Article pubs.acs.org/EF
Sensitivity and Identification Indexes for Fuel Oils and Crude Oils Based on the Hydrocarbon Components and Diagnostic Ratios Using Principal Component Analysis (PCA) Biplots Peiyan Sun,† Mutai Bao,*,‡,§ Fujuan Li,† Lixin Cao,† Xinping Wang,† Qing Zhou,† Guangmei Li,† and Hongxia Tang† †
Key Laboratory of Marine Spill Oil Identification and Damage Assessment Technology, North China Sea Environmental Monitoring Center, State Oceanic Administration, Qingdao 266033, China ‡ Key Laboratory of Marine Chemistry Theory and Technology, Ministry of Education, Ocean University of China, Qingdao 266100, China § College of Chemistry & Chemical Engineering, Ocean University of China, Qingdao 266100, China ABSTRACT: In total, 164 oil samples, including 56 crude oils (OILs), 12 light fuel oils (LFOs), 66 heavy fuel oils (HFOs), 12 weathered fuel oils (WHFOs), and 18 weathered crude oil (WOILs) samples, were used to screen the sensitivity and identification indexes using principal component analysis (PCA) biplots. Also, 31 oil samples, including 1 LFO, 15 HFOs and 15 OILs, were used for validation by PCA biplots. The sensitivity indexes were determined from PCA biplots based on 143 components; light n-alkanes exhibited good positive correlation with LFOs, aromatic hydrocarbons displayed positive correlations with HFOs, and terpanes and steranes were indicative of OIL characteristics. Among 43 diagnostic ratios, the sensitive diagnostic ratios were screened for LFOs, HFOs, and OILs, using PCA biplots. Thirty-eight (38) diagnostic ratios that could discriminate three types of oils were chosen: 21 diagnostic ratios were chosen for the classification of OILs and HFOs, 19 diagnostic ratios could identify OILs and fuel oils (FOs) (including LFOs and HFOs), and 6 diagnostic ratios could discriminate between LFOs and the other two types of oils (OILs and HFOs). In addition, all of the ratios could resist weathering, which means that all of the results included weathered oil samples. All of the identification indexes, including 143 hydrocarbon components, and 43, 38, 21, 19, and 6 diagnostic ratios, were validated by PCA biplots based on 31 oil samples.
1. INTRODUCTION Oil spill pollution has become a serious threat to the marine environment. Shipping is the largest source of oil spill accidents, accounting for 63.9%. Fuel oils (FOs) play an important role in shipping oil spills. In the China Sea area, fuel oil spills occur more often than crude oil spills. However, there is great complexity inherent in the investigation and identification of shipping fuel oil spill accidents, especially for heavy fuel oil, because there is a great similarity to crude oil, with respect to their physical properties and chemical fingerprints. When a mysterious oil spill occurs, determining whether the oil spill involves fuel or crude oil will guide the oil spill source identification. There are broad and narrow concepts of FOs. From the broad sense, all oils can become FOs when used as fuel. From the narrow sense, FO can refer to specific types of heavy fuel oils (HFOs), such as marine HFOs. In Europe, the concept of HFOs generally refers to a black viscous residue oil obtained via distillation. Mixtures with lighter fractions are mainly used as steam boiler and furnace fuels, as a large, slow speed diesel fuel, and as a variety of industrial fuels. However, in the United States, the concept of FOs refers to any combustible liquid or liquefied petroleum product that has a flash point of no less than 37.8 °C; it can refer to a residual FO, but also a distillate FO. The chemical compositions of FOs are not only dependent on their “parent” crude feed, but also on the refining methods, a © XXXX American Chemical Society
