Continuous logging of data every 0.02 seconds for six months yields an outrageous 180 Gbytes of passive seismic data. Each of the stations log vertical and northerly and easterly oriented shear motions. All time signals are synchronized with GPS clocks.
The Mark Products (now owned by Sercel) L22 short period seismometer is ubiquitous in the seismological community. Therefor, it is reasonable to understand its characteristics. W. Menke (1991) performs a comprehensive analysis of the performance of the L22. It has a resonance frequency of 2 Hz, which is significantly lower than that of an exploration geophone. However, as seismologists are normally not interested in higher frequencies, the response functions shown never extend above about 30 Hz. The authors claim to have seen significant cross-axis coupling of the shear and compressional channels over frequency bands near the natural frequency, but this seems to be of little concern to this use of the data. Of more concern, the authors identify ``one of the main instrument defects'' of the L22 being strong amplitude resonance peaks centered at 28 Hz in over 20% the instruments.
Preliminary manipulation of the SCVSE data show that traces are indeed white show no coherence in their raw form. Cross correlating the records at this time provides no useful information as I have not been able to implement the code with the irregular geometries required with the data shown in Figure 2.
At this stage, it is unclear whether strong earthquake energies will help or hinder the experiment. While we desire strong incident wave fields, over representation of energy from particular azimuths and incidence angles may be detrimental. The underlying question here is whether or not teleseismic events (earthquake signature from long distances) will be the predominant energy source to illuminate the subsurface by reflecting from the free surface. If so, focusing our efforts in time around the arrival times of known events (from published earthquake catalogs) may significantly reduce the length of the time series that need be processed. Rather than long, continuous time records, we can isolate discrete time windows that can be treated analogously to single shot experiments. The price to pay for this however will be in resolution. Due to the geometry of the radial structure of the earth, we can only expect incident waves in a limited window of incidence angles from below. In addition, the usable period of these events is centered around one or two seconds which will greatly reduce the resolution of the image.
Figure 3 shows the earthquake and blasting events within 500 km of the survey location during the time the recording units were deployed. As these events and their times are readily available, it will be easy to window data series within and between major quake events to address this question.
Alternatively, it may be possible to use the earthquake energy in both contexts within the framework of an illumination study. Because the timing, azimuth and ray parameter of the earth-quake energy is available, it may be possible to normalize or otherwise manage what could be over-abundant energy.
Of benefit to this type of survey is that those who design and utilize these surveys are principally interested manufacturing tomographically derived velocity models from earthquake events utilizing both vertical and shear components of ground motion. This fact results in the availability of initial velocity models for migration studies and rudimentary practices in separating incident and scattered wave fields. However, it is my sincere hope that ambient noise will provide sufficient images as to not need to focus on teleseismic events. Due to the large offset between stations (three to five kilometers in this instance) the transmission losses of some of the ambient noise will undoubtedly prevent correlated signal from spanning the entire breadth of the survey layout.
Despite the outcome of this question however for this particular training set, the issue needs addressed specifically with an experimental mobilization tailored for our interests. This means that it should have receiver arrays designed to attenuate surface waves, station spacing on the order of a few meters (rather than the kilometers associated with seismologic data sets), and a roughly square map view (as suggested by Artman (2002b)) with a regular station spacing. The harder the near surface that the receivers are coupled to, the less high frequency surface noise will conflict with our desired signal.