Abstract: The bottom of the deep water area of the Indian Ocean is rather flat, and the main multiples are usually related to the multiples created by strong energy and wide frequency wave band, which are difficult to be removed by the conventional generalized free surface multiples prediction technique. In order to solve the problem, the multi-wave model is improved with the generalized free surface multi-wave prediction technique, and thus the original data are separated into two parts, i.e. low frequency data and high frequency data. The low-frequency data may be converted into curvelet domain to optimize the multi-wave model. Reducing the multi-wave model directly from the original data, multiples are suppressed. The technique has been successfully applied in the deep waters of the Indian Ocean, as the correlation multiples of the seabed are suppressed, the effective signals highlighted, and the signal-to-noise ratio and quality of the profile obviously improved.