Beruflich Dokumente
Kultur Dokumente
Summary mura et al. 1995; Bahonar et al. 2007; Das 2007; Manrique et al.
The First-Eocene heavy-oil reservoir (1E) in the Wafra field is a 2007; Penney et al. 2007). The screening studies were confirmed
candidate for steamflooding because of its world-class resource by dynamic modeling and economic analyses, which led to the de-
base and low-estimated primary recovery. However, industry has cision for a steamflood pilot (Barge et al. 2009). The initial pilot
little experience in steamflooding carbonate reservoirs, which has small scale test (SST) consisted of a single 1.25-acre inverted five
prompted the staging of several 1E steamflooding tests, the latest spot with central injector and one temperature observation well
of which is the large-scale pilot (LSP) started in 2009. To assist in (TOW). The SST’s primary objectives were to test long-term
facilities design, to help understand expected performance in a injectivity and water-treatment technology as a prelude to the
very heterogeneous reservoir, and to provide input to early-deci- larger 16-inverted-five-spot-pattern pilot (2.5-acre spacing) known
sion analyses, numerical thermal simulation was used to generate as the LSP. The location of the two pilots is shown on the structure
probabilistic forecasts. When adequate pilot history was available, map for the 1E in Fig. 3. Steam injection at the SST began in 2006
the model was validated with probabilistic methods. and currently continues primarily to evaluate scale- and corrosion-
The LSP model contained 1.5 million cells, which allowed the mitigation options (Barge et al. 2009). Steam injection into the
maintenance of adequate resolution and proper boundary condi- LSP C-Zone, which is the focus of this paper, began in 2009. After
tions in the pilot area. Parallel computation enabled a probabilistic the demonstration of reservoir response to continuous steam injec-
workflow to be implemented with this large thermal model. tion, the project was recompleted to the B-Zone. Steam injection
In this paper, we highlight the methodologies and inputs used into the LSP B-Zone was scheduled to begin in October 2011. Fig.
to generate the probabilistic forecasts and validate the model. 4 shows a type log for the LSP and a crossplot of porosity and per-
Major results of this work include the following: In contrast to meability for all available core-plug data. Note that core-perme-
many greenfield forecasts, the LSP forecasts were conservative, ability values range up to 15,000 md.
likely because of the unique aspects of the forecasting methodol-
ogy, proper selection of uncertainty ranges, and the relatively
high density of input data for model construction; wide variations Geology
in production metrics were forecast, indicative of a highly hetero- The 1E is composed mainly of highly dolomitized, meter-scale
geneous reservoir; results indicated that the validated model depositional cycles of basal peloid dolopackstones (subtidal) that
adequately captured the global or statistical pilot heterogeneity, were capped by peloid dolowackestones and dolomudstones (Bach-
enabling proper capture of steamflood flow/drainage mechanisms; tel et al. 2011). The 1E production largely comes from dolomitized
and despite this heterogeneity, forecast oil-recovery levels were peloidal wackestones, packstones, and grainstones. The abundance
comparable with those observed in steamfloods in sandstone of evaporites and the low abundance of normal marine fauna suggest
reservoirs. that 1E deposition took place on a shallow, gently dipping, and gen-
erally restricted ramp environment. Subsequent chemostratigraphy,
magnetostratigraphy, and biostratigraphic work confirmed the initial
Introduction core and well log with stratigraphy for the 1E (Montgomery et al.
The 1E is in the Wafra field in the Partitioned Zone (PZ), Saudi 2011).
