Unsupervised hierarchical sequence stratigraphy framework of carbonate successions

Abstract

Performing the high-resolution stratigraphic analysis may be challenging and time-consuming if one has to work with large datasets. Moreover, sedimentary records have signals of different frequencies and intrinsic noise, resulting in a complex signature that is difficult to identify only through eyes-based analysis. This work proposes identifying transgressive-regressive (T-R) sequences from carbonate facies successions of three South American basins: (i) São Francisco Basin - Brazil, (ii) Santos Basin - Brazil, and (iii) Salta Basin - Argentina. We applied a hidden Markov model in an unsupervised approach followed by a Score-Based Recommender System that automatically finds medium or low-frequency sedimentary cycles from high-frequency ones. Our method is applied to facies identified using Fullbore Formation Microimager (FMI) logs, outcrop description, and composite logs from carbonate intervals. The automatic recommendation results showed better long-distance correlations between medium- to low-frequency sedimentary cycles, whereas the hidden Markov model method successfully identified high-resolution (high-frequency) transgressive and regressive systems tracts from the given facies successions. Our workflow offers advances in the automated analyses and construction of lower- to higher-rank stratigraphic framework and short to long-distance stratigraphic correlation, allowing for large-scale automated processing of the basin dataset. Our approach in this work fits the unsupervised learning framework, as we require no previous input of stratigraphical analysis in the basin. The results provide solutions for prospecting any sediment-hosted mineral resource, especially for the oil and gas industry, offering support for subsurface geological characterization, whether at the exploration scale or for reservoir zoning during production development.