wide range of applications, and different demands. These factors result in FOs having different fingerprint characteristics than crude oils (OILs). Studies on the fingerprint characteristics of FOs have been previously carried out, and some of their features have been summarized. For example, Benlahcen et al.1 proposed that a “phenanthrene/anthracene ratio of 1” can be used to indicate that the contamination by polymeric aromatic hydrocarbons (PAHs) was due to combustion processes. Wang et al.2 summarized and proposed a new index, “∑other three- to sixring PAHs/∑five alkylated PAHs”, for all types of oil, which is considered to be an aromatic hydrocarbon combustion source index. Other indicators, such as “phenanthrene (phenanthrene and anthracene)”, “phenanthrene (methyl phenanthrene)”, “benzo[a]pyrene/benzo ([a]pyrene and benzo[e]pyrene)”, “benz[a]anthracene/chrysene”, “indeno(1,2,3-cd) pyrene/ (indeno(1,2,3-cd) pyrene and benzo(GHI) perylene)”, have also been used as an index for the source of hydrocarbon contamination. A European oil spill identification standard pointed out that a rich aromatic content, obvious methylphenanthrene isomers, and retene deletion are all fingerprint features of HFOs. Received: March 2, 2015 Revised: April 16, 2015
A
DOI: 10.1021/acs.energyfuels.5b00443 Energy Fuels XXXX, XXX, XXX−XXX
Article
Energy & Fuels Table 1. Hydrocarbon Compounds Used in PCA n-alkanes
PAH
abbrev
steranes and terpanes
abbrev
C9 C10 C11 C12 C13 C14 C15 C16 C17 Pr C18 Ph C19 C20 C21 C22 C23 C24 C25 C26 C27 C28 C29 C30 C31 C32 C33 C34 C35 C36 C37 C38
naphthalene C1-naphthalenes C2-naphthalenes C3-naphthalenes C4-naphthalenes phenanthrene C1-phenanthrenes C2-phenanthrenes C3-phenanthrenes C4-phenanthrenes dibenzothiophene C1-dibenzothiophenes C2-dibenzothiophenes C3-dibenzothiophenes fluorene C1-fluorenes C2-fluorenes C3-fluorenes chrysene C1-chrysenes C2-chrysenes C3-chrysenes 2-methyl naphthalene 2-methyl naphthalene biphenyl acenaphthylene acenaphthene anthracene fluoranthene pyrene benz(a)anthracene benzo(b)fluoranthene benzo(k)fluoranthene perylene benzo(a)pyrene benzo(e)pyrene indeno(1,2,3-c,d)pyrene dibenz(a,h)anthracene benzo(g,h,i)perylene C20-triaromatic sterane C21-triaromatic sterane C26,20S-triaromatic sterane C26,20R- + C27,20S-triaromatic steranes C28,20S-triaromatic sterane C27,20R-triaromatic sterane C28, 20R,triaromatic sterane retene benzo(a)fluorene benzo(b)fluorene 2-methylpyrene 4-methylpyrene 1-methylpyrene 3-methyl phenanthrene 2-methyl phenanthrene 9/4-methyl phenanthrene 1-methyl phenanthrene 2-anthracene 4-methyl dibenzothiophene 2/3-methyl dibenzothiophene 1-methyl dibenzothiophene
N C1N C2N C3N C4N P C1P C2P C3P C4P D C1D C2D C3D F C1F C2F C3F C C1C C2C C3C 2MN 1MN Bp Acl Ace An Fl Py BaA BbF BkF Pe BaP BeP IP DaA BgP C20TA C21TA SC26TA RC26TA+SC27TA SC28TA RC27TA RC28TA retene B(a)F B(b+c)F 2Mpy 4Mpy 1Mpy 3-MP 2-MP 9/4-MP 1-MP 2-MA 4-MD 2/3-MD 1-MD
C14-sesquiterpane C14-sesquiterpane-1 C15-sesquiterpane C15-sesquiterpane-1 C15-sesquiterpane-2 C15-sesquiterpane-3 C16-sesquiterpane C16-sesquiterpane-1 C16-sesquiterpane-2 C16-sesquiterpane-3 C21 tricyclic diterpane C22 tricyclic diterpane C23 tricyclic diterpane C24 tricyclic diterpane C25 tricyclic diterpane C26 tricyclic diterpane C26 tricyclic diterpane 18α-22,29,30-trisnorhopane 17α-22,29,30-trisnorhopane 17α,21β- 25-norhopanehopane 17α,21β-30-norhopane + 18α-30-norneohopane 15α-methyl-17α-27-norhopane (diahopane) 17β,21α-30-norhopane (normoretane) 18α-oleanane 17α,21β- hopane 17β,21α--hopane (moretane) 17α,21β, 22S-homohopane 17α,21β, 22R-homohopane Gammacerane 22S-17α(H),21β(H)-bishomohopane 22R-17α(H),21β(H)-bishomohopane 22S-17α(H),21β(H)-trishomohopane 22R-17α(H),21β(H)-trishomohopane 22S-17α(H),21β(H)-tetrakishomohopane 22R-17α(H),21β(H)-tetrakishomohopane 22S-17α(H),21β(H)-pentakishomohopane 22R-17α(H),21β(H)-pentakishomohopane 20S-10α(H),13β(H),17α(H)diasterane 20R-10α(H),13β(H),17α(H)diasterane 20S-5α(H),14α(H),17α(H)-cholestane 20R-5α(H),14β(H),17β(H)-cholestane 20S-5α(H),14β(H),17β(H)-cholestane 20R-5α(H),14α(H),17α(H)-cholestane 20S-5α(H),14α(H),17α(H)-ergostane 20R-5α(H),14β(H),17β(H)-ergostane 20S-5α(H),14β(H),17β(H)-ergostane 20R-5α(H),14α(H),17α(H)-ergostane 20S-5α(H),14α(H),17α(H)-stigmastane 20R-5α(H),14β(H),17β(H)-stigmastane 20S-5α(H),14β(H),17β(H)-stigmastane 20R-5α(H),14α(H),17α(H)-stigmastane
SES1 SES2 SES3 SES4 SES5 SES6 SES7 SES8 SES9 SES10 TR21 TR22 TR23 TR24 TR25 TR26A TR26B Ts Tm NOR25H H29+C29TS DH30 M29 OL H30 M30 H31S H31R GAM H32S H32R H33S H33R H34S H34R H35S H35R DIA27S DIA27R C27ααS C27ββR C27ββS C27ααR C28ααS C28ββR C28ββS C28ααR C29ααS C29ββR C29ββS C29ααR
B
DOI: 10.1021/acs.energyfuels.5b00443 Energy Fuels XXXX, XXX, XXX−XXX
Article
Energy & Fuels
were used for validation and it proved to be effective for the identification of FOs and OILs.
Historically, monitoring of oil degradation processes for weathered crude oils (WOILs) that persist after a spill have relied primarily on GC-MS analysis of 17α,21β-hopane (H30), which was determined by Prince et al.3 to be relatively resistant to biodegradation. Therefore, most biodegradation studies have been based on hopane ratios to ascertain the extent of degradation for readily degraded compounds. PCA based on 10 diagnostic PAH ratios was used to distinguish between HFOs and OILs for 47 oil samples from 15 different countries and refineries,4 but not including the LFOs, WOILs, and WHFOs. GC-MS combined with the PCA method was applied to 101 chromatograms of m/z 217 (tricyclic and tetracyclic terpanes) from oil spill samples and source oil identification, including 16 weathered spill samples.5 Principal component analysis (PCA) is a multivariate technique that analyzes a data table in which observations are described by several intercorrelated, quantitative dependent variables.6 Its goal is to (i) extract the important information from the table, (ii) represent it as a set of new orthogonal variables called principal components, and (iii) display the pattern of similarity of the observations and of the variables as points on a map. PCA has been used in many areas as an “unsupervised” method describing a data set without acknowledging it, and its application has been reported in spilled oil fingerprint identification.5 Statistical evaluation ensured an objective matching of oil spill samples with suspected source oils as well as classification into positive matches, probable matches, and nonmatches. The data analysis is further refined if two or more source oils are classified as probable match by using weighted least-squares fitting of the principal components, local PCA models, and additional information relevant to the spill case. The discriminative power of PCA was enhanced by deselecting the most uncertain variables or scaling them according to their uncertainty, using a weighted least-squares criterion.7,8 Multiway PCA was used on large numbers of specific chemical components resolved with comprehensive two-dimensional gas chromatography−flame ionization detection (GC × GC-FID) to determine multimolecular