Retrieval of Subsurface Resistivity from Magnetotelluric Data Using a Deep-Learning-Based Inversion Technique

Inversion is a fundamental step in magnetotelluric (MT) data routine analysis to retrieve subsurface geoelectrical model that can be used inform geological interpretations. To reduce the effect of non-uniqueness and local minimum trapping problems improve calculation speeds, data-driven mathematical method with deep neural network was developed estimate resistivity. In this study, learning (DL) inversion technique using revised multi-head convolutional (CNN) architecture investigated for MT analysis. We created synthetic datasets consisting 100,000 random samples resistivity layers train network’s parameters. The trained validated independent noised datasets, predicted results displayed reasonable accuracy reliability, which demonstrates potential application DL real-world data. analyze collected southwestern Athabasca Basin, Canada. calculated from detailed distribution compared traditional iterative inversion. Since approach predict without multiple forward modeling operations after CNN created, framework suitable speed up computation multidimensional