Arabia and Kuwait (Fig. 1). Discovered in 1953, it is the shallow- The diagenetic history of the reservoir is very complex and
est of five significant producing reservoirs at Wafra and contains begins with an early period of dolomitization/dissolution of depo-
world-class volumes of heavy oil—more than 10 bbl of 13 to sitional CaCO3. This episode was followed by a period of the dis-
19 API oil (Meddaugh et al. 2007). A simplified stratigraphic col- solution of dolomite and precipitation of anhydrite cement. The
umn for the PZ is shown in Fig. 2. The 1E averages a 750-ft thick- final late-stage period includes anhydrite dissolution and calcite
ness with a gross average porosity of 35% and a gross average cementation. Pollitt and Saller (2011) concluded that, on the basis
permeability of 250 md on the basis of core-plug measurements. of petrographic and isotope analyses, bacterial sulfate reduction
The 1E is a depletion-drive reservoir with partial-solution gas (BSR) was responsible for dissolving the sulfate nodules, the pre-
drive and limited aquifer support. The aquifer support is not suffi- cipitation of calcite cement and native sulfur, and the biodegrada-
cient to maintain reservoir pressure at the current production rate. tion of the oil. BSR also likely contributes to the high sulfur
Oil recovery by primary production to date is approximately 5%, content of the oil. The very high porosity in this reservoir is likely
which makes the 1E a large enhanced-oil-recovery (EOR) target. related to its shallow burial and early oil emplacement (Pollitt and
Screening studies (Hale 2002) showed that continuous steam- Saller 2011). Oil emplacement is not fully understood, but it is
flooding was the best EOR method for the reservoir, which placed believed to have occurred in one or more relatively recent phases
it among the few steamfloods to date in carbonate rock (Naka- and may be continuing at present (Patience et al. 2011). The per-
meability variability in the reservoir is related to depositional fa-
cies and stratigraphic position. There is little evidence of areal or
Copyright V
C 2013 Society of Petroleum Engineers
geographical variations in diagenetic-facies distribution at the
This paper (SPE 150580) was accepted for presentation at the SPE Heavy Oil Conference overall field scale (Pollitt and Saller 2011).
and Exhibition, Kuwait City, Kuwait, 12–14 December 2011, and revised for publication.
Original manuscript received for review 7 November 2011. Revised manuscript received for Some fractures occur in the reservoir, but a core and formation
review 31 October 2012. Paper peer approved 10 December 2012. microimage (FMI)-based interpretative study together with
Kuwait
Ara
Mina
South Umm Gudair
bia
Saud
nG
Wafra
ulf
Kuwait Partitioned Kuwait
Saudi Arabia
Zone Saudi Arabia
South Khafji
Fuwaris
Humma
Arq
Saudi Arabia
0 5 10 20
Kilometers
Fig. 1—Map showing the approximate location of major fields in the PZ. The large, central field is Wafra.
Miocene
Oligocene Kuwait G
p
Eocene
Tertiary
Rus
First E
ocene
Seco
nd An
Paleocene hydrite
Seco
nd Eo
cene
Maastricht. Tayra
t
Sad
Senonian i an
dR
um
aila
Cretaceous
Ah
ma
di
Wa
Cenomanian ra
Bur
gan
Albian
Rat
awi
Rat
awi
ooli
Aptian Arab te
S
Hithulaiy
Barremian
Neocomian
Fig. 2—A simplified stratigraphic column for the Wafra field, PZ, Saudi Arabia and Kuwait, showing major producing intervals. The
First-Eocene reservoir is the shallowest reservoir at the Wafra field. The other major producing zones are the Second Eocene
(includes the informally named 1[3/4] eocene reservoir), Maastrichtian, Wara, and Ratawi oolite reservoirs. All the major reservoirs
are carbonate reservoirs except for the Wara sandstone reservoir (after Dvoretsky et al. 2012).
Static Model
SST
A detailed static model for the LSP was generated to support
dynamic modeling. The reservoir properties by stratigraphic inter-
val for the LSP area of the 1E ranged as follows: unit thickness—
7 to 57 ft; average porosity—24 to 43%; average water saturation
Ma
(Sw)—27 to 59%; and permeability 20 to 960 md (Meddaugh
in
et al. 2007). Table 1 provides a summary of the semivariogram-
Ar
model parameters used to build the LSP area static model. Geo-
ea
statistical analysis of the pilot project and surrounding wells
W
Core Porosity Vs. Permeability (log10) Type Log for First Eocene
5
VOL_UWAT_1
01 v/v
4 DEPTH CSW_2
CPOR_2
01
VOL_UOIL_1
v/v
10
TVDSS
VOL_GYPSUM_1
TOPS
0.50 V/V
SWIRR_WFS_2 01 v/v
FEET
FEET
CALI_1
3 020 IN
RXO
0.22000 OHMM
CPERM_1
0.110000 MD
10 V/V 0.50
VOL_UOIL_1
V/V 01
VOL_ANHYDR_1
v/v
GRN_1 RT PERM_3 SWT_1 VOL_UWAT_1 VOL_DOLOM_1
080 GAPI 0.22000 OHMM 0.110000 MD 10 V/V 0.50 V/V 01 v/v
2
Log10 Perm
400
B1A
1
0
450
1100 E1
SZ1
–1
–2 500
–3
–4 550
SZ1
1200
E12a
–5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 600
Porosity E12
Steamflood 650
1300 SZ3
Average Porosity and Sw Pilot TT
1.00
Completions 700
Mean PHIE of Stratigraphic Interval
0.90
0.80 Producers 750
1400
0.70 Injectors
0.60 800
0.50
0.40 850
1500
0.30
0.20 900
0.10
0.00 950
1600
0.00 100.00 200.00 300.00 400.00 500.00 600.00 700.00 800.00
Relative Stratigraphic Depth (Feet; EOC000 = 0)
1000 BODC1
Fig. 4—Right—Type log at right shows the vertical variability of the reservoir interval. Curves shown (left to right) are gamma ray
and caliper; depth with 2007 stratigraphic interval names, resistivity; core and regression-based calculated permeability; core and
calculated well-log water saturation; core and calculated well-log porosity; and log-derived mineralogy in which purple indicates
dolomite, yellow indicates gypsum, pink indicates anhydrite, green indicates oil saturation, and blue indicates water saturation.