differences between oil samples as well as provide insight into the overall molecular relatedness between various OILs.9 Corilo et al.10 highlighted the utility of PCA in evaluating elemental composition data obtained by high-resolution FT-ICR mass spectrometry to correctly identify the source of environmental contamination caused by the unintended release of HFOs. Here, we mainly use PCA biplots to explore oil samples in more detail. The biplot11 provides a useful data analysis tool and allows the visual appraisal of the structure of large data matrices. It is especially revealing in PCA, where a biplot can display interunit distances and indicate the clustering of units, as well as display variances and correlations between variables. Similarities between species or sites may be gleaned from the types of plots. It is also common to interpret the axes in a biplot and treat the coordinates as scores on these axes. To understand FO fingerprint features more deeply and to be able to distinguish FOs and OILs clearly, we collected 164 oil samples and studied the sensitivity and identification indexes to discriminate different oils, using PCA biplots. The sensitivity indexes first screened for each oil sample identification, which has a positive or negative response with the oil sample. The indexes then are screened to identify different oil samples from sensitivity indexes of each oil sample (i.e., identification indexes). The sensitivity and identification indexes screened
2. EXPERIMENTAL SECTION 2.1. Oil Samples for Screening the Sensitivity and Identification Indexes. A total of 164 oil samples, including 56 crude oil samples (O1−O56), 12 light fuel oils (LFOs) (L1− L12), and 66 HFOs (H1−H66) collected from ships, as well as 12 WHFOs (WH1−WH12) collected from a ship collision accident in 2007 and sampled in 2008 and 2009, and 18 weathered crude oils (WOILs) (WO0−WO17) sampled from bioremediation field simulated experiments in 2009, were used for screening the sensitivity and identification indexes. 2.2. Oil Samples for Validation of the Sensitivity and Identification Indexes. Thirty one (31) oil samples, including 15 crude oil samples (UO1−UO15), 1 LFO (UL1), and 15 HFOs (UH1−UH15), were used for the validation of the sensitivity and identification indexes. 2.3. Sample Processing Method. The oil samples were processed in the following procedure: 0.8 g of oil samples were weighed, the oil was diluted and dissolved with hexane into 10 mL in a 10 mL volumetric flask. Then, 200 μL of the resulting solution was placed into the sample bottle, to which was added 500 μL of hexane, 100 μL of the 100 μg mL−1 n-alkane internal standard, 100 μL of the 10 μg mL−1 sterane and terpane internal standard, and 100 μL of the 10 μg mL−1 PAH internal standard. The resulting solution was then mixed well prior to GC-MS and GC-FID analyses.12,13 2.4. GC-FID and GC-MS Analysis. Analyses of n-alkane distribution were performed on a Shimadzu (Kyoto, Japan) Model GC-2010 with an FID detector. Analyses of PAHs and biomarkers (steranes and terpanes) compounds were performed on a Shimadzu Model GC/MS-QP2010 system. System control and data acquisition were achieved with GC solution and GC-MS solution software, respectively.13,14 2.5. PCA Method. PCA was employed and enabled the simultaneous analysis of many diagnostic ratios. Weathering and biodegradation were taken into account. Statistical evaluation ensured an objective matching of different types of oils. The PCA biplots were constructed by selecting over 143 characteristic components peaks (Table 1) and 43 diagnostic ratios (Table 2).
