Combining Stratigraphic Forward Modelling with Multiple-point Statistics - A Case Study from Seismic to Tracer Response

Abstract

Summary Stochastic reservoir modeling is an integral part of quantifying subsurface uncertainties. Classical geostatistical methods like Gaussian random function and multi-point geostatistics (MPS) are robust and cheap in computing time. However, these methods are based on mathematical/statistical concepts and therefore lack geological plausibility. Physical modeling with stratigraphic forward modeling (SFM), on the other hand, is capable of generating detailed 3D simulations of the geological realm. Conditioning SFM to e.g. well log data is expensive and not always successful. A hybrid approach of SFM with MPS can support the conditioning. This approach generates concept driven models that match the well data while also keeping geological continuity. Experiments were done on the mature 7th Tortonian oil reservoir in, Austria. Classical geostatistical approaches failed to generate enough dynamically diverse prior models to envelop the production data. First one geological process (SFM) model was generated and conditioned to well data. The result was then used as a training image (TI) for MPS. These results better match the wells while still preserving the geological information from SFM. All simulation models have been initialized and dynamically simulated. In comparison with the common geostatistical approach, they are dynamically more diverse while being more constrained by geological concepts.