SZ labels are shown at far right; blue arrow corresponds to approximate depth of only known vertical barrier to steam at the SST
pilot. Bottom left—Plot showing the variation of average porosity and Sw by stratigraphic depth. Red bar shows the location of SZ-
3 injection perforations, and green bar shows location of SZ-3 producer completions. Black and gray bars show the same for the
SZ-2 completions (injection into SZ-2 began in late 2011). Top left—Gives a summary of the core-plug data for First-Eocene reser-
voir. Note the large number of plugs with measured porosity greater than 50%. Maximal measured core-plug permeability is
approximately 15,000 md (from Meddaugh et al. 2011a, b).
core-plug data for intervals below the EOC900 surface; and Sw (2005a) and Meddaugh et al. (2007). The use of petrophysically
distributed by colocated cokriging with SGS constrained by strati- derived lithofacies as an additional constraint was investigated
graphic layer by use of the previously distributed porosity as soft but not ultimately used because of a relatively low correlation
data and the layer- specific correlation coefficient between poros- between predicted and actual lithofacies in the validation wells.
ity and Sw. The layer-specific correlation coefficients varied from The LSP model covers a 1.7 1.7-km area (Fig. 5), and it was
0.5 to 0.8. Additional details regarding the workflow and its generated with a 5-m areal cell size. The 5-m areal grid was used
application to this reservoir are provided by Meddaugh et al. to ensure that there were at least 10 to 15 cells between wells in
EOC000 EOC000
EOC050 EOC050
EOC100 EOC100
EOC180 EOC180
EOC200 EOC200
EOC210 EOC210
EOC300 EOC300
EOC400 EOC400
EOC500 EOC500
EOC510 EOC510
EOC600 EOC600
EOC620 EOC620
EOC700 EOC700
EOC800 EOC800
EOC900 EOC900
0.2 Porosity 0.6 0.0 Sw 0.8
EOC000 EOC000
EOC050 EOC050
EOC100 EOC100
EOC180 EOC180
EOC200 EOC200
EOC210 EOC210
EOC300 EOC300
EOC400 EOC400
EOC500 EOC500
EOC510 EOC510
EOC600 EOC600
EOC620 EOC620
EOC700 EOC700
EOC800 EOC800
EOC900 EOC900
Stratigraphic Layers in Model 1 Perm2 (log10) 4
Fig. 5—West-to-east cross-sections from LSP model (only immediate LSP area is shown; about 1600 ft horizontal, 300 ft vertical).
Shown clockwise from upper left are porosity, water saturation, permeability (log10), and stratigraphic layer. Ball markers show C
Zone (Steam Zone 3) perforations.
500
700
800
900
1000
1100
1200
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Saturation
Oil Match and 20% History Error Band Water Match and 20% History Error Band
3.5E+07 3.0E+07
Sim Cum Oil Sim Cum Water
3.0E+07 History Cum Oil 2.5E+07 History Cum Water
Cumulative Water (stb)
Cumulative Water (stb)
2.5E+07
2.0E+07
2.0E+07
1.5E+07
1.5E+07
1.0E+07
1.0E+07
5.0E+06
5.0E+06
0.0E+00 0.0E+00
1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010
DATE DATE
Fig. 8—Comparison of forecasts from the deterministically validated model and historical primary production oil (left) and water
(right) from 1958 to the startup of the LSP pilot. The gray areas represent error bands (620%) for the historical data.
Uncoded Uncoded
Factor Minimum Maximum Description
0.5
tions. Uncoded-factor minima and maxima were 1 and 1, respec-
0.4 tively. Uncoded-factor peak levels were zero, with the exceptions
0.3 of the f_kx_mult, f_ky_mult, f_soi_mult, and f_sorg_mult, which
0.2 were set to 0.3, 0.3, 0.8, and 0.333, respectively. A custom Excel
application was written to simplify/automate error-free generation
0.1 of the very large simulation decks used in this work.
