INTRODUCTION
Oil recovery in Block Shen95 is estimated at 11.96% in 2006 based on 1998 volumetric estimate of oil originally inplace. The block is marked by low water injectivity and the inability to properly history match field performance. These are likely due to the high pour point nature of the crude oil in the block. The occurrence of paraffin precipitation at the injection well vicinity is due to cold waterflooding and regions of pressure reduction below bubblepoint pressure may cause permeability reduction for this highly compartmentalized block by postdepositional faults.
As a first step, permeability has to be properly characterized towards obtaining
a reasonable and acceptable history match against performance forecast and simulation
studies of various enhanced oil recovery schemes. Permeability can be estimated
to an appreciable degree from the concept of hydraulic (flow) unit at cored
and uncored wells with the advantage of over 200 logged wells. Traditional regression
based approach results in average permeability with low data scatter compared
to core data (Wendt et al., 1986). Permeability
prediction from logs tend not to wholly honour permeability trend but is better
off with the application of discriminant analysis based on petrological variables
such as grain size (Jennings, 1999; Adams,
2005).
Hearn et al. (1984) introduced the concept of
flow unit to determine the distribution of rock types that most strongly control
fluid flow and defined a flow unit as a reservoir zone that is continuous both
laterally and vertically and has similar permeability, porosity and bedding
characteristics. In their study, stratigraphic sequence provided the initial
framework for flow unit delineation with petrophysical properties and petrographical
analysis. Flow unit has evolved from the initial conception by Hearn
et al. (1984) with the use of different parameters for delineation
which include reservoir process/delivery speed and k/Φ plotted as constant
lines on Pickett plot (Aguilera and Aguilera, 2002);
stratigraphic modified Lorenz plot (Gunter et al.,
1997); R_{35} (pore throat radius measured at 35% mercury saturation)
from Winland’s equation with 4 basic flow units classified by Megaports
(R_{35}>10 μm), Macroport (210 μm), Mesoport (0.52 μm)
and Microport (R_{35}<0.5 μm) (Martin et
al., 1997) and combined delineation parameters kh/μ (transmissibility),
Φc_{t}h (storativity) and nettogross (Ti
et al., 1995). Other flow unit delineation parameters and models
are flow zone indicator (FZI) discussed below; a generalized approach by Lawal
and Onyekonwu (2005), integrating FZI, R_{35} and r_{b}
(grain size) and statistical zonation based on permeability heterogeneity (Testerman,
1962). All these delineation methods are static based. Flowbased technique
exist but staticbased are widely used. The short coming of a recent flowbased
technique (Schatz and Heinemann, 2007) is the use of
upscaled reservoir properties in a geological model. This is unlike the use
of core and well log data in static based techniques.
The flow unit discriminator parameter presented by Amaefule et al. (1993) has theoretical footing from the concept of bundle of capillary tubes considered by Kozeny and Carmen. Kozeny and Carmen both cited in Amaefule et al. (1993). This incorporated the Darcy’s and Poiseuille’s laws for flow in a porous media and tubes and the concept of mean hydraulic unit. FZI is applied for flow unit delineation and no distinction shall be made between flow unit and Hydraulic Unit (HU), or hydraulic flow unit. Both flow unit and hydraulic unit shall be used interchangeably.
This study covers the FZI delineation concept from theory to practical application of HU delineation and prediction. Bayesian statistical method is used to predict HU at uncored wells having welllogs by considering different relationships between the welllogs. Furthermore, sensitivity analysis was carried out on different number of HU delineated categories with different combinations of welllog response discretization. The choice of optimal HU delineation and prediction scheme is verified. Verification of predicted HU in uncored wells is with respect to reservoir performance, plan view status map of HU distribution, interwell HU crosssection and perforation thickness linked to HU categories. As well as other traditional verification methods specifically for cored wells.
MATERIALS AND METHODS
HU theory: Amafule et al. (1993) presented
the method for the use of hydraulic flow units to divide rock facies as a result
of the considerable variation of permeability even in well defined rock type.
This method is based on mineralogy and texture which defines similar fluid flow
characteristics that is independent of lithofacies. Amafule
et al. (1993) used the concept of bundle of capillary tubes and gave
an equation which was rearranged to isolate the variable that is constant within
a HU. This presented the optimal use of the equation. The rearranged Eq.
1 is:
where, k is permeability (md); Φ_{e} is effective porosity; F_{s} is shape factor; τ is tortuosity; and S_{gv} is surface area per unit grain volume (μm).
where, RQI is reservoir quality index (μm); Φ_{z} is pore volumetograin volume ratio and FZI is flow zone indicator (μm).
Equation 24 are used to compute the functions
for plotting a loglog plot based on (Eq. 5) for RQI versus
Φ_{z}. All data that have similar FZI will be on a straight line
slope equal to one and the value of FZI is determined at the intercept of the
slope at Φ_{z}=1. Thus, all data on the same straight line can
be taken to have similar pore throat attributes (the same hydraulic unit) governing
flow. Moreso, from Eq. 6 below, permeability can be computed
for all those points on the same straight line with same FZI.
FZI incorporates mineralogy and texture to efficiently discriminate rock facies having similar fluid flow attributes as demonstrated by statistical techniques. Particularly, pore geometry attributes that influence zoning was verified with petrographic data, in situ condition on RQI and capillary pressure to characterize pore throat structure.
HU classification, prediction and verification: Classification follows
the detailed methodology presented by Abbaszadeh et al.
(1996) using graphical analysis of probability and histogram plots of log
FZI and Ward’s hierarchical clustering method. This is verified with R_{35}.
The traditional approach at predicting HU at uncored wells that have welllogs
relies on the Bayesian method (Abbaszadeh et al.,