Table 2. Diagnostic Ratios Used in PCA
C
diagnostic ratio
diagnostic ratio
nC17/Pr nC18/Ph Pr/Ph (C19+C20)/(C19−C22) C23TER/C24TER Ts/Tm C29αβHOP/C30αβHOP C31αβ(S/(S+R)) C32αβ(S/(S+R)) C33αβ(S/(S+R)) C34αβ(S/(S+R)) C35αβ(S/(S+R)) C27STERαββ/(αββ+ααα) C28STERαββ/(αββ+ααα) C29STERαββ/(αββ+ααα) C29STERααα(S/(S+R)) C27STERαββ/(C27−C29)STER αββ C28STERαββ/(C27−C29)STER αββ C29STERαββ/(C27−C29)STER αββ GAM/H31 OL/H30 ∑TR/H30
DH30/H30 C2-D/C2−P C3-D/C3−P C3-D/C3−C ΣP/ΣD 2-MP/1-MP 4-MD/1-MD C21TA/RC28TA SC26TA/SC28TA RC27TA/RC28TA retene/C4−P B(a)F/4-Mpy B(b+c)F/4-Mpy 2Mpy/4-Mpy 1Mpy/4-Mpy 2-MA/ΣC1P Fl/Py P/An BaA/C (3 + 2)MP/(9/4 + 1)MP pyrogenic index
DOI: 10.1021/acs.energyfuels.5b00443 Energy Fuels XXXX, XXX, XXX−XXX
Article
Energy & Fuels
3. RESULTS 3.1. Screening the Sensitivity and Identification Index Based on the Hydrocarbon Components. A total of 143
Figure 2. PC1/PC2 biplots on three types of hydrocarbon components: (a) n-alkanes, (b) polycyclic aromatic hydrocarbons (PAH), and (c) terpanes and sterances. Figure 1. PCA biplots on all hydrocarbon components: (a) PC1/PC2, (b) PC1/PC3, and (c) PC2/PC3.
the biplots (PC1/PC2 biplot, PC1/PC3 biplot, and PC2/PC3 biplot) in Figure 1. From the PC1/PC2 biplot (Figure 1a), three types of hydrocarbon components were distributed in different zones and distinguished well by PC1 and PC2. The PAHs and light components of the n-alkanes from n-C9 to n-C18 were mainly distributed in the left of the vertical coordinates and FOs were also distributed in this region, especially the LFOs mainly
hydrocarbon components (peak areas), including n-alkanes, PAHs (including five series and individual aromatics), terpanes, and steranes (listed in Table 1), were chosen for analysis by PCA. The first three PCs (PC1, 36%; PC2, 22%; PC3, 16%), which accounted for 74% of the variance, were used to generate D
DOI: 10.1021/acs.energyfuels.5b00443 Energy Fuels XXXX, XXX, XXX−XXX
Article
Energy & Fuels
Figure 3. PCA biplots based on 43 diagnostic ratios: (a) PC1/PC2, (b) PC1/PC3, (c) PC1/PC4, (d) PC2/PC3, and (e) PC2/PC4.
gathering around the light n-alkanes. The variation in the aromatic hydrocarbons was parallel to the HFOs, which means they have good correlation to each other. Other n-alkanes, terpanes, and steranes were distributed in the right of the vertical coordinates. This area was the crude oil principal
distribution zone, which demonstrated the ability of these compounds to identify crude oil. From the PC1/PC3 biplot (Figure 1b), the distribution zones of n-alkanes, PAHs, terpanes, and steranes were clearer, and were located in the left, middle, and right areas, E
DOI: 10.1021/acs.energyfuels.5b00443 Energy Fuels XXXX, XXX, XXX−XXX
Article
Energy & Fuels
Figure 4. PCA biplots based on 38 diagnostic ratios for the discrimination of three types of oil.
Figure 6. Distribution of oil samples based on P/An, (3 + 2)MP/9/(4 + 1)MP.
Figure 7. PCA biplots based on 21 diagnostic ratios for the discrimination of OILs and HFOs.
Figure 5. PCA biplots based on 19 diagnostic ratios for the discrimination of OILs and FOs.
and OILs, based on the n-alkanes, