|
Intellectual Merit
|
The overall goal of the project was to use ALBACORE data processing techniques to inform the optimal conditions and design of an offshore Central California deployment. This was pursued in the context of different data products that may be needed for research activities towards reducing the uncertainties in path effects such as development of the next-generation statewide seismic velocity model.
From August, 2010 through September, 2011, a deployment of 34 OBSs took place off the coast of Southern California comprising the data gathering component of the ALBACORE project. The primary goal of the ALBACORE (Asthenosphere and Lithosphere Broadband Architecture from the California Offshore Region Experiment) project was to study the deformation history of the Pacific side of the tectonic boundary with North America. The OBSs sat on the seafloor in a 150 km north-south by 400 km east-west array and provide the 50 sps ambient noise and earthquake data. These data have been combined with Southern California Seismic Network (SCSN) stations that are on the islands and on the coastline for more complete near-shore coverage. Ambient noise cross-correlation functions (NCFs) have already been calculated between all possible station pairs for 12 months of data, after initial processing that included instrument response correction and narrowband filtering. Noise on the OBSs peaks in the ~0.15 to 1 Hz range and provides useful measurements from about 0.02 to 0.25 Hz.
The OBS locations were chosen such that approximately uniform station spacing could be achieved, taking advantage of existing SCSN stations installed on Catalina, San Clemente, San Nicolas, Santa Barbara, Santa Rosa, and San Miguel Islands. In the shallow-water Continental Borderland region, station spacing was approximately 40-50 km, and in the deep-water oceanic plate region, station spacing was approximately 75 km. SCSN data for the same time periods from Southern California island and coastline stations have been incorporated into the calculations.
Task #1: The first task consisted of determining optimal OBS station spacing and sensor type for different methods of data inversion for a future, offshore extension of a Central California seismic velocity model. In considering the trade-offs between the potential number of OBSs available from IRIS OBSIP for an offshore Central California seismic experiment, and the footprint of the area over which raypath coverage is desired, we conclude that the optimal station spacing is ~30 km. This assumes that we have access to ~25 broadband OBSs that should cover a region between Pt. Arguello to the south and San Simeon to the north. The OBSs would primarily sit on the continental shelf, extending 100 km from the coastline. This suggested configuration was used in a proposal for ship time to deploy OBSs as part of the CCSP project. The result is that ~30 km station spacing accomplishes the necessary spacing for a new velocity model based on inversion of earthquakes and NCFs for station pairs that include onland stations. The OBSs would complement a contemporaneous land seismometer deployment. Station spacing of ~30 km corresponds to appropriate seismic structure wavelengths and approximately matches the land seismic station density. Uniform spacing and distribution would be a more optimal configuration for seismic tomography.
Task #2: The second task consisted of quantifying the optimal use of horizontal components and co-located pressure gauge waveforms to increase the signal-to-noise ratios (SNR) of teleseismic, local earthquake, and ambient noise cross-correlation function signals produced by OBSs. OBS data are notoriously noisy due to different non-elastic (aseismic) sources of vibrations. One of the most prominent is the effect of infragravity waves (or wave loading) on the seismometer waveforms. One way in which the effect of wave loading can be reduced on the vertical component OBS is to remove coherent signals between the vertical component and a co-located pressure gauge that records contemporaneous pressure waveforms associated with infragravity waves. The non-elastic noise can be estimated by calculating the transfer function between the vertical-to-pressure components, and subtracting the coherent signal between the two from the vertical component time series. It has been shown that this process reduces low-frequency noise levels for OBS stations on soft sediments, near high currents, in shallow water or for poorly leveled seismometers (Webb and Crawford, 1999; Crawford and Webb, 2000). When OBSs are requested from the IRIS OBS Instrument Pool, the broadband OBS package always includes a low-frequency pressure gauge sensor. It is usually a differential pressure gauge (DPG), as in the case of ALBACORE, whose waveforms can be converted to absolute pressure through application of a DPG instrument response function. We investigated the wave loading effect on each ALBACORE OBS by quantifying the frequency range over which the coherence is large and for which wave loading corrections can be made. We investigated the vertical component-DPG coherence individually for each ALBACORE station, determined if there is a relationship between OBS deployment conditions and coherence, and showed how earthquake and ambient noise cross-correlation functions (NCFs) can be cleaned up by this correction technique. An analogous method can be applied to remove tilt noise on the vertical components using co-located horizontal component OBS data. Transfer functions between one horizontal component and the vertical were used to predict and remove coherent signal from tilt in that direction. The second horizontal component was also corrected for coherent signals from the first horizontal, since tilt in any direction other than the principle axes will induce signals on both, and then used to remove any remaining coherent tilt signal on the vertical component.
We followed the procedure of Webb and Crawford (1999) and Crawford and Webb (2000) to perform tilt corrections and DPG corrections. These types of corrections are traditionally applied to longer-period signals (i.e., greater than 50 s) than those of our study, but we found that they improve some of our measurements nonetheless. For a given non-vertical component (either horizontal or the DPG), a transfer function to the vertical component was determined from a 12-hour period of time known to be quiet and free of earthquakes. This transfer function describes the frequencies and associated phases at which signals are coherent, and can be used to predict the signal that a given pressure signal or tilt event will contribute to the vertical-component OBS channel. The coherencies vary strongly with location and water depth, so we tapered the transfer function to zero outside the period range of 5 to 15 s where the signals are most coherent. We applied all corrections in sequential steps. We first determined and applied a transfer function from one horizontal component to all other components before proceeding. Next, transfer functions from the second horizontal component were applied to both the pressure gauge and vertical component, and finally we applied the transfer function from pressure gauge to vertical. This sequential processing (i.e., also correcting one horizontal based on the other) ensured that any effect of the water column which affects both components coherently would not be mistakenly corrected twice. The transfer function is independent of units, so we did not apply the vertical instrument response until after the entire process is complete.
The effects of the wave loading and tilt corrections were often dramatic and potentially useful for earthquake detection. In contrast, we found that for the 5-9 s period range of noise cross correlations, the application of the corrections reduced the strength of the fundamental surface mode observation relative to the first overtone. It is likely that at the shorter periods our fundamental mode measurements are so sensitive to the water layer that removing signals coherent with the DPG and tilt also removes much of the useful signal. The first overtone, however, is sensitive to deeper structure and is relatively easier to measure with the correction. Thus, we considered both the uncorrected and corrected sets of noise correlations when measuring dispersion curves, using whichever set shows a stronger signal at a given period. Thus our recommendation is that an offshore component of CCSP uses co-located absolute or differential pressure gauges to be able to make a wave loading correction on the OBS waveform data.
Task #3: This task consisted of determining convergence times for stacked ambient noise cross-correlation functions, and correlations with the occurrence of large teleseismic and local earthquakes as well as seasonal meteorological events such as storms. The reason for this task was to determine the minimum and maximum number of weeks or months of OBS data recording time were necessary to achieve NCFs with good SNR, especially if an OBS was not available or does not record for a full 12 months. For an offshore component of CCSP, it may be necessary to deploy OBSs in a piecemeal fashion because the total number of available OBSs will likely be more limited compared to on-land deployments. A possible model is that a number of OBS stations would be deployed for a certain number of months, retrieved for data download, then redeployed in a nearby region for another multi-month duration. It is often assumed that at least one year of ambient noise data collection is needed in order to obtain useful cross-correlation functions. However, the choice of one year is driven primarily by the fact that battery life is usually the determining factor in experiment duration. In the case of an OBS deployment where deployment time needs to be optimized (and possibly minimized in anticipation of subsequent deployments), it is desirable to know the minimum amount of time needed for useful NCFs.
We examined our stacked ALBACORE NCFs from individual station pairs, for behavior as a function of time over the course of the deployment year. We examined NCFs for each week of the experiment, i) independently on a week-by-week basis, and ii) cumulatively over the year on weekly intervals. We also observed how large teleseismic and local earthquakes (especially their coda) affected the resulting NCFs even when amplitude normalization is applied. Based on this, we quantify the minimum amount of time needed for NCF convergence, as well as illustrate how the presence of a water layer (at thicknesses between 1200 and 4800 m in ALBACORE) results in additional low-velocity arrivals of energy.
We conducted a wide search of ALBACORE OBS-OBS and OBS-SCSN station pairs to determine the average amount of time needed to obtain convergence to useful NCFs from the stacked cross correlations. The station settings consisted of deep (>2000 m) water, shallow water (1000-2000 m), and southern California island settings. Surprisingly, we find that in the most extreme cases, no more than approximately four months of OBS data are necessary for achieving convergence. In many cases, only 4-8 weeks of stacked, cumulative NCFs are needed to measure travel times.
