Case study of a heavy oil reservoir interpretation using Vp/Vs ratio and other seismic attributes Carmen C. Dumitrescu * Sensor Geophysical Ltd, Larry Lines, CHORUS, University of Calgary

Summary The north-eastern part of Alberta contains an estimated 200 billion cubic meters of bitumen. The Lower Cretaceous McMurray Formation forms one of the largest petroleum deposits in the world. The focus of this study is the southern portion of the Long Lake lease located approximately 40 km southeast of the City of Fort McMurray in the Athabasca Oil

Sands Region. In this case study, identifying the sand-dominated point-bar deposits and the mud-dominated abandoned channel fill deposits were the primary interpretive focus. First, we calculated and analyzed several edge detector attributes, such as semblance and volume curvature attributes (most negative, most positive and dip curvature)

in an effort to understand the reservoir and non-reservoir depositional elements. Next, we performed rock physics and neural network analysis. Rock physics highlighted Vp/Vs ratio and density as good lithology discriminators, identifying the estimation of Vp/Vs and density volumes from seismic data as useful and important objectives. Neural network analysis was used

in an effort to refine the interpretation of the reservoir properties derived from deterministic inversion. Neural network estimation of reservoir properties has proven effective in significantly improving accuracy and vertical resolution in the interpretation of the reservoir. Neural network analysis is performed on logs and pre- and post-stack seismic attributes. The results provide detailed identification of the sand and shale within the heavy oil

reservoir of Long Lake South. Finally, we integrated all available attributes to characterize and map the reservoir heterogeneities impacting the SAGD operation, i. e. extent of bitumen-saturated sand.

Introduction

The Oil Sands reservoir related to the Long Lake South (LLS) Project is contained within the McMurray Formation, which is the basal unit of the Lower Cretaceous Mannville Group. The McMurray Formation directly overlies the Subcretaceous Unconformity, which is developed on Paleozoic carbonates of the Beaver Hill Lake Group, and overlain by the Clearwater and Grandrapids Formations of the Mannville Group.

The study area (Figure 1) is located along the axis of the McMurray Valley system, which was localized by the

dissolution of underlying Devonian evaporates, creating the preferred depositional fairway for the Lower Cretaceous McMurray sediments. The most significant bitumen reservoirs within the McMurray Formation are found within the multiple channels that represent lowstand system tracts, incised into the regional, prograding parasequence sets that represent highstand system tracts. During sea level rise, these incised channel systems were filled with a transgressive

estuarine complex, consisting of sandy to muddy estuarine point bars. In the Long Lake area, the McMurray Formation is dominantly composed of these multiple, sand rich, fluvial and estuarine channels, which are incised into each other and stacked along a preferred path of deposition. This preferred path is aligned north-northwest to south-southeast in the Long Lake area (Long Lake Project, 2006). Depending on their size and configuration, non-reservoir shale bodies can

impede steam chamber growth and fluid drainage within a SAGD production process.

Figure 1: Map showing the Long Lake South area, Alberta, Canada Volume curvature attributes complement and improve the ability of semblance to reveal geologically meaningful features. These attributes can reveal stratigraphic features (i.e. channels) with significant impact on reservoir evaluation

but are based on morphology, not on rock physics. Petrophysical analysis has determined that Vp/Vs and density are key discriminators between sand and shale. Therefore we used neural network analysis to estimate a Vp/Vs and density volume. Inversion is a necessary step in imaging and interpreting the reservoir and there is a

1765SEG Houston 2009 International Exposition and Annual Meeting

Case study of a heavy oil reservoir

continuous struggle to improve the resolution of the inverted volume. Depending on the seismic data and the number of wells available, a Vp/Vs ratio volume can be obtained from: (i) traveltime measurements on the vertical and radial components of the multicomponent records (Lines et al.,

2005), (ii) AVO analysis and simultaneous inversion using only the PP component (Dumitrescu and Lines, 2006), or (iii) joint inversion of the PP and PS (registered in PP time) poststack seismic data (Dumitrescu and Lines, 2007). Neural network analysis (NNA) is used to estimate seismic volumes by integrating well information and existing seismic volumes (e.g. deterministic inversion results). For

this project, the estimated Vp/Vs ratio and density volumes are used successfully in mapping oil sands in heavy oil reservoirs. Finally, we integrate all available attributes to characterize and map the reservoir heterogeneities impacting SAGD operation, i. e. extent of bitumen-saturated sand, water and gas saturated zones.

