Production-focused seismic: insights from the Eagle Ford

FIGURE 1. In this outcrop image of the Eagle Ford from Lozier Canyon in southwestern Texas, a theoretical lateral has been drawn to show that wellpath planning must take into account significant vertical variation. (Images courtesy of Global Geophysical)

Geoscientists may cringe when they hear that the unconventionals are an “engineering” play or a “factory” development that is “all about the completion.” However, they can apply their understanding of complex and subtle heterogeneity to help address practical engineering issues such as well and completion stage spacing, fault avoidance, water production, and well and stage productivity.

Geologists, geophysicists, and engineers can come close to agreeing on a list of the key resource/reservoir characteristics that control well prospectivity and productivity in various unconventional plays. Such a list may look like this:

• Total organic carbon (TOC);
• Porosity;
• Thickness;
• Facies;
• Brittle/ductile quality;
• Differential stress;
• Stress field orientation;
• Faults and natural fractures; and
• Reservoir/pore pressure.

Note that this list includes both static and dynamic reservoir properties.

FIGURE 2. In this image of high-resolution South Texas 3-D seismic in which amplitude and a fault probability attribute are corendered, numerous vertical/sub-vertical low-throw faults and large fractures are clearly visible. These can negatively impact the effectiveness of fracture stimulation and also provide conduits for pressure loss. Note the complexity of the fault geometries in the map view sense.

All disciplines will note the obvious, that multiple characteristics (or variables) influence well performance; however, elaborate and passionate arguments often ensue about which variable is most influential and should be used to guide decision-making on well locations, well paths, and even stage spacing and location. Vertical and lateral heterogeneity are known and can be observed in the significant variations in well performance (even in closely spaced developments) and in outcrops (Figure 1).

Geophysicists will note that many of these characteristics have proxies among the many prestack and post-stack attributes that can be derived from recent high-resolution full-azimuth long-offset 3-D seismic data. For example, structural attributes like curvature, fault probability, and incoherence can highlight structural complexities such as faults and fractures (Figure 2); frequency attributes like spectral decomposition can correlate well with stratigraphic thickness and provide indications of gas distribution; acoustic and elastic inversion can provide attributes that relate to porosity, TOC, and rock mechanical properties; and azimuthal anisotropy can indicate fractures, differential stress, and overpressure. These seismic attributes provide 3-D understanding of the spatial variations of these characteristics. However, the challenge facing geologists and geophysicists is that the “sweet spot” revealed by one key characteristic may be totally different from that of another key characteristic. In addition, when the seismic attributes are correlated to production data, the resulting correlation coefficients for each attribute are typically in the 0.3 to 0.6 range. So when presenting these attribute proxies for key resource/reservoir characteristics, geoscientists often are presenting information that only “sort of” relates to production, and their maps or suite of attributes often show conflicting sweet spots (Figure 3).

FIGURE 3. In these oblique views of three seismic-based 3-D geological property models for three key resource/reservoir characteristics, bright colors represent the ‘good’ areas of each property. Note that the sweet spots do not all occur in the same locations in each model.

However, the “sort of related to production” nature of these key characteristics is simply a reflection of complexities seen in unconventional reservoirs and the multivariable nature of the factors that control productivity. No single variable is the primary driver of productivity, so no one characteristic provides a strong enough correlation to predict well performance by itself. In addition, having multiple maps with different sweet spots does not make the information practical or particularly useful to geoscientists or engineers.

What is needed is a technique to integrate the 3-D spatial understanding of each key characteristic into 3-D volumes that can be used to make well and completion decisions and to address production-related issues such as unexpected water production, pressure depletion, and erratic stimulation performance.

Regression modeling

One approach is to use multivariate regression modeling, which performs a simultaneous statistical analysis of multiple variables (attributes) to understand how they relate to what scientists are trying to predict (a reservoir property or production metric).

FIGURE 4. This map view shows the predicted maximum monthly gas production generated from the multivariate nonlinear regression with the more prospective areas in the warmer colors (higher values). The new wells are blind tests of the model (they were not used in the modeling exercise). Shown is a 207-sq-km (80-sq-mile) area in the Eagle Ford.

In practical terms, they must first assess how all of the available seismic attributes, completion variables, geologic attributes, and other engineering data types are

related to the property – porosity, brittleness, water-oil ratio (WOR) – or production metric – maximum month, six-month cumulative, etc. – in which they are interested. A simple way to do this is via linear regression analysis. A linear regression is performed between each attribute/data type and the property or production metric of interest. As already noted, many attributes have poor correlation coefficients in this one-to-one correlation technique. However, this initial filtering step will reveal a subset of attributes that have moderate correlations, and it is this group of “sort of related” attributes from which a portfolio of key performance indicators to be used in the multivariate regression modeling can be selected.

In the multivariate statistical analysis, a suite of seismic attribute volumes (and perhaps other data types) are mathematically related to a reservoir property or production metric, and the resultant transform is used to create a single volume that is a prediction of expected property or production values based on the seismic-property/production data relationship. One such model predicting gas production is shown in map view in Figure 4 and can be used to identify well prospectivity. In that example, lateral length, a brittle/ductile elastic inversion volume, 10 Hz and 32 Hz spectral decomposition, and azimuthal anisotropy comprised the portfolio of performance indicators, and their cumulative effect on production was integrated via the multivariate modeling to produce the single volume and map identifying the production potential of the Eagle Ford in the study area.

FIGURE 5. The image on the left shows the predicted WOR model in the background with the actual Qwater log displayed along the lateral. The model shows a good correlation with the Qwater log generated from the production logging tool. This model was extracted along the borehole and compared with the Qwater log as shown to the left. This blind test validated the seismically constrained WOR model at the stage level.

In addition to well prospectivity and production prediction, this approach can be applied to understanding the spatial distribution of static and dynamic reservoir properties. For example, Global recently helped an operator understand water production in its Eagle Ford development. One step in the analysis was to build a predictive model of WOR using data from 41 wells and seismic over 518 sq km (200 sq miles). The portfolio of performance indicators for WOR was determined to include azimuthal anisotropy, curvature, mu from elastic inversion, and thickness. Integrating these attributes in multivariate analysis created a 3-D model of WOR that had a correlation coefficient of 0.88 between actual and predicted WOR (with no single attribute correlating more than 60%). This model had sufficient resolution to analyze WOR performance at the stage level (Figure 5). The models predicted values for WOR matched production logging tool results at the stage level on a blind well left out of the analysis. This means that engineers can use this model for well planning and for the planning of individual completion stages.

A real, simple, and practical value of the multivariate analysis described above is that models of static and dynamic properties and models of production can be created that incorporate and integrate the cumulative effect of multiple key variables from both geoscience and engineering. This provides a single map or volume that can be used to pick prospective well locations and plan well and lateral paths. In fact, lateral and vertical resolution can be sufficient to also plan completion locations and designs (Figure 6). This modeling approach is a technically robust and production-focused application of seismic data that allows geoscientists to work directly with engineers to incorporate multivariable heterogeneity into well and completion planning.

FIGURE 6. These views show wellpath and fracture stage spacing options based on the production potential predicted by a multivariate nonlinear regression model of the Austin Chalk.