Consider first a station pair that consisted of a deep water OBS-deep water OBS (e.g., OBS07-OBS15) for which the water depths and lithospheric structure between stations are the least heterogeneous. For this pair, we find variations in NCF features throughout the 12 months of stacked NCFs. For 4 s filtered waves, the first, main arrival corresponding to the fundamental mode appears strong with good SNR after only about 4 weeks. The 11 s filtered waves take about 2 months of stacked NCFs to show good SNR. We observe possible higher modes (which clearly arrive earlier than the main arrival) emerging with good SNR after about 3-4 months. The coda in the NCFs (later arrivals that probably correspond to body wave arrivals) take 2-4 months of data to emerge with good SNR. Thus, if body wave arrival times or waveforms are needed in inversions for velocity structure, more cumulative data (2-4 months) are needed than for the fundamental Rayleigh wave only (1 month). In the 23.4 s NCF data, there is almost immediate convergence, i.e., no more than about 4 weeks of data are needed.
Consider next a station pair that consisted of a deep water OBS-shallow water OBS (OBS03-OBS18). We see the influence of wave paths crossing the continental shelf. Sampling both continental and oceanic lithospheric velocity heterogeneity. The final, stacked, cumulative NCF main arrivals emerge after 3-4 months. We observe body wave type arrivals, similar to those observed in the deep water pair, for 11 s filtered data. For the 23.4 s filtered data, observations are similar to the deep water pair, i.e., convergence is almost immediate, except that energy from the 3/11/11 Mw9.0 Tohoku earthquake dominates the waveforms. Although we apply normalization to down-weight the amplitudes of earthquake arrivals, the Tohoku effect dominates the 23.4 s waves for about three months into the cumulative stacks.
Task #4: This task was to characterize offshore Central California continuous data from OBSs deployed in 2013 by PG&E (cabled and temporary stand-alone stations off Point Buchon in San Luis Obispo County) for earthquake SNR and ambient noise signals, and comparison with ALBACORE results. We examined ambient noise cross correlations from OBSs installed by PG&E off the coast of Central California. This include data that were recorded by the cabled OBSs and by a temporary OBS array deployed by PG&E off Point Buchon in San Luis Obispo County. Four temporary OBS units were deployed for 8 weeks during October-November 2013. Data from the cabled array were limited due to hardware issues; however, we obtained enough to calculate NCFs and noise characteristics. As part of this task we calculated power spectral densities (PSDs) of individual sensors, and coherence values from OBS and DPG waveform information.
Related to the previous task, we also examined convergence rates for a PG&E OBS-PG&E OBS pair. We obtained data for two PG&E stations: POBS and 1616, intermediate-band Guralp velocity sensors. The stations are near each other – approx. 3.5 km apart; thus the spatial sampling lengths are not large enough to adequately sample long periods. However, we find that for the 0.5 - 1.0 Hz range, the SNR of these waves is surprisingly good, even though this falls within the main microseismic noise band. In this shallow water station pair, we observe NCF convergence after only 8-10 days for 1 Hz NCFs. We observe similar behavior for frequencies slightly below and slightly above this range.
PSDs were computed for 24 hours from both ALBACORE and PG&E OBS data from similar time periods. The ALBACORE data are from September 16, 2010, and the PB&E data are from September 15, 2013, both quiet days with no storms or earthquakes. Welch’s method of using overlapping time series windows was applied to the PSDs calculation. Both the ALBACORE and PG&E OBS data exhibit large values in the PSDs in the primary microseismic noise peak between 1 and 10 seconds. They both peak between 3 and 4 seconds, likely corresponding to ocean waves traveling in opposite directions. Another microseismic peak is seen only in the PG&E data at between 10-20 seconds, with a peak at 16 seconds. This likely corresponds to surface gravity waves in the shallow water, i.e., ocean waves interacting with the shallow water shelf and coastline, that is not seen in the ALBACORE stations because these are much further from any coastline.
The overall goal of these activities is to use ALBACORE data processing techniques to inform the optimal conditions and design of an offshore Central California deployment. This will be pursued in the context of different data products that may be needed for research activities towards reducing the uncertainties in path effects such as development of the next-generation statewide seismic velocity model.
In summary, the data processing products that result from this project will be used to provide guidelines on strategies for a future OBS deployment that could comprise an important component of the Central California Seismic Project. Elements of this strategy will need to include assessment of local and teleseismic earthquake OBS data SNR in the context of event detection and inversions for seismic velocity models; methods that use ambient noise cross-correlation functions for seismic velocity inversions; optimal design of OBS deployments in relatively shallow water; and taking advantage of multiple data streams to optimize SNR. |
