Magnetotelluric impedance estimation Based on clustering analysis

Abstract

De-noising or removing noise the magnetotelluric (MT) data using the conventional methods is dependent on noise ratio in the time series collected. To treat this problem, we propose a clustering analysis method based on Impedance Euclidean Distance (IED) and Improved K-Means Cluster Method (IKCM). Firstly, the time series is divided into same length segment and estimated the impedance of interest frequency. Secondly, the IED is evaluated by the real and imaginary part of impedance, which conduct the sample of IKCM. Thirdly the IKCM is performanced and the optimal class is determined according to the coherence, the compactness and the smoothness criterion. The resistivity and phase are estimated by the result of the best class. Finally, we evaluate the proposed method using synthetic data set and field data set collected at Tianba Village of Yunnan Province, China. Experimental results demonstrate that the presented approach can be used to select out free noise or less noise impedance from noise data set, the evaluation results also indicate that the proposed method is superior to the robust method in terms of the smoothness and continuity of the resistivity and phase curves.