Frequency-selection LSRTM

According to equation 3.13, the maximum frequency sampling for this field dataset is larger than 1 Hz, assuming average water depth 0.5 km, maximum depth 2.5 km. However, due to the fact that the shot spacing 37.5 m is larger than half of the dominant wavelength (30 m) in the water, denser sampling is needed than what equation 3.13 predicts. Empirical tests suggest a frequency sampling of 0.3 Hz. Therefore, each shot over iterations visits 208 frequency samples in the frequency domain. To form a supergather, all the shots are encoded with frequency-selection functions and blended together. For the next iteration, the frequency of each shot is shifted by 0.3 Hz to form a new supergather. The LSRTM algorithm iterates for 208 iterations to go over all the frequencies and the image is shown in Figure 3.14(b). The low-frequency artifacts in the shallow part and in the right bottom corner are removed. Furthermore, the resolution of the LSRTM image is enhanced compared to shot-domain RTM image, which leads to better delineation of the faults as shown in the zoom views in Figures 3.15 and 3.16. The drawback is that there are still some high-frequency noise present in the LSRTM image.