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Results and Discussion
SW
Probabilistic Forecasts. Math proxies were generated for the 19
Fig. 10—Relative permeability curves represent the minimum responses shown in Table 3 by fitting the factors to those re-
and maximum wettability used in the design of experiment. sponses from the 60 Latin-hypercube cases with Gaussian-process
modeling. The modeling code also generates and ranks the impact
of factors on these responses, which are listed in Table 4. Charts
The eight uncertainty factors used in this work (Table 2) were showing the relative magnitude of factor impacts on some key
found during screening work to significantly affect one or more steamflood responses are shown in Fig. 11. Discounted cumula-
LSP steamflood responses. Factor ranges were determined by ei- tive oil (DCO) is controlled mainly by soi_mult, viso, and swir_-
ther physical measurements or the author’s experience with other mult. Peak oil rate is dominated by viso. Cumulative steam/oil
steamflood projects including the SST pilot. Those factors deter- ratio (SOR) is controlled mainly by soi_mult, swir_mult, and
mined by experience are generally much larger than those factors viso. SBT is controlled almost exclusively by kx_mult. Although
that could be constrained by sufficient physical measurements. interesting, such information is also valuable in that it shows the
This screening work included additional factors such as porosity factors that should be used in the upcoming model-validation
and Sw uncertainty, which were found to have little impact com- exercise.
pared with the impact caused by the uncertainty associated with Next, the proxies were used with Monte Carlo simulation to
the final eight factors listed in Table 2. Note that our analysis calculate P10/P50/P90 levels for each of the 19 responses. Desir-
showed that the porosity uncertainty is no more than 61 porosity ability-function-based MRO was then used to determine factor
unit, and the Sw uncertainty is on the order of 63 saturation units sets that, when plugged into the LSP simulation model, produced
(Meddaugh et al. 2005b). Some uncertainties normally included levels for all 19 responses that were nearly equal to the prescribed
in probabilistic studies, such as net reservoir cutoffs (e.g., poros- P10, P50, or P90 levels calculated previously with Monte Carlo
ity, Sw, Vshale), are not relevant to this study. Additional uncertain- simulation. These three factor sets are listed in Table 5.
ties (such as the number of and areal/vertical distribution of steam The three sets of factors from Table 5 were plugged into the
barriers and baffles) were not considered because there was no LSP model to generate P10/P50/P90 forecasts. Probability levels
Response Description
Factor Effect for DCO Interior 9 Producers Yr. 7 Factor Effect for Peak Oil Rate Interior 9 Prod. Yr. 7
NORMALIZED MAIN EFFECT
1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0
f_ f_ f_ f_ f_ f_ f_ f_ 0.0
so vis sw kz so kx so kr f_ f_ f_ f_ f_ f_ f_ f_
i_ o ir_ _m rw _m rg w vis so kx kz sw so kr so
m _m o i_ _m _m ir_ rw w rg
ul m ul ul m ul ul m _m
t ul t ul t ul t t ul
t t t t ul
t
FACTOR FACTOR
Factor Effect for Cum. SOR Interior 9 Producers Yr. 7 Factor Effect for 90% Producers Achieve Steam Breakthrough
NORMALIZED MAIN EFFECT
NORMALIZED MAIN EFFECT
1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
f_ f_ f_ f_ f_ f_ f_ f_ f_ f_ f_ f_ f_ f_ f_ f_
so sw vis kz so kx so kr kx sw kr so so vis kz so
i_ ir_ o _m rw _m rg w _m ir_ w rg i_ o _m rw
m m ul _m ul ul m m ul _m
ul ul t t t ul ul t
t t ul t t ul
t t
FACTOR FACTOR
RESPONSE RESPONSE
0.9
RESPONSE PROBABILITY LEVEL
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
RESPONSE
Fig. 12—LSP P10-50-90 models derived using method of Osterloh (2008) have multiple responses all near their desired levels.
heterogeneous. Several areally extensive bedded evaporative flood the B-Zone. This provided the opportunity to validate the
zones may act as barriers to the vertical migration of steam, but LSP C-Zone steamflooding model. In contrast to the all-too-often
this is highly uncertain at this time. As seen in Fig. 15, the nodular result that models overpredict oil production (Meddaugh et al.
