We compared the effect of varying the time-extents for prediction filters in the presence of high-amplitude noise. In a simple noise-only case, when we used the same number of coefficients in the lateral direction for both processes, f-x prediction passed about twice the noise energy as a short time-length t-x prediction. As the time-length of the t-x prediction filter increased, the t-x prediction passed more random noise. When filters with a time-length comparable to the data time-length were used, we found almost no difference between the filters or between the results of t-x prediction and f-x prediction. Thus, because of its ability to limit the filter length in time, t-x prediction has a definite advantage over f-x prediction in removing random noise.
This difference in passing random noise is shown in Figure 2, where the results of the t-x prediction and f-x prediction are applied to a two-dimensional stacked section. We used a time length of five coefficients for the t-x prediction result and a spatial length of five coefficients for both the t-x and f-x prediction results shown in Figure 2. The window sizes in both cases were 30 traces by 300 samples, or 0.6 seconds in time. This data set contains a moderate amount of noise and was one of the few nonsynthetic data sets showing recognizable differences between t-x prediction and f-x prediction. While the results are similar, the f-x prediction passes somewhat more noise than t-x prediction.