SPE-139032-PP Field Development Strategies for Bakken Shale Formation S.Zargari, SPE, S.D. Mohaghegh, SPE, West Virginia University

Cop y righ t 20 10, S o ciet y of P etr oleu m E ngine er s Thi s p ape r wa s pr epa re d fo r pre s ent ation at t he S P E E a ste rn Re gional M e et ing held i n M or gant ow n, W e st V i rginia, U S A , 1 2– 14 O ct obe r 201 0. Thi s pap er wa s s ele cte d for p re s ent ation b y an S P E pro gr am co m m itt ee follo wing r e view of inf or m atio n c ont ained in an a b str a ct s ub m itted b y th e aut ho r( s ). Co nte nt s of the p ape r ha v e not be en re vie wed by t he S o ciet y of P etr oleu m E ngi nee r s an d a re subje c t to co rr e ction b y t he a utho r ( s). Th e m a te rial doe s n ot n ec e s sa ri ly r efle ct a n y po sitio n of th e S o ciet y o f P et role um E ngine er s, it s offic er s, o r m e m be r s. E le ctr oni c rep ro du ctio n, dist ribu tion, o r sto ra ge of an y pa rt of t his pa pe r witho ut t he w ritte n con s ent of the S o cie t y of P e trol eum E n gine er s i s pr ohibit ed. P e rm i s sion to rep ro du ce i n p rint i s re st rict ed to a n a b str a ct o f no t m or e th an 300 w ord s ; illust ra tion s m a y n ot be co pied. T he ab st ra ct m u s t c ont ain con s pic uou s a c kno wle dgm ent of S P E co p yrig ht.

Abstract Bakken shale has been subjected to more attention during the last decade. Recently released reports discussing the high potential of the Bakken formation coupled with advancements in horizont al drilling, increased the interest of oil co mpanies for investment in this field. Bakken fo rmation is comp rised of three layers. In th is study upper and middle parts are the core of attention. Middle member which is believed to be the main reserve is most ly a limestone and the upper member is black shale. The upper member plays as a source and seal which has been subject to production in some parts as well. In this study, a Top-Down Intelligent Reservoir Modeling technique has been imp lemented to a part of Bakken shale format ion in Williston basin of North Dakota. This innovative technique utilizes a combination of conventional reservoir engineering methods, data mining and artificial intelligence to analyze the available data and to build a full field mo del that can be used for field development. Unlike conventional reservoir simulat ion techniques which require wide range of reservoir characteristics and geological data; Top-Down modeling utilizes the publicly availab le data (production data and well logs ) in order to generate reservoir model. The model accuracy can be enhanced as more detail data becomes available. The model can be used for proposing development strategies. The model is then used to identify remain ing reserves and sweet spots that can h elp operators identify infill locations. Furthermore, a pred ictive model was generated, history matched and economical analysis for some proposed new wells is performed.

Introduction Unconventional Resource s Oil and gas have been supplying a significant energy demand of societies during the last centuries. A considerable fraction of industrial and houshold energy demand has been supplied fro m oil and gas resources. Exploit ing energy resources has always been restricted to economy of the process and available technology. Oil and gas production has been started fro m shallo wer resources requiring fewer amount of investment and lower level of technology. As time passed, technology improvement and rapid demand of energy with increase in oil and gas price motivated the oil and gas industries to exploit deeper and more challenging resources. Conventional resources are those that are producible using the most developed technologies of present time (1). Unconventional resources are those that have not been consid ered economically feasible to be produced for decades. Unconventional resource plays require special development strategies and must meet growing challenges of water availability and transportation to produce.

2

Field Development Strategies for Bakken Shale Formation

SPE 139032

Bakken Formation Bakken shale formation is the most significant tight oil play in the United States. The Bakken formation is an oil-bearing stratum which covers parts of Montana and North Dakota in its US share. Production from Bakken started more than 50 years ago. Bakken formation is comprised of three distinct layers (Upper, Middle and Lower members). The Middle member of the Bakken Formation is very fine- to fine-grained arg illaceous, dolomitic sandstone to siltstone (2). The M iddle member lies between the two black Upper and Lower shale members of Bakken. Based on the latest release of USGS in 2008, the undiscovered resources of Bakken format ion in its US share is estimated to be 3.65 b illion barrels of oil and relevant amount of associated gas. The latest estimate was 25 times higher than the previous estimate in 1995 of 151 million barrels. The cause of increase in this estimated value was the unique success of hydraulic fracturing and horizontal drilling in this field. The Bakken format ion belongs to late Devonian / early Mississippian age, covering 200,000 square miles of Williston basin in North Dakota and Montana continued up to Canada. Figure below shows the location of Bakken formation in Williston basin.

