the algorithm need only solve for model
cells sufficiently close to MT sites. The validated neural network model was tested for reliability with a single conductor model. The application to a real world MT data set from the Mount Meager volcanic area displays consistent profiles and
conductor locations, which can be related to a magma body. |
Summary | (Plain Language Summary, not published) The magnetotelluric (MT) method is a natural source electromagnetic geophysical survey technique for the exploration of natural resources. The
computation time for running 3-Dimensional MT inversions is huge. In this study, an alternative technique utilizing artificial neural networks has employed to predict subsurface physical properties. Building on the previous research, we propose a
pseudo-3D inversion algorithm within a convolutional neural network (CNN) framework to generate a sparse subsurface resistivity distribution model from MT response in less time than conventional geophysical inversion. The validated network model was
tested for reliability with a single conductor synthetic model. The application to a real world MT data set from the Mount Meager volcanic area displays consistent profiles and conductor locations, which can be related to a magma body. |
GEOSCAN ID | 329724 |
|
|