1996). Other methods are Artificial Neural Network (ANN) (Aminian
et al., 2002, 2003) and the recent classification
tree concept (Perez et al., 2005). Of all the
applications of the classification tree concept to electrofacies, lithofacies
and HU prediction from welllogs, HU result was worst off. The Bayesian method
is applied to predict HU category from welllog based on established HU from
cored wells. It involves inferring the probable HU of a well at a particular
depth using welllog responses based on the HU probability database of discretized
welllog responses. The database is from the link of established HU at cored
wells to their respective welllog responses collectively. A problem of accuracy
arises as stated by Abbaszadeh et al. (1996)
on the number of bins required from the welllog discretisation exercise. It
is an issue for concern since accuracy increases with increase in number of
bins but increases computation time. Equation 7 shows the posterior
conditional probability p(HU_{j}x) for each HU category with respect
to a welllog response. And Eq. 8 is the corresponding HU probability
for independent multiple welllogs given a simplified expression for posterior
conditional probability. Equation 9 stands for intersection
of multiple welllogs. The application of the Bayesian technique to facies identification
similar to HU prediction in uncored wells by the study of Kapur
(1998) states a limit for increase of bins as regards to that study. Optimal
number of bins is related to the ratio of data size to number of bins. Furthermore,
the particular HU chosen is that of the highest probability. The application
of Eq. 7 and 8 are clearly stated by Abbaszadeh
et al. (1996) and Li and AndersonSprecher (2006).
While Eq. 9 is shown by Kapur (1998):
where, p(xHU_{j}) is the conditional probability of welllog response
x, for a particular HU and p(HU_{j}) is prior probability of occurrence
of HU from core data classification. The use of Eq. 7 is by
the addition of the posterior probability computed for each welllog suggesting
mutually exclusive multiple welllogs:
where, x_{i} represents welllog responses x_{1}, x_{2},…x_{n};
p(HU_{j} x_{1}, x_{2},…x_{n}) is posterior
HU probability with respect to independent multiple welllog responses; the
other variables are similar to Eq. 7:
where, P(x_{i}) is p(x_{1}∩x_{2}…….∩x_{n}) the probability of intersection of multiple welllog responses; p(x_{i}HU_{j}), p(x_{1}∩x_{2}…….∩x_{i}HU_{j}) prior and conditional probability of intersection of multiple welllog responses for jth HU_{j}; p(HU_{j}x_{i}), p(HU_{j} x_{1}∩x_{2}…….∩x_{n}) posterior jth HU_{j} probability considering intersection of multiple welllog responses.
Verification is largely a qualitative process. Optimal HU definition is first
of all selected based on statistical analysis. The optimal HU involves the comparison
of core permeability and that derived from various HU definitions with different
combinations of welllog responses discretization. Secondly, the chosen definition
and discretization is applied for prediction in all uncored wells. The accuracy
of predicted HU is examined qualitatively by tying reservoir performance to
HU distribution from interwell crosssection obtained from interwell HU correlation.
This approach was initially applied by Martin et al.
(1997) as a process of linking reservoir performance to identify flow unit
quality and distribution in the reservoir. It was basically tying performance
to the individual delineated flow unit thickness at well locations (Gunter
et al., 1997). In this study, the link is tied to performance curves
with interwell HU crosssection, connectivity, perforation face HU thickness
and nearby injection well influence. Thirdly, the tie of HU to lithology.
GEOLOGICAL SETTING
Block Shen95 is in Liaohe basin of the Jinganbao oilbearing structure belt
of Damintun depression in Xinmin city, Liaoning Province, NorthEast China (Fig. 1). The block is faulted, anticline in nature, oilbearing strata dipping NW,
major axis 6.8 km and minor axis 3.3 km, area of 15.8 km^{2} and structurally
above a naturally fractured carbonate reservoir. The block is divided into two
main subblocks North and South with respect to one of the major sealing faults
in the NE direction (Fig. 1). The block has a complex system of faults with
sealing, partial sealing and non sealing faults (faults are normal and growth
faults). Oilbearing strata is of the Shahejie formation of the Paleogene period
(Eocene) within the subsurface depth of 18002230 m and consisting of an average
porosity of 19% and low permeability of 3 mD. Reservoir depositional environment
is fluviodeltaic (having 12 depositional sedimentary environments) with relatively
small sand body and poor connectivity due to its complex depositional structure.
Sedimentation took place when the basin was stable prior to depression. From
geological model, the reservoir has 3 zones and 17 layers which are heterogeneous
with high variation in thickness and poor horizontal connectivity. The top zone
is marked by downward fining and the 2 other zones by downward coarsening grain
size. Vertical reservoir heterogeneity from Lorenz plot is about 0.8082 from
5 cored wells within the block under study. There is no gas cap but alternating
cycles of oilwater interval and edge water with negligible pressure support.
Formation water is composed mainly of fresh water.
Oil Originally InPlace (OOIP) estimate stands at 2084x10^{4} t by 1984 and 1440x10^{4} t by 1998 based on volumetric estimates. Exploration of the block commenced in late 1988 and by late 2006 the block had 78 production (47 active producing wells) and 31 injection wells (8 active injection wells) with an average daily injection rate of 33 m^{3} day^{1} and monthly instantaneous Voidage Replacement Ratio (VRR) of 1.04 compared to the peak injection rate of 125 m^{3} day^{1} and VRR of 4.1 with 9 injection wells (7 active injection wells) by early 1991. By late 2006, watercut was 56.98% and cumulative oil production was 172.29x10^{4} t.
RESULTS
Six cored well data are available for analysis (J17, J3769, J4367, J4975,
J53 and J74, Fig. 1). J53 is out of the boundary under study
and cored zone is not known and has no SP welllog. The other wells have SP,
AcousticAC (SWave) and true formation resistivity (RT) welllogs except J17
with deep resistivity electrical log (RL3D). J3769, J4367, J4975 and J74
are chosen for HU classification. All log data have already been depth shifted
and cored data was not filtered to honour vertical resolution with welllog.

