Uncertainty in
Reservoir Parameter Estimation From Attributes
Swan, Herbert1, Rod K. Nibbe2,
Michael Faust3, Brian Russell4, Dan Hampson5
(1) ConocoPhillips Alaska Inc, Anchorage, AK (2) formerly ConocoPhillips Alaska
Inc, Anchorage, AK (3) ConocoPhillips Alaska Inc, Anchorage, (4)
Hampson-Russell, a Veritas Company, Calgary, AB (5) Hampson-Russell Software
Services Ltd, Calgary, AB
Seismic attribute calibration is a
commonly used procedure that identifies linear relationships between seismic
interval attributes and an averaged reservoir property, e.g. porosity, to
predict that property away from the wells. This paper focuses on criteria that
may be used to select which attributes should be used, and on a method to
quantify the statistical uncertainty of the resulting estimates, assuming the
attributes are normally distributed.
The standard criteria used to select an
optimal set of attributes are the correlation coefficient between predicted and
measured properties, the maximum prediction error at a single well, and the
root mean-square error at all the wells. These standard criteria are important
and valid but ignore the likelihood that one or more chosen attributes are in
fact uncorrelated from the reservoir property, despite an apparent (albeit
spurious) correlation with reservoir samples from a limited set of wells.
Therefore, the standard criteria must be balanced against the probability of
spurious correlation when weighing the various attribute combinations.
The standard criteria also fail to quantify
the inherent uncertainties of the predicted properties estimate. This can be
accomplished by computing a map of the prediction interval, which displays the
ranges of uncertainty of the prediction to a particular level of significance.
Combinations of attributes never seen at the wells (e.g., at a gas cap or
facies change) result in more uncertainty and correspondingly wider prediction
intervals. Attribute combinations that are highly correlated with each other
also yield larger prediction intervals.
Examples of the successful application of
this technique will be shown from the Alpine field of
AAPG Search and Discover Article #90063©2007 AAPG Annual Convention, Long Beach, California