Stochastic Modeling & Petrophysical Analysis of Unconventional Shales: Spraberry-Wolfcamp Example

Transcription

1 Stochastic Modeling & Petrophysical Analysis of Unconventional Shales: Spraberry-Wolfcamp Example Fred Jenson and Howard Rael, Fugro-Jason Introduction Recent advances in fracture stimulation techniques along with strong oil prices have revived interest in the Spraberry-Wolfcamp formations in West Texas. The Spraberry-Wolfcamp intervals have been the objective of traditional exploration and development for an extended period of time and recently these new developments in drilling and completion technologies have made zones with lower quality rocks commercially viable. The potential productive area is estimated at over 400,000 acres. These reservoirs are low porosity, low permeability rocks comprised of quartz, clays, and calcite along with significant quantities of dolomite, pyrite, and kerogen and require more sophisticated interpretation methods. The petrophysical properties of the Spraberry- Wolfcamp interval are analogous to that of unconventional shale plays like the Barnett formation. Stochastic modeling is a useful technique for volume based interpretation of these types of unconventional reservoirs. A particular advantage to this type of analysis is that it can be done without sophisticated new logging techniques. Instead, existing logs from conventional wells can be analyzed along with core data to generate probable solutions. This paper presents the results of a stochastic modeling study undertaken to demonstrate the ability to use conventional logs to evaluate the potential of unconventional reservoirs. A typical well with both log and core data was selected for the study. Two separate stochastic models were created to determine if independent analyses could generate similar results. These models were calibrated to X-ray diffraction (XRD) analysis of drill cuttings, which showed the primary mineral components. One of the models was then applied to a second well in the trend to demonstrate the validity of the interpretation. Finally, results were compared and reviewed in light of production data. The study shows that two independently defined stochastic models generated using conventional logs and core analysis did an excellent job of estimating the mineral volumes. In addition, a modified Passey method was used to compute the volume of TOC and this estimate was good across the Spraberry portion of the borehole, but badly underestimated the TOC volume across the lower Wolfcamp. The stochastic results from either of the independent models were more accurate than the modified Passey method in predicting the volume of TOC. Spraberry-Wolfcamp Wells Two wells were used in this modeling project. Bourne #3 was the control well and Pence #1 was used to test the model. One of the key aspects of stochastic modeling is that it can be applied to existing boreholes for evaluation of exploration potential. Cuttings were available for Bourne #3 for confirmation of results and the model was validated by applying it to the Pence #1 borehole. The suite of logs used in this study included a litho-density curve along with a natural gamma ray tool (NGT) curve, which increases the strength of the model. This tool measures natural gamma ray radiation and identifies the concentrations of potassium, thorium and uranium. Clays typically have potassium and thorium as their primary radioactive elements. If a litho-density NGT string Page 1

2 was not deployed in the field, a conventional density gamma ray log is sufficient for modeling purposes. A review of logs shows significant gamma ray content in both wells. The Density-Neutron logs show no obvious productive zones based on criteria used for conventional wells. Figure 1 shows the three main logs for Bourne #3 and Pence #1. Modified Passey Method The Passey Method and Modified Passey Methods can be used to identify the volume of total organic carbon (TOC). Under certain conditions, the result of the method can be correlated to Kerogen content. Figure 1: Main logs used as sources for stochastic modeling. The GR Curve is an NGT and the Density is a lithodensity tool. The method works best in traditional shale sections where there is high clay content and no permeability. In unconventional shale reservoirs, the best results come from reservoirs that are self-sourcing and selfsealing. The methods require at least one well with core data in each area of interest for calibration purposes. Figure 2: The F-Overlay Method for this well used Density-Neutron shale corrected Porosity. There was good agreement with XRD data for the upper zone but poor agreement in the lower zone. There are a variety of conditions that can create an apparent Passey effect, such as when overlaying sonic resistivity or density resistivity. It is important that certain conditions be met when using Passey type methods to generate a valid estimation of the TOC. The most critical factors are that the reservoir be self-sourcing and self-sealing, such that all hydrocarbons are generated in place and the zones are so tight that the hydrocarbons cannot escape. If this is the case, the TOC is directly proportional to the volume of Kerogen and the volume of Kerogen can be determined in an area by calibrating log responses to core data for at least one well in that area. Page 2

