Traditional CUBE (combined uncertainty bathymetry estimation) filter generally fails to estimate the depth of the water accurately, and produces blur terrain feature in underwater slope riprap area by the multi-beam sounding data filtering with poor ability to remove gross error.Based on the sounding data characteristics of slope riprap area, a secondary CUBE filtering algorithm is proposed. The key idea is to estimate the plane position and depth synchronously, taking account of the transfer model of plane and water depth uncertainty related to terrain gradient. It is based on the second depth estimation model which considers geometric gradient and an optimization method of reference depth to accurately estimate the depth of water in underwater riprap areas and to detect gross error automatically.A set of meaningful experiments are also presented. And the experimental results have shown that the proposed method provides an automatic and accurate elimination for gross error. Specifically, the accuracy of the proposed algorithm is a few times higher than that of the traditional CUBE. It also demonstrates that the proposed filter shows superiority in preserving terrain features with better efficiency and robustness.