The Property Model - Reservoir Model Design - page 88

Table 3.3 Example object dimensions and correlation lengths from the Lajas outcrop model study
Facies Object length (m) Object width (m) Object thickness (m)

Meandering channels 1,000 300 2

Trough cross-bedded cannels 1,000 100 1.5

Mixed tidal flats 500 400 0.5

Correlation lengths for properties
within objects

Horizontal correlation length ʻ x , ʻ y (m) Vertical correlation length, ʻ z (m)

All facies 50-500 0.5-5.0

very simplistic manner - such as the assumption
that all channel objects have a constant porosity
and permeability - or can be done by “filling”
the objects with continuous properties using a
Gaussian simulation method. Each model element
is assigned the statistical parameters required to
define a continuous property field (using Sequen-
tial Gaussian Simulation) which applies only to
that element. This process can become quite com-
plicated, but allows enormous flexibility and the
ability to condition geological reservoir models to
different datasets for multiple reservoir zones.
This process is illustrated for an object-based
model used for stochastic simulation of perme-
ability for an outcrop model (Brandsæter et al.
2005 ). A section from the Lajas Formation in
Argentina (McIlroy et al. 2005 ) was modelled
due to its value as an analogue to several
reservoirs in the Hatenbanken oil and gas prov-
ince offshore mid Norway. The model is 700 m
by 400 m in area and covers about 80 m of
stratigraphy - it is thus a very-high resolution
model highly constrained by detailed outcrop
data. The model illustrates how an object model
(based on outcrop data) can be combined with
Gaussian simulation of petrophysical properties
(based on well data from the Heidrun oilfield
offshore Norway). Note that in this case we
assign reservoir properties to the outcrop model,
whereas normally we would make a reservoir
model assuming geological object dimensions
derived from outcrop studies.
Table 3.3 summarises some of the dimensions
assumed for the geological objects in this model.
Object lengths are in the range of 500 m to
1,000 m, with widths slightly smaller, while
object thicknesses are in the in 0.5-2 m range.
Thesevalueshavesomeuncertaintybutare

relatively well known as they are based on
detailed study of the outcrop. However, the corre-
lation lengths, ʻ z , that are required to con-

trol property distributions within objects are much
less well constrained. In this study, a plausible
range of correlation lengths (Table 3.3 )was
assumed and used as input to a sensitivity analy-
sis. The range was chosen to test the effects of
highly-varying or gradually-varying properties
within objects (Fig. 3.24 ). Sensitivity analysis
showed that oil production behaviour is very sen-
sitive to this value, alongside the effects of anisot-
ropy and facies model (Brandsæter et al. 2005 ).
In general, we expect there to be some property
variation within geological objects, therefore

ʻ x , ʻ y ,

ʻ x ,

ʻ y , object dimension. The question is how

much variation? The choice of correlation lengths
for property modelling is therefore very uncertain
and also rather important for flow modelling. In
practice, sensitivity to this parameter needs to be
evaluated as part of the model design. The value
range should be constrained to any available geo-
logical data and to evidence from dynamic data,
such as the presence or absence of pressure com-
munication between wells in the same facies or
reservoir unit.
A useful guideline is to test the following
hypotheses:
(a) Properties are relatively constant within geo-
logical objects: ʻ object dimension.
(b) Properties are quite variable within geologi-
cal objects:

ʻ z , <

1/3 object dimension.
(c) Properties are highly variable within geolog-
ical objects:

ʻ

1/10 object dimension.
Note that the grid size needs to significantly
smaller

ʻ

than the correlation length being

ʻ

modelled (e.g. 1/10 object dimension

would require a very fine grid).