We will show two field-data examples. First, let us look at a common receiver gather recorded in a Walk-Away marine survey. The gather is displayed in Figure 6a. This gather is collected with many shots on the surface and a common receiver down into a well. The first arrival and several events that follow it are down-going waves. These events are strong and are expected to be picked correctly. After 0.75 seconds, the up-going waves reflected from a deep subsurface start to interfere with the down-going waves. We are curious to know the results of our algorithm when events intersect. Figure 6b shows the result of picking. The picked curves are overlain directly on the data. We see that the algorithm picks the first arrival and several early arrivals quite well. Carefully looking at the picked curve for the first arrival, we see that a ``pull-down'' of the first arrival at offset -0.6 km is correctly picked. This ``pull-down'' probably is caused by a low velocity anomaly of the medium. This observation proves the high resolution and high accuracy of our algorithm. At the lower-left corner of Figure 6, events interfere with one another. We see that the picked curves more or less follow the relatively strong events. Since this situation violates our assumption, we do not expect to have good picking. The same thing happens at offset 0.7 km and arrival time 0.76 sec, where two events merge to one.

Our second example uses a common midpoint (CMP) gather. We want to demonstrate the power of our algorithm for moveout corrections. In Figure 7a, we show this CMP gather together with the picked curves. The events between the second and third picked curves are strongly aliased at far offset. The result of moveout correction is show in Figure 7b. We see that events are flattened nicely. To compare our method with normal moveout (NMO) corrections, we apply two methods to the same CMP gather. The results, displayed in Figure 8, are similar except that NMO correction makes the wavelet stretch out at far offset whereas our method does not. In this example, the method described in this paper gives good result because the signal-to-noise ratio of data is high. When noises are strong, NMO will give better result than does our method.

Figure 6

Figure 7

Figure 8

1/13/1998