Multiple-Point Geostatistical Lithofacies Simulation of Fluvial Sand-Rich De-Positional Environment: A Case Study from Zubair Formation/South Rumaila Oil Field

Abstract

Abstract Modelling accurate geological structure is an important step in reservoir simulation. The common pixel-based algorithms are incapable of generation the most realistic geospatial patterns. However, multiple-point Facies Geostatistics (MPFG) procedure is more efficient to capture complex geological structures as it mainly based on utilizing the training images that reflect geological architecture. In this paper, we present a detailed MPFG of fluvial sand-rich depositional environment in South Rumaila Oil Field in Iraq. The fluvial depositional system was created through a 2D user-defined training image that was sampled to create a surface map. Then, neural networks algorithm was adopted to train the surface map in order to create the discrete template for pattern generation of 3D Facies distribution along with the 3D Grid Construction. The resulted training image represents a numerical geological model that contains three-dimensional information about a geological heterogeneity and structure. The facies pattern was then considered for 3D Lithofacies modelling conditioning to upscaled discrete facies distribution for 20 wells in the reservoir. Based on core data analysis, three lithofacies have been modeled and predicted for all 20 wells that were distributed main at the reservoir crest. The lithofacies are Sand, Shaly Sand, and Shale. The training image body and channels were pointed to the three aforementioned facies and it replaced the variogram to create the 3D Facies system. The resulted integrated model through multiple-point statistics has much more reasonable facies architecture than indicator simulation as it is clearly noticeable how sand channels are distributed with continuous direction towards the sea shoreface. To check how validate our model and capture the true depositional environment, many realizations were generated based on leave-one-out cross-validation technique. The sand channels continuity was achieved through spatial posterior probability of Lithofacies distribution that were incorporated in MPGF approach as 3D trend model. However, including the geometrical trend models of thickness or depth has no effect on the channels continuity. The spatial posterior probability trends were modeled from the predicted probability distributions of Lithofacies that have been created by Kernel Support Vector Machine for all 20 wells in the formation. The resulted reasonable MPGF model was considered for core permeability and log porosity distribution given each facies through Gaussian simulation. In reservoir simulation, the MPGF model has led to fast reaching the minimum error of history production matching. That reflects how is efficient considering multiple-point statistics to reconstruct complex geological structures.