Reservoir Modeling of Complex Thin-Bedded Deep Water Deposits Integrating Well Data, Seismic Inversion and Depositional Model from Offshore Sabah, Malaysia

Abstract

The studied field, located offshore Sabah, Malaysia is composed of a succession of thick-bedded sand lobes, thin-bedded heterolithics and shale dominated mass transport deposits (MTDs). While the thinly bedded units contain significant hydrocarbon reserves, the distribution and continuity of sand bodies within these units cannot generally be derived from well data alone. Modern analogues suggest that these thin beds are likely composed of small scale sandy lobes embedded into a shale background. In order to better understand the vertical and lateral connectivity of these turbiditic elements, a numerical model integrating well data and seismic inversion results was built, constrained by a conceptual depositional model based on analogue data. Multi-point geostatistics (MPS) were used as a platform to combine geological knowledge - in the form of a realistic training image, well measurements and geophysical trends. In order to reproduce observed facies proportions and architecture, the training images were generated using a novel user-guided, semi-automated workflow based on object-distance simulation. Reproducing the vertical heterogeneity of the model observed in thinly bedded units while retaining the lateral variability evidenced by seismic inversion was achieved by creating first set of facies probability cubes from the seismic inversion results using supervised neural-network estimation, then refining these probabilities by calibrating them to vertical proportion curves extracted from high-resolution facies logs. This technique enabled reproducing complex depositional patterns such as compensational stacking and hierarchical distribution of facies. It also allowed sedimentary bodies directly visible on the seismic inversion results to be integrated explicitly into the facies model. The simulated facies distribution was then used to constrain petrophysical property population, yielding detailed and realistic dynamic models. While the presented approach is specifically tailored to the modeling of thinly bedded deep water environments, the innovative techniques proposed for generating a "photorealistic" training image and for integrating seismic-scale results with high resolution well data could also be used for representing a varied set of depositional settings.