Facies And Permeability Prediction From Wireline Logs And Core - A North Sea Case Study

This paper details work on a field in the Southern North Sea gas province. The primary lithology is aeolian sandstone. Within the reservoir there are a variety of lithofacies including dune, inter dune, fluvial and others. These lithofacies cannot be readily identified from wireline logs alone. Two primary objectives were cited for this case study, namely facies and permeability prediction from wireline log data in uncored and partly cored wells. These objectives apply to many areas worldwide. The geological model used lacked definition in the uncored wells and permeability prediction can prove to be extremely difficult using conventional regression analysis techniques. An understanding of the geological history of the area was essential to understand this lack of conventional predictive capability. The tectonic history of the study areas is complex with several phases of subsidence and uplift with associated faulting leading to breaching of the reservoir hydrocarbon traps. Further, the tectonic history has played a part in the compartmentalization seen in the area. The presence of clay minerals is compartment dependent and is a function of temperature and burial depth. Consequently, present day depths and fluid contents were of little use in assisting with permeability prediction. Several methods were used in order to give a comparison between the conventional explicit probabilistic and less well known implicit probabilistic techniques. Conventional explicit techniques were attempted for permeability prediction and correlation coefficients of the order of 0.2 were achieved. Implicit probabilistic techniques, such as n-dimensional histograms, gave a marked improvement - up to 0.65. For facies prediction, implicit probabilistic methods such as cluster analysis and n-dimensional histograms and a less conventional explicit probabilistic technique, discriminant analysis were used. These results were promising given the paucity of data and the complexity of the fields. Of the three methods, n-dimensional histograms proved to be the most successful with greater than 70% success rate. These results are presented together with improved techniques for successfully using the implicit probabilistic methods.