evaporite zone that separates the C-Zone from the shallower B- 2011a), the P10/P50/P90 models all underpredicted oil and over-
Zone did impede the vertical migration of steam in sections of the predicted water (Fig. 19). This increased confidence that the deci-
model, but there were locations in which the permeability was not sion analyses that had been derived from these forecasts were
as low (i.e., a baffle), and steam did rise above the nodular evapo- perhaps conservative. However, this is tempered by the fact that
rite interval. In addition, within the C-Zone (and others), there are the more mature stages of the steamflooding could not be com-
areas with high permeability contrast, but which are not so great pared. Four possible reasons that the forecasts were not overly op-
that they act as barriers to steam vertical migration, as did the timistic were the following: the high input-data density for
evaporite. Instead, this characteristic causes steam to propagate constructing the static model, the relatively fine gridblock resolu-
laterally more extensively than in reservoirs without such high per- tion (e.g., minimum of 10 to 15 cells or more between wells), the
meability contrast, in which steam typically rises vertically very proper identification and inclusion of key uncertainty factors and
near the injector before eventually reaching a barrier and then good assignment of uncertainty ranges, and the models being
spreading laterally. The example TOW trace shown in Fig. 16 more physically realistic because multiple responses had the cor-
demonstrates this flow characteristic—two “convective events” rect probability level.
are seen, but they are vertically close together and merge quickly.
The convective events in the 1E, which occur in zones between
two evaporites, are thus not the same as the classic convective Level of Validation—Pilot or Individual Wells. The LSP vali-
events often described in conceptual steamflood models (i.e., dation was performed at the pilot level, not at individual wells.
steam flow in two or more hydraulically isolated zones). The num- The basis for this decision is briefly discussed.
bers in the blue boxes in Fig. 16 indicate the number of 1E-type An earlier project was conducted to determine the impact of
convective events observed at TOW locations in the model. semivariogram length on model forecasts (Meddaugh et al. 2010).
The highly heterogeneous nature of the 1E is also demonstrated The models used in this work were very similar to the LSP model.
by the wide variation in SBTs at model producers (Fig. 17). Addi- The models were in the same 40-acre location as the LSP, but
tional analysis revealed that several adjacent producers had among they included only the B-Zone and were not scaled up. Otherwise,
the earliest and latest SBTs. Yet another response indicative of the same workflow used to generate the LSP static model was
high heterogeneity is the large variability in oil-production fore- used to construct models for this project. To enable analysis that
cast for the 25 individual producers (Fig. 18). would result in statistically valid conclusions, multiple (25) geo-
statistical-model realizations were used to generate forecasts.
Comparisons of cumulative-oil-production responses for indi-
LSP-Model Validation. A decision was made in February 2011 vidual wells from the 25 models (Fig. 20) indicated that the
to halt the steamflooding in C-Zone and recomplete and steam- response variation at individual wells was very large. Coefficients
RESPONSE RESPONSE
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
RESPONSE
Fig. 13—LSP P50 models derived using older method have only one response (DCO) at the desired level; levels of other responses
vary seemingly randomly.
of variation (CV) for responses sampled in 2024 (standard devia- nically reasonable to use a single model realization to validate his-
tion/mean) had ranges of 0.25 to 0.62 for oil. The results suggest torical production at the pilot level.
that the effort to validate historical production at individual wells Further work demonstrated that the large variation in
with a single model realization would be of limited use. responses at individual wells was largely caused by permeability
In contrast, variation in oil responses for the pilot (sum of all heterogeneity resulting from the use of a cloud-transform (CT)-
producers) from the 25 models (Fig. 21) was small. CVs for based workflow to distribute permeability, with constraint at only
responses sampled in 2024 were 0.035 and 0.001 for oil and the few cored wells. Note that the CTs were constrained by strati-
water, respectively. This low variation indicated that it was tech- graphic layer.
P10 Model
Well Steam Injection Rate (stb/day)
0.20
50 200
0.10
0 0 0.00
2009 2011 2013 2015 2017 2019 2021 2023 2025 2009 2011 2013 2015 2017 2019 2021 2023 2025
Date Date
Fig. 14—P10-50-90 LSP forecasts with initial steam-injection rate of 500 stb/day. Left, mean oil rate for interior nine producers.
Right, oil recovery factor in the four interior C-Zone patterns. Oil rate curves indicate that the oil recovery mechanism in early years
is displacement but in later years is gravity drainage.
