High-throughput screening of oil fingerprint using FT-IR coupled with chemometrics

Abstract

An important element of the oil spill emergency response is the ability to rapidly identify the properties of oil spilled. Chemometrics provides large numbers of multivariate analysis tools that allow for more extensive use of data. Fourier transformed infrared spectroscopy coupled with classification and prediction models such as partial least square (PLS) and PLS-DA (discriminant analysis) allows the rapid identification of oil type and characteristics. By searching for the maximum covariance with the variables of interest, PLS allows the visualization of relations between samples and variables. The framework of this study is based on two main steps: The first is classification of oil and the second is prediction of physicochemical properties. Separated into four main categories: crude, light fuel, heavy fuel, and lubricant, spectrums of 92 oils were calibrated to predict the oil type and physicochemical properties of 26 oils. The predictability and robustness of the model was further validated using weathered oil. The classification and prediction models have accuracy of >95%. Most of the PLS models have root mean square error of calibration and prediction ranging from 0.10–3.07 and 0.3–2.8, respectively. External cross validations using weathered oils showed high prediction accuracy (relative standard deviations <5%). By increasing the number of oil type and samples, this approach is a promising method and can be included as part of the oil spill fingerprinting protocols.