Figure 1- Location of the Bakken Form ation in Williston Basin (3) Two different sections of the formation were selected for modeling Middle and Upper members of Bakken. One of the study areas is having all the wells completed in Upper Bakken and the other one in Middle Bakken. Locations of selected areas are shown in the following figure.

Location of M iddle Bakken M odel

Location of Upper Bakken M odel

North Dakota

Figure 2- Location of sections sele cted for modeling Upper and Middle Bakken layers in st ate of North Dakot a (4) Two different models are generated for the sections. The Upper Bakken model has an area of 68,000 acres and the Middle Bakken model’s area is 165,000 acres. Bakken format ion may be observed in a wide range of depths over North Dakota state. Bakken format ion in the study area of this research was found at the depth range of 9,000 to 10,6 00 ft. The thickness of Middle Bakken member increases as we go more North-West. The thickness of Middle Bakken member in this study area varies fro m 20 to 70 ft. The thickest part of Upper Bakken member in its North Dakota share is located at the center o f the basin. The thickness of

[SPE 139032]

Zargari, Mohaghegh

3

Upper Bakken member in the study area varies fro m 4 to 12 ft. Bakken fo rmation is a relatively tight sedimentary rock with lo w porosity and permeability. As a result of such low porosity and permeability, the recoverable reserve was estimated too low until latest report of USGS. The latest reports stated a significant change in recoverable reserve, influenced by technology advancement in the last decade. Well Completion and Stimulation in Bakken Formation The oil production from Bakken formation has significantly increased after the intensive application of horizontal drilling and hydraulic fracturing during the last decade (5). Horizontal lateral opens up greater exposure to the formation and hydraulic fracturing generates fractures which facilitate flu id flow to the wells. Hydraulic fracturing technology has been applied to latterly drilled wells (such cases present in this study are mostly comp leted in Middle Bakken member). Recent wells drilled in Bakken fo rmation are having as long as 20,000 ft of horizontal leg and several stages of hydraulic fracturing. Lateral length of the wells present in this study ranges fro m 1200 ft to 13,000 ft in Middle Bakken layer and hundreds of feet to 10,000 ft in Upper Bakken layer. Application of advanced drilling and completion technologies in Bakken formation is so expensive. Hydraulic fracturing process requires huge amount of injecting fluid wh ich is mostly water or gasoline base. Several pump trucks inject the fracturing fluid at high rate and pressure in order to reach the yield po int of target format ion. The most recent released reports of drilling cost in Bakken format ion obtain a cost of between 3,500,000 to $5,000,000 per well (6). Therefore, economical considerations are the mos t important issues in development of the field. Top-Down Intelligent Reservoir Modeling Reservoir simulation and modeling is the science of understanding the reservoir behavior and predicting the future of the fie ld. Reservoir models are set of measured data which are correlated by certain functions . In traditional reservoir simulat ion the flu id flow correlat ions keep functional relationship between reservoir characteristics, completion data and production constrains. In this study, Top-Down Intelligent Reservoir Modeling has been utilized in order to generate a cohesive reservoir model which is coordinated such that it can be implemented in field development. In Top -Down Intelligent Reservoir Modeling, solid reservoir engineering techniques are coupled with geostatistic, data mining and artificial intelligence ( 7, 8). History matching is an essential part of reservoir modeling. By history matching, the reservoir model will be tuned with the production behavior of the reservoir in the past. In conventional reservoir simulat ion, history matching is the process in which the predictive result of the model is compared with actual data. A recurrent process of modify ing the input parameters takes place until the best match is achieved. The history matched model will be used for forecasting the future behavior of the reservoir. Top-Down Intelligent Reservoir Modeling technique has a different approach to history matching than conventional reservoir simu lation. In conventional reservoir simu lation, flu id flow equat ions govern the relation between static and dynamic aspects of the reservoir, whereas in Top-Down Intelligent Reservoir Modeling the intelligent model estimates relat ionship in data. Reservoir characteristics and results of production analysis are emp loyed for building an intelligent model. Top-Down Modeling is comprised of several tasks which can be listed as follows:

