Noise reduction for broad-band, three-component seismograms using data-adaptive polarization filters

• Published in: Geophysical Journal International
• Main Authors: Du, Zhijun, Foulger, G. R., Mao, Weijian
• Format: Text
• Language: English
• Published: Oxford University Press 2000
• Subjects: Articles; Iceland
• Online Access: http://gji.oxfordjournals.org/cgi/content/short/141/3/820; https://doi.org/10.1046/j.1365-246x.2000.00156.x

We develop a data-adaptive polarization filter that can spectacularly reduce microseismic noise contamination in three-component broad-band seismograms. The filter uses a multitaper spectral analysis method for computing the data spectral density matrix, which is defined as an ensemble average of outer products of the spectrum and its Hermitian adjoint. Under the assumption that strong noise in three-component, broad-band seismograms is additive white noise, and that its spectral density can be determined from seismogram segments without signals, that is, a pre-signal arrival time window, we construct a data-adaptive filter from a spectral density matrix that has been decontaminated of noise. Since the noise corrupting the seismograms is complicated and stochastic, the resulting residual due to the real, non-stationary nature of microseismic noise can leave small-amplitude, quasi-sinusoidal, background oscillations after filtering. These oscillations can be removed by subsequent application of an optimum Wiener filter. Application of the filter to synthetic data with real noise superimposed suppresses the noise by about three orders of magnitude at the expense of less than 5 per cent corruption of the original seismogram in amplitude. Application to several real recordings of teleseismic earthquakes on a three-component broad-band seismic station in Iceland shows that excellent signal-to-noise recovery is possible, rendering such data usable for both arrival time and waveform analysis. This technique may potentially increase by an order of magnitude the volume of usable data collected in seismic experiments in noisy environments, for example, on oceanic islands.