What is Relevant to Flow? A Comprehensive Study Using a Shallow Marine Reservoir

These days “estimating uncertainty” is the mantra. As we do this, we ask ourselves which is better: an array of geologically simple rapidly history-matched models, or a single geologically comprehensive, carefully history-matched model. After all, uncertainty, which is normally characterized by a range of forecasts from techniques such as Experimental Design, is difficult to quantify using just one model however comprehensive it may be. Yet if forecasts are obtained from a series of simple models, how good are they? Choosing one over the other also has significant implications in the time required for modeling, and also reservoir management. Specific questions, that directly affect the cost of modeling, come to mind. These are: What is the optimal level of geological detail, especially if the uncertainty management plan includes history-matching and simulating a series of models? Can the oil trapped behind the flood front be estimated by a series of simple lateral (shales) barriers/baffles or do we always need an extensive sequence stratigraphy framework? A detailed model may not be as amenable to uncertainty estimation by the virtue of its size. Is field-scale history matching adequate or is well-by-well history matching a must? Perhaps the brute force approach of probabilistic forward modeling provides the panacea. After all the proof-ofthe-pudding lies only in the model’s ability to accurately predict field performance. Finally, we also ask should the modeling strategy, i.e., comprehensive vs. simple, be dependent on the response variable of interest, e.g. ultimate recovery factor vs. infill drill locations. As such, if ultimate recovery is the objective, a simple model may suffice. To answer questions like these, we re-visit current reservoir modeling paradigms. As a datum, we use a comprehensively modeled waterflood from Western Africa. This reservoir was modeled using extensive sequence stratigraphic techniques. The model was scaled-up from about 14 million cells to about 280,000 cells using a flow-based scale-up algorithm, carefully preserving all the mappable mudstones above flooding surfaces. History-matching for a 30-year period was systematically conducted with a team of field engineers and simulation specialists. The whole process took about a year to complete. Against this datum, we compare a series of rapidly built geological models that still honor all the data and the overall depositional architecture, and yet are significantly different from the datum geological model by the virtue of the modeling strategies implemented. The different modeling strategies vary in complexity from changes in variogram lengths and direction to simple tank models with stochastic sandstones and mudstones conditioned by well data. Various geostatistical algorithms were also investigated for facies modeling and petrophysical properties population within facies. The new models were history-matched using the conventional manual method and two separate assisted-history matching methods that use sensitivity coefficients. The question being addressed was: does constraining the geological models to the same dynamic data always create an imprint over the underlying geological variation and result in similar predictions? Preliminary results indicate that the history-matching overprint tends to mask some of the dramatic geological variations. This can have significant ramifications in modeling strategies, especially when assessing uncertainty in presence of substantial history.