Enhancement of predictive deconvolution with a generalized stationarity transformation

The application of predictive deconvolution to seismic data corresponding t o intermediate water depths (i.e., 100-400 m) fails on nonzero offsets unle ss the data is preprocessed by a stationarity transformation, which regular izes the primary-multiple time separation. Currently, the use of a stationa rity transformation to aid predictive deconvolution is limited to first-ord er multiples. In this paper, we re-derive the first-order stationarity tran sformation using Wiener prediction theory and with the notation of statisti cal mechanics, show how the transformation can be extended to all orders of multiples. We show that when the multiples are nonoverlapping, the applica tion of the generalized stationarity transform to predictive deconvolution is a simple linear process, which allows one to use existing predictive dec onvolution software. Using a simple model data set, we compare predictive d econvolution results for the first-order stationarity transformation and th e generalized stationarity transformation. Results demonstrate a significan t improvement in the reduction of second-order multiples when the generaliz ed approach is used.