Super-resolution magnetotelluric data inversion with seismic texture constraint

ABSTRACT Magnetotelluric (MT) data inversion reconstructs the subsurface resistivity structure using measured natural electromagnetic (EM) fields. Due to the attenuation of low-frequency EM waves in geologic conductors, MT inversion has lower sensitivity and resolution for subsurface structures compared with seismic reflection imaging. Seismic reflection provides rich, detailed subsurface information that complements EM data, such as the spatial locations of geologic boundaries and fine-scale geophysical attributes. This work develops a super-resolution MT inversion approach incorporating the seismic texture constraint. We develop a multiresolution model parameterization scheme to represent the unknown model with the coarse-scale resistivity background and fine-scale resistivity details. A modified, pretrained VGG19 neural network (NN) serves as an effective texture extraction operator. This inversion method requires no NN training and simultaneously optimizes MT data misfit, texture discrepancy, and various regularization penalties using the adaptive moment estimation optimizer. Experiments show that our inversion can reduce or at least maintain the data misfit and effectively improve MT inversion resolution to a seismic-like level. This approach has promising application potential in the oil industry for highlighting quasi-layered subsurface structures.