Robust Regression Estimators in Magnetotelluric Data Processing: Performance and Evaluation

Magnetotelluric (MT) transfer functions are commonly estimated using statistical regression methods. Reliable MT-derived geoelectrical structures are essential for investigating deep geodynamic processes and metallogenic systems; however, noise contamination can distort impedance estimates and lead to misleading geological interpretations. In this study, we introduced an MM-estimator into MT data processing and evaluated its performance alongside M- and S-estimators using synthetic examples and an MT data survey from northeastern China. The results indicate that all three estimation methods can effectively suppress complex background noise and improve the stability and reliability of MT impedance estimation. The methods produced generally consistent apparent resistivity and phase responses, with MM-estimator yielding relatively the lowest root-mean-square (RMS) values. However, the deviations–uncertainty trade-off should be carefully evaluated in practical applications, particularly when processing long-period MT data with MM-estimator. Finally, we suggest combining any of the three robust estimation methods with the remote reference technique to help suppress noise in both predictor variables (magnetic-field data) and dependent variables (electric-field data).