![[pdf]](icons/pdf.png){kind=icon}
|
|
|
|
| Least-squares reverse time migration for the Cascadia ocean-bottom dataset | |
|
Although our modeling operator in LSRTM-HM can generate higher-order surface-related multiples, we do not see major improvement in the inverted images. One reason is related to the fact that there is limited signal coming from the double-mirror event in our recorded data. This dataset only has recording up to five seconds which results in some truncation in the double-mirror signal. It will be interesting to test on datasets with either a longer recording time or a shallower sea bottom. Another reason is related to the constant density assumption in our ascoutic modeling. In Figure 9 (b), we can see that the double-mirror signal is generated in the modeling. However, its amplitude is significantly weaker than that in the field dataset (Figure 9 (a)). With a constant density propagation kernel, the higher-order multiples are generated solely from reflection off sharp contrasts/interfaces in the migration velocity model. For the Cascadia dataset, it turns out that the seabed's reflectivity is attributed significantly to the density contrast, which is not accounted for in our modeling. As an result, the velocity constrast alone does not account for the full reflectivity of the sea bottom. We are currently trying to match the correct reflectivity by implementing our LSRTM using the full ascoutic wave equation. That is, without the constant density assumption. Nearby well log data will be used to estimate the density constrast at the sea bottom.
The computational cost of LSRTM is higher than that of RTM, with a factor proportional to the number of iterations. For an OBS survey, such additional computational costs can still be affordable. This is because the number of pre-stack migrations needed equals the number of OBS receivers in the survey. The number of migration needed is substantially smaller than that in a towed-streamer survey. Recently, Dai et al. (2010) suggested using phase encoding in LSRTM, which can reduce the computation substantially but introduces crosstalk into the image.
|
|
|
|
| Least-squares reverse time migration for the Cascadia ocean-bottom dataset | |
|