You are on page 1of 1

Spatial uncertainty, frequency uncertainty and their impact

on volumetrics and field development


Statistics that have great impact on reservoir modeling
and resource evaluation include frequency statistics and
spatial statistics (Ma et al., 2008). Frequency statistics is
especially important for the overall heterogeneity and mass
balance; spatial statistics is especially important in describing
the continuity, local heterogeneity, facies pattern, and
connectivity of the reservoir properties (Journel and Alabert,
1990). These two schools should be coupled in reservoir
modeling and uncertainty analysis (Ma et al., 2011).
Reproduction of the histogram in stochastic simulation is
such an issue that involves both frequency and spatial
statistics, and it has drawn significant attention (Soares,
2001; Robertson et al., 2006). However, honoring the
histogram of the data is generally not a good idea when a
sampling bias exists (Ma, 2010).
Figure 4a shows a good histogram match between the
porosity model in Figure 3g and the well-log porosity data.
But the model is actually biased as a result of the sampling
bias in the wells. The same can be said to the model in Figure
3d. Specifically, more wells were drilled in the eastern part,
wherein reef facies are dominant and porosity is generally
higher. Propensity analysis and subsequent facies modeling
can mitigate sampling bias, such as the models in Figures 2c
and 2d. Thus, use of such a facies model as a constraint to the
porosity model enables the mitigation of sampling bias.
Figures 4b and 4c show two porosity model realizations
generated using GRFSbased collocated cosimulation
constrained to the facies model in Figure 2d and the seismic
attribute (Figure 3e). Although the histogram of the model
does not match the well-log porosity histogram (Figure 4d),
the histogram matches between the model and the data are
actually good for each facies (Figures 4e, 4g, 4i). On the other
hand, although the global histogram match between the
model and the data is good for the model without mitigation
of the sampling bias (Figure 3g), the histogram matches are
not good for each facies (Figures 4f, 4h, 4j).

You might also like