Comparison of deep learning models for 1D magnetotelluric inversion

This paper presents a comparative study of three deep learning (DL) architectures for one-dimensional magnetotelluric (MT) data inversion: long short-term memory (LSTM), gated recurrent unit (GRU), and Informer models. We developed a comprehensive framework for generating realistic synthetic MT data, training the models, and evaluating their performance through multiple quantitative metrics. Our synthetic dataset comprised 100,000 samples with 25 periods spanning 10 −3 to 10 3 seconds, created using statistical parameters derived from real MT data. Each model was trained on apparent resistivity and phase responses to recover subsurface resistivity profiles. The results show that the recurrent neural network architectures (LSTM and GRU) slightly outperform the attention-based Informer model, with the LSTM achieving the best performance (MSE of 0.06455 Ω m 2 , R 2 of 0.45234). Despite their differing architectures, all the models successfully captured the major subsurface resistivity contrasts. When applied to real MT data from the UAE, the tested models showed promising results in terms of reconstructing subsurface structures. Overall, this study demonstrates the viability of DL approaches for MT inversion, with potential applications in efficient field-based subsurface imaging.