To see the power of a non-stationary filter to characterize model covariance
let 's return to the fault model missing
data problem (Figure 8).
For our interpolation we
will use the known model as the basis for our PEF, and have a micro-patch
size of one sample in *x* and *z*.
The resulting interpolation, Figure 10 isn't
quite as high-frequency as our initial model and we have a minimal
amount of continuation of the layers over the fault, but we generally
do an excellent job recovering the model.

fault.cont
The result of finding
non-stationary PEF using the correct model and then applying it to
the missing data problem. The returned image is almost exact and does
a better job with bed discontinuities than Figure 8.
Figure 10 |

4/28/2000