550 550
Z (Depth)
Z (Depth)
600 600
650 650
700 700
Fig. 15—Left, vertical permeability in a cross section (exaggerated vertical dimension) through a row of LSP injectors; blue arrow
indicates location of a low-permeability evaporite zone. Right, saturations in a vertical cross section through a row of LSP injec-
tors indicate steam has migrated vertically above the depth of the evaporite layer where the evaporite is not contiguous (a baffle);
steam vertical migration was stopped by the evaporite zone where it was contiguous (a barrier).
ST-900 ST-901
2 4
SI-880 SI-881 SI-882 SI-883
ST-914 ST-915
3 3
SI-892 SI-893 SI-894 SI-895
620
640
660
680
700
720
740
760
780
800
820
840
860
880
900
80 160 240 320 400 480
Temperature
Fig. 16—Top, number of convective events observed at temperature observation wells in the P50 LSP model; bottom, example tem-
perature observation well trace indicating occurrence of two convective events in the C-zone.
Cumulative Distribution
0.8
0.8
1-Jul-2024
0.7
Breakthrough 0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.1
0.2
0.0 0.1
0 12 24 36 48 60 72 84 96 108 120
0.0
Months for Producer Steam Breakthrough 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Well Cumulative Oil
Fig. 17—Wide variation in distribution of steam breakthrough
time (SBT) at producers in the P50 model is indicative of a
Fig. 18—Wide variation in well cumulative oil distribution (P50
highly heterogeneous model.
model; 500 stb/day initial injection rate) is indicative of highly
heterogeneous model.
Model Validation—Probabilistic Workflow. Injection wells
were rate constrained by distributing daily pilot steam-injection shown in Fig. 24. Proxies were generated for 57 (19 3)
rate (Fig. 22) and steam quality equally across all 16 injectors. responses by the monthly sampling of forecast cumulative pro-
Production wells were rate constrained by distributing daily pilot duced oil, water, and fluid. Response targets for the MRO were 57
fluid-production rate (Fig. 23) equally across all 25 producers. (19 3) historic monthly sampled pilot cumulative produced oil,
Starting model conditions were the same as those used previously water, and fluid. MRO produced numerous factor sets that, when
in the forecasting work, with the following exceptions. plugged back into the simulation model, produced forecasts that
The probabilistic validation workflow was very similar to the bracketed the historical data (not shown). Taken as a whole, the
workflow described previously for the P10/P50/P90 forecasts. factor ranges in these sets were still quite broad. However, the
The difference was that targets in Step 3 were the historic steam- results changed significantly when the sets were culled down to
flooding responses instead of Monte-Carlo-derived percentile lev- the two that provided the best matches between forecast and his-
els. Five uncertainty factors (Table 6) were used to generate a 40- torical data. As seen in Figs. 25 and 26, two of the factor sets pro-
run Latin-hypercube design. The five uncertainty factors were duced model forecasts of cumulative oil and water that matched
chosen because the aforementioned forecasting work showed that the historical data very closely. The factor levels for the two best-
they significantly affected early-time oil and water production. validated models are also shown in Figs. 25 and 26. The high
The other three factors in Table 6 with gray background were quality of the match is also evident in Fig. 27, in which simulated
held fixed at their P50 levels. and historical oil and water rates are compared. Interestingly, the
Note that the oil-viscosity range changed from the values used factors in both sets are quite similar, which when considered
in the probabilistic forecast; the maximal value was substantially along with the high quality of the match perhaps indicates that a
lower. This was performed after viscosity measurements of prop- distinctive solution was found.
erly cleaned and dried oil samples from throughout the 1E became Oil-recovery factors were plotted for forecasts generated by
available. None of these samples had viscosities as high as the ear- three factor sets from the MRO step—the most optimistic, the
lier data. Use of the lower maximal viscosity was further supported best-validated, and the most pessimistic. As seen in Fig. 28, the
by a lack of quality-control information for the samples with the recovery factor for the LSP C-Zone steamflooding ranged
much higher viscosity. Although not quantified in this study, the between 6 and 9% of oil in place present at the start of steam
use of the initial, too high maximal oil viscosity also likely contrib- injection. Instantaneous SOR was derived from the best-validated
uted to the conservative nature of the probabilistic forecasts. model plotted in Fig. 29. SOR during the last few months, when
steam injection was higher and more sustained, was approxi-
Model Validation—Results and Discussion. Cumulative pilot mately 4 to 5. This is a reasonable level for an early-stage
oil and water forecasts from the 40 Latin-hypercube cases are steamflooding.