3 Determining Shale Volume Shale volume is a critical component in determining how much TOC is present. Shale is a fine-grained sedimentary rock that forms from the compaction of silt and clay-size mineral particles commonly called mud. This composition places shale in a category of sedimentary rocks known as mudstones. Shale is distinguished from other mudstones because it is fissile and laminated. Laminated rock is made up of many thin layers. Fissile rock readily splits into thin pieces along the laminations. Figure 3: The VCL generated using the KTH Curve Data matches VCL from XRD Analysis of the cuttings quite well. An F-Overlay, a Modified Passey Method, was used to evaluate the Bourne #3 well. Figure 2 shows the Kerogen evaluation from this Modified Passey Method, which used the Density-Neutron shale corrected Porosity. The results were in good agreement with the XRD data for the upper zone. However, results for the lower zone were in poor agreement. The discrepancy between the Modified Passey TOC and the core cuttings TOC near the bottom of the Bourne #3 well (around feet) was really interesting. It demonstrates that the Passey method is not necessarily accurate in all cases and highlights the need to understand what conditions create these log responses. The F-Overlay method results imply that the majority of the porosity in this section is filled with salt water while the core cuttings analysis does not support this conclusion. One possible explanation is a mineral component affecting the resistivity measurement that has not been properly modeled. It is known from the cores that there is a significant volume of pyrite present, which might be suppressing the measurement of resistivity. There is no known accepted method of correcting resistivity measurements for conductive minerals. While there are many ways to determine shale volume, Gamma Ray and Natural Gamma Ray logs are preferred. Other methods include: SP (Spontaneous Potential) Log Resistivity (works great in some plays, such as the Haynesville) Neutron Density Neutron Sonic Neutron Sonic Density To determine shale volume, for the Spraberry- Wolfcamp, the Potassium Thorium (KTH) curve from the NGT was used (Figure 3). Potassium is an elemental component of most shales and the source for the gamma ray response typically seen in shales. While thorium is not part of the chemical composition of shales it is often deposited in shale rich environments. For these reasons the KTH curve is an excellent shale volume indicator and justifies increasing the use of the NGT tool in open-hole logging. Equally important is that Uranium is clearly associated with the presence of Kerogen. There is a close relationship between organic carbon and deposition of uranium salts [1]. Organic carbon generates a reducing environment that is conducive to chelation of uranium salts. This mechanism explains the high gamma ray values often associated with unconventional shale reservoirs. Ironically, this high response indicator of Kerogen was ignored for many years because the shales were considered nonproductive. Page 3

4 The shale volume (VCL) generated using the KTH Curve Data matched quite well the VCL from XRD Analysis of the cuttings (Figure 4). The conductivity of dry Clay is very low. Hingle Plots crossplots of porosity and resistivity were used to provide an estimate of Dry Clay Bulk Density for input to stochastic modeling. Stochastic Modeling Background Stochastic models combine statistical analysis in a framework where the outcome is anticipated but unknown. One or more random variables are included in the stochastic calculus, and chance or probability is an integral component of the model. Using this type of modeling, geoscientists build probability functions and then work to determine the most probable solution. Figure 4: Log view of the VCL generated from XRD Analysis of the cuttings from Bourne #3. The key to successful stochastic modeling is that you know the answer for at least one well before the modeling begins. In order to validate the model that is constructed there has to be some method of defining success. Core data is the optimum control mechanism but cuttings, mud logs, or years of experience also has value. The interpreter can minimize the number of variables by calculating some of the mineral volumes external to the Figure 5: Two models for the same well. Model A generated six minerals and 3 fluids from eight input curves including uranium and VCL. Model B used seven inputs including gamma ray and TOC and was also an under-modeled solution. Page 4

