Premium
Advanced classification of carbonate sediments based on physical properties
Author(s) -
Insua Tania L.,
Hamel Lutz,
Moran Kathryn,
Anderson Louise M.,
Webster Jody M.
Publication year - 2015
Publication title -
sedimentology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.494
H-Index - 108
eISSN - 1365-3091
pISSN - 0037-0746
DOI - 10.1111/sed.12168
Subject(s) - geology , well logging , lithology , sedimentology , identification (biology) , support vector machine , remote sensing , geophysics , artificial intelligence , computer science , petrology , geomorphology , botany , biology
Abstract Physical properties such as bulk density (gamma ray attenuation), P‐wave velocity (primary or compressional wave acoustic velocity), electrical resistivity and magnetic susceptibility are related to characteristics of the marine sediments that, in turn, are indicative of the lithology. Non‐destructive physical properties are routinely measured during Mission Specific Platform expeditions conducted by the Integrated Ocean Drilling Program using a multi‐sensor core logger on whole cores. The goal of this study was to develop linear and non‐linear relations among physical properties and different types of carbonate sediment to identify relevant information that may aid in the classification of carbonates. The database and model presented here integrate sedimentology with physical properties data. Data were analysed using three techniques: Linear Discriminant Analysis, Random Forest and Support Vector Machines. The models that best describe the nature of the data are Random Forest and Support Vector Machines, reaching up to 79% and 74% total accuracy, respectively. This article presents an application of machine learning as a potentially useful tool for classifying sediment types, developed specifically for assisting with the challenging identification of the lithologies in coral cores. This technique can also be used for provisional core description prior to splitting, thereby enabling identification and preservation of potentially critical intervals for special analyses and studies. These methods of data analysis can also assist with sample selection for specific studies. Other applications include the interpretation of lithotypes from wireline geophysical logging data, particularly in boreholes where core recovery is poor or sampling is limited to drill cuttings.