The results of the Q inversion described above applied to the SCS data are shown in Figure 3C. The analysis was performed on groups of 20 traces divided into four layers by five interfaces shown in Figure 3A. The time thicknesses for the selected layers remained constant across the section, and hard lower and upper bounds of Q = 5 and 2000 respectively were used to limit the models. Q values of 2000, adjacent to the hard bound, were assumed to be erroneous and were omitted. Values of Q = 5 were not encountered.
Immediately apparent in the Q profiles for each layer is that Q in the HSZ (layers C and H) is systematically higher (Q = 90-600) than in the GSZ (layers B and G, Q = 10-200). Also apparent is the gradual decrease in Q from the southwest to northeast (left to right) until a minimum is reached, after which Q increases more sharply. At Site 995, curve C is high, ~500-600, sloping downward to a minimum of 80-90 at Site 997, and rising to 150-200 at the northeast edge of the section. Although these values are not remarkably low, they may be lowered by the effects of intrabed multiples, which we expect to play a more dominant role in the layers with more reflective sediments like those in layer C.
Q for layer H nearly parallels Q for layer C, starting again at 500-600 at Site 995, but is systematically higher, reaching a minimum at 100-200, 1-2 km southwest of Site 997, and then rising very sharply back to 500-600 before reaching Site 997. Layer H is where most of the hydrates were found (and where the least reflectivity is observed) and is where we expect to find the highest values of Q. The lower two layers are known to contain some free gas and are therefore expected to exhibit lower values of Q, which the results confirm. Layer B in the GSZ starts at Q = 200-300 in the southwest and decreases to Q = 30-50 at Site 995 with a very slight increase to Q = 70-150 in the northeast. Similarly, layer G in the GSZ starts at Q = 150-250 in the southwest, decreasing sharply to near Q = 10 between Sites 994 and 995, rising gradually to Q = 40-60 by Site 997, and systematically showing the lowest Q of the four layers. Although the uncertainties mentioned above are larger than many of the differences we see at any given point in Figure 3C, the existence of such systematic behavior suggests that the trend across the section is more certain than the analysis at any single location.
Although the travel times obtained from the VSPs were quite useful in delineating the Vp structure across the ridge (Holbrook et al., 1996), the actual waveform data was somewhat disappointing. Although accurate traveltime picks were obtained, an unpredictable clamping arm malfunction and a generally noisy environment significantly degraded the waveforms. Several attempts were made to identify sources of both electrical noise in the area of the underway geophysics lab and acoustic noise emanating from the hull of the drill ship (Hoskins and Wood, 1997), but little could be done to mitigate either. The cleanest displays of the time domain VSP data are shown in the Initial Reports volume (Paull, Matsumoto, Wallace, et al., 1996): Site 994 on p. 136, Site 995 on p. 207, and Site 997 on p. 306. To make these displays, only the clearest traces from the most powerful source (the air gun) were filtered stacked at each clamp depth. Because Site 995 was significantly less noisy than the other two sites, our efforts were concentrated on that data set.
The revised processing included editing out visibly bad traces and then time shifting the data so that the first arrivals at each clamp could easily be windowed from the rest of the trace, which was frequently quite contaminated with noise. The data were bandpass filtered, keeping only the range with the most power, 10-120 Hz. Spectra of the windowed first arrivals were computed and stacked over each clamp depth with a median stacking algorithm to mitigate the effect of outliers. Noisy spectra in the final stack were edited out leaving the display shown in Figure 5A. A similar procedure was applied to the water-gun data resulting in the display shown in Figure 5C. Unfortunately the data were too noisy to obtain Q estimates from the spectral modeling technique used on the SCS data.
Despite the noise, a significant shift in the frequency content is resolvable in the spectral data at the depth of the BSR. The effect is accentuated by the lobate nature of the spectrum, in which higher frequency lobes are more acutely bent than are the lower frequency lobes. This bending can be modeled to estimate Q. The most prominent lobe (Fig. 5B) was used to invert for a two-layer Q model (HSZ over GSZ). In the top layer, almost no frequency shift was visible. The two lines indicate the expected trajectory of this lobe in depth given a Q of 10 and 20. Clearly, any higher Q than 20 results in no shift, so we may only put a lower limit on this layer of Q = 20. These data are less sensitive to Q not just because of the increased noise, but primarily because of the low frequency at which the modeling is done (55 Hz vs. 200 Hz for the SCS data). Because Q for the GSZ is significantly lower, we may identify Q quite readily at a value of 6; the three lines show the expected result when the upper layer is modeled with Q = 20 and the lower layer is modeled with Q = 5, 6, and 7 (Fig. 5B). The final model is drawn over the somewhat lower power water-gun data and appears to be consistent, but the water-gun spectra do not exhibit the lobate character that facilitated the previous analysis and accurate modeling is not possible.