Fig. 1: 
Cored well location with major faults indicated and block
location 
This is to preserve reservoir heterogeneity or it may have a negative impact
on the inherent pore geometrical attributes related to FZI. Table
1 shows the distribution of core sample points in the 3 zones and available
welllogs.
HU classification: Probability plot of FZI for each well shows common
division into 4 HU’s apart from that of J17 and J3769 with 23 HU’s
by visual observation with a straight line drawn through it (not shown due to
space restriction). To encompass the reservoir heterogeneity from the cored
wells, J3769, J4367, J4975 and J74 were combined for classification to enable
reasonable HU prediction from welllogs. Probability plot from this indicated
4 visually distinguishable HU categories (Fig. 2) and histogram
plot of FZI and logarithm of FZI showed superposition of more than 4 lognormal
and normal distributions. However, with Ward’s hierarchical cluster analysis
on SPSS^{®} optimal number of clusters were selected (5, 6 and 9
HU categories identified as HU5, HU6 and HU9, Fig. 2). The
selection process took into account the distribution of sample points to reasonably
prevent the domineering or overshadowing of larger data set of a particular
HU over another. RQI vs Φ_{z} were plotted for the 3 HU definitions
(HU5, HU6 and HU9) and for each HU category in the definition straight lines
with slope 45^{0} were drawn to Φ_{z} = 1 at the mean FZI
for that category (Fig. 3). The earlier plot coupled with
permeability (k) vs porosity (Φ) (Fig. 4) for each HU
definition illustrates the narrowing of data scatter in each HU category with
the increase of HU categories from 5 to 9 and the improvement of coefficient
of correlation for k_{HU} vs. k_{core} (HU5 = 0.9590, HU6 =
0.9679 and HU9 = 0.9856) especially for high permeability values. Table
24 are summary of core sample point distribution in each
HU category for each HU definition. Also, the superposition of pore throat radius
(R_{35}) with Winland’s equation on the classical division into
4 flow units shows encouraging results of the clustering process (Fig.
4).
HU prediction: Table 3 shows the Spearman’s rank
correlation of 5 cored wells within the boundary of interest with J17 least
appreciable compared to others. Both AC and SP logs are quite low.
Table 1: 
Well core data and logs 

