next up previous print clean
Next: CONCLUSIONS Up: Curry: Iteratively re-weighted least-squares Previous: Synthetic Example

CASC93 Data Set

The previous example used noise that presents specific problems with existing $\ell^2$ based optimization methods. For a more realistic test of this method, a real data set with severe noise problems is needed. I have now interpolated a single event from the CASC93 Nabelek and Zandt (1993) data set. The CASC93 data set is passive seismic data experiment performed in 1993-1994 in central Oregon, with an irregular 2-D array of three-component seismometers. Arrivals from distant earthquakes were recorded, and can be used to image the lower crust of that region Shragge (2003). As shown in Figure [*], the data is very noisy, containing broadband noise, as well as artifacts from the preprocessing of the data. The data is on a 2-km grid, with approximately 30 percent of the bins filled.

While the results are definitely more subtle than in the test case, the same conclusions are supported. In the left side of Figure [*], we can see that switching to an IRLS PEF estimation improves the estimation in some areas, most notably in the lower-right corner patch, where the PEF does not benefit as much from regularization due to end effects. Smaller spikes in the difference between only IRLS PEF estimation and IRLS PEF estimation and interpolation are shown in the lower right. As before, it appears that certain spikes appear, although they aren't as obvious as in the previous example.

 
casc93
casc93
Figure 6
The CASC93 land data set. Top right: original data; top left: L-2 interpolated data; mid left: IRLS PEF estimation; bottom left: IRLS PEF estimation and IRLS interpolation; mid right: difference between L-2 interpolation and IRLS PEF estimation; bottom right: difference between IRLS PEF estimation and IRLS PEF estimation and IRLS interpolation.
view burn build edit restore


next up previous print clean
Next: CONCLUSIONS Up: Curry: Iteratively re-weighted least-squares Previous: Synthetic Example
Stanford Exploration Project
7/8/2003