Coordinate Attention-Temporal Convolutional Network for Magnetotelluric Data Processing

Magnetotelluric (MT) has significant value in earthquake prediction, space weather monitoring, mineral resources exploration, and deep Earth structure detection. However, due to the complexity of the environment, MT data collected often are of low data quality due to noise pollution. The noisy data seriously affect the accuracy of the detection of underground structures. Therefore, we propose a magnetotelluric noise suppression method based on a coordinate attention-temporal convolutional network (CA-TCN). First, the CA-TCN is trained with a large dataset of artificially created data to learn the nonlinear mapping relationship between the noisy data and noise contours. Then, the CA-TCN model achieves the mapping transformation from noisy data to noise contours in the MT data. Finally, we subtract the noise contours obtained from the CA-TCN mapping model from the original noisy data to achieve signal-to-noise separation and reconstruct high-quality data. In simulated experiments, the similarity between denoised data and known high-quality data from Qinghai reaches 98%. The results demonstrate that the proposed method exhibits significant advantages compared to convolutional neural network (CNN) methods and so on. These findings validate the feasibility of the proposed approach. We applied the proposed method to the real measured data collected from the LuZong mining area, resulting in smoother and more continuous apparent resistivity curves. This indicates that noise in the MT data has been effectively removed, leading to a significant improvement in the quality of the MT data.