The CO2 storage operation at Sleipner in the Norwegian North Sea provides an excellent demonstration of the application of time-lapse surface seismic methods to CO2 plume monitoring under favorable conditions. Injection commenced at Sleipner in 1996 with CO2 separated from natural gas being injected into the Utsira Sand, a major saline aquifer of late Cenozoic age. CO2 injection is via a near-horizontal well at a depth of about 1012 m below sea level (bsl) some 200 m below the reservoir top, at a rate approaching 1 million tonnes (Mt) per year, with more than 11 Mt currently stored.

In time-lapse analysis, we have to distinguish the seismic response changes due to oil and gas production at a reservoir over the years from several other causes, such as the recording signature and random noise. In this paper, we focus our attention on the velocity macromodel provided by seismic tomography, which is a basic tool for the data regularization, its depth or time migration, anda possible final subtraction among different vintages. We show first that we cannot use just a single velocity model for all data sets, because of seasonal variations of the overburden velocity (which is mainly due to seawater temperature in marine cases and to the water table depth in land cases). However, we can exploit the basic assumption of time-lapse analysis for constraining reflection/refraction tomography, i.e., by imposing the constraint that the layer structure and the local velocities do not change outside the reservoir (and in the shallowest part) over time. We thus get coupled models that are physically consistent, with a better spatial coverage and higher information redundancy. The new method is illustrated by a marine case history from the North Sea.