Monitoring and Predicting Steam Chamber Development in a Bitumen Field

Monitoring and Predicting Steam Chamber Development in a Bitumen Field


6MB Sizes 0 Downloads 11 Views



18 Kelsey Schiltz1, David Gray2


Colorado School of Mines, Golden, CO, USA ; Nexen Energy ULC, Calgary, AB, Canada2

18.1 INTRODUCTION 18.1.1 GENERAL Although the word “unconventionals” has almost become synonymous with production from shales in the United States, heavy oil and bitumen constitute another category of unconventionals with massive resource potential. According to standards set by the United States Department of Energy, oil is considered heavy if it has an API gravity of less than 22.3 and bitumen if it is less than 10 (Chopra et al., 2010). Approximately one trillion barrels of heavy oil and bitumen are estimated to be recoverable worldwide, with the largest accumulations in Canada and Venezuela (Rigzone, 2014). The overwhelming majority of Canada’s heavy oil and bitumen resources are found in three deposits within the province of Alberta. The Albertan government estimates that its oil sands contain 1.8 trillion barrels of original bitumen in place, 80% of which is located in the Athabasca Oil Sands deposit (Fig. 18.1) (Alberta Energy, 2012). While the resource potential of heavy oil and bitumen is enormous, the cost of producing and refining the crude is high. In particular, bitumen requires special effort to produce because it does not flow under normal subsurface conditions. Although bitumen reservoirs tend to be shallower than conventional reservoirs, surface mining is rarely a viable option. For the majority of bitumen deposits, in situ recovery methods that are more expensive and generally less effective than surface mining must be used. Optimization at every stage in the development process is crucial in order to maintain the economic viability of the project. Time-lapse seismic surveying has proven to be one of the most useful ways to monitor the effectiveness of in situ recovery methods, particularly those that involve steam flooding. It can be used to improve operating strategies in both producer and injector wells, change well completions, and find infill drilling opportunities. For example, at ConocoPhillips and Total’s Surmont Field in Alberta, seismic amplitude changes were used to identify areas along the wellbore where steam injection had been effective and those where it had not (Byerley et al., 2009). This insight led to a modified injection scheme that improved steam efficiency within just four months. The success of time-lapse monitoring Unconventional Oil and Gas Resources Handbook. Copyright © 2016 Elsevier Inc. All rights reserved.




FIGURE 18.1 The field is located within the Athabasca Oil Sands deposit, the largest of three major deposits in Alberta (Wikipeda, 2014).



of bitumen recovery is attributable to several factors, including the shallow depth of the reservoir, the high porosity and low bulk modulus of the sands, and the large change in fluid compressibility during production. These factors all increase the likelihood of seeing a time-lapse seismic signature (Lumley et al., 1997). What has proven more challenging than the time-lapse reservoir monitoring is reservoir characterization aimed at distinguishing good quality sands from shales that can limit production. Some studies have attempted to predict reservoir sands though inversion or neural networks (e.g.,Roy et al., 2008 and Tonn, 2002), but generally the authors do not go as far as to compare their results to observed time-lapse effects for validation. In this study of a bitumen field in Alberta, the distribution of steam in the reservoir determined through time-lapse analysis is leveraged to investigate the accuracy of sand and shale prediction using a neural network.

18.1.2 THE ATHABASCA FIELD The focus of the study is a bitumen field covering 63,000 acres within the Athabasca Oil Sands deposit in Alberta, Canada (Fig. 18.1). Bitumen has such a high viscosity at in situ conditions that it is immobile and cannot be produced using conventional methods. Production at this field is enabled by a thermal recovery method called steam-assisted gravity drainage, or SAGD. The SAGD process involves two vertically-stacked horizontal wells separated by approximately 5 m. The upper well injects high-pressure, high-temperature steam into the reservoir. The steam heats the bitumen and lowers its viscosity so that it can flow via gravity down to the lower horizontal well where it is produced (Smalley, 2000). As the bitumen drains from the reservoir, it is replaced by steam in the pore space, creating what is known as the steam chamber. The evolution of the steam chamber is highly controlled by the presence of low-permeability intervals within the reservoir (e.g., shale stringers and mud plugs), as depicted in Fig. 18.2.

