One-Dimensional Modeling of Magnetotelluric Data using Convolutional Neural Network-Gated Recurrent Unit Based Inversion and Its Implementation on Field Data

The magnetotelluric method is a geophysical method that utilizes natural variations in the electromagnetic field to map the resistivity distribution beneath the surface. In this method, inversion is the primary process used to estimate the resistivity structure from field data. This study proposes a deep learning-based approach for one-dimensional magnetotelluric inversion, combining Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) as an alternative inversion method. The dataset consists of 20 layers with resistivity and thickness values randomly selected within a specific range at 4000 meters depths and a probability value. Apparent resistivity and phase are obtained through forward modeling based on selected resistivity and thicknesses as input, while the resistivity structure is used as output, with a large data sample. The dataset was standardized and normalized using a logarithmic scale and the MinMax method to map values into the 0-1 range. The dataset was used to train the proposed CNN-GRU model, which is capable of mapping the resistivity distribution in the subsurface. The results show that the CNN-GRU model could map the resistivity distribution model and predict its thicknesses with small error based on the apparent resistivity and phase data, indicating that it can be used for one-dimensional inversion in magnetotellurics. Nevertheless, the model performed quite well on several field datasets, showing a good fit between predicted and true values.