The shallow multiples are better eliminated in the image space. Since there is a very small coverage in offset for these multiples, when transformed to the image space and mapped into angle-domain common-image gathers, there is a wider distribution along the angle axis in the image space compared with the offset axis in the data space; therefore, subtraction performs better in the image space than in the data space.
For deeper multiples, although we were able to do a good job in the data space, there is still some energy remaining. The coherent noise is strong enough to produce an event in the image space that might interfere with future processing, such as, migration velocity analysis. Although, more detailed work can always be done in the data space to remove the multiples more carefully, the final stage and result are going to be in the image space. If we see coherent noise in the image space, we will be obligated to go back to the data space and re-process the data.
This re-processing is not needed if we do all our processing in the image space. The image space is where we want our final result to be coherent and interpretable. Furthermore, it is ideal for performing target-oriented processing and/or analysis, as, for example, focusing on an specific event or area to improve the image, like velocity or illumination problems.