Selecting parameters of predictive deconvolution and applications

Deconvolution process, one of the stages of stable stages of seismic data flows, is a mathematically inverse process, and it is usually used to increase the temporal resolution of seismic data. Predictive Deconvolution, a special type of general deconvolution, is frequently used to filter reverberation in seismic data particularly and short and long period multiples. The applicability and the performance of predictive deconvolution are based on two significant parameters, prediction distance and operator length. Predictive deconvolution filter is designed with these two parameters and it is applied to seismic signal to filter any types of multiples as convolutional.Both parameters are determined by analyzing the auto-correlation of seismic signal. Prediction distance is usually the parameter which controls the temporal resolution of output signal. If it is shorter, temporal resolution increases. However, it might damage primary reflections and therefore, Signal/ Noise (S/N) ratio decreases. Operator length is the parameter which controls the part to be filtered with predictive deconvolution and the performance.On the other hand, with traditional approach accepting defining these parameters as constant cannot provide desired filtering.A new approach has been developed to solve the problems encountered in the determination of the two significant parameters of predictive deconvolution in this study. The basic thought is formed with the decrease of time differences between primary reflection events on the autocorrelation of a shot gather passing the stages of pre-data process toward far offsets and the enlargement of source wavelet period. Detailed analysis of this approach was done on synthetic and real data and the applicability and the performance of predictive deconvolution were discussed in terms of cause and effect relations. Therefore, prediction distance was increased, while the operator length was shortened toward far offsets. As a result, it was observed that the use of the parameters based on offsets increased the performance rather than constant parameters to filter multiples.