Identifying and Removing Noise from the Monterey Ocean Bottom Broadband Seismic Station (MOBB) Data

David Dolenc, Barbara Romanowicz, Bob Uhrhammer, Paul McGill (MBARI), Doug Neuhauser, and Debra Stakes (MPC)

Introduction

The Monterey ocean bottom broadband station (MOBB) was installed in April 2002, 40 km offshore in the Monterey Bay, at a water depth of 1000 m, through a collaborative effort between the Monterey Bay Aquarium Research Institute (MBARI) and the Berkeley Seismological Laboratory (BSL; Romanowicz et al., 2003). The MOBB includes a 3-component Guralp CMG-1T broadband seismometer, a differential pressure gauge (DPG) and a current meter. Prior to the deployment, the seismometer system was extensively tested and insulated at the BSL to minimize the long-period noise due to the sensitivity of the instrument to air movement within the titanium pressure vessel. During the deployment, further steps were taken to minimize the background noise that could be generated by the flow of water around the pressure vessel housing the seismometers (Uhrhammer et al., 2002; Dolenc et al., 2006). The remaining long-period background noise observed at MOBB is primarily due to pressure forcing from the infragravity ocean waves (Dolenc et al., 2005). Infragravity waves are ocean surface waves in the frequency band between 0.002 and 0.05 Hz. Results from a previous study (Dolenc et al., 2005) showed that the long-period noise data recorded at MOBB can be used to better understand where and when the infragravity waves are generated. But for the study of seismic signals, additional processing is needed to remove the long-period noise from the MOBB data.

Background noise removal

To remove the long-period background noise from the MOBB data we employed two methods. In the first one we subtracted the simultaneously recorded ocean bottom pressure signal from the vertical seismic acceleration in time domain. The scale factor by which the pressure signal was multiplied was assumed to be frequency independent and was linearly estimated from the data. This method has previously been used to remove atmospheric pressure signal from the vertical seismic data recorded at land stations (Zürn and Widmer, 1995). Figure 13.1a,b shows that this method can also be used to remove long-period background noise from the ocean bottom seismic data. The presented results are for a 5.5-hour period for which the pressure signal was removed from the vertical seismic acceleration signal in time domain. The result is shown in frequency domain to illustrate the successful removal of the infragravity ``hump''.

In the second approach we combined the pressure observations with measurements of the transfer function between vertical seismic and pressure recordings to predict the vertical component deformation signal. The predicted signal was then removed from the vertical seismic data in either frequency or time domain (Crawford and Webb, 2000). The transfer function was calculated from periods without earthquakes. Since it is only a function of structure at the MOBB location, it does not change with time and can be applied to all data from this site. Figure 13.1c-f shows an example of the transfer-function method to remove noise from the earthquake free vertical MOBB data. An example of the long-period background noise removal for a period that included an earthquake is shown in Figure 13.2a-d. The 1-hour period used in the calculation now included the 12/06/2004 $M_{w}$ 6.8 Hokkaido, Japan event. The same transfer function as described above and shown in Figure 13.1d was used. The result shows that the method successfully recovers the seismic phases that were previously hidden by the long-period background noise and that the result is similar to the waveforms from the nearby land station SAO (Figure 13.2d).

Figure 13.1: Left: Example of the time-domain method to remove noise from the earthquake-free vertical data. (a) Power spectral density (PSD) calculated for a 5.5-hour period without earthquakes for the vertical seismic channel (black) and the DPG (gray). At periods longer than 20 s the infragravity ``hump'' is observed for both datasets. (b) PSD for the vertical seismic channel before (black) and after (gray) the time-domain subtraction of the DPG signal. Right: Example of the transfer-function method to remove noise from the earthquake-free vertical data. (c) Power spectral density (PSD) for 1-hour period without earthquakes for the vertical seismic channel (black) and DPG (gray). (d) Transfer function between vertical seismic and DPG signal calculated from 144 1-hour long data windows within 2005.034-056 period. (e) Coherence between the vertical seismic and DPG channel for the selected 1-hour period on day 2005.035. (f) PSD for the vertical seismic channel before (black) and after (gray) the noise removal using the transfer function shown in (d).

Signal-generated noise removal

The other type of noise observed at MOBB is the signal-generated noise. It is due to reverberations of seismic waves in the shallow sedimentary layers and may be unavoidable in buried ocean bottom installations. It is particularly strong following the arrival of sharp and strong seismic phases that are characteristic of large deep teleseismic events. We used two methods to remove the signal-generated noise. The first one employs the empirical transfer function constructed from MOBB data and nearby land station data that do not show the signal-generated noise. Results obtained with this method for the Fiji Islands event are presented in Figure 13.2g. Island station FARB was used as a reference land station to obtain the empirical transfer function. To demonstrate that the empirical transfer function from one event can be used to remove signal-generated noise from another event, we used the empirical transfer function obtained from the 11/17/2002 $M_{w}$ 7.3 Kurile event to ``clean'' the Fiji Islands event data. The results in Figure 13.2h show that most of the signal-generated noise at MOBB was removed. This suggests that to routinely clean smaller events, we will not need to compute the empirical transfer function every time, but rather use one from a previous strong event.