       

Data Preparation Reservoir Boundary Identification and Delineation Volumetric Calculation Geostatistics Decline Curve Analysis Field Wide Pattern Recognition Intelligent History Matching Infill Location Determination

4

Field Development Strategies for Bakken Shale Formation



SPE 139032

Economical Analysis

Data preparation is one of the most crucial tasks when performing Top -Down Intelligent Reservoir Modeling. Reservoir characteristics and production data are the most frequent type of the data utilized in reservoir simu lation. Forecasting the production of existing and upcoming wells is a key to successful development strategies. Production forecasting vastly depends on the extent of our understanding about productivity and accessibility of hydrocarbon all around the field. Decline curve analysis technique is imp lemented to analyze the production behavior of the wells. The results of decline curve analysis will afterward be part of a data set which forms the res ervoir model. In Top-Down modeling we usually start with collecting the data and preparing the comp rehensive data set. Static and predictive models will be built afterwards and predictive model is history matched. Using different design tools such as fuzzy pattern recognition or estimat ive models, we are able to make development decisions and forecast the production. The result ed prediction will be afterwards utilized in the model for further predict ions (Figure 3).

Figure 3- Schematic flowchart of Top-Down Modeling process Reservoir Management Petroleu m reservoir management is the application of state of-the-art technology to explo it a reservoir while minimu m capital investment and operation cost is used to achieve the maximu m economic recovery of oil o r gas fro m the field ( 9). Reservoir management is co mprised of set of operations and decisions, by which a reservoir is identified, estimat ed, developed and evaluated from its explorat ion through depletion (1 0). Develop ment of a field primarily starts with drilling some exp loratory wells in the field. More wells will be drilled when productivity of the field is proved. There are many aspects which should be considered when making decisions for development of a field. Reservoir properties, geological and environmental consideration are the keys to field development. In this study, Top-Down Intelligent Reservoir Modeling has been utilized in o rder to generate a cohesive reservoir model which is coordinated such that it can be imp lemented in field development. Methodology Two concurrent approaches of Top-Down modeling were carried out in this study. One “Static Reservoir Model” and an “Intelligent History Matched-Predictive Model” were generated. In the static reservoir model, a set of volumetric and geo-models were built, then fuzzy pattern recognition was performed. This part of study is a combination of production analysis (decline curve an alysis (DCA)), production statistics, volumetric analysis, geostatistics and data clustering. Static models can provide us with general maps of the reservoir and verify distribution of reservoir characteristics. Geostatistical maps of DCA parameters and clustered cumulative production has been generated. Geostatistical results of production behavior wh ich provides us with grid based models can be used as estimative

[SPE 139032]

Zargari, Mohaghegh

5

models to determine production of new wells in the field. Econo mical analysis was also perfor med in order to obtain estimated rate of return fo r new wells. Reservoir characteristics, completion data and production strains and reports are employed for training an intelligent model. The model correlates static information with production data. A sp ontaneous process of history matching and creating a predictive model is carried out. An intelligent model is built wh ich carries comp letion informat ion and reservoir characteristics. The intelligent model is trained, calibrated and verified using the stat ic and dynamic informat ion of the field. The process of generating an appropriate trained intelligent model can be carried out consecutively until the least error in calibrat ion results is achieved. Static Re servoir Model Location of the existing wells is shown in the maps below. The reservoir model was delineated into 5 acre grids. Reservoir boundary was identified considering location of the wells. Using voronoi graph theory, the reservoir has been delineated into segments around the wells; each section has been assigned to be drainage zone of associated well.

Figure 4- Steps of reservoir boundary identification and delineation (Middle Bakken model) Production data from the wells have always been a valuable data for petroleum engineers. The production from a field is the final goal and brings lots of evidences with it to the surface. Flow rate is a response of nature to the producer. Trend of changes in production tells us about depletion of the reservoir. Declination models of production have been studied widely in the past. Arps (11) presented a set of rate-time decline curves. Arps decline equation is

Reservoir characteristics are measured from logs. Different log readings at the locatio n of the wells were measured to generate a well based data set for reservoir characteristics. Using geostatistics, well based data set was transformed to respective g rid based models. Reservoir characteristic maps have been generated as shown in following figures.

