Seismic Repeatability Benchmarks

Abstract The effects of various parameters on repeatability can be considered in making decisions on acquisition and processing parameters. For this reason we apply repeatability benchmarks based on the RMS difference of two data sets to various repeat marine seismic data. The datasets considered here come from two different areas, Foinaven and Viking Graben. We have two vintages of 3D Streamer and OBH data from Foinaven ('95 and '98) and 2D multi-component data, with and without cable redeployment, from Viking Graben. The repeatability study of these data sets provides quantitative measurements of the effects of acquisition parameters, data processing and survey matching methods. Introduction The oil and gas industry is using various types of time-lapse seismic data for 4D analysis to monitor reservoir changes. The most common data type is surface towed streamer data, which may have inconsistent parameters between base and monitoring surveys. Another type of time-lapse data is from seabed sensors. We apply the NRMS (Normalized RMS) repeatability benchmark to various repeat marine seismic data sets. This benchmark is based on the relative difference between each of two compared data sets. Another often-used repeatability benchmark is Predictability (Ref. 3). In this paper NRMS is computed at various steps of a matching procedure during which the amplitudes of each pair of data sets are matched and phase and time shifts between the two data sets are compensated. NRMS is evaluated both in the time and frequency domains, to determine the effects of acquisition and processing parameters both globally and for different frequency bands. Besides NRMS, for time lapse survey design the needed parameter is the Noise to Signal Ratio (NSR); NSR should be lower than production effects on seismic data. We estimate NSR from Normalized RMS. Repeatability benchmarks The repeatability benchmark used here are based on the RMS difference of two data sets normalized by the average RMS energy of the two parent data sets. We call this benchmark Normalized RMS (NRMS). Given two data sets A and B, NRMS for traces aj and bj would be: Mathematical equation (1) (Available in full paper) where j denotes trace location and i time samples, i=1,...,Nt. The lower bound of the NRMSj is of course zero in the case that aj = bj. The upper bound is 2. If traces ajj and bj are not related, but have the same average amplitude, it can be shown that the expected value of the NRMS is ?2. Once NRMS is computed, it can be shown that the Noise-to-Signal Ratio (NSR) is: Mathematical equation (2) (Available in full paper) This expression for NSR is strictly valid only if in data sets A and B noise has the same variance and signal has the same power. NSR tends to infinity when the two data sets are made of random noise and is 0 when data sets A and B are equal. Equations (1) and (2) are computed in the time domain but NRMS and NSR can also be computed in the frequency domain, applying (1) and (2) to the Fourier transform of ajj and ajj.