*RL3D deep resistivity electrical log 


Fig. 2: 
HU classification for HU5, HU6 and HU9 



Fig. 3: 
HU delineation for HU5, HU6 and HU9 


Fig. 4: 
kΦ delineation and relationship plot with R_{35}
verification 
However,
SP and RT logs are used. Firstly, SP was linearly normalized to give shale volume
(Vsh) as welllog response range, minimum and maximum values vary considerably.
RT and Vsh are used for HU inference.
A database consisting of probability of occurrence for each HU category corresponding
to welllog response discretized data range is assigned to all HU definitions
of HU5, HU6 and HU9.
Table 2: 
HU5 classification data distribution 

Table 3: 
HU6 classification data distribution 

Table 4: 
HU9 classification data distribution 

This depends on the use of either intersection of multiple
welllogs, independent or mutually exclusive multiple welllogs. Different combinations
of Vsh and RT number of bins were used. The numbers of bins are RTA, 6; RTB,
8; VshA, 3 and VshB, 5. For RT, bin size was based on multiple of 2 and then
on logarithm of RT since it is not normally distributed and by observing the
distribution of the logarithm of RT for all wells collectively.
Predicting HU for each HU definition depends on the combination of welllog
data discretization based on 1 (VshA and RTA), 2 (VshA and RTB) and 3 (VshB
and RTB) for constructing a probability database for HU inference. In other
words, HU prediction was carried out on HU51 (VshA and RTA), HU52 (VshA
and RTB), HU53 (VshB and RTB); HU61 (VshA and RTA), HU62 (VshA and
RTB), HU63 (VshB and RTB) and HU91 (VshA and RTA), HU92 (VshA and RTB)
HU93 (VshB and RTB). The outcome of HU prediction leading to k prediction
for some wells and application to predict HU and k for well J17 (not included
in combined cluster analysis and probability database buildup) and the corresponding
k prediction from kΦ relationship based on each sequence stratigraphy
zone is presented in Fig. 5 and 6 for correlation of predicted
permeability to core permeability. Figure 5 represents the
mutually exclusive concept (from each welllog by Eq. 7) and
Fig. 6 that of intersection of multiple welllogs by Eq. 9.
Comparing Fig. 5 and 6 give a glaring preference for the mutually
exclusive multiple welllogs over that of the intersection of multiple welllogs
consideration. Optimal k prediction is obtained with HU51 and HU63 having
the latter as the best combined k prediction. HU histogram for both is shown
in Fig. 7. Selection of the best HU definition and combination of welllog responses
discretization scheme with the aid of Fig. 5 is highly subjective. The choice
of HU definition can be narrowed down to a few but not essentially the best.
A good HU prediction fit from HU histogram plot should be a spread of data frequency
narrowed to the main diagonal which for the plots shown may be reasonable and
acceptable.
The major distribution of the frequency plot is within the vicinity of HU category
with high frequency as shown in Table 13.
HU histogram plot is rather a qualitative means approach.