6

Field Development Strategies for Bakken Shale Formation

Density Porosity

SPE 139032

Neutron Porosity

Pay Thickness

Bulk Density

Gamma Ray

Resistivity

Figure 5- Geo models of measured log values for Middle Bakken member

TOC

Porosity

Pay Thickness

Figure 6- Geo models of measured log values for Middle Bakken member Vo lu metric analysis, which acquires its rudiments from results of decline curve analysis and reservoir characteristics, has been implemented to calculate original oil in place, remaining reserve and recovery factor. Well based and field wide recovery factors were calcu lated.

[SPE 139032]

Zargari, Mohaghegh

7

Fuzzy Pattern recognition After complet ion of abovementioned steps, fuzzy pattern recognition has been performed. This technology enables us to delineate the reservoir qualitatively. This technology is also applicable to recognition o f sweet spots in the reservoir. As result of pattern recognition approach, the reservoir will be delineated into different quality sections. Respective relative reserv oir quality index (RRQI) will be assigned to each delineated portion. In Fuzzy Pattern recognition, actual data is plotted along latitude and longitude of the model. Gray dots in figure below show actual data. As we can see in figure below, there is no specific trend recognizable in actual data. Using Fuzzy Logic, a patt ern is discovered in data (pink dots) as shown in the figure. Discovered pattern will be delineated into three clusters of High, Medium and Lo w qualities (both longitudinally and latitudinal). The intersection of delineated pattern and delineation lines are thereafter superimposed on a two dimensional map. The highest quality index region is where high quality longitudinal delineate meats high quality latitudinal delineate, is called High -High region.

Figure 7- Process of delineation in Fuzzy Pattern Recognition technology By superimposition of all the intersecting point on the two dimensional map, we will get all quality indexes. Different relat ive reservoir quality indexes will be recognized and the reservoir will be delineated qualitatively. Qualit y regions are painted in different colors to be distinguished easier. Darker co lors show higher quality regions.

Figure 8- Sample tornado diagram shows average values and number/percentage of wells existing in different RRQIs Statistical analysis on production rates and results of decline curve analysis provides valuable data. Clustered cumulative production in the forms of first few months and first few years of p roduction can generate a useful set of information relate d to production behavior. By mapping these clusters of production history as a model, a grid based estimated model of production behavior will be generated. Intelligent History Matching - Predictive Modeling Reservoir characteristics are components of fluid flow behavior of the reservoir. Porosity and permeability of the rock and fractures (natural or induced fractures) control the flu id migration fro m entire reservoir toward producing wells. A wide range of data might be availab le to engineers with clear eviden ce of effectiveness. In some cases, the data might not be

8

Field Development Strategies for Bakken Shale Formation

SPE 139032

in the form that traditional reservoir simu lators can manage to process nor is any direct method of interrelating them into mean ingful records is available. Intelligent process of modeling and history matching provides us with an unlimited capability of treating raw data. Organic nature of intelligent models provides us with such an unlimited liberty in feeding the model with any sort of information that could be influential in rate of flow. The model was provided with log readings of neutron and density porosity, bulk density of the formation, gamma ray and resistivity. Lateral length of the wells was also included in the data set. Effect of hydraulic fracturing process was also considered. Intensity of fracturing process was considered by utilizing measure of injected fracturing fluid’s volume and weight of injected proppant. For the case of Middle Bakken member, co mplet ion reports and well files were thoroughly studied to identify lateral length, number of lateral legs and intensity of fracturing process. Volu me o f injected fluid and weight of injected proppant were considered as a measure of intensity of fracturing. Production of two offset wells was also included in data set for each well. Co mplet ion and stimu lation reports coupled with production data and reservoir characteristics generate a comprehensive data set. Table below shows the list of data imp lemented for generating intelligent model of M iddle Bakken member. Table 1- List of data implemented in generation of comprehensive data set for the purpose of building intelligent model in Middle Bakken layer’s model Implemented Data Time

Pay Thickness(ft)

Latitude

Resistivity

Longitude

Vol. of Injected Fluid

Days of production(t)

Weight of Inj. proppant

Days of production(t-1)

q(t-1)-Gas

Days of production(t-2)

q(t-1)-Oil

Density Porosity

q(t-2)-Oil

Bulck Density

DOFP-Oil

Depth(ft)

q(t-1)-Water

Fracturing Job Index

Days of Production(t-1)(1P)

Gamma Ray

q(t-1)-Oil(1P)

Lateral length

Days of Production(t-1)(2P)

Neutron Porosity

q(t-1)-Oil(2P)

Generated data set was partitioned into three parts to be implemented in train ing, calib ration and verificat ion of intelligen t model. Part itioning was made in a time based manner. It means different periods of production history were considered to be used in different steps of the process. In the case of Middle Bakken model, production data was available as early as April 2006 until very recent (Ju ly 2010). Production data from April 2006 to February 2010 was used for training the network. This data set is comprised of 60% of available data. The rest of the data was utilized fo r the purpose of calibrat ion (3 months) and verification (3 months). Figure below shows the data partitioning in generation of M iddle Bakken model.

[SPE 139032]

Zargari, Mohaghegh

9

Figure 9- Data Partitioning for building and history m atching the Middle Bakken model Total organic carbon (TOC) is an indicator of potential hydrocarbon source rock in shales or silty shales. The best indicators of high TOC are high resistivity and relatively considerable porosity. There are different sources of organic matters in shales which generate different type of organic material. Tab le below shows some of the different types (12). Table 2- Name, type and source of different kerogens (12) M aceral Alginite Exinite Cutinite Resinite Liptinite Vitrinite Inertinite

Kerogen Type I II II II II III IV

Original Organic M atter Fresh water algae Pollen , Spores Land-plant cuticle Land-plant resins All land-plant lipids; marine algae Woody and cellulosic material from land plants Charcoal; highly oxidized or reworked material of any origin

In the case of Upper Bakken format ion, comp letion informat io n and reservoir characteristics were utilized as listed in table below. TOC as a measure of organic matter existence was also implemented. Table 3- List of data imp lemented in generation of co mprehensive data set for the purpose of building intelligent model in Upper Bakken layer’s model Implemented Data Time

Porosity from log

Latitude

q(t-1)-Gas

Longitude

q(t-1)-Oil

No. of Days of production(t)

q(t-2)-Oil

No. of Days of production(t-1)

DOFP-Oil

No. of Days of production(t-2)

q(t-1)-Water

Depth(ft)

No. of Days of Production(t-1)(1P)

Lateral length

q(t-1)-Oil(1P)

Pay Thickness(ft)

No. of Days of Production(t-1)(2P)

TOC

q(t-1)-Oil(2P)

In the case of Upper Bakken model, a long period of production data was available. Produc tion fro m the first wells has been started from June 1988. Field production was available until very recent (August 2010). Partit ioning of data was performed in a time based manner. Production from June 1988 until December 2001 was considered for training purpose. Production of years 2002 to 2004 were considered for calibrat ion and the rest (2005 to Aug. 2010) for verification purpose. Figure belo w shows the data partitioning in generation of Upper Bakken model.

10

Field Development Strategies for Bakken Shale Formation

SPE 139032

Figure 10- Data Partitioning for building and history mat ching the Upper Bakken model

Results and di scussion Static Re servoir Modeling The static reservoir modeling was performed for decline curve parameters (Q i , Di and b) and clustered cumulative production data for both strata (Upper Bakken and Middle Bakken). The static models of cu mu lative production show the statistical success of existing wells all around the reservoir (Figures 11 and 12). Generated estimated models are shown as simple geo -models in the following figures. Based on the resulting maps of estimated models sweet spots can be identified.

First 6 M onths Cumulative Production

First Year Cumulative Production

Qi

Di

b

Figure 11- Results of Statistically Estimated Models of Middle Bakken

[SPE 139032]

Zargari, Mohaghegh

First 3 Years Cumulative Production

11

First 10 Years Cumulative Production

Qi

Di

b Figure 12- Results of Statistically Estimated Models of Upper Bakken Results of Fuzzy Pattern Recognition In the case of Middle Bakken model, an extensive volumetric calcu lation was performed. Fu zzy p attern recognition for remain ing reserve in Middle Bakken model was performed for different time targets. Remain ing reserve is a function of original o il in place and decline rate. The fuzzy pattern recognition model of remaining reserve highlights the mos t prolific areas with higher initial oil in place (Figure 13). Well ranking analysis was also performed which provides us with ranking of the wells based on recognized RRQIs. The result of well ranking is also included in fuzzy maps in wh ich wells are show n in different colors of RRQIs that they belong to.

