Sensitivity of LSRTM to Errors in Migration Velocity

To study the sensitivity of multisource LSRTM to the background velocity model, another smoothed velocity model is generated by applying a triangle smoothing filter to the original slowness model with a window size of 800 m along both X and Z directions. Figure 2.10 shows (a) the smoothed velocity model and (b) the corresponding relative slowness perturbation. This model contains a larger slowness perturbation compared to Figure 2.3 and is used to migrate the same dataset with the conventional RTM method and the result is shown in Figure 2.11(a). The image quality is considerably degraded compared to Figure 2.4(a), which suggests that the conventional RTM method is sensitive to the accuracy of the migration velocity. The multisource LSRTM algorithm is applied with the new smooth migration velocity to one, four and eight supergather(s) to produce the images shown in Figures 2.11(b), (c), and (d) with the linear inversion scheme and 30 iterations. All the images are filtered with the same high-pass filter. Compared to Figures 2.4(b), (c), and (d), the new images are of lower quality, but the image obtained with eight supergathers (Figure 2.11(d)) shows suppressed migration artifacts, balanced amplitudes and improved resolution at shallow depths, but little or no crosstalk noise. The above results suggest that LSRTM is sensitive to the accuracy of the migration velocity, but in this example is more robust than the conventional RTM method.