3.0E+06 3.0E+06
Sim P10 Factors Sim P10 Factors
Sim P50 Factors Sim P50 Factors
2.5E+06 2.5E+06
Sim P90 Factors Sim P90 Factors
Cumulative Water (stb)
Cumulative Oil (stb)
Actual Actual
2.0E+06 2.0E+06
1.5E+06 1.5E+06
1.0E+06 1.0E+06
5.0E+05 5.0E+05
0.0E+00 0.0E+00
0.0E+00 5.0E+05 1.0E+06 1.5E+06 2.0E+06 0.0E+00 5.0E+05 1.0E+06 1.5E+06 2.0E+06
Cumulative Steam Injected (stb) Cumulative Steam Injected (stb)
Fig. 19—Left, comparison of actual (historical) and P10-50-90 forecast oil production vs. steam injected. Right, comparison of
actual and P10-50-90 forecast pilot water production vs. steam injected. Note that forecasts underestimated actual oil production
and overestimate water production. Resolution between simulation levels is slight because of immature nature of the pilot.
300000 300000
Cum Oil Prod (STB)
100000 100000
0 0
2009 2011 2013 2015 2017 2019 2021 2023 2025 2009 2011 2013 2015 2017 2019 2021 2023 2025
Date (YEARS) Date (YEARS)
400000
300000
Cum Oil Prod (STB)
200000
100000
0
2009 2011 2013 2015 2017 2019 2021 2023 2025
Date (YEARS)
Fig. 20—Cumulative oil production for three interior wells from 25 LSP-like P50 B-zone geostatistical models. Note the large
response variation at individual wells as a function of model realization. CV range at 2024 5 0.25 to 0.62.
As expected and explained previously, the differences between permeability is statistically constrained by separate porosity vs.
historic and best-validated cumulative oil at the end of the C- permeability CTs for each stratigraphic layer. The CTs are derived
Zone steamflooding were very large for all but a few wells. from the five cored wells in the pilot area.
Importantly, as seen in Fig. 30, the distributions of wells’ historic Comparisons of pilot and validated model TOW counts
and forecast cumulative oil were very similar. This suggests that grouped by temperature range are listed in Table 7. The compari-
the model global heterogeneity was very similar to that in the son shows that the well counts for pilot and model were equal,
pilot and that the model captured the global flow/drainage physics which provides additional evidence that the model captured the
of the steamflood pilot without the need of fractures or other high- global flow/drainage physics of the steamflooding pilot. As seen
extreme-permeability pathways. The results also suggest that, at in Fig. 31, steam did not propagate to any of the producers in the
the local level, the model does capture an overall statistically best-validated model, which is consistent with field observation
valid description of the local heterogeneity. Porosity is reasonably and provides additional evidence that the model captured the
well-constrained locally by well-log data from 56 wells, and the global flow/drainage physics of the steamflood pilot.
4E+06
7000 700
Cum Oil Prod (STB)
6000 600
3E+06
5000 500
3000 300
1E+06
2000 200
1000 100
0
2009 2011 2013 2015 2017 2019 2021 2023 2025
Date (YEARS) 0
Uncoded Uncoded
Factor Minimum Maximum Description
4E+06 1.2E+06
Water Produced (STB)
1E+06
3E+06
800000
Cumulative WPC_Norm)
OPC_Norm)
2E+06 600000
Cumulative
400000
1E+06
200000
0 0
2009.5 2009.75 2010 2010.25 2010.5 2010.75 2011 2011.25 2009.5 2009.75 2010 2010.25 2010.5 2010.75 2011 2011.25
Date (YEARS) Date (YEARS)
Fig. 24—Cumulative water (left) and oil (right) production from the five-factor 40-count Latin hypercube design cases used in the
LSP probabilistic model validation work. Math proxies were created by fitting the factors and 40 responses (sampled at monthly
intervals) using Gaussian process modeling; 57 proxies total, 19 proxies each for cumulative produced oil, water, and liquid).
These proxies were then used with desirability-function-based multiple response optimization (MRO) to find sets of factors that
gave the best overall match to the historical response targets.
1.5E+06
2.E+05
1.0E+06
1.E+05
5.0E+05
0.E+00 0.0E+00
Fig. 25—Historic oil production, two forecasts that best Fig. 26—Historic water production, two forecasts that best
matched it, and levels of the corresponding five uncertainty matched it, and levels of the corresponding five uncertainty
factors. Note factor levels are highly constrained. Shaded area factors. Note factor levels are highly constrained. Shaded area
is historic measurement error. is historic measurement error.