12

Field Development Strategies for Bakken Shale Formation

SPE 139032

Figure 13- Fuzzy pattern recognition of Remaining reserve in Middle Bakken model as of 2010 to 2020 These models provide us with another imp lement for recognition for sweet spots. The High-High region is where more productive wells and considerable amount of oil in place exist. There are also several High quality wells in High -Mediu m region next to High-High region which shows good potential in that area as well. Infill Drill Locations Co mparing results of statistical models with fuzzy pattern recognition may not be entirely co mpatible. We need to keep in mind that results of statistical models come fro m experience of dealing with the field in the past. Making the best interpretation by considering both models in order to better identifying location of sweet spots also requires considering some operational issues. Since we do not have access to operational data, decisions should be made based on available informat ion. Based on resulting models of fuzzy pattern and static models, several new wells were proposed which are shown in follo wing figures. Brown squares show the location of new wells.

Figure 15- Location of proposed new wells in Upper Figure 14- Location of proposed new wells in Middle Bakken model (15 new wells)

Bakken model (12 new wells)

Estimated results of production behavior of new wells in terms of decline curve components are obtained. Recovery factor and EUR of the wells and the field is also calculated. Estimated results from statistical models are shown in following tables. Table 4- Estimated results of production and recovery factor for the new wells in Middle Bakken Member

[SPE 139032]

Zargari, Mohaghegh

13

Table 5- Estimated results of production and recovery factor for the new wells in Upper Bakken Member

Economical analysis for new wells was performed in order to calculate net present value (NPV) as a measure of rate of return. Investment in any field should be associated with acceptable rate of return. Economical feasibility of field development depends upon amount of investing resources and rate of making benefits in the paying off period. Investments in an oil field are comprised of exp loration and drilling, surface facilit ies and transportation. Considering all the resources that we invest in a field at one hand, production rate and oil price on the other hand, we are able to perform economical analysis. Oil p rice was assumed to be $50 in these calculations. A range of 3,500,000 to $5,000,000 of drilling cost was utilized in analysis. Results of economical analysis for proposed new wells are shown in fo llo wing tables. Table 6- Economical analysis in Middle Bakken model for new wells

14

Field Development Strategies for Bakken Shale Formation

SPE 139032

Table 7- Economical analysis in Upper Bakken model for new wells

The results of economical analysis for Middle Bakken member are very pro mising. Estimated production rate for the new wells comp leted in this layer is considerable enough to have reasonable investment justification. Economical analysis of new wells completed in Upper Bakken member is quite challenging. Production fro m Upper Bakken layer in the area of study has been started from late eighties. Application of most recent drilling, co mpletion and reservoir stimu lation technologies may increase the production rate significantly. Intelligent History Matching – Predictive Modeling Intelligent models were trained, calibrated and verified using comp rehensive data sets. Co mparing the results of prediction and actual production is the best validator of history matching part of process. The min imu m but sufficient number of data should be used in building the model such that model is not overloaded by unnecessary data neither influential parameters are not supplied. Few examples of results of the predictive modeling in Middle Bakken model are shown below.

[SPE 139032]

Zargari, Mohaghegh

15

Oil Rate Well 17040 Actual Oil Rate

Predicted Oil Rate

Oil Rate Well 16905 Actual Cum

Predicted Cum

Actual Oil Rate

30,000.0

250000

Predicted Oil Rate

Actual Cum

Predicted Cum

20,000.0

140000

18,000.0 120000 200000

16,000.0 100000

14,000.0

20,000.0 150000

15,000.0 100000 10,000.0

Oil Rate (bbl/month)

Oil Rate (bbl/month)

25,000.0

50000

5,000.0

12,000.0

80000

10,000.0 60000

8,000.0 6,000.0

40000

4,000.0

20000 2,000.0

0.0 Sep-08

0 Dec-08

Mar-09

Jul-09

Oct-09

Jan-10

May-10

0.0

Aug-10

Jun-08

0 Sep-08

Dec-08

Time (month)

Oct-09

Jan-10

May-10

Aug-10

Figure 18- Results of history mat ching for the well number 16905 in Middle Bakken

Oil Rate Well 16671 Predicted Oil Rate

Jul-09

Time (month)

Figure 16- Results of history mat ching for the well number 17040 in Middle Bakken