FIGURE 18.2 Steam chamber growth can be impeded by a low-permeability interval such as shale (McDaniel and Associates Ltd, 2006).



The bitumen (w8 API) is produced from the Lower Cretaceous McMurray Formation. The McMurray Formation was deposited on top of Devonian limestones in a north-west to south-east trending paleo-valley and was subsequently capped by marine shales of the Clearwater Formation (Wightman and Pemberton, 1997). The McMurray is often informally separated into three units known as the lower, middle, and upper McMurray. As a general approximation, these three subdivisions represent a transgressive progression from a fluvial, to an estuarine, to a marine depositional environment. Clean, blocky reservoir sands are generally found toward the base of the McMurray in sand-filled channels or point bars deposited in low-lying estuaries. Toward the top of the formation, increasing marine influence resulted in the deposition of more fine-grained silts, clayey sands, and silty clays (Dusseault, 2001). Throughout the McMurray, lateral facies variations can be quite abrupt due to the various depositional elements found in the fluvial setting. These depositional elements, including sand-filled channels, mud-filled channels, point bars and counter-point bars, have variable porosities and permeabilities which need to be well-understood in order to optimize production. This study demonstrates how seismic reservoir characterization can be used to create a geologic reservoir model that can predict steam chamber development.

18.1.3 SEISMIC AND WELL DATA The time-lapse seismic data used in this study cover 2.5 km2 and encompass four SAGD well pads (Fig. 18.3). The baseline survey was acquired in 2002 before any steam injection had taken place and therefore provides a true snapshot of initial reservoir conditions. The monitor survey was acquired in 2011 after approximately four years of production. One of the most important considerations in the interpretation of time-lapse seismic is the repeatability of the surveys. Repeatability is a measure of how similar two or more seismic surveys are in areas where no subsurface changes have taken place. The more repeatable the surveys, the easier it is to isolate and interpret production-related changes in the subsurface. The first step to ensure good repeatability was to acquire both surveys using similar acquisition parameters. Next, the data were processed in parallel to reduce the chance of introducing spurious differences during the processing stage. Parallel time-lapse processing was performed to create a poststack compressional volume, poststack radial converted-wave volume, and angle stacks for both the baseline and monitor surveys. Despite taking precautions to ensure repeatability, nonproduction related differences will inevitably exist between the surveys. To further eliminate these differences and increase the repeatability of the data, a cross-equalization workflow was performed on the monitor survey. Cross-equalization is a multistep process that attempts to identify and eliminate unwanted differences between the baseline and monitor, thereby allowing accurate time-lapse interpretation of the production-related changes (Ross et al., 1996). Throughout the cross-equalization process, the normalized root-mean-square (NRMS) ratio was used as a statistical indicator of repeatability. This ratio, defined in Eq. (18.1), reaches a minimum value of 0 when two traces are identical (i.e., perfectly repeatable) and a maximum of two when two traces are identical but opposite in polarity (Kragh and Christie, 2002). An overall reduction in the NRMS ratio in an area that is expected to be free of production effects indicates that the repeatability has been improved by the cross-equalization process.



FIGURE 18.3 Map of the SAGD well configuration within the study area. The red outline shows the boundary of the four-dimensional (4D) seismic survey covering well Pads 1 (pilot), 3, 5, and part of Pad 2.

NRMS ratio ¼



As a starting point, the mean NRMS was calculated over a window beneath the reservoir in which time-lapse changes were not expected. The initial NRMS in this window was 1.61 (in other words, highly nonrepeatable). Next, a number of cross-equalization steps were applied to the monitor survey in sequence, including phase and time matching, static time shifts, time-variant time shifts, amplitude normalization and a shaping filter. Although the details of these operations are beyond the scope of this paper, the overall effect was a decrease in the mean NRMS in the analysis window from 1.61 to 0.53, which represents a significant improvement in repeatability. A line through each of the cross-equalized time-lapse volumes is shown in Fig. 18.4. The first thing to note is the dramatically lower frequency content of the converted-wave data compared to the compressional data. This is one major limitation of interpreting converted waves. The second observation that can be made, however, is that there are changes in reflectivity within the reservoir (between the blue and green horizons) that are likely related to SAGD production. These cross-equalized time-lapse volumes were used in the subsequent timelapse analysis and reservoir characterization work. In addition to the seismic data, there are 43 vertical wells within the study area. Each has a full suite of standard logs, including caliper, gamma ray, density, neutron, and resistivity. Several calculated curves such as neutron porosity, water saturation, and volume of shale were also provided by the operator. All 43 wells within the study area have P-wave (compressional) sonic logs



