Evaluation of a New 5D Seismic Volume Reconstruction Method: Tensor Completion versus Fourier Reconstruction

Multi-dimensional Fourier interpolators have become the industry standard for 5D seismic volume reconstruction. However, room for improvement exists and a few key aspects of seismic data reconstruction do require additional study. The latter includes stability in the presence of coherent noise and statics, recovery conditions for extremely sparse data sets and computational efficiency. This presentation addresses some of the aforementioned problems by introducing a new technique for 5D interpolation based on multilinear algebra. Prestack seismic data is organized in a tensor, which is assumed to be a low rank structure when the data is properly sampled. A practical algorithm is presented where tensor rank reduction permits to recover the missing traces and increase the signal-to-noise-ratio of the seismic volume. The technique is compared to a multi-dimensional Fourier interpolator. We have obtained encouraging results with synthetic volumes. In particular, numerical tests indicate noticeable gains in computational efficiency and reconstruction fidelity for very sparse data sets.