Actual Oil Rate

Mar-09

Oil Rate Well 16936 Actual Cum

Predicted Cum

30,000.0

Actual Oil Rate 350000

Predicted Oil Rate

Actual Cum

Predicted Cum

25,000.0

160000

300000

25,000.0

140000 20,000.0 120000

200000 15,000.0 150000

10,000.0 100000

Oil Rate (bbl/month)

Oil Rate (bbl/month)

250000 20,000.0

100000

15,000.0

80000 10,000.0

60000

40000

5,000.0

50000

5,000.0

20000 0.0 Apr-07

0 Nov-07

Jun-08

Dec-08

Jul-09

Jan-10

Aug-10

Feb-11

Time (month)

0.0 Sep-08

0 Dec-08

Mar-09

Jul-09

Oct-09

Jan-10

May-10

Aug-10

Time (month)

Figure 17- Results of history mat ching for the well number 16671 in Middle Bakken

Figure 19- Results of history mat ching for the well number 16936 in Middle Bakken

Predicted and actual field wide cumu lative production of Middle Bakken format ion is shown in figure below. The production in the field starts increasing until maximu m nu mber of wells comes to production. Thereafter, the field production declines as time passes.

Figure 20- Field wide cumulative production (upper chart) and number of producing wells (lower chart) in Middle Bakken Few examp les of results of history matching in Upper Bakken model are shown below.

Oil Rate Well 9273 Actual Oil Rate

Oil Rate Well 12542

Predicted Oil Rate

Actual Cum

Predicted Cum

12,000.0

Actual Oil Rate 300000

7,000.0

10,000.0

250000

6,000.0

8,000.0

200000

6,000.0

150000

4,000.0

100000

Predicted Oil Rate

Actual Cum

Predicted Cum 300000

250000

Oil Rate (bbl/month)

Oil Rate (bbl/month)

5,000.0 200000 4,000.0 150000 3,000.0 100000 2,000.0 2,000.0

50000

0.0

0

May-88

Jan-91

Oct-93

Jul-96

Apr-99

Jan-02

Oct-04

Jul-07

50000

1,000.0

0.0

Mar-10

May-88

0 Jan-91

Oct-93

Jul-96

Time (month)

number 9273 in Upper Bakken

Oct-04

Jul-07

Mar-10

Figure 23- Results of history mat ching for the well number 12542 in Upper Bakken

Oil Rate Well 12484 Predicted Oil Rate

Oil Rate Well 12570 Actual Cum

Predicted Cum

Actual Oil Rate

7,000.0

160000

6,000.0

140000

Predicted Oil Rate

Actual Cum

Predicted Cum

10,000.0

600000

9,000.0 500000

8,000.0

120000

5,000.0

7,000.0 100000

4,000.0 80000 3,000.0 60000 2,000.0

40000

Oil Rate (bbl/month)

Oil Rate (bbl/month)

Jan-02

Time (month)

Figure 21- Results of history mat ching for the well

Actual Oil Rate

Apr-99

400000

6,000.0 5,000.0

300000

4,000.0 200000

3,000.0 2,000.0

1,000.0

100000

20000

1,000.0 0.0

0

May-88

Jan-91

Oct-93

Jul-96

Apr-99

Jan-02

Oct-04

Jul-07

Mar-10

Time (month)

Figure 22- Results of history mat ching for the well number 12484 in Upper Bakken

0.0 May-88

0 Jan-91

Oct-93

Jul-96

Apr-99

Jan-02

Oct-04

Jul-07

Mar-10

Time (month)

Figure 24- Results of history mat ching for the well number 12570 in Upper Bakken

Figure 25- - Field wide cumulative production (upper chart) and number of producing wells (lower chart) in Middle Bakken Uncertainty Analysi s Effect of different co mpletion and reservoir stimu lation on production rate was investigat ed. Simu ltaneous effect of pay thickness and depth was considered in this part of study. Effect of depth and lateral length on production is shown in figure below. The model shows direct relationship between production rate and lateral length and reverse relationship with depth. The decline in production in long laterals is an indicator of operational difficulties associated with back production of the wells right after co mp letion.

