The similarity-weighted stacking approach can obtain a much improved stacking result than the equal-weight stacking considering the increased SNR, however, will still cause
energy damage when an inappropriate reference trace is used to calculate the similarity based weights. We proposed a new weighted stacking method that is based on principal component analysis (PCA). The principal of the method is to prepare an ideal NMO-corrected data matrix via low-rank approximation. The low-rank approximation, or in other words the principal components, is obtained via solving a low-rank constrained optimization problem via singular value decomposition (SVD). The proposed PCA-based stacking method can help alleviate the negative effects caused from abnormal trace, erratic and non-gaussian random noise existing in the data matrix, and thus is robust in field data processing. The proposed approach is tested via a synthetic CMP gather and a field data example, which shows very promising performance. Future research topics include substituting the current PCA framework with more sophisticated algorithms to make the obtained components statistically as independent as possible, such as Independent Component Analysis (ICA), where higher-order statistics rather than second-order moments are used to determine basic vectors and is proven to be stronger than PCA (Hyvärinen et al., 2001).