The second method uses a synthetic transfer function computed by modeling the response of shallow layers at the MOBB location. The response of the sedimentary layers is modeled using the propagator matrix approach (Kennett and Kerry, 1979). To obtain the response of the 1-D structure we used a previously published 1-D crustal model for this region (Begnaud, 2000) and replaced the top 350 m with a slower sedimentary layer. The synthetic transfer function was obtained by the spectral division of the result obtained with the 1-D model with the additional sedimentary layer and the result obtained with the original 1-D crustal model. The synthetic transfer function was then used to deconvolve the signal-generated noise from the MOBB vertical channel (Figure 13.2i).

Results presented in Figure 13.2e-i show that both methods can successfully remove the signal-generated noise from the MOBB data and that the obtained results are similar to the waveforms observed at the nearby land station (Figure 13.2f).

Figure 13.2: Left: Example of long-period background noise removal for the 12/06/2004 $M_{w}$ 6.8 Hokkaido, Japan earthquake using the transfer function shown in Figure 13.1d. (a) Original MOBB vertical data. (b) MOBB data bandpass filtered between 0.001 and 0.1 Hz. (c) MOBB data after removal of the coherent DPG signal and bandpass filtered between 0.001 and 0.1 Hz. (d) Land station SAO data bandpass filtered between 0.001 and 0.1 Hz. Right: Three examples of deconvolution of the signal-generated noise at MOBB for the Fiji Islands event. (e) Original MOBB data. (f) Original FARB data. (g) MOBB data after removing empirical transfer function constructed using MOBB and FARB data. (h) Same as (g), only that empirical transfer function obtained from the 11/17/2002 $M_{w}$ 7.3 Kurile Islands event was used. (i) MOBB data after removing a synthetic transfer function obtained by 1-D modeling of the shallow structure with a 350 m sedimentary layer ( $v_{p}=0.324 km/s$, $v_{s}=0.196 km/s$, and $\rho =1.3 g/cm^3$) .

Conclusions

The described methods are an important tool to remove noise from the seismic data recorded at the ocean bottom buried broadband installations. Both methods used to remove the long-period background noise require the locally recorded pressure signal at the seafloor. It is therefore important to have a reliable pressure sensor collocated with the ocean bottom seismometer. It is also important that the sampling rate for the DPG and other environmental data (e.g. temperature, ocean current speed and direction) is high enough so that they can be used in the post-processing for the complete seismic frequency band.

Acknowledgements

The MOBB instrumentation, deployment, and maintenance were supported by Lucile and David Packard Foundation funds to MBARI, NSF grant OCE9911392, and UC Berkeley funds.

References

Begnaud, M. L., K. C. McNally, D. S. Stakes, and V. A. Gallardo, A crustal velocity model for locating earthquakes in Monterey Bay, California, Bull. Seism. Soc. Am., 90, 1391-1408, 2000.

Crawford, W. C. and S. C. Webb, Identifying and removing tilt noise from low-frequency ($<0.1$ Hz) seafloor vertical seismic data, Bull. Seism. Soc. Am., 90, 952-963, 2000.

Dolenc, D., B. Romanowicz, D. Stakes, P. McGill, and D. Neuhauser, Observations of infragravity waves at the Monterey ocean bottom broadband station (MOBB), Geochem., Geophys., Geosys., 6, (9), 1-13, doi:10.1029/2005GC000988, 2005b.

Dolenc, D., B. Romanowicz, B. Uhrhammer, P. McGill, D. Neuhauser, and D. Stakes, Identifying and removing noise from the Monterey Ocean Bottom Broadband seismic station (MOBB) data, Submitted to Geochem. Geophys. Geosys., 2006.

Kennett, B. L. N., and N. J. Kerry, Seismic waves in a stratified half space, Geophys. J. R. Astr. Soc., 57, 557-583, 1979.

Romanowicz, B., D. Stakes, R. Uhrhammer, P. McGill, D. Neuhauser, T. Ramirez, and D. Dolenc, The MOBB Experiment: A prototype permanent off-shore ocean bottom broadband station, Eos Trans., AGU, 84, (34), 325, 331-332, 2003.

Uhrhammer, R., B. Romanowicz, D. Neuhauser, D. Stakes, P. McGill and T. Ramirez, Instrument testing and first results from the MOBB Observatory, Eos Trans., AGU, 83, (47), Fall Meet. Suppl., F1008, 2002.

Zürn, W. and R. Widmer, On noise reduction in vertical seismic records below 2 mHz using local barometric pressure, Geophys. Res. Lett., 22, 3537-3540, 1995.