5 stochastic model and using these volumes as input parameters. The volume of clay and/or the total organic carbon are often determined and then used in the modeling process. A Stochastic Model for Spraberry Wolfcamp Reservoirs Two stochastic models were created. Figure 5 shows the inputs and outputs for Model A and Model B. Model A generated six minerals and three fluids for a total of nine output curves. Of the eight input curves, four were externally computed and four were log curve measurements. Included among these inputs were uranium and VCL. A model built with one less input than outputs is considered an optimally determined model. In Stochastic modeling geoscientists can also over-model by using more inputs than outputs, or under-model by using fewer inputs than outputs. Model B used gamma ray and TOC among its seven inputs and it is an undermodeled solution. Model results for each mineral were then calibrated to XRD core cuttings analysis (Figure 6). The relative presence of quartz is particularly important, as the more of this mineral present, the easier it is to stimulate the reservoir. For Bourne #3, the mineral volumes from the stochastic models match the core cuttings XRD Analysis quite well. The stochastic model matched the volume of Kerogen reasonably well and certainly performed better than the Modified Passey. Pyrite and calcite were also reasonably well matched. Porosity was determined to be below 10% with fairly low water saturation. Some areas had significant water, so the reservoir would be expected to produce oil with quite a bit of water. The only mineral that did not match XRD results well was dolomite and the volume of dolomite generated from the stochastic models was slightly higher than that observed in core cuttings. The authors do not view this discrepancy as overly significant. The model was then run on Pence #1 with very plausible results. Figure 7 shows the results for both wells as defined by Model A. Figure 6: Stochastic model results for Model A were calibrated to XRD curve cuttings analysis. The relative presence of quartz makes it easier to stimulate the reservoir. Page 5

6 Figure 7: Model built on Bourne #3 and then applied to Pence #1 well. Discussion of Model Results The true benefit of stochastic modeling is that the model can be applied to other wells in the area without revision. Construction of a valid model that correctly predicts mineral, fluid, and Kerogen volumes is a time consuming process. If the model does not apply to more than one well in an area there should be serious questions about its validity and additional modeling will likely be needed. A key aspect of stochastic modeling is that there is no unique solution. A deterministic approach is an attempt to predict a real world scenario. The rock being evaluated has porosity, calcite, quartz, clays, kerogen, and other minerals in specific proportions at every depth. It is easy then to understand why different models end up with similar results when considering both models are trying to predict the same reality. There were significant differences between the two stochastic models used in this study, yet the results were very similar, as shown in Figure 8. The only noted differences between these models were that Model A predicted slightly more dolomite and Model B predicted a slightly more Kerogen. Once a model can be applied to multiple wells, geoscientists can compare water saturation, porosity, and mineralogy with confidence when designing completions and fracture stimulation programs. Consistent interpretation methods are an important aspect of field development and can be invaluable when evaluating new areas for exploration potential. Table I shows production data for the two wells evaluated. Of note is the ratio of water to oil produced. These wells were perforated and fractured in a standardized manner rather than in a custom manner to avoid the areas of heavy water saturation. If the engineers had a tailored frac program based on the stochastic model generated in this study, water production and its associated cost would likely have been reduced. Table I: Relative profiles of project wells Bourne #3 Pence #1 Production 22 months 35 months Bbls oil 18,705 38,464 Mmcf gas 18,817 51,518 Bbls water 39,797 56,012 Current oil production bbls/mo Page 6

7 Figure 8: Models A and B produced very similar results. There was more dolomite in Model A (left) and more Kerogen generated in Model B (right). Conclusions This study shows that stochastic methods can provide accurate models of unconventional reservoirs. Further, conventional logs are sufficient input for stochastic modeling. Geoscientists evaluating existing conventional fields for their unconventional potential can therefore make use of existing logs and potentially avoid expensive new logging programs. Reservoir engineers should pay more attention to petrophysics when it comes to unconventional reservoirs, given the predictive powers demonstrated in this study. Better discrimination of zones could enable better targeted perforation and fracturing jobs and minimize the water to hydrocarbon production ratio. This could lead to lower cost with equal or better total production. including how brittle the formation is, how the fractures will propagate and recommended frac height. References [1] Spears, R. W., and Jackson, S. L., 2009, Development of a Predictive Tool for Estimating Well Performance in Horizontal Shale Gas Wells in the Barnett Shale, North Texas, USA, Petrophysics, 50, p The stochastic models can be further detailed by applying rock physics analysis to predict the most and least productive zones. Geoscientists can compute Poisson s ratio, Young s modulus and bulk compressibility. These rock mechanics can then be used to predict important fracture properties Page 7