Geostatistical simulation is now the routine approach used in petroleum geosciences to model reservoir heterogeneities. The generation of 3D facies distributions is usually based on one of the four following approaches: indicator simulation, pluri-gaussian models, object-based models or multi-point statistics approaches. The chosen method should be the one which is best capable of representing the quantified geological information associated with the reservoir’s depositional environment: indicator variograms, rock-type rules, genetic units geometry or training images are the quantification tools respectively associated with each of the four modelling approaches.
It is fair to say that the above quantification tools are not necessarily familiar and easy to approach by geologists, who often prefer to discuss geological architecture in terms of cyclicity, coarsening/fining-upward sequences, or probability of transitions between facies (for instance). As a result the geostatistical reservoir models are not always as realistic as the geologists would like.
This project will first consist of comparing the various quantitative assumptions made by the different geostatistical methods and express those in relations with the quantitative rules used by geologists. For instance, what are the implicit and explicit architectural controls imposed by a training image vs pluri-gaussian assumption? An important subject to address will be that of the modelling of vertical vs lateral variations and geostatistical uses that can be made of geological principles such as Walther’s Law and sequence stratigraphy considerations.