Introduction

Time-lapse (or 4D) seismic monitoring of hydrocarbon reservoirs has seen tremendous growth throughout the 1990s and during this decade. In general, 4D seismic is based on the premise that changes in fluid content (e.g. due to production) cause changes in the acoustic properties of rocks which are detectable in recorded seismic data. A detailed review of the seismic properties of reservoir pore-fluids is given by Batzle and Wang (1992). Lumley (1995) gives a comprehensive review of the theory, caveats and applications of time-lapse seismic in reservoir monitoring. Also, in his review of methods and current applications of 4D seismic, Calvert (2005) outlined many of the acquisition, processing modeling and integration requirements for successful application of the technology. Since its adoption as a monitoring tool, many successful applications of time-lapse seismic monitoring have been published Biondi et al. (1998); Lefeuvre et al. (2003); Whitcombe et al. (2004); Zou et al. (2006).

Repeatability is a major consideration for successful application of 4D seismic monitoring, especially in reservoirs with very low seismic responses. Non-repeatability may result from differences in survey acquisition geometry and binning, cable feathering, tides, source-wavelet bandwidth and phase variability, differential static time-shifts, ambient noise, residual multiple energy, and relative mispositioning of imaged reflection events Johnston (2005); Rickett and Lumley (2001). Laws and Kragh (2000) and Eiken et al. (2003) discuss acquisition techniques that may reduce some of these uncertainties. Recent advances in time-lapse processing have improved the success rate in 4D seismic monitoring. Some of the common processing issues are discussed by Ross and Altan (1997) and Eastwood et al. (1994). Rickett and Lumley (2001) outline a cross-equalization scheme involving spatial re-alignment, matched filtering, amplitude balancing and warping. Co-processing or parallel processing (involving controlled amplitude and phase, early geometry equalization, and application of the same algorithms and parameters) of the different seismic datasets is now common practice Johnston (2005).

Although some of the best practices in time-lapse reservoir monitoring help improve the the reliability of time-lapse responses and confidence in their interpretations, many loopholes still exist. Most of these shortcomings may not be important in reservoirs with large seismic responses -- where such unwanted effects are submerged by the much stronger time-lapse response. However, in many scenarios (e.g. sub-salt reservoirs), slight inaccuracies may cause considerable spurious 4D effects. We envisage that with the gradual increase in demand for more optimal reservoir management (hence the need for more accurate amplitudes), and current changes in acquisition patterns (hence the need to utilize surveys with potentially widely varying geometries), circumventing the shortcomings in current time-lapse imaging practice will be necessary.

We briefly discuss some of the challenges in sub-salt reservoir monitoring and show preliminary (raw) 2D synthetic time-lapse images in the presence and absence of salt. Our goal is to attenuate contamination by artifacts caused by differences in acquisition geometries and also to correct for the weakened time-lapse response due to limited illumination. We discuss two target-oriented least-squares inversion approaches that may overcome some of these challenges.

In the first approach, based on previous work by Valenciano et al. (2006), the time-lapse image is given as the difference between two least-squares inverse images. In the second approach, we pose time-lapse imaging as an inverse problem and propose to directly solve for the time-lapse image by least-squares inversion. By solving for the time-lapse image through inversion rather than as a difference between migrated seismic surveys, we believe many undesired effects in seismic monitoring of sub-salt reservoirs -- and reservoirs with very complex overburdens -- could be removed.