Evaluating the Application of Process-Based Models as Training Images for Multiple Point Statistics
Process-based forward models recreate stratal architecture following a series of predefined rules, coupled with user defined input parameters. A significant challenge for this approach when modelling subsurface reservoirs is that the inherent complexity of depositional systems typically means that conditioning to subsurface data is challenging or impossible. This problem can be addressed by using the accurate and realistic geometries produced by the process-based models as 3D training images for Multi-Point Statistical (MPS) modeling. To validate this combined “process based and MPS” approach, a base case geocellular model was built from the well exposed deltaic deposits of the Cretaceous Ferron Sandstone which crops out in Ivie Creek, Utah. This model was built from a UAV (drone) derived virtual outcrop and a series of sedimentary logs from the area and is 1 km2 and 50 m thick. The model is well constrained in 3D and the detail of the stratigraphic architecture is well captured, including clinoforms, bedsets and distributary channels. A process-based model of a comparable delta complex was generated using Delft 3D software. The Delft 3D model was imported into Petrel and used as a training image to condition a series of subsequent MPS simulations using different numbers of conditioning wells. The resulting simulations were validated against the base case model by visual comparison and a flow simulation. A key challenge in the use of MPS is the availability of suitable training images. Results suggest that the process-based models make excellent training images and that when combined with regional knowledge, such as depositional trends, then the combined approach is a valid methodology for producing better reservoir models.
AAPG Datapages/Search and Discovery Article #90291 ©2017 AAPG Annual Convention and Exhibition, Houston, Texas, April 2-5, 2017