Fig. 5: 
Correlation between predicted and core permeability for different
HU inference combination (mutually exclusive multiple welllogs case) 

Fig. 6: 
Correlation between predicted and core permeability for different
HU inference combination (intersection of multiple welllogs case) 


Fig. 7: 
Statistical HU Comparison for cored wells J3769 and J4367
including well J17 by HU51 and HU63. (a) HU51: J3769, (b) HU51: J4367,
(c) HU51: J17, (d) HU63: J4367, (e) HU63: J3769 and (f) HU63: J17 
Figure
6 shows the HU histogram for individual wells with HU51 and HU63. And Fig. 810 shows HU prediction, k prediction
and correlation plots for some wells including J17 not included in the combined
HU classification and probability database construction. Prediction of k for
J17 is reasonable considering that it was not included in HU classification
and inference buildup, but exhibited less number of HU categories from FZI
probability plot discrimination analysis.
On observation of the statistical HU comparison plot for cored wells and Table
24 one can draw a conclusion. The HU distribution (Fig.
7) depends on the core sample point distribution in each HU category for
each HU definition as seen in Table 24.
Also, correlation between predicted k and core k which depicts predicted HU
accuracy is hinged on the degree of correlation between welllog responses and
either FZI or k. Wells with high correlation coefficient between welllog responses
and FZI showed better prediction power as shown in Table 5
and Fig. 5. HU51 may be a better predictor from the cluster
of histogram to the main diagonal shown in Fig. 7 compared
to HU63. So, HU51 is chosen. Further analysis, prediction and verification
are therefore based on this HU definition.


Fig. 8: 
HU and permeability prediction Well J4367 by HU51 and HU63.
(a) HU51: J4367, (b) HU51: J4367, (c) HU63: J4367 and (d) HU63: J4367 


Fig. 9: 
HU and permeability prediction Well J3769 by HU51 and HU63.
(a) HU51: J3769, (b) HU51: J3769, (c) HU63: J3769 and (d) HU63: J3769 

Fig. 10: 
HU and k prediction Well J17 by HU51. (a) HU51: J17 and
(b) HU51: J17 
HU VERIFICATION
The choice of HU definition was based on the predictive power of permeability on cored wells. However, the verification of predicted HU at uncored wells having welllogs and further validation at cored wells relies on the link to dynamic and static data. The use of dynamic data of reservoir performance curves is linked to the perforation thickness in each HU category and interwell HU crosssection. This also takes into cognizance injection well influence based on connectivity to production well. It generally accounts for HU connectivity and distribution. Static based comparison is linked to lithology. This is applicable to only cored wells.
HU is computed for all logged wells numbering over 100 from the probability
database for HU inference. In 15 interwell crosssections covering the entire
block, 100 wells were used for HU correlation using the depositional layer structure
as a framework which was not entirely honored. A structural model based on this
correlation was built consisting of 51 layers. This model serves as the HU model
(Fig. 11).
Dynamic HU verification: For the purpose of dynamic HU verification
12 wells were used to correlate reservoir performance to HU. These wells are
grouped into H (high productive wells), M (medium productive wells) and L (low
productive wells) by qualitative analysis. Classification was basically on initial
production rate, cumulative oil production and watercut. This was tied to perforation
thickness in each HU category, HU distribution, thickness and connectivity from
interwell crosssections (Fig. 1214).
The classification into H (4 wells), M (4 wells) and L (4 wells) were clearly
discernable except that of 2 wells. All apart from 1 well classified into H
has relatively high perforation thickness in the highest quality HU that is
HU #1.
Table 5: 
Spearman’s rank correlation well log responses with FZI 

