Understanding and Predicting Fractures at Tengiz – A Giant, Naturally Fractured Reservoir in the Caspian Basin of Kazakhstan

Wayne Narr¹, Dennis J. Fischer², Paul M. (Mitch) Harris¹, Thomas Heidrick², Ben T. Robertson², Karen Payrazyan^1
1ChevronTexaco Energy Technology Company, San Ramon, CA
²Tengizchevroil, Atyrau, Kazakhstan

Tengiz oil field in Kazakhstan produces from an isolated carbonate platform (areal extent of 160 km2) of Devonian and Carboniferous age consisting almost exclusively of limestone. Seismic and well data clearly show two principle regions within the buildup, platform and slope (or flank) that directly relate to reservoir quality and production characteristics. Natural fractures significantly impact producibility of the flank portion of the reservoir. 

Characterization of fractures in the Tengiz reservoir has two primary objectives:

• To advance a consistent, qualitative, geological conceptual model that allows us to understand the fracture distribution.
• To build a quantitative model for use as the basis for fluid-flow simulation.

Most Tengiz fractures formed syndepositionally due to gravitational collapse of the laterally expanding Tengiz carbonate platform. Many reservoir fractures are equivalent to neptunian dikes, which originate as syndepositional extension fractures. Syndepositional faults may also be present, but in smaller abundance. The Tengiz fractures strike parallel to the depositional margin and are in greatest abundance in the vicinity of the paleo-shelf-margin (rim) and slope. Facies showing the highest lateral growth-rate have the greatest fracture density. The Permian Capitan shelf margin of the Guadalupe Mountains of New Mexico presents a genetic analog for this style of fracturing.

Constructing a model for flow simulation involves progressing from discrete fractures to computation of effective medium flow properties for model cells. Fracture data for our reservoir model come primarily from image logs and core. These discrete fractures are converted to fracture density logs (fracture surface area/m³). We use a neural-net approach for spatial distribution of fracture properties throughout the model. This allows various distributed properties (matrix porosity, facies, seismic attenuation, etc.) to determine the spatial distribution of fracture density. The approach is similar to non-linear multiple regression; the input parameters predict the output distribution. The choice of distributed properties (input parameters) is where geologic knowledge and intuition come into play.

The final step combines fracture density, fracture geometry, and matrix permeability to compute permeability tensors for each grid cell. This step uses a boundary-element model that combines the interacting effects of fracture-flow and matrix-flow.