Wavelet based adaptive thresholding technique to remove earthquakes and other transient signals from continuous ambient noise records

Ambient noise tomography technique to image the earth structure has emerged rapidly in the past two decades. The earth is always at unrest which causes ground vibrations known as ambient noise. The recordings of continuous ambient noise may obscured by earthquakes and other transient signals which are unwanted signals for performing ambient noise tomography. To remove these unwanted signals, temporal normalization techniques like 1-bit, running absolute mean, clipping threshold etc are available in the literature. The 1-bit and running absolute mean temporal normalization methods are effective to remove unwanted signals but every sample of ambient noise is also modified, while in case of clipping threshold unwanted signals are not eliminated efficiently. To overcome this shortcoming, discrete wavelet transform (DWT) based thresholding method is proposed in this paper. The raw ambient noise records are decomposed into different frequency sub-bands, where the threshold is computed for each sub-band using universal threshold value estimation method, and the signal is thresholded at each sub-band using the corresponding threshold value. The clean ambient noise is reconstructed using inverse DWT. In this paper, synthetic test data is processed using the proposed as well as the conventional methods to show the performance of the proposed method. Mean square error between the raw ambient noise and the clean ambient noise data is also computed to quantify the modifications caused in the ambient noise samples due to processing. Further, results are shown for the real data recorded at different stations of USArrav using the proposed method.