*RL3D (deep resistivity) 

Fig. 11: 
Fence diagram HU model 

Fig. 12: 
(ad) Link of reservoir performance to perforation thickness
for each HU category and HU distribution for high production wells 

Fig. 13: 
(ad) Link of reservoir performance to perforation thickness
for each HU category and HU distribution for medium production wells 

Fig. 14: 
(ad) Link of reservoir performance to perforation thickness
for each HU category and HU distribution for low production wells 
This was due to high connectivity with respect to 3 nearby injection
wells. The well in question, J3757 has no HU #1. This is a good account for
verifying prediction, in particular, the influence of interwell connectivity.
Classification of wells J4567 and J4668 into M does not justify the perforation
thickness in each HU category and also HU distribution. However, for well J4567
with respect to HU distribution, cumulative production and perforation distribution
this well may be classified into H. The isolation of this well by the partially
sealing faults may account for the low but steady production trend giving rise
to a possibly acceptable HU prediction. For J4668, both HU distribution from
cross section and perforation thickness at various HU categories can not explain
the inability to tie these to performance. In this case, predicted HU may be
questionable.
Static HU verification: This seeks to ground truth HU with available lithology from core. The well considered show high correlation to clay, silty clay and silt, specifically with the lowest quality (HU #5) and silt to a lesser degree with HU #3. HU#1 and #2 are rather correlated to gravel and fine sand and to a lesser extent with very fine sand. Considering well J74 shown in Fig. 15, it has significant correlation to lithology. However, HU does not entirely correspond to lithology but the flow characteristic of lithology. In addition, shown in Fig. 15 is the integration of R_{20} with a good link to HU. R_{20} is obtained from data local to the block. R_{20}, pore throat radius at 20% mercury saturation was the best fit to 35 sample points from two wells with a correlation of 0.948.
Figure 16 gives a summary of the workflow from HU classification
to verification as applied in this study.
DISCUSSION
There is a dearth on the discussion of the Bayesian approach to multiple welllogs
and its simplified applications on HU. Most references limit presentation to
a welllog without further explanation on application to multiple welllogs.
HU inference in this study was largely done by addition of HU probability from
individual welllog responses for similar HU category at the same depth. This
connotes that HU probability estimation is independent of each welllog as in
Tang and White (2008) application to facies, but mutually
exclusive.

Fig. 15: 
Link of HU to lithology for well J74; Track 1 HU, Track 2
Lithology, Track 3 R_{20} 
Therefore, each welllog can determine HU at a particular depth irrespective
of the fact that HU probability from other welllog might be zero or negligible
towards contributing to the sum of probability for a particular HU category.
Using the additive rule is a simplified approach. However, in comparison to
the intersection of multiple welllogs at the same depth, the simple additive
rule was slightly better off.

