|dc.description.abstract||Time-lapse seismic data quality has improved over the past decade, which makes
dynamic interpretation of the reservoir changes possible. To push the limits of this
technique further, this thesis studies the time-lapse seismic noise generated by overburden
heterogeneities, as well as its influence on quantitative seismic interpretation.
This is done by testing the accuracy of a multi-attribute pressure and saturation
inversion method in this context to gain insight into its performance in the case
of seismic acquisitions not being perfectly repeated. Extensive seismic modelling
studies are conducted in order to quantify the accumulated error for three different
Channels in the overburden above the Nelson Field, North Sea, are found to cause
errors in the time-lapse amplitudes. The magnitude of these amplitude errors decreases
with increased repeatability of the monitor survey’s source and receiver
positions. On average, saturation change is estimated to an accuracy of less than
6% when affected by amplitude errors only. However, these mean errors significantly
increase to more than 20% if the residual time shifts caused by the channels
are not removed from the seismic data. Moreover, the maximum saturation change
estimation error can exceed the production induced signal locally. In addition, a
major finding of this study is that the shape of the channel in conjunction with
the acquisition direction has a significant impact on the spatial distribution of the
errors at the reservoir level. It is also shown that the commonly used repeatability
measures of NRMS or Δsource+ΔReceiver do not correlate well with the spatial
distribution of areas with increased saturation change estimation error.
Consequently, a layer stripping method is presented which reduces the amplitude
errors caused by the overburden channel and the acquisition non-repeatability by a
factor of two. Nevertheless, the limits of using post-stack data to invert for timelapse
changes become apparent and, as a result, it is strongly advised to do further
research into applying this method to pre-stack seismic data.
Production-induced amplitude changes inside the stacked reservoirs of a deepwater
West of Africa field constitute the second overburden complexity studied. These
changes imprint on the lower reservoir channel and reduce the time-lapse amplitude
change locally by up to 42%. Furthermore, time-lapse amplitude errors are as large
as 38% in case that the velocity change inside the upper reservoir is not included in
the monitor migration velocity model. In addition, an important conclusion of this
study is that due to the high frequency assumption ray-tracing based seismic modelling
does not perform well for cellular models such as this West of Africa example.
Finite-difference modelling methods are strongly advised to be used instead.
Finally, the effect of overburden changes above the highly compacting Ekofisk chalk
reservoir, North Sea, is investigated by combining reservoir simulation, geomechanical
and ray-tracing models. The velocity change of the overburden rocks reduces
the time-lapse amplitudes at the top reservoir predominantly in the zone of vertical
displacements greater than six metres. In this zone, the mean time-lapse amplitude
errors in the full and far offset stack data are 9.4% and 4.23%, respectively. These
errors decrease below 2.3% in areas of less than six metres vertical displacement.
Consequently, the full and far offset stack amplitudes are not suited for quantitative
time-lapse interpretation. The time-lapse amplitudes for the near and mid
offset stacks are significantly less affected and the mean errors are smaller than 1.5%
across the entire reservoir. Therefore, these two partial stacks are recommended for
quantitative time-lapse interpretation.
Three different overburden complexities in the North Sea and West of Africa are
studied and prove to have a measurable impact on the time-lapse amplitudes. It is
shown that these errors affect the ability to estimate the saturation change and in a
way that is not entirely predictable from inferences using commonly used repeatability