Abstract: Adaptive subtraction cannot fully attenuate surface-related multiples without damaging primaries when the two are difficult to separate. To address this issue, this study proposes an adaptive subtraction method for surface-related multiples based on local orthogonalization. A conservative subtraction strategy is first adopted to preserve primaries while retaining residual surface-related multiples. Using the local primary-multiple orthogonalization (LPMO) algorithm, weight coefficients are then estimated from the local correlation between residual and predicted multiples, which enables further suppression of residual multiples without damaging primaries. To improve the computational efficiency of the LPMO algorithm, a windowing strategy is introduced to divide seismic data into independent windows for separate processing, and then splice and fuse them based on their original spatial positions. This strategy enhances both processing accuracy and computational efficiency. Synthetic and field data tests demonstrate the method's ability to effectively suppress surface-related multiples while protecting primaries.