Abstract: The extremely low frequency electromagnetic network records electromagnetic signals all the time, and automatically processes them into magnetotelluric impedance information, monitoring the changes in underground resistivity and electromagnetic signals. The inaccurate impedance estimation caused by electromagnetic noise hinders the normal operation of the system. Magnetotelluric sounding is one of the important methods for deep exploration of the earth. For a long time, magnetotelluric has been widely criticized for its low resolution and strong inaccuracy. With the development of magnetotelluric impedance tensor analysis technology and three-dimensional inversion technology, the resolving power of magnetotelluric has been greatly improved. However, the inaccuracy of the inversion results due to errors in the observation data is still a difficult problem to solve. The electromagnetic signal of the natural field is very weak compared to the electromagnetic signal caused by human activities. With the development of human society, suppressing human noise has become an inevitable problem in magnetotelluric data processing. Traditional magnetotelluric data processing methods use far-reference and robust estimation algorithms to suppress noise and improve data quality. However, our recent research has found that robust estimation algorithms have little effect in data processing in strong interference areas, and sometimes even lead to worse data. We conducted an in-depth analysis of the magnetotelluric measured data in the southern segment of the Tanlu fault zone. By studying the characteristics of magnetotelluric interference signals and the effects of various data processing methods, and then comparing the results of different data processing methods, we have summarized a set of magnetotelluric data processing schemes in strong interference areas. First of all, the remote reference measuring point that is far away from the measuring point, with weak noise, and weak coherent noise is essential for data processing. Second, for measuring points with strong coherent noise and long duration, robust estimation processing techniques should not be used. Robust estimation technology will strengthen near-field interference and suppress effective signals. Third, for dead band data, robust estimation technology is a necessary method to obtain effective signal impedance estimation. Without using robust estimation techniques, it is impossible to extract valid signals from weak deadband signals. Fourth, for data that has both near-field interference and deadband, put the power spectrum processed by the above two methods together, and hopefully obtain better results from the fine editing of the data. In addition, no matter what kind of interfering data, extending the data collection time, especially the data collection time during the night with less interference, can effectively improve the data quality. During data processing, reducing the superimposition before power spectrum selection and increasing the number of selections can also help obtain better processing results.

Key words: Magnetotelluric, strong interference, data processing, remote reference, robust estimation

摘要： 天然源的大地电磁信号易受干扰影响，导致阻抗估计的结果不准确，这种难以克服的缺陷制约着大地电磁法的应用。随着工业化发展，人文干扰越发严重，单一的数据处理方法已经不能改善强电磁干扰条件下的数据质量。本文结合大地电磁数据处理的基本理论对中国东部强干扰地区实测数据进行了处理。分析了稳健估计（Robust）处理、分段叠加处理和分时处理等技术的处理效果，总结了不同干扰情况下的数据处理最佳方案。对于不同干扰特征的数据，要综合分析Robust处理对数据的影响，灵活应用Robust处理。为了得到更好的处理结果，应适当增加数据采集时间，特别是夜间干扰较弱时段的数据。增加数据分段的个数，减少每段中的数据量，提供更多可供编辑的数据也是得到优质数据的必要条件。

关键词: 大地电磁法, 强干扰, 数据处理, 远参考, 稳健估计