Depositionally Conditioned Training Images for Fluvial Sandsheet Reservoir Models: Examples From the Lower Castlegate Sandstone, Utah, USA and Jamuna River, Northern India

Abstract

The recovery of hydrocarbon reserves from fluvial reservoirs is extremely low, despite their high net-to-gross nature. Low recovery rates are attributed to a lack of geological realism within reservoir models. Training images are typically used to populate such 3D reservoir models and are governed by outcrop analogue data. However, such data is usually locally controlled by its geological context, and thus its applicability can be brought into question. This research generates more realistic, depositionally conditioned training images and compares the resulting reservoir models to a conventional sequential indicator simulation base-case model. Depositionally conditioned training images use logical rules, to create geologically realistic multi-point statistic models. This study uses both an outcrop analogue from the Upper Cretaceous Lower Castlegate Sandstone at Tuscher Canyon, Utah, and satellite data from the modern-day Jamuna River, northern India.The outcrop analogue is a ~5 km² terrestrial photogrammetric dataset of the 20 m to 60 m thick Lower Castlegate succession, which comprises an extremely high net:gross fluvial sandsheet. This dataset is supplemented with the modern satellite data of a compound barform from the Jamuna River, northern India. Key stratigraphical surfaces are used to create a deterministic framework for a pixel-based training image to be developed. The training image uses quantified statistical properties of the sedimentary architecture from both datasets, with field and digitally measured palaeocurrent indicators, to develop the empirically-based depositional conditions. The newly developed training images provided the basis for multi-point statistic models, which are then compared to more conventionally-generated, sequential indicator simulation models using the same input parameters. Results from the depositionally conditioned models show a much improved constraint on the location, size and distribution of heterogeneity-bearing architectures. Thus depositionally conditioned, multi-point statistic models provide a more geologically realistic representation of the distribution of sub-modelling-unit scale baffle heterogeneities, and their subsequent impact on fluid flow. This study therefore has implications for reducing uncertainty in secondary and tertiary phases of production, which are commonly more susceptible to the influences of sedimentary heterogeneity.

AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019