Modeling Heterogeneity in a Low-Permeability Gas Reservoir Using Geostatistical Techniques, Hyde Field, Southern North Sea

Hyde field is a small (133 Gcf (FOOTNOTE *) reserves) gas field in the United Kingdom southern North Sea. The reservoir, the Permian Rotliegende Group, contains eolian, fluvial, and sabkha facies. The eolian facies constitute the dominant flow units. Illite cementation results in low reservoir quality. In the eolian facies permeability averages 2.5 md and rarely exceeds 10 md. To maximize recovery from this field, with its thin gas column and low reservoir quality, Hyde has been developed with three long-reach horizontal wells. A reservoir model that could accurately predict future production had to include a full range of heterogeneities from the effects of thin muddy laminations at the core-plug scale to the spatial distribution of eolian, fluvial, and sabkha facies over the entire field at the largest scale. By carefully defining the model's layering scheme, we incorporated strongly deterministic elements where facies trends were influenced by laterally extensive stratigraphic surfaces. The distribution of facies within stratigraphic units was modeled using geostatistical techniques. Average permeability values were assigned to the poorest quality fluvial and muddy sabkha facies. Permeability values for the eolian and sandy sabkha facies were determined by upscaling models of their internal permeability structure. The resulting distribution of permeabilities across the field, present in a 2.2-million-cell model, was upscaled into a coarser grid suitable for simulating the full field. This new model produced a significantly better match to dynamic data than did an earlier simple layer model that was used to take the field to sanction. The results of our work suggest that stochastic modeling allowed us to represent a level of heterogeneity that was not captured by the earlier simple layer model. By capturing this level of heterogeneity, we were able to achieve a significantly better match of model predictions to production data.