2,500 10,000
Best Validation Model PILOT WATER PRODUCTION RATE (STB/DAY) Best Validation Model
Historical 9,000 Historical
PILOT OIL PRODUCTION RATE (STB/DAY)
2,000 8,000
7,000
1,500 6,000
5,000
1,000 4,000
3,000
500 2,000
1,000
0 0
Fig. 27—Comparison of historic and best-validated LSP forecast oil (left) and water (right) production rates.
reservoir area), (4) using a modern dynamic probabilistic method large static and dynamic models, and (7) securing availability of
that created physically realistic models in which forecast levels dense, high-quality historic pilot production data.
for dozens of responses had relatively equal probability levels, (5) The validation process took advantage of earlier work that pro-
properly identifying and quantifying the impact of key process vided an understanding of what can and cannot be attempted with
uncertainties, (6) securing the availability of state-of-the-art paral- geostatistical models of the 1E—and why. This knowledge saved
lel computational facilities and codes that enabled use of very substantial time and effort by avoiding validating the model at the
Instantaneous Steam Oil Ratio (SOR)
0.1 30
Oil_Recovery_Factor
Recovery from most optimistic validation case
Oil_Recovery_Factor-Conf_3MRO_7_mtb
0.09 Recovery from best validation case
Oil_Recovery_Factor-Conf_MRO_9_mtb
Recovery from most pessimistic validation case 25
0.08
Oil_Recovery_Factor
0.07
20
0.06
0.05 15
0.04
10
0.03
0.02 5
0.01
0
0
2009.5 2009.75 2010 2010.25 2010.5 2010.75 2011 2011.25
Date (YEARS)
Cumulative Distribution
TEMPERATURE-OBSERVATION-WELL COUNTS GROUPED
0.8 BY TEMPERATURE RANGE
0.7
0.6 No. TOWs No. TOWs
(>140 F) (> 190 F)
0.5
0.4 Pilot 9 6
0.3 Best validation 9 6
0.2
Historical Data
0.1
Best Validation Model Steam Injection Rate HCPV Steam Injected Instant SOR
0 12000 6
0 10000 20000 30000 40000 50000
400
Validated
OPR
Model Forecast Validated
OPC Model Forecast
OPR-p10_2ft_700_nomaxstm_ovis_1 OPC-p10_2ft_700_nomaxstm_ovis_1
400000 P90 Model Forecast
P90OPR-p50_2ft_700_nomaxstm_ovis_1
Model Forecast OPC-p50_2ft_700_nomaxstm_ovis_1
Mean Interior Well Cum Oil (STB)
OPR-p90_2ft_700_nomaxstm_ovis_1
P50 Model Forecast P50 Model Forecast
OPC-p90_2ft_700_nomaxstm_ovis_1
320
P10 Model Forecast P10 Model Forecast
Oil Prod Rate (STB/DAY)
300000
240
200000
160
80 100000
0 0
2009.5 2011.5 2013.5 2015.5 2017.5 2019.5 2021.5 2023.5 2025.5 2009.5 2011.5 2013.5 2015.5 2017.5 2019.5 2021.5 2023.5 2025.5
Date (YEARS) Date (YEARS)
Fig. 33—Left, comparison of mean oil rate forecasts for nine interior LSP C-zone wells using best-validated and P10-50-90 models.
Right, comparison of mean cumulative oil forecasts for nine interior LSP C-zone wells using best-validated and P10-50-90 models.
Initial steam injection rate for these cases was 700 stb/D, and heat management was used.
Y X Y X
7 Months 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 5 Years 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Z Z
SP-820 SI-892 SP-816 SI-889 SP-812 SI-886 SP-808 SI-883 SP-804 SP-820 SI-892 SP-816 SI-889 SP-812 SI-886 SP-808 SI-883 SP-804
Y X Y X
10 Years 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
15 Years 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Z Z
SP-820 SI-892 SP-816 SI-889 SP-812 SI-886 SP-808 SI-883 SP-804 SP-820 SI-892 SP-816 SI-889 SP-812 SI-886 SP-808 SI-883 SP-804
Fig. 35—Steam saturation in NE-SW cross sections (exaggerated vertical dimension) through the best-validated LSP model at sev-
eral times; initial injection rate was 700 stb/day. In a few places, steam did rise through a baffle in the low permeability evaporite
zone (dashed yellow line). Steam flowed mainly laterally in early months, but in later years (post steam breakthrough) steam flow
was gravity dominated. Gravity drainage of oil is evident in later years around the three interior producers.