Physical and financial constraints often cause poor sampling of seismic data. These missing data create problems while processing and compromise the final image quality. Different techniques have been proposed to reconstruct these missing data. Projection Onto Convex Sets (POCS) has shown to be an effective method for seismic data reconstruction. However, the main disadvantage of POCS method is its high computational cost. In the case of 5D POCS, each iteration needs two 4D Fourier transforms and the number of iterations is directly related to the threshold model. The traditional linear threshold model converges very slowly. The exponential model improves the convergence but it is only an approximation of the optimal threshold model. The data driven threshold model allows to obtain high quality interpolation results in a few iterations but it can be computationally expensive for 5D POCS method. We introduce a piecewise linear threshold approach adapted to the distribution of spectral amplitudes. The methodology is based on an efficient selection algorithm on distributed memory. Results demonstrate that this piecewise linear threshold model provides a good compromise between precision and computational cost. Presentation Date: Tuesday, October 18, 2016 Start Time: 4:10:00 PM Location: 148 Presentation Type: ORAL