Carbonate Lithofacies Prediction Using Neural Network and Geostatistical 3D Modeling of Oolite Shoals, St. Louis Limestone, Southwest Kansas

Kansas Geological Survey

KGS Open-file 2006-04

Summary and Conclusions

Neural network analysis coupled with lithofacies descriptions and available well data provide a basis to construct improved fine-scale facies models using stochastic simulation methods. These models improve our understanding of the distribution and controlling factors on the external geometry of relatively thin ooid grainstone intervals in the St. Louis Limestone of southwest Kansas. Object-based simulation and indictor simulation with kriging were used to build facies distributions of oolitic deposits based on geometric data and lithofacies curves predicted from well logs using a neural network. Within the facies framework, the model illustrates the internal and external geometry and distribution of St. Louis Limestone oolitic reservoirs and associated lithofacies. The fine-scale distribution of rock properties can be assigned to each facies according to the spatial relationship built from variogram analysis. The defined static geological models can be upscaled to run simulations in order to explore responses to fluid flow.

 

The following specific conclusions are drawn from this study:

 

  1. Three-dimensional spatial distributions of the St. Louis Limestone oolitic reservoirs are modeled and explored with object-based simulation and indicator simulation with kriging. Results illustrate that the location of structural highs and general trends at the time of deposition had an impact on distribution and orientation of the St. Louis Limestone oolitic complexes. Oolitic deposits formed linear belts along the edge of a local embayment trending approximately northwest-southeast, while additional shoals were formed on the platform adjacent to, but not on, local highs.
  2. The observed pattern change in the lateral extent and geometric connectivity of modeled oolitic deposits from zone D to zone A suggests the overall agradation of oolitic deposits during relative rise in sea level (zone D to zone B) followed by deepening near the top of the St. Louis Limestone prior to erosion at the base of the Ste. Genevieve Limestone.
  3. Oolitic deposits are generally absent in the center of the embayment. Modeled oolitic complexes illustrate an agradational and the slightly progradational depositional pattern as they accumulated from zone D through zone A. Eolianite deposits show a back-stepping depositional pattern and some were accumulated on the local structural highs and some in the local lows on the slope. Eolianite deposits abundance and thickness increase upward from zone D to zone A. Tidal flat deposits are thinner, laterally extensive, and accumulate and thicken into the structural lows and toward the embayment. The content of tidal-flat deposits decreases from the bottom to the top within each zone, and is interpreted as indicating transgression and followed by agradation of shallow-marine to non-marine facies within each zone.
  4. Eolianite complexes are better represented at locations where oolitic complexes are relatively thin (i.e., structural highs and localized low areas on the slope). The processes that deposited the oolitic complexes eroded the previously deposited eolianite complexes and tidal-flat deposits.
  5. The cemented oolitic deposits have been modeled as barrier/baffle layers among the oolitic complexes and are concentrated at their edges.
  6. Uncertainty and variability of the external and internal geometry of the St. Louis Limestone oolitic reservoirs were explored using stochastic simulation methods, producing multiple equally probabilistic realizations.
  7. Object-based simulation produced more geologically reasonable facies distributions than indicator simulation. The different geostatistical approaches produce similar facies patterns and provide support to the interpretation. Because of relatively high well density, object-based simulation required longer computation times than indicator simulation.
  8. The results and statistical models are useful for understanding the geometry of oolitic complexes or other shallow marine sand bodies. The methodology can be implemented in complex depositional systems without seismic data.

Acknowledgements

The Kansas Geological Survey provided financial support as part of Lianshuang Qi's dissertation research. Thanks to Geoff Bohling and Marty Dubois for discussions, assistance, and comments. Special thanks to Dana Adkins-Heljeson for his assistance and comments on the open-file reports. We would like to thank Roxar Corporation for providing access to Reservoir Modeling System (RMS) and Geoplus Corporation for access to PETRA.

 

 


http://www.kgs.ku.edu/PRS/publication/2006/OFR06_06.html/p3-07.html

Last Modified April 2006