Fig. 16: 
Summary of HU classification, prediction and verification
workflow 
This is based on permeability prediction. Also,
independent multiple welllogs gave higher HU prediction accuracy but with the
inability to predict 2 HU categories out of the 5 HU categories from the chosen
HU5 definition. Both simplification schemes clearly depicts the success of independent
multiple welllogs as applied by Tang and White (2008) and Li and AndersonSprecher (2006) to facies prediction.
The increase in the number of bins of well log discretization yields better
overall correlation until a certain limit where further increase as in the case
of HU63 inference gives an overall poor result from predicted permeability
correlation with core derived permeability. This is in line with the study of
Kapur (1998) for facies prediction by the Bayesian method
and the statement by Abbaszadeh et al. (1996).
However, the increase in number of bins resulted in remarkable improvement of
accuracy for certain HU categories. In particular, that of the highest quality
HU category compared to reduction in accuracy for the low quality categories.
This may be related to the poor welllog responses correlation to FZI and/or
permeability. There is a drift towards the nondiscretization of welllog responses
with the independent multiple welllogs concept. The application results in
successful facies classification by the Bayesian technique in fitting prior
probability with probability distribution function (Beta and Gaussian distribution
function and Kernel density estimation, Tang and White (2008)
and Li and AndersonSprecher (2006)).
There remains, the non explanation of the failure to delineate HU #4 with welllogs during HU prediction for the chosen HU definition, HU51. The choice of HU51 definition over that of same number of categories (HU52 and HU53) and definitions with 6 and 9 HU categories (HU61, HU62, HU63; HU91, HU92 and HU93) is based on coefficient of determination of core permeability to permeability computed from the respective definitions.
The reasonable link of reservoir performance with HU distribution depicts the
relationship of the integration of reservoir performance to quality of flow
unit and its distribution. It attests to tying the spatial distribution of data
patterns of flow based attributes to similar performance regions by Martins
et al. (1997). As such, this significantly serves as a reasonable
means of qualitative HU prediction accuracy in uncored wells before the tedious
reservoir simulation process. From this point, it can be said that HU prediction
is relatively acceptable. However, for lithology, it was not entirely successful.
HU #5, the lowest quality hydraulic unit showed the best match with lithology
(clay, silty clay and silt). Not surprising, Amaefule et al. (1993) had
concluded that multiple flow unit exists within a particular rock type (lithofacies)
as pore geometry attributes can efficiently characterize HU. Ebanks
(1987) likewise stated that flow unit does not always coincide with lithofacies,
while Biniwale and Benhrenbruch (2004) stated that depositional
trends and diagenesis are the major factors that control HU quality.
The R_{20} correlation was suitable for computing pore throat radius using core permeability and porosity for verifying predicted HU. But, its use for verifying the classification of HU based on Ward’s hierarchical cluster analysis was not successful. This was as a result of the capillary data set not encompassing wide variability of reservoir heterogeneity. The situation is due to the low correlation of porosity to pore throat radius. Hence, Winland’s R_{35} equation was used. It is widely accepted as it contains large data set of varying sandstone particle size. The geological time of the sandstone is within the same range as that of Block Shen95 making it suitable for use.
The basic verification of the strength of flow unit delineation in uncored
wells is by simulation study, examples of case studies are Ti et al.
(1995), Gunter et al. (1997) and Svirsky
et al. (2004). This is time consuming. Svirsky et al. geostatistically
populated HU, stochastically across a reservoir and selected 3 realizations
based on P_{10}, P_{50} and P_{90}% recovery from uncertainty
analysis using streamline simulation. It may not be a true representation of
HU distribution across the reservoir and likewise it is time consuming. Interwell
HU correlation, though subjective as applied in this study seems a better approach
to populating the reservoir with further assessment of established HU by welllogs
and facies or depositional environment structural model.
CONCLUSIONS
Hydraulic Unit (HU) was delineated, predicted from well logs and validated
based on KozenyCarmen equation, Bayesian statistical method, reservoir performance,
petrophysical properties and lithology. The outcomes are:
• 
The mutually exclusive probability approach proved better
than the intersection of multiple welllog responses approach for prediction
of HU by the Bayesian technique. Whereas, independent multiple welllog
responses had better overall HU category prediction though failed to predict
2 HU categories out of the 5 HU categories 
• 
HU prediction in uncored wells with the aid of well logs is
to a larger extent an art based on finding the optimal welllog responses
discretization for HU inference and the number of HU categories with its
corresponding core sample point distribution 
• 
Permeability estimation may be considered satisfactory by
HU with respect to high reservoir heterogeneity, number of cored wells available
and poor welllog responses correlation to permeability and flow zone indicator 
• 
The integration of reservoir performance and HU distribution
at both well locations and interwell crosssections showed reasonable and
acceptable qualitative match for the wells considered. This is a means of
verifying predicted HU at both cored wells and uncored wells having welllog
data. The HU distribution was made possible by interwell HU correlation
using depositional cycles as a framework 
• 
The validation of HU with lithology was reasonable for high
quality HU and had high accuracy for the lowest quality HU which corresponded
to R_{20}, pore throat radius at 20% cumulative mercury saturation 
ACKNOWLEDGMENT
The researchers thank Liaohe Oilfield Company for the funding of this research.