Abstract
Reservoir properties are mainly determined based on well log information.
However, wells in most reservoirs are sparse and widely spread compared
to the size of the reservoir. Seismic data is thus one of the most
important complementary sources of information used to build 3D models
of hydrocarbon reservoirs. The need for a high quality reservoir
description starts as soon as a discovery is made. In the appraisal
phase, hydrocarbons in place and the amount of recoverable reserves
are estimated based on the reservoir model. Improved structural models
are also needed in optimal well placement during the production and
development phase of a reservoir. Knowledge about saturation and
pressure distributions in a reservoir are valuable both in the exploration
and development phase of a reservoir. This knowledge is used to evaluate
the size of a field, determine an optimal drainage pattern, and decide
on optimal well design to reduce risks for blow-outs and damage on
production equipment. Reducing uncertainty in reservoir property
estimates from seismic data have large economic impact on the development
of a hydrocarbon reservoir.Quantitative reservoir property information
can be obtained either through direct estimates of reservoir properties
from seismic data or through estimates of elastic properties (velocities
and densities) that are related to reservoir properties. The relationship
between physical properties of rocks and fluids and P-wave seismic
data are often empirical and non-unique. This leads to large uncertainties
in reservoir models derived from pressure wave seismic data alone.
Since shear waves do not propagate through fluids, combined use of
pressure wave seismic data and shear wave seismic data might increase
our ability to derive fluid and lithology properties from seismic
data. One way to obtain information about shear wave velocities over
a large area is to acquire multicomponent seismic data (for instance
x, y, and z component geophone data). Parts of this thesis focus
on methods to combine the information from multicomponent seismic
data with pressure wave (hydrophone) seismic data. In this way we
improve the accuracy in the estimates of pressure wave velocity,
shear wave velocity and density in the subsurface.To obtain information
about changes in reservoir parameters like fluid saturation and pore
pressure during production, comparisons between different vintages
of seismic data acquired over the field can be performed. Differences
in the seismic signal from the same area over a time period (time-lapse
seismic data) can be interpreted as changes in reservoir properties.
Benefits of improved reservoir characterization include ability to
locate bypassed oil and mapping of fluid fronts. This leads to saved
costs due to reduced number of misplaced wells, and increased production
because of optimized well placement. In the early days of seismic
reservoir monitoring, the analyses were qualitative, e.g. to identify
undrained areas, analyzing the sealing capacity of faults, and detect
drainage patterns. Today, time-lapse seismic analysis is still mainly
qualitative. To be able to obtain more quantitative estimates of
changes in reservoir properties from the time-lapse seismic data,
we need to establish links between the rock parameters and the seismic
data. I have used both time-lapse surface seismic data and time-lapse
multicomponent seismic data to estimate production related changes
in fluid saturation and pressure.Finally, to be able to utilize rock
physical information obtained from seismic reservoir characterization
in reservoir modelling, information about uncertainties in the estimates
are essential. One way to do this is to use deterministic models
(rock physics models) that relates reservoir properties to seismic
data, and assume that the model parameters are independent. However,
the variables in these estimations are inherently dependent and should
be treated as such. By formulating the problem in a Bayesian framework,
dependencies between the different variables and spatial dependencies
can easily be included. I have used both deterministic uncertainty
analysis and Bayesian estimation methods to quantify uncertainties
in the estimates. urn:nbn:no:ntnu:diva-564
Users
Please
log in to take part in the discussion (add own reviews or comments).