Quantitative evaluation of vitrinite reflectance and atomic O/C in coal using Raman spectroscopy and multivariate analysis

Vitrinite reflectance (VRo) is a standard petrographic method for assessing thermal maturity (rank) of coal. The vitrinite reflectance technique, however, requires significant petrographic experience, can be time-consuming, and may be biased by analyst subjectivity. Correlations between coal rank and Raman spectral properties are a promising alternative that can supplant some of the limitations inherent in the VRo protocol. The traditional peak-fitting methodologies for quantifying metrics from Raman spectra, however, also suffer from analyst subjectivity that can affect correlations between analyte and spectral properties.

This research combines high-throughput Raman spectroscopy with multivariate analysis (MVA) to create calibration models for the prediction of coal rank though VRo and atomic O/C ratio. MVA techniques eliminate the ambiguous subjectivity prevalent in peak-fitting methods by evaluating the full Raman spectrum, then identifying the integral vibrational modes for constructing accurate models. Partial least squares (PLS) regression models were developed using Raman spectra and VRo values (0.23–5.23%) for 68 geographically diverse coal samples. The calibration set was validated using one-half of the samples to rigorously assess the model’s predictive accuracy. The root mean standard error of prediction was 0.19 for the VRo model and 0.014 for the atomic O/C model. Both models exhibited linear correlations, with coefficients of determination (R²) for the validation set of 0.99 (VRo) and 0.93 (atomic O/C), despite the geographic and rank diversity of the samples. This study demonstrates the applicability and power of using PLS models for the prediction of both the VRo and atomic O/C ratio from Raman spectra. The quantitative MVA protocol contained herein provides a Raman alternative to the VRo industry benchmark for coal rank that is not subject to the limitations and subjectivity of peak-fitting methods.