%0 Journal Article %T Inversion of well logs into rock types, lithofacies and environmental facies, using pattern recognition, a case study of carbonate Sarvak Formation %+ School of Mining, Petroleum and Geophysics Engineering %+ School of Mining Engineering, University College of Engineering %+ Géosciences Rennes (GR) %A Moradi, Majid %A Tokhmechi, Behzad %A Pedram, Masoudi %< avec comité de lecture %@ 0891-2556 %J Carbonates and Evaporites %I Springer-Verlag %V 34 %N 2 %P 335-347 %8 2019-06 %D 2019 %R 10.1007/s13146-017-0388-8 %K Sarvak Formation %K Naïve Bayes classifier %K Nearest Neighbor %K Well logs interpretation %K Facies classification %Z Sciences of the Universe [physics]/Earth SciencesJournal articles %X The “facies” is a frequently used term for describing sedimentary units. In the literature, this term has been used for different purposes, as depositional environment, rock type, lithofacies, etc. In subsurface geology, the core samples are essential for facies studies. While lacking cored intervals, the well logs provide precious subsurface information, but the complexity of well log responses leads most of the time to the complexity of interpretations. In this paper, a method is proposed to study the facies types through well logs. The case study is a carbonate platform system, deposited in the upper Cretaceous, named Sarvak Formation, in one of Iranian onshore oilfields, located in the Abadan Plain. For this purpose, parametric and non-parametric (k-nearest neighbor) classifiers were applied to the dataset. Detailed petrography, assisted by core descriptions, led to 37 microfacies, grouped into three main lithofacies, four carbonate rock types and six environmental facies. Classifiers could not identify the microfacies due to a limited number of observations and high variations. However, the environmental facies were truly classified. In addition, lithofacies classification and Dunham carbonate rock typing were carried out correctly. It is discussed that k-nearest neighbor is clearly the outperformed classifier, and the classical parametric models were inadequate due to the nature of the input well logs (dependency of well logs and may be their distribution), and the overlapping of the input feature space. %G English %L insu-01612951 %U https://insu.hal.science/insu-01612951 %~ INSU %~ UNIV-RENNES1 %~ UR2-HB %~ CNRS %~ GR %~ OSUR %~ UR1-HAL %~ UR1-SDLM %~ UNIV-RENNES2 %~ TEST-UNIV-RENNES %~ TEST-UR-CSS %~ UNIV-RENNES %~ INRAE %~ TEST-HALCNRS %~ UR1-ENV %~ TEST3-HALCNRS %~ TEST4-HALCNRS %~ TEST5-HALCNRS