Figure 26- Effect of depth and lateral length on production r ate (Middle Bakken model)

Figure 27- Effect of volume of injected fracturing fluid and lateral length on production rate (Middle Bakken model) Effect of volu me of in jected fracturing flu id and lateral length is sho wn in figure 27. The model shows direct effect of both injected fluid and lateral length on production. The same effect of long laterals on production drop is observed in this model as well. It can be concluded that a limit for length of lateral exists for Middle Bakken formation wh ich is about 17,000 ft. Model below shows simultaneous effect of pay thickness and lateral length on production rate. The model shows non significant effect of lateral length at thinner parts of the reservoir and more significan t effect of lateral length at thicker pay thicknesses.

Figure 28- Effect of pay thickness and lateral length on production r ate (Middle Bakken model)

Figure 29- Effect of TOC and porosity log re ading on production r ate (Upper Bakken model)

Spontaneous effect of TOC and porosity in Upper Bakken model was also investigated. The result of this un certainty analysis is shown in Figure 29. Th is model shows direct effect of TOC in production at higher poro sity values. Conclusions Two different Top-Down approaches in reservoir simulat ion were performed for Upper and Middle members of Bakken formation. Simple geo maps were generated for reservoir characteristics from log record. Application of production st atistic and reservoir geological maps in generating Statistical Estimated Models was conducted. Fuzzy pattern recognition technology was employed in order to analyze remain ing reserve as a function of time in Middle Bakken. Sweet spots were identified and new wells were proposed to be drilled. Production behavior of new wells, recovery factor and EUR of the field was calculated. Econo mical analysis was performed based on the results of estimated models. NPV as an indication of rate of return was calculated assuming different investment costs. A history matched model of production and reservoir characteristics was generated for both Upper and Middle members of Bakken format ion. This new approach to history matching generate a predictive model wh ile history matching. The history matched predictive model has been built by utilizing all the reservoir characteristics, co mpletion data and production rate. The intelligent predictive model is capable of predicting the future production for existing and new wells.

18

Field Development Strategies for Bakken Shale Formation

SPE 139032

Sensitivity analysis for several reservoir characteristics and completion extents were performed which guides us to better design wells and stimu lation process. This type of informat ion can be used in order to reduce the investment cost. The combination of these two approaches of reservoir modeling is a very good tool in reservoir management. By performing all these analysis, sweet spots were identified and an extensive comp rehension about future of the field was achieved. Aknowledgment Authors would like to acknowledge NETL/DOE for financially supporting this project (Project # 4000.4.650.920.004), Intelligent Solutions Inc. for providing the IPDA, EPIQ and IDEA software packages and PEARL members at WVU for their technical support. References

1. www.pttc.org; Forum for Transfer of Technology and Best-practices within the Oil & Gas Community 2. J.N. Fox C.D. Martiniuk; Reservoir Characteristics and Petroleum Potential of the Bakken Formation, Southwestern Manitoba; JCPT, Vol. 33, No. 8, October 1994. 3. USGS 4. Julie A. LeFever; North Dakota Geological Survey Geologic Investigations No. 59 Bakken Formation Map Series , 2008 5. N. Buffington , J. Kellner, J.G King, B. David, A. Demarchos, L. Sheperd; New technology in the Bakken Play Increases the number of stages in Packer/Sleeve Completion; SPE western regional meeting, Anaheim, California, 27-29 May 2010.

6. www.OilandGas Investors.com; October 2009. 7. R. Gaskari, S.D.Mohaghegh, J.Jalali; An Integrated Technique for Production Data Analysis with Application to Mature Fields; SPE 100562; SPE Gas Technology Symposium held in Calgary, A lberta, Canada, 15–17 May 2006.

8. A. Kal antari-Dahaghi, S.D. Mohaghegh; Top-Down Intelligent Reservoir Modeling of New Albany Shale; SPE 125859; SPE Eastern Regional Meeting, Charleston, West Virg inia, 23–25 September, 2009.

9. G.C.Thakur; What is reservoir management?; Journal of Petroleu m Technology; Volu me 48, Nu mber 6; P 520-525; June 1996.

10. M.L. Wiggins and R.A. Startzman; An Approach to Reservoir Management; SPE 20747; 65th Annual Technical Conference and Exh ibition of the Society of Petro leu m Engineers, New Orleans, LA, September 23-26, 1990.

11. Arps, J.J.; Analysis of Decline Curves, Trans, AIM E. 1945. 12. R. N. Heistand, H. G. Humphries; Direct Determination of Organic Carbon in Oil Shale, Analytical Chemistry, Vol. 48, No. 8, July 1976, p 1193.