<p>Deep learning (DL) methods have emerged as a powerful tool for the inversion of geophysical data. When applied to field data, these models often struggle without additional network fine-tuning. This is because they are built on the assumption that the statistical patterns in the training and test datasets are the same. To address this, we propose a DL-based inversion scheme for Radio Magnetotelluric data where the subsurface resistivity models are generated using Gaussian random fields (GRFs). The network’s generalization ability was tested on two distinct datasets, each with a significant distribution shift from the GRF used for training. One dataset consisted of a homogeneous background with various rectangular anomalous bodies. The second dataset, designed to be more challenging, represented a geologically realistic scenario with multiple layers and faults. After end-to-end training with the GRF dataset, the pre-trained network successfully identified board anomalies in the unseen data. Synthetic experiments confirmed that the diversity in the GRF dataset enhances generalization compared to a homogeneous background dataset. The network accurately recovered structures in the resistivity model and demonstrated robustness to noise, outperforming traditional gradient-based methods. Finally, the developed scheme is tested using exemplary field data from a waste site near Roorkee, India. The proposed scheme enhances generalization in a data-driven supervised learning framework, suggesting a promising direction for DL based inversion methods in Radio magnetotelluric data.</p>