Can thin beds be identified using statistical phase estimation?

Abstract

Robust interpretation of seismic data relies upon identification and understanding of wavelet phase. Typically, well logs are used for the estimation of seismic wavelets, whereby the phase is obtained by forcing a well-derived synthetic seismogram to match a seismic trace. However, well logs are not always available, can predict different phase corrections at nearby locations and, due to sparse spatial sampling, cannot be used to accurately estimate phase variation in 3D. We introduce an extension to a statistical kurtosis-based phase estimation technique, proven able to estimate phase in agreement with seismic-to-well ties, without the use of well data. Our extension allows temporal and spatial phase variation to be estimated directly from the seismic data. Application of this method to a series of synthetic datasets demonstrates its ability to estimate true non-stationary phase, as well as apparent local phase anomalies, from clean or noisy data. Further tests on real seismic data show that the method can be used to locate apparent local phase anomalies caused by geology, such as multiple thin beds. We conclude that phase anomalies can be detected using statistical phase estimation techniques and could supplement the standard amplitude interpretation methods.