Random noise reduction by FXY prediction filtering

Abstract

Random noise can cause problems with the interpretation of seismic sections and can degrade the performance of deconvolution, velocity analysis and migration. When the noise occupies the same frequency band as the signal the only way to attenuate the noise is to perform some 'averaging' process across adjacent traces. A very successful way of doing this on 2D data is with the method of FX prediction filtering. This paper describes the advantages that are gained by extending FX spatial prediction filtering for random noise reduction to 3D. First of all we give an outline of spatial prediction filtering in both 2D and 3D, pointing out the detrimental effects of the process on weak events, curved events and faulted events. It is then shown how the true 3D version of the method (which we call FXY prediction) performs better than the 2D version with respect to the above problems, with examples given using synthetic data. We find that FX and FXY filtering are both capable of reducing random noise on 3D data to the same extent, but that FXY is preferable because it gives less distortion of the geology.