An automatic sediment-facies classification approach using machine learning and feature engineering

The delineation of sediment facies provides essential background information for a broad range of investigations in geosciences but is often constrained in quality or quantity. Here we leverage improvements in machine learning and X-ray fluorescence core scanning to develop an improved approach to automatic sediment-facies classification. This approach was developed and tested on a regional-scale high-resolution elemental dataset from sediment cores covering various sediment facies typical for the southern North Sea tidal flat, Germany. We use a machine-learning-built classification model involving simple but powerful feature engineering to simulate the observational behavior of sedimentologists and find that approach has 78% accuracy, followed by error analysis. The model classifies the majority of sediment facies and also, importantly, highlights critical sections for further investigation. Research resources can thus be allocated more efficiently. We suggest that our approach could provide a generalizable blueprint that can be applied and adapted for the research question and data type at hand. Detection of sedimentary facies and their boundaries can be automated effectively by a combination of a machine learning classification model and feature engineering, suggesting analyses of X-ray fluorescence profiles of sedimentary cores from North Sea tidal flats in Germany

Automated summary

This study develops an improved approach to automatic sediment-facies classification using machine learning and X-ray fluorescence core scanning. The approach shows high accuracy in classifying sediment facies and identifies critical sections for further investigation, providing a more efficient allocation of research resources.