FIGURE 18.4 Line through the cross-equalized time-lapse seismic data. The McMurray reservoir is located between the blue (black in print versions) and green (gray in print versions) horizons.

and four wells also have an S-wave (shear) sonic log. The well control was leveraged extensively for horizon interpretation and for training the neural network.

18.2 MAPPING STEAM In the theoretical case of a perfectly homogenous reservoir, the steam chamber would develop uniformly along the entire length of the borehole. On the other hand, when heterogeneities such as shale



stringers are present, the steam cannot fully penetrate these tight intervals and an irregular steam distribution develops in which certain areas of the reservoir are not being swept. Mapping the distribution of steam is an important first step in determining whether heterogeneities are influencing production from this field. The SAGD process causes significant fluid and saturation changes that affect the seismic response and enable the visualization of the steam chamber through time-lapse analysis. First, the bitumen transforms from a quasisolid to a liquid as its viscosity is lowered through heating. Then, the bitumen is replaced by steam as it drains from the reservoir rock down to the production well. Overall, these changes result in a decrease in the P-wave velocity within the steamed zones. The lower velocity in turn causes seismic events beneath the steamed reservoir to be shifted downward in the monitor survey relative to the baseline. The map in Fig. 18.5(a) shows the time shifts observed in the Devonian reflector at the base of the reservoir. Significant time shifts of up to 8 ms can be seen where the steam chamber has been effectively developed. Similar time-lapse anomalies were also observed when the amplitude changes within the reservoir were mapped (Fig. 18.5(b)). In both timelapse maps, the anomalies over well Pad 5 are elongated along the wellbore and it appears that adjacent steam chambers have begun to coalesce. These wells provide an example of good conformance; in other words: a large portion of the borehole has a well-developed steam chamber. The overall lack of time-lapse effects seen along the wells of Pads 1 and 2NE, however, indicates poor steam chamber development that may be related to a higher percentage of low-permeability shales. The results of the reservoir characterization will be compared to these time-lapse anomalies to verify this theory. Time-lapse analysis of the compressional data indicates that the steam chamber is well-developed in some areas and poorly developed in others. To further solidify the connection between steam chamber development and production, a crossplot was constructed between the length of the steam anomalies in the time shifts map (Fig. 18.5(a)) and cumulative oil production for each well. Wells from Pad 1 were excluded from the plot since these wells have been active over a longer time period. The resulting plot (Fig. 18.6) shows a clear positive correlation between the degree of steam chamber conformance and production (a similar result was found by Byerley et al., 2009 at the Surmont project). This relationship provides confirmation that the 4D seismic anomalies are indicative of steam and that maximizing steam chamber development leads to increased production. Performing reservoir characterization to identify zones of low reservoir quality is one way to optimize well pair placement for improved steam chamber development. As a final step in the time-lapse analysis, a 4D inversion of the PP (compressional) poststack data was performed to create a three-dimensional (3D) representation of the steam chamber. As previously mentioned, steamed reservoir has a decreased P-wave velocity and bulk density, which in turn lowers the P-impedance. In the 4D inversion process, the baseline and monitor surveys are both inverted for P-impedance and the results are used to calculate the percent change in impedance from 2002 to 2011. A cross-section through the result of this calculation, shown in Fig. 18.7(a), shows that steam-filled reservoir corresponds to an impedance decrease of 10–20%. Observations from wells logged before and after steaming confirm that P-impedance changes on the order of 20% can be expected. Next, regions of the P-impedance difference volume corresponding to a reduction of 15% or greater were extracted to isolate the 3D volume that has been filled with steam. The 3D representations of individual steam chambers are referred to as steam geobodies and can be seen in Fig. 18.7(b). From an aerial perspective, the geobodies have the same distribution as


(a) Time-shifts observed between the baseline and monitor survey in the reflector marking the base of the reservoir (b) Amplitude difference after cross-equalization (monitor-baseline). The map shows the minimum amplitude between the McMurray horizon (McM) and 15 ms below the Devonian horizon (Dev).





