Modeling total organic carbon and organic facies: process-based and machine learning algorithms for the Espírito Santo basin

• Ano de defesa: 2024
• Autor(a) principal: Araújo, Fellippe Roberto Alves Bione de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/87559/00130000094pk
• Idioma: eng
• Palavras-chave em Português: TOC; Source rocks; Marine environment; Geological modeling; Geological data; Geochemistry; Well logs; XGBoost; Carbono orgânico; Fácies sedimentares; Modelagem geológica; Rocha geradora; Modelagem computacional; Geoquímica; COT; Rochas geradoras; Ambiente marinho; Dados geológicos; Perfis geofísicos de poços
• Link de acesso: https://app.uff.br/riuff/handle/1/33962

Understanding the spatial distribution of potential source rocks in a sedimentary basin is fundamental for paleoenvironmental interpretations, petroleum system modeling and exploration risk assessment. One way to efficiently assess source rocks' potential and distribution is through total organic carbon (TOC) and organic facies modeling. In this thesis, Sections 1 – 5 are focused in presenting process-based algorithms that were developed from a conceptual model of marine organic carbon deposition, including empirical equations for primary productivity, carbon flux and burial efficiency that allow to estimate the autochthonous and allochthonous fractions that compose TOC. Based on the calculated component fractions of organic carbon, a classification algorithm using fuzzy logics and reference values of hydrogen index and oxygen index is applied to obtain the correspondent organic facies classes. These algorithms are presented as an alternative for estimating TOC and organic facies classifications in marine environments of the Brazilian marginal basins. To test the applicability of the proposed methods, two simulation approaches are presented for the Espírito Santo Basin (ESB) offshore region: the first, a series of 2D simulations utilizing modern surficial sediments data and the second, 1D simulations of deep geological time using available data of 24 wells. The simulations show good adherence with a similar evaluation conducted for the whole South Atlantic Ocean but provide new approaches, specifically targeting the ESB. Different statistical methods and the Dynamic Time Warping technique are used to evaluate simulations' performance and obtain the optimal configurations for the ESB. The results indicate the predominance of transgressive regimens through the analyzed stratigraphic intervals and organic facies classifications of mixed continental and marine contributions, both in accordance to previously known characteristics of ESB stratigraphic record. The occurrence of TOC anomalies interpreted as related to Oceanic Anoxic Events, previously reported for some of the analyzed wells is successfully captured by the conducted process-based simulations. Ultimately, the findings presented in this work provide insights about the practical application of the proposed methods in the Brazilian continental margin. Alternatively, in Section 6, a different simulation approach, focusing on creating a generalized model for TOC prediction is presented. In this case, the XGBoost machine learning algorithm was applied to a compiled comprehensive data set containing well log and geochemical data from the ESB to run multiple solutions of parameter tuning and effectively predicting TOC for unconstrained stratigraphic intervals. This approach is then compared the traditional ΔlogR method, outperforming the latter. XGBoost effectively predicted TOC, yielding a coefficient of determination R2 of 0.71, RMSE of 0.55 and MAE of 0.30, based on the average of all 10-fold cross-validation test sets for a large dataset, containing 6353 observed TOC entries, thus, indicate the potential of machine learning for TOC prediction in large, heterogeneous data sets, configuring a promising tool for the usage of available public data sets in similar applications, such as the oil and gas (O&G) industry's exploration phase or field reassessment.