Method Part A: Semblance and volume curvature attributes Volume curvature attributes complement and improve the ability of semblance to reveal geologically meaningful features. These attributes can reveal stratigraphic features

(i.e. channels) with significant impact on reservoir evaluation but are based on morphology, not on rock physics. Semblance and volume curvature attributes are calculated from the pre-stack time migrated volume. Semblance is a statistical measure of waveform similarity given by the ratio of the energy of the average data trace to the average energy of each data trace . All volume curvature attributes used in this study (most negative, most positive and

dip curvature), are second order derivatives. Most positive and most negative curvature are derived from maximum curvature by setting some coefficients to zero. It is now possible to see almost every single lineament contained within the surface. The magnitude of the lineament is also preserved, allowing better lineament discrimination. However, the shape information is not preserved in this attribute, making it comparable to the first derivative

methods (Roberts, 2001). Dip curvature displays the rate of change of curvature, in the direction of the maximum dip.

Part B: Neural network analysis for rock property volumes The LLS Project includes 3-D seismic and approximately 50 logged and cored wells over a 50 square kilometer surface area. The 3-D seismic was acquired in 2005 using the Sercel

digital multi-component recording system. Out of 42 wells

with dipole sonic logs only 31 were used in the NNA. Petrophysical analysis was performed on all the wells in order to provide a trustworthy set of logs for inversions and neural network analysis.

The neural network was used in an effort to account for non-linear relationships between logs and seismic after first testing the linear multi-attribute method alone. Neural network analysis has four steps: 1. perform a multi-attribute step-wise linear regression and its validation, 2. train neural networks to establish the nonlinear relationships between seismic attributes and reservoir properties at well locations, 3. validate results on the wells withheld from the training, 4.

apply trained neural networks to the 3D seismic data volume. The NNA was performed for 31 wells and eight seismic volumes (PSTM stack, AVO-derived P-wave and S-wave reflectivity, Fluid Factor, P-wave impedance, S-wave impedance and Vp/Vs from inversion and Density from a previous NNA) and a seven-point operator. Using the ranking process available within the software and after checking the errors, we selected the attributes for training the

network.

Results Part A: Semblance and volume curvature attributes A typical seismic line in the pre-stack time migrated volume, is presented in Figure 2. The reservoir is between McMurray

and devonian horizons.

McMurray

Devonian

#1#2 Figure 2: Seismic line and two features within McMurray reservoir: #1. Sand-dominated point-bar deposit, #2. Mud-dominated abandoned channel fill deposit.

Semblance and volume curvature attributes are calculated on the prestack time migrated volume. Figure 3 and Figure 4 present a horizon time slice of the semblance and most positive curvature at McMurray+10ms. Both attributes highlight the edge of the abandoned channel (#2).

