Because petrophysical properties vary among lithofacies,fundamental to reservoir
characterization and the construction of a cellular reservoir model is the population of
cells with lithofacies.Determining the number of lithofacies classes and the criteria for
defining classes involved four criteria:(1)maximum number of lithofacies recognizable
by neural networks using petrophysical wireline log curves and other variables;(2)
minimum number of lithofacies needed to accurately represent lithologic and
petrophysical heterogeneity;(3)maximum distinction of core petrophysical properties
among classes;and 4)the relative contribution of a lithofacies class to storage and flow.
An optimal solution using these criteria resulted in eleven lithofacies distinguished on the
basis of rock type(siliciclastic or carbonate),texture(Folk(1954)grain size for
siliciclastics and Dunham(1962)classification for carbonates),and principal pore size
(visual estimate).In classifying dolomite rocks we did not consider depositional texture
but rather the present texture and pore size that is primarily a function of crystal size and
the presence or lack of molds of leached carbonate grains.Classes based on differences in
core petrophysical properties coincided well with major lithofacies classes of rocks,and
have fairly distinctive wireline log response to petrophysical properties,the principal
variables used for neural network prediction of lithofacies.Although defining more
classes might have improved petrophysical prediction accuracy,the inability of neural
networks to effectively recognize distinguish finer lithofacies classes discouraged limited
finer class distinctions(e.g.:discriminating between fine grained packstone and coarse
grained packstone).