FIGURE 18.6 For each well pair, the length of the borehole associated with significant time shifts in Fig. 18.5(a) is plotted against the cumulative production.

the anomalies in the time-lapse maps in Fig. 18.5, but now the vertical position of the steam chamber has also been defined. Ultimately, the geometry of these steam geobodies will be compared to that of the interpreted shales to determine whether the shales exhibit control over steam chamber development.

18.3 RESERVOIR CHARACTERIZATION USING A PROBABILISTIC NEURAL NETWORK In an ideal case, one seismic survey acquired before steam injection begins could be used to map shale zones and accurately predict the success of proposed SAGD well pairs. Therefore, the objective of reservoir characterization in this study was to use the seismic data to distinguish between clean reservoir sands and shales that may inhibit the growth of the steam chamber. Seismic inversion is one method commonly used to pursue such goals and was the first method attempted in this study. Prior to initiating the seismic inversion work, a petrophysical analysis was performed using 11 wells



FIGURE 18.7 (a) Percent change in P-impedance between the baseline and monitor surveys. The presence of steam is indicated by P-impedance decreases of 10–20%. (b) 3D steam geobodies created by extracting the areas of the P-impedance percent change volume correspond to impedance decreases of greater than or equal to15%.

within the 4D study area to determine which of the seismic inversion products, P-impedance, S-impedance, or density, would be the most useful sand-shale indicator. In the analysis, computed logs of each inversion product were plotted against the volume of shale (Vsh) log. While P-impedance was shown to have a weak correlation to Vsh, density had by far the strongest correlation at R2 ¼ 0.88 (Fig. 18.8). Density is notoriously difficult to estimate, however, from the



FIGURE 18.8 Crossplot showing a strong correlation (R2 ¼ 0.88) between density and volume of shale well logs.

inversion of seismic data with useable reflection angles of less than 50 (Roy et al., 2008). After attempting both a prestack and joint inversion on the seismic dataset (maximum angle of 45 ), it was found that density could not be accurately constrained and another approach would have to be taken. It was decided to use a probabilistic neural network (PNN) to predict density from multiple seismic attributes. A PNN is an algorithm that simulates the nonlinear way in which the human brain learns (Hampson and Russell, 2012). The PNN tries to determine the relationship between a set of seismic attributes and a target well log. Once the transformation between the attributes and the target log has been determined at the training wells, the same transform is applied to the seismic data to create a volume of the target log property. The set of seismic attributes with the best correlation to the target density log was determined through multiattribute analysis. Seismic attributes were derived from the full stack compressional and converted-wave volumes, a poststack P-impedance inversion result, and a prestack P-impedance inversion result. Then, the multiattribute list is arrived at using stepwise regression. Stepwise regression is an approximation method that takes much less computation time than testing every possible combination of attributes. The method works by first searching through the list of available attributes to find the single attribute that best correlates with the target log. This attribute is listed as attribute 1. Next, the program finds the best combination of two attributes, assuming that one of these



Table 18.1 The Six Attributes Selected by Multiattribute Analysis to be Used in PNN Training 1. 2. 3. 4. 5. 6.

