Skip to main content

Research Repository

Advanced Search

Statistical significance of three-dimensional stochastic fluvial reservoir modelling

Davies, Chester Huw Christopher

Statistical significance of three-dimensional stochastic fluvial reservoir modelling Thumbnail


Chester Huw Christopher Davies


Stuart Clarke

Andrew J Mitten


Fluvial systems, characterized by complex structures at multiple scales, often serve as excellent reservoirs for both hydrocarbons and carbon storage. Locating such reservoirs, and assessing their quality is challenging due to their sub-seismic nature. Reservoir modelling plays a crucial role in the prediction of distribution, and the consequent assessment of the reservoir’s viability. This study will focus on stochastic, geocellular reservoir models due to their industry usage and uncertainty associated with stochastic processes.
Whilst current research employs ten to twenty realizations for developing stochastic threedimensional fluvial reservoir models, this is derived from two-dimensional experiments, and so the applicability is questionable given the extra complexity associated with a third dimension. Little research has been done surrounding how these two-dimensional results can be applied to three-dimensions.
This novel methodology determines the optimal number of realizations by creating a sample reservoir model population. It compares the distribution of properties within this smaller population to the entire dataset, using two boundary conditions: the lower bound set by the number of realizations required to model the maximum standard deviation of the whole population, and the upper bound determined by the number of realizations required for reservoir property repetition. This search window identifies the size of the sample population that best matches the whole population, providing the total number of realizations for a statistically significant dataset.
This methodology uses three different reservoir modelling algorithms with a wide range of input parameters to generate suites of synthetic reservoir models to develop and test the proposed methodology, which is then applied to a previously established example (Tuscher Canyon). Three output parameters are retrieved from the Schlumberger™ Petrel v.2020 software representing the properties of the modelled reservoir: target fraction, average geobody thickness, and standard deviation of geobody thickness. The average of all the synthetic reservoir model suites indicated that an average of 32 realizations is required to sufficiently reduce the uncertainty of the models, independent of algorithm or model input parameters, but showed significant variability. When Tuscher Canyon is considered, the number of realizations for a statistically significant dataset is markedly different from that suggested by the synthetic dataset, meaning that there is no standardized number of realizations. The standard deviation of geobody thickness is important when reservoir modelling as it gives insight into the variation of how connected individual geobodies are and is highly variable between realizations, making it the most suitable reservoir property to be used for determining the number of realizations to use. This methodology will help to reduce uncertainty of fluvial reservoir models, resulting in better characterization, de-risking, and better assessment of economic viability.


Davies, C. H. C. (2024). Statistical significance of three-dimensional stochastic fluvial reservoir modelling. (Thesis). Keele University. Retrieved from

Thesis Type Thesis
Deposit Date Jun 13, 2024
Publicly Available Date Jun 20, 2024
Public URL
Award Date 2024-06


Downloadable Citations