1766SEG Houston 2009 International Exposition and Annual Meeting

Case study of a heavy oil reservoir

low

high

#1

#2

Figure 3: Horizon time slice of the semblance at

McMurray+10ms (showing features #1 and #2)

low

high

#1#2

Figure 4: Horizon time slice of the most positive curvature at McMurray+10 ms (showing features #1 and #2) Part B: Neural network analysis for rock property volumes Petrophysical analysis has determined that Vp/Vs and density are key discriminators between sand and shale. The

cross-plot in Figure 5 is between Vp/Vs (X-axis) and density (Y-axis) logs from wells in the study area. The two zones, sand and shale, are defined based on the associated histograms.

sand

shale

Vp/Vs

Rho (kg/mc)

GR (API)

Figure 5: Cross-plot of the Vp/Vs (X-axis) and Density logs (Y-axis) colored by GR Deterministic inversion and neural network analysis were performed in an attempt to get a better definition of the bitumen sand and to better differentiate between gas sands

and shale in the McMurray Formation. In a previous paper (Dumitrescu and Lines, 2009) we discussed how neural network analysis for estimating Vp/Vs volume can improve the results from deterministic inversion. Figure 6 shows the estimated Vp/Vs results on a horizon time slice at McMurray+7ms and Figure 7 shows a similar time slice but from the density volume.

high

low

#2#1

Figure 6: Horizon time slice at McMurray+7ms on Vp/Vs volume obtained from neural network analysis.

1767SEG Houston 2009 International Exposition and Annual Meeting

Case study of a heavy oil reservoir

high

low

#2#1

Figure 7: Horizon time slice at McMurray+7ms on density

volume obtained from neural network analysis. Finally, by integrating all available attributes we characterize and map the reservoir heterogeneities impacting SAGD operation, i.e. extent of bitumen-saturated sand, water and gas saturated zones. We investigate two ways to integrate all these attributes: (i) by cross-plotting the rock physics volumes and one of the edge detector attributes and (ii) by using the RGB technique on a color

cube where R (red) is used for Vp/Vs, G (green) is used for density and B (blue) is used for most positive curvature. In Figure 8 we present a co-rendered slice between Vp/Vs and semblance on a horizon slice at 10 ms below McMurray.

Figure 8: Horizon slice at McMurray+10 ms of the (top) Vp/Vs results from neural network analysis and (bottom) semblance.

Conclusions

We presented a case study of a heavy oil reservoir interpretation using rock physics attributes, i.e. Vp/Vs and

density volumes (obtained from neural network analysis), and edge detector attributes, i.e. semblance and volume curvature. Using the Vp/Vs and density volumes computed with

neural network analysis minimizes the uncertainty in gas sand, bitumen sand and shale identification thereby contributing to optimal horizontal well placement. This will in turn have the ultimate effect of increasing production and economic efficiency. Integration of edge attributes with rock physics attributes allows for a realistic geological interpretation for oil sands

development purposes.

Acknowledgements

We acknowledge and thank the following organizations for their support: Sensor Geophysical Ltd., Nexen Inc. and the Consortium for Heavy Oil Research by University Scientists (CHORUS).

1768SEG Houston 2009 International Exposition and Annual Meeting

EDITED REFERENCES Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2009 SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web. REFERENCES Dumitrescu, C. C., and L. Lines, 2006, Vp/Vs ratio of a heavy oil reservoir from Canada: CSPG - CSEG - CWLS Convention,

Abstracts, 10–15. ———, 2007, Heavy oil reservoir Characterization using Vp/Vs ratios from multicomponent data: 69th Annual Conference and

Exhibition, Extended Abstracts, P301. Dumitrescu, C. C., L. Lines, D. Vanhooren, and D. Hinks, 2009, Characterization of heavy oil reservoir using Vp/Vs ratio and

neural network analysis: CSEG Convention, 1–4. Lines, L., Y. Zou, A. Zhang, K. Hall, J. Embleton, B. Palmiere, B., C. Reine, P. Bessette, P. Cary, and D. Secord, 2005, Vp/Vs

characterization of a heavy-oil reservoir: 75th Annual International Meeting, SEG, Expanded Abstracts, 1134–1136. Long Lake Project, 2006, http://www.longlake.ca/project/technology.asp. Roberts, A., 2001, Curvature attributes and their application to 3D interpreted horizons: First Break, 19, 85–99.

1769SEG Houston 2009 International Exposition and Annual Meeting