Geoscience Reference
In-Depth Information
Generate concepts
(work beyond the data)
Forecast
Identify uncertainties
Generate models
using geostats
Model build
Rich statistical algorithms
Forecasts
Work up data
Fig. 2.18
The data-driven approach to reservoir
modelling
Fig. 2.19
The concept-driven approach to reservoir
modelling
statistical guidelines for a model build. In reser-
voir modelling we are typically dealing with
much more sparse data, an exception being direct
conditioning of the reservoir model to high qual-
ity 3D seismic data (e.g. Doyen
2007
).
Analternativeistotakeamoreconcept-driven
approach (Fig.
2.19
). In this case, the modelling
still starts with an analysis of the data, but the
analysis is used to generate alternative conceptual
models for the reservoir. The reservoir concept
should honour the data but, as the dataset is statis-
tically insufficient, the concepts are not limited to
it. The model build is strongly concept-driven, has
a strong deterministic component, and less empha-
sis is placed on geostatistical algorithms. The final
outcome is not a single forecast, but a set of
forecasts based on the uncertainties associated
with the underlying reservoir concepts.
The difference between the data- and concept-
driven approaches described above is the expecta-
tion of the geostatistical algorithm in the context
of data insufficiency. The result is a greater
emphasis on deterministic model aspects, which
therefore need some more consideration.
2.5.2 Different Generic Approaches
To emphasise the importance of user choice in
the approach to determinism and probability, two
approaches to model design are summarised
graphically (Fig.
2.18
).
The first is a data-driven approach to
modelling. In this case, the model process starts
with an analysis of the data, from which statistical
guidelines can be drawn. These guidelines are
input to a rich statistical model of the reservoir
which in turn informs a geostatistical algorithm.
The outcome of the algorithm is a model, from
which a forecast emerges. This is the approach
which most closely resembles the default path in
reservoir modelling, resulting from the linear
workflow of a standard reservoir modelling soft-
ware package.
The limit of a simple data-driven approach
such as this is that there is a reliance on the rich
geostatistical algorithm to generate the desired
model outcome. This in turn relies on the statis-
tical content of the underlying data set,
yet for
most of our reservoirs, the underlying data set is
statistically insufficient
. This is a critical issue
and distinguishes oil and gas reservoir modelling
from other types of geostatistical modelling in
earth sciences such as mining and soil science.
In the latter cases, there is often a much richer
underlying data set, which can indeed yield clear
2.5.3 Forms of Deterministic Control
The deterministic controls on a model can be
seen as a toolbox of options with which to realise
an architectural concept in a reservoir model.
