Abstract: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 fine-tuning of the network. 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 (GRF). The network's generalization ability was tested with an out-of-distribution (OOD) dataset comprising a homogeneous background and various rectangular-shaped anomalous bodies. After end-to-end training with the GRF dataset, the pre-trained network successfully identified anomalies in the OOD dataset. Synthetic experiments confirmed that the GRF dataset enhances generalization compared to a homogeneous background OOD dataset. The network accurately recovered structures in a checkerboard 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 OOD generalization in DL methods.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the generalization ability of deep - learning methods in processing geophysical data (especially radio - magnetotelluric sounding, RMT) inversion. Specifically, when these models are applied to actual data, additional fine - tuning of the network is often required because they are based on the assumption that the training and test data sets have the same statistical pattern. To overcome this challenge, the author proposes a deep - learning inversion scheme for generating underground resistivity models based on the Gaussian Random Field (GRF). ### Main problems and solutions: 1. **Insufficient generalization ability**: - **Problem**: Traditional deep - learning methods perform poorly when dealing with actual data because the distributions of training data and test data are inconsistent. - **Solution**: Enhance the diversity of training data by using GRF to generate diverse resistivity models, thereby improving the generalization ability of the model. 2. **Missing data processing**: - **Problem**: Missing data often exists in actual data, which will affect the performance of the model. - **Solution**: Introduce Compressed Sensing (CS) technology to reconstruct the original signal from sparse or incomplete data, so as to improve the model's ability to process incomplete data. ### Specific implementation methods: - **Generate diverse training data**: Use GRF to generate complex resistivity distributions to ensure the diversity and complexity of training data. - **Compressed sensing technology**: Reconstruct complete RMT data from partially missing data through the L1 - regularized least - squares condition and convex optimization strategy. - **Network architecture**: Adopt U - Net as an encoder - decoder network, combined with convolutional layers, batch - normalization layers and ReLU activation layers to handle the 2D RMT data inversion task. ### Experimental verification: - **Synthetic experiment**: It is verified through the synthetic data set that the data generated by GRF can significantly improve the generalization ability of the model, especially when dealing with out - of - distribution (OOD) samples. - **Noise robustness**: The performance of the model under different noise levels is tested. The results show that U - Net performs well under low - noise conditions, but its performance degrades under high - noise conditions. - **Compressed sensing experiment**: The effectiveness of the CS method in processing missing data is verified. The reconstructed data is very close to the original data, and its performance in the inversion task is also relatively consistent. ### Summary: This research effectively improves the generalization ability and the ability to process missing data of deep - learning models in RMT data inversion by introducing GRF to generate diverse training data and applying compressed sensing technology, providing new directions and methods for practical applications.

Enhancing Deep Learning based RMT Data Inversion using Gaussian Random Field

3D cooperative inversion of airborne magnetic and gravity gradient data using deep learning techniques

Two-dimensional magnetotelluric inversion method based on deep learning

Improving Deep Groundwater Aquifer Characterization with Deep Learning Inversion of Audio-Frequency Magnetotelluric Data

Physics-Informed Deep Learning Inversion with Application to Noisy Magnetotelluric Measurements

DMRF-UNet: A Two-Stage Deep Learning Scheme for GPR Data Inversion under Heterogeneous Soil Conditions

A Deep Learning Gravity Inversion Method Based on a Self-Constrained Network and Its Application

Two-stage Deep Learning for 3D Joint Inversion of Gravity and Magnetic

Solving Electromagnetic Inverse Problem Using Adaptive Gradient Descent Algorithm.

Estimating Three-Dimensional Resistivity Distribution with Magnetotelluric Data and a Deep Learning Algorithm

Deep learning electromagnetic inversion with convolutional neural networks

Inversion of Gravity Anomaly Based on AlexNet Deep Neural Network

Physics-Driven Deep Learning Inversion for Direct Current Resistivity Survey Data

One-dimensional deep learning inversion of electromagnetic induction data using convolutional neural network

3-D Magnetotelluric Deep Learning Inversion Guided by Pseudo-Physical Information

Physics-guided Deep Learning-Based Inversion for Airborne Electromagnetic Data

Deep Learning for 3-D Inversion of Gravity Data

Physics-guided deep-learning inversion method for the interpretation of noisy logging-while-drilling resistivity measurements

Joint Gravity and Magnetic Inversion Using CNNs' Deep Learning

Lévy Gradient Descent: Augmented Random Search for Geophysical Inverse Problems

3-D Joint Inversion of Gravity and Magnetic Data Based on a Deep Learning Network With Automatic Recognition of Structural Similarity