Classification of weathered petroleum oils by multi-way analysis of gas chromatography–mass spectrometry data using PARAFAC2 parallel factor analysis
Introduction
Gas chromatography–mass spectrometry (GC–MS) used in the analysis of petroleum oils [1], [2] provides an extracted ion chromatogram (EIC) for a number of prescribed m/z channels. Each EIC expresses hydrocarbon constructional isomers having the same m/z. Petroleum oils are distinguished based on the differences of patterns (known as ‘hydrocarbon fingerprints’) they exhibit in their EICs. Among thousands of different compounds that exist in oils, those which are “source specific” and “weathering stable” are used for identification of source of oil spills [1], [2], [3], [4]. Two methods, published by the American Society for Testing and Materials (ASTM D 5739-00) [2] and the Nordisk Innovations Centre (Nordtest) [4], [5] are the major approaches for source identification of oil spills by GC–MS. In the ASTM method, matching the oil spill to the suspected source(s) is through overlaying and visually comparing the EICs. With Nordtest methodology a number of diagnostic ratios (DRs, ratios of chosen isomer peak areas) of the oil spill spectra are plotted against those from each suspected source. A conclusion is then reached based on the approach of DR points to the straight-line (spill ratio = suspect ratio) within their measurement uncertainties [4]. Both methods must contend with possible degradation of a spilled oil by exposure to the environment. Important mechanisms of degradation include bacterial action and photo-oxidation. It has been shown that bacteria degrade some isomers in preference to others; for example with methyldibenzothiophenes (MDBTs), the isomers 2- and 3-MDBT are preferred by bacteria above others and 4-MDBT is most stable with respect to this mechanism; [4], [6] or in the case of methylphenanthrenes (MPhs), the greater photo-oxidation of 1-methylphenanthrene (1-MPh) over 2-MPh has been reported by different groups [7], [8]. Different weathering mechanisms will cause spectral discrepancies in the EICs and might be misleading in conventional visual comparison of chromatograms. Moreover with visual comparison of spectra there could be a risk of subjective errors [9], particularly in the cases that are not clear-cut, such as comparisons of highly similar diesel oils. The effect of biodegradation is kept to a minimum in the Nordtest methodology by exclusion of the peaks of those isomers preferred by the bacteria. However photo-oxidation [7], [8], [10] is not considered in assigning DRs, perhaps because the majority of the DRs calculated in the Nordtest approach have been first offered by geochemists as biomarkers for petroleum exploration where there is a greater interest in long term weathering mechanisms, such as biodegradation. Compared to biodegradation which is the last fate of petroleum oil in the environment [10], photo-oxidation is short term and its depletion effect on hydrocarbons is opposite to that in biodegradation. It has been shown that bacteria target 2- and 3-MP more than 9- and 1-MP, while the latter two exhibit more sensitivity to photo-oxidation [7], [8] or when MDBTs are concerned, the 4-MDBT is the most stable in terms of biodegradation and the least stable in terms of photo-oxidation [6], [7]. This could make some of the proposed polycyclic aromatic hydrocarbon (PAH) diagnostic ratios such as 4-MDBT/1-MDBT less informative in photo-oxidized oil spills. It has also been reported that while biodegradation of PAHs is less with increasing ring numbers and branches, photo-oxidation is greater [8], [10]. Therefore, the use of other PAH diagnostic ratios such as C2-DBT/C2-Ph, C3-DBT/C3-Ph, C3-DBT/C3-Chr and Retene/C4-Ph could also be misleading in photo-oxidized oils (prefixes C2 to C4 indicate the number of substituted carbons; Ph and Chr are abbreviations for phenanthrene and chrysene, respectively). It is apparent from Prince et al.'s work [10] that those ratios do not remain unchanged during the photo-oxidation process. An alternative would be the use of resistant biomarkers (such as hopanes, steranes and triaromatic steroids) to form diagnostic ratios [1], [2], [4]. They have long been utilized in assigning crude oil (or other heavy oil) spills to their responsible sources, but issues arise when lighter and refined petroleum products such as diesel oils are concerned. Many of these stable diagnostic molecules are removed (or remain in a low and so non-measurable concentration) from the oil during refining [11]. In contrast to the ASTM method, successful assessment in the Nordtest methodology requires reasonable separation of isomer chromatographic peaks and knowledge of the origin of each peak in an EIC.
Identification of various biodegraded heavy oils using the Nordtest method has been extensively reported [1], and therefore is not reiterated in this paper. In this study a number of oil samples were weathered using a simple experiment to ensure that the weathering process was dominated by photo-oxidation (and evaporation) processes, and then an objective chemometrics methodology was applied to classify the weathered oils. The reported method does not require the peaks to be separated or identified and all the isomers (i.e. the entire EIC) are included in the analysis. It is demonstrated that exploratory analysis of oil samples is possible by fitting a suitable multi-way model (PARAFAC2) to GC–MS data. Supervised pattern recognition using a proposed analysis of variance (ANOVA)-least significant difference (LSD) variable selection applied to the baseline corrected and smoothed EICs is shown to improve the discrimination among very similar diesel oils.
Section snippets
Exploratory data analysis
A GC–MS analysis of a typical oil sample provides an array of data of size (number of samples × elution time points × number of m/z channels). This three-way structure (samples × elution times × m/z channels) of GC–MS data implies that fitting a proper three-way model would be more successful than an unfolded two-way principal component analysis (PCA) in which one mode (e.g. m/z) is coalesced with another (e.g. elution time) [12]. In PARAFAC [13], the data cube is decomposed into one scores and two
Oil samples
A set of 17 different petroleum oils, ranging from transformer and lubricating oils to crude and diesel oils, as well as mixtures of them, were obtained from the New South Wales Department of Environment and Conservation (NSWDEC). A list of all studied oil samples with their kind and origin is given in Table 1.
Weathering procedure
Oil samples were subjected to a regime of weathering by placing a 2–5 mm thick slick of each oil over water in a beaker, which was then exposed on the roof of a building for 2, 7 and 15
Weathering effects
The weathering procedure applied in this work was dominated by evaporation and photo-oxidation processes. No effects due to biodegradation were anticipated as the samples were not in direct contact with microorganisms. The percentage loss of hydrocarbons (analytes) within the oil can be calculated relative to 17α(H)21β(H) hopane [10] which acts as an internal standard:where AF, HF, Aw and Hw are the concentration (or peak area) of the analyte and 17α(H)21β(H)
Conclusion
This study represents the first application of PARAFAC2 for exploratory and supervised GC–MS analysis of different types of petroleum oils. The method can form the basis of an objective identification of the source of oil spills. The weathering experiment performed here was set up in a way to be dominated by photo-oxidation and evaporation which mimics the short term fate of an oil spill in a hot sunny weather (summer in Sydney). The Environmental Protection Authorities regularly monitor places
Acknowledgements
The authors thank Mr. Stephen Fuller from the NSW Department of Environment and Conservation (now Department of Environment and Climate Change) for the analysis of oil samples and his instructive comments. D.E. thanks the Australian Government for an International Postgraduate Research Scholarship.
References (31)
- et al.
J. Chromatogr. A
(1999) - et al.
Environ. Forensics
(2002) Chemosphere
(1993)- et al.
Chemosphere
(1996) - et al.
Anal. Chim. Acta
(2001) - et al.
Spill Sci. Technol. Bull.
(2003) Chemometr. Intell. Lab. Syst.
(1997)- et al.
Chemometr. Intell. Lab. Syst.
(2006) - et al.
J. Chromatogr. A
(1998) - et al.
J. Chromatogr. A
(2002)