Facies Modelling Based on Multi-Point Statistics (MPS) in Submarine Fan Deposits

Master's thesis in Petroleum geosciences engineeringMulti-point statistics (MPS) is a relative new geostatistical facies modelling technique that was welcomed by oil industry as a promising alternative to the classical methods (object-based and pixel-based modelling). However today, this method is yet not widely used by geo-modeller. This thesis is a conceptual study focusing on the development of methodologies and best practices for MPS application to submarine fan deposits. For the conceptual study, 6 wells are available. The facies logs are re-interpreted following published work of turbidite systems. Based on analogue data, a conceptual model is created that integrates the available well data. Training images that are used as the main driver of MPS can be regarded as the link between the conceptual model and the facies simulation. One serious limitation of training images is the request of stationary, because this can be in conflict with complex depositional environments such as turbidite systems. However this limitation has been avoided in this study thru subdividing the conceptual model into several regions and assigning to each region a different training image. The vertical dimension of the turbidites of the training images is derived from analogue data that delivers the relationship between the submarine channel width and the channel depth. Vertical and horizontal gradual changes of the facies are addressed thru setting up probability distributions. Based on the conceptual model, six facies models are carried out using different methods. Four methods are based on multi-point statistics (MPS) using different groups of training images: single layer training images of two facies, multilayer training images of two facies, multilayer training images of three facies and multilayer object modelling training images. Two additional methods are carried out using Gauss indicator simulation (GIS) and object modelling. Based on the experience gained by this study best practices are given