Integrating Deep Learning and Discrete Cosine Transform for DC-Resistivity Inversion: Application to Synthetic and Real Data

Electrical resistivity tomography (ERT) is a geophysical method crucial for subsurface imaging, offering key insights into fluid saturation and porosity. Large-scale ERT campaigns can pose logistical and time challenges related to data acquisition and processing. The ERT inverse problem is nonlinear and ill-posed, and it is usually solved through deterministic gradient-based algorithms. In this work, we develop a novel DC resistivity inversion approach that integrates a deep-learning framework with Discrete Cosine Transform (DCT). Furthermore, we incorporated a high-quality dataset featuring realistic resistivity models derived from real data. We first apply our approach to synthetic data not previously used during the training stage. Then, we demonstrate the applicability of the method through inverting real data. Our results show very good model prediction performance associated with low data misfit when applying the network to both synthetic and real data cases. Our inversion proposal provides an opportunity to obtain real-time resistivity models with minimal data misfit, thereby influencing decision-making during the acquisition stage and substantially reducing ERT-related costs for both acquisition and processing in large campaigns.