next up previous print clean
Next: Conclusions Up: Effects of and niter Previous: Data

Results

Given the velocity model, we expect to see symmetrical shadow zones under the edges of the low velocity lens. They are evident in the result of our downward continuation migration (Figure [*]). They are the most clear in the depth slice (the top panel), enclosed by black ovals, as narrow black ``V''s. They can be seen as a slight dimming in two places along the event in the common ray parameter section (lower left panel). The common image gather (lower right panel) also shows a gap at high offset ray parameters.

 
mod.mig
Figure 2
Result of downward continuation migration. The display is a flattened cube in which the top panel is a depth slice, bottom left panel is a common ray parameter section, bottom right is a ray parameter gather. The shadow zones can be seen in the depth slice as thin black ``V''s inside the 2 black ovals. The black circle (uppermost) indicates strong artifacts.
mod.mig
view burn build edit restore

To examine the effects of $\epsilon$ and niter, I ran four experiments. For the first, I set niter=3 and $\epsilon=1$. This result can be seen in Figure [*]. The shadow zones are still visible inside the black ovals but have begun to fill in. The artifacts at the edges of the event (inside the black circle) are reduced and the imaging artifacts above the flat event are also diminished.

 
mod.1dprec.3it1eps
Figure 3
Result of inversion with 3 iterations and $\epsilon=1$. The display is a flattened cube in which the top panel is a depth slice, bottom left panel is a common ray parameter section, bottom right is a ray parameter gather. There is little change in the shadow zones (black ovals) but the artifacts are reduced (black circle).
mod.1dprec.3it1eps
view burn build edit restore

The second experiment set niter=3 and $\epsilon=10$. This result (Figure [*]) shows that the shadow zones are mostly filled in. This combination of $\epsilon$ and niter seems to be a good choice.

 
mod.1dprec.3it10eps
Figure 4
Result of inversion with 3 iterations and $\epsilon=10$. The display is a flattened cube in which the top panel is a depth slice, bottom left panel is a common ray parameter section, bottom right is a ray parameter gather. The shadow zones are largely filled in (inside the black ovals). The artifacts inside the black circle are diminished.
mod.1dprec.3it10eps
view burn build edit restore

The third experiment had niter=12 and $\epsilon=1$. Figure [*] shows an unexpected result. The shadow zones have actually become larger. I believe that the high number of iterations has allowed the data fitting goal to clean up artifacts that creep into the shadow zones because of approximations in the imaging operator. The low $\epsilon$ keeps the regularization operator from trying to fill the shadow zone.

 
mod.1dprec.12it1eps
Figure 5
Result of inversion with 12 iterations and $\epsilon=1$. The display is a flattened cube in which the top panel is a depth slice, bottom left panel is a common ray parameter section, bottom right is a ray parameter gather. The shadow zones are actually getting larger (inside the black ovals) as the high number of iterations cleans up artifacts while the low $\epsilon$ doesn't help fill the shadow zones.
mod.1dprec.12it1eps
view burn build edit restore

The final experiment used niter=12 and $\epsilon=10$. Figure [*] shows that this combination has filled the shadow zones nicely. Unfortunately, there is also some noise introduced, faintly visible as a wide vertical stripe in the center of the common ray parameter section.

 
mod.1dprec.12it10eps
Figure 6
Result of inversion with 12 iterations and $\epsilon=10$. The display is a flattened cube in which the top panel is a depth slice, bottom left panel is a common ray parameter section, bottom right is a ray parameter gather. The shadow zones are filled in (inside the black ovals), although this combination of iterations and $\epsilon$ seems to be causing noise away from the reflector.
mod.1dprec.12it10eps
view burn build edit restore


next up previous print clean
Next: Conclusions Up: Effects of and niter Previous: Data
Stanford Exploration Project
7/8/2003