Improving a Numerical Sequence Stratigraphic Model through a Global Sensitivity Analysis; Giant Carbonate Offshore Field, Abu Dhabi

Abstract

Abstract Forward stratigraphic modelling is based on the deterministic reconstruction of depositional processes in a sequence of time steps moving forward in time. This approach is usually hindered with various, uncertain parameters. Today, uncertainty analysis using experimental design and response surface techniques is commonly used in the field of dynamic reservoir simulation. This study presents the innovative application of these techniques on forward stratigraphic modelling of a giant carbonate field from offshore Abu Dhabi, leading to the generation of multiple realizations to be used as the starting point for better geomodel construction. A variety of environmental and stratigraphic parameters are used, some of which carry an important uncertainty with regards to their range of possible values. It is therefore critical to assess their impact on the development of the basin fill – a tedious exercise for subsurface fields, whereby the only physical data come from well cores. The Experimental Design and Response Surface techniques have been innovatively applied at reservoir scale to improve the calibration of the model and to produce alternative facies distribution scenarios in the study reservoir. The idea behind this approach is, first to perform a global sensitivity analysis with a large number of parameters and simulations, and then to narrow down the uncertain domain in order to select the best stratigraphic models according to criteria of calibration quality and geological consistency. Input data for this model calibration consisted mostly of an extensive sedimentological core study carried out on several wells, and a high resolution sequence stratigraphic analysis. The quality of calibration (simulation vs core data) was assessed by two user-defined quantitative functions called Thickness Calibration Indicator and Rock Texture Calibration Indicator. Following a first manual calibration of a reference case, specific uncertain parameters (e.g. eustasy; carbonate production versus depth; carbonate production vs. time; wave parameters; gravity and wave transport; erosion rates) were selected and their ranges of values defined based on experience and knowledge of geology over the study area. Latin Hypercube Experimental Design was used to ensure a uniformly distributed sampling of the parameters. Sensitivity analysis based on the responses (texture and thickness calibration indicators) was carried out and allowed to identify the most influential parameters as well as their ranges of values yielding good calibration indicator values. A second set of simulations was then launched considering only the most influential parameters and their refined ranges. Other parameters were assigned with constant values used in the reference case model. This generated a collection of various, well calibrated models. A last filtering of simulations with the highest calibration indicator values and good geological consistency was performed to provide a handful of acceptable multi-realizations. Finally, confidence maps were computed based on the facies distribution variation of the multi-realizations compared with the reference case. This study enhanced the understanding of major controlling parameters on carbonate production and allowed modelling alternative geologically meaningful scenarios of carbonate facies distribution across the investigated reservoir.