Integrate (PP full stack) Amplitude envelope (PP full stack) Average frequency (PS full stack) Time (PP full stack) Filter 35-40-45-50 (PS full stack) Second derivative (PP full stack)

attributes is attribute 1. The method continues in this way until either all of the attributes have been ranked or a specified validation criterion has been met. Table 18.1 shows the six attributes that were selected by the analysis. Note that two out of the six attributes are calculated from the converted-wave (PS) volume, indicating that converted-wave data adds significant value to density prediction. A detailed description of each of the attributes can be found in Russell (2004). The six attributes were then used to train the PNN to predict the density logs within the reservoir zone defined by the McMurray (McM) and Devonian (Dev) horizons. Once the relationship was established at the training well locations, the same transform was applied to the entire baseline seismic volume to create a predicted density volume. Figure 18.9 shows the results along an arbitrary line. All the wells shown in this line are blind wells and were not included in the multiattribute analysis or PNN training in any way. For each well, the filtered density log is displayed in color along with the Vsh curve in black. Although the magnitude of the predicted density in this section is not always accurate, the relative density variations are very consistent with the well log data. Ultimately it is the ability to capture the relative highs and lows in the density that is important for distinguishing sands and shales. The sands and shales interpreted from the PNN can also be related to geologic features. Figure 18.10 shows the predicted density at one inline along with a blind well. The schematic below the inline illustrates some of the common morphologic features in a fluvial–estuarine depositional environment (Leckie et al., 2009). In particular, sandy lower point bars (LPBs) and shaley inclined heterolithic strata (IHS) are often found at the inner bends of meandering channels. A channel-like profile was identified within the density inline, and adjacent features bear strong resemblance to IHS and LPB deposits. The geologic appearance of the PNN density volume provides further support for its accuracy and lends credibility to lithologic interpretations. The primary goal of reservoir characterization for a bitumen reservoir undergoing SAGD is to accurately identify shales. This is because the distribution of shales highly controls the development of the SAGD steam chamber. Therefore, a test of the utility of the predicted density volume is to see if a relationship exists between the distribution of high density zones interpreted as shales and the distribution of the steam chamber. The map in Fig. 18.11(a) shows the 3D shale geobodies (blue) predicted by the PNN in the lower to middle part of the reservoir. These shales are particularly important because they limit production more severely than shales closer to the top of the reservoir. The map in Figure 18.11(b) shows the same shales along with the steam geobodies (yellow) determined through time-lapse analysis. By comparing the two maps, it can be seen that the steam chamber has developed better in areas that lack shales in the lower half of the reservoir and is poorly developed in areas that have a significant number of shales (Pad 1, for example). In a few areas, most noticeably on Pad 5, the

An arbitrary line through the PNN density volume. Inserted blind wells show the filtered density log in color (0-150-160 Hz) and the Vsh log in black. The reservoir is between the McM and Dev horizons. cwB/C, Clear Water B/C; McM, Top McMurray Formation; Dev, Top Devonian.





FIGURE 18.10 Inline through the PNN density result along with filtered density at a blind well. The schematic at the bottom shows depositional elements commonly found in this fluvial-estuarine environment (Leckie et al., 2009). Similar elements have been interpreted in the density result.


FIGURE 18.11 (a) Distribution of shales in the lower half of the reservoir interpreted from the predicted density volume (b) Interpreted shales (blue, gray in print versions) along with the steam geobodies (yellow, white in print versions). The shales appear to be controlling the development of the steamchamber.




FIGURE 18.12 View looking directly toward the heels of Pad 5 wells. The steam chamber (yellow, white in print versions) has risen and encountered shale in the middle of the reservoir (blue, gray in print versions) but is interpreted to have moved around it over time (red (dark gray in print versions) arrow) to access the good quality reservoir above.

steam chamber seems to have developed regardless of the presence of shales. From a different perspective, however, it appears as if steam that was trapped up against the base of the shale has managed to move around or through it over time to access the reservoir sands above (Fig. 18.12). Whether or not this happens depends on the lateral extent and continuity of the shale body and whether or not good quality sands exist above it. From this view, it can be seen that there is an interval of clean sand that the steam has filled between the shale in the middle of the reservoir and the shales at the top of the reservoir. Overall, the good correlation between interpreted shales and the steam chamber distribution indicates that the PNN density volume is a useful tool for predicting the success of SAGD well pairs.

18.4 CONCLUSION This study demonstrates the value seismic data can add in the SAGD development of bitumen resources as both a monitoring tool and a reservoir characterization tool. In the first part of the study, time-lapse seismic data were used to visualize steam chamber development by analyzing timeshifts, amplitude changes, and P-impedance changes. The development of the steam chamber was found to be nonuniform and a correlation was observed between poor steam chamber development and lower production. Because the efficiency of SAGD production is known to be greatly influenced by low-permeability shales, a reservoir characterization study of the baseline multicomponent survey was undertaken with the goal of distinguishing between high-quality reservoir sands and low-permeability shales.



A petrophysical analysis of 11 tied wells within the study area identified density as the acoustic parameter most closely associated with the volume of shale. After attempts to constrain density through prestack and joint inversion were unsuccessful, a PNN was used to predict density from multiple seismic attributes. The PNN density prediction was geologic in appearance, accurately captured relative density, and allowed the interpretation of reservoir shale bodies. Perhaps most importantly, a relationship was observed between the interpreted shales and the distribution of the steam chamber in the reservoir. This relationship provided confirmation that the PNN density prediction can be used to predict the success of SAGD in a given area. Since the study, shale barriers and baffles are being mapped in two ways. The first is by using PNN analysis to predict high density intervals prior to steam injection. The second is by using the boundaries of the 4D steam geobodies to infer thin shales that are below seismic resolution. The net effect is a much better understanding of the behavior of the reservoir.

LIST OF ABBREVIATIONS IHS Inclined heterolithic strata LPB Lower point bar NRMS Normalized root-mean-square PNN Probabilistic neural network SAGD Steam-assisted gravity drainage

REFERENCES Alberta Energy, 2012. Oil Sands- Facts and Statistics. Byerley, G., Barham, G., Tomberlin, T., Vandal, B., 2009. 4D seismic monitoring applied to SAGD operations at Surmont, Alberta, Canada. SEG Technical Program Expanded Abstracts 2009, 3959–3963. Chopra, S., Lines, L., Schmitt, D.R., Batzle, M., 2010. Heavy-oil reservoirs: their characterization and production. In: Heavy Oils: Reservoir Characterization and Production Monitoring: Society of Exploration Geophysicists, Geophysical Developments 13 (1), 1–69 Dusseault, M., 2001. In: Comparing Venezuelan and Canadian Heavy Oil and Tar Sands: Proceedings of Petroleum Society’s Canadian International Conference. 2001-061, pp. 1–20. Hampson, D.P., Russell, B.H., 2012. Neural Network Theory. Hampson-Russell Software, 9.1.3. Kragh, E., Christie, P., 2002. Seismic repeatability, normalized rms, and predictability. The Leading Edge 21 (7), 640–647. Leckie, D., Fustic, M., Seibel, C., 2009. Geoscience of one of the largest integrated SAGD operations in the World- a case study from the Long Lake, Northeastern Alberta. CSPG Luncheon Talk. Lumley, D., Behrens, R., Wang, Z., 1997. Assessing the technical risk of 4-D seismic project. The Leading Edge 16 (9), 1287–1292. McDaniel and Associates Consultants Ltd., 2006. Rigzone, 2014. How It Works: Heavy Oil. Ross, C., Cunningham, G., Weber, D., 1996. Inside the cross-equalization black box. The Leading Edge 15 (11), 1233–1240. Roy, B., Anno, P., Gurch, M., 2008. Imagine oil-sand reservoir heterogeneities using wide-angle prestack seismic inversion. The Leading Edge 27 (9), 1192–1201.



Russell, B., 2004. The Application of Multivariate Statistics and Neural Networks to the Prediction of Reservoir Parameters Using Seismic Attributes (Ph.D. thesis). University of Calgary. Smalley, C., 2000. Heavy Oil and Viscous Oil in Modern Petroleum Technology. Institute of Petroleum (Chapter 11). Tonn, R., 2002. Neural network seismic reservoir characterization in a heavy oil reservoir. The Leading Edge 21 (3), 309–312. Wightman, D., Pemberton, S., 1997. The lower cretaceous (Aptian) McMurray formation: an overview of the Fort McMurray area, Northeastern Alberta. In: Petroleum Geology of the Cretaceous Mannville Group, Western Canada: Canadian Society of Petroleum Geologists, Memoir 18, pp. 312–344. Wikipedia, 2014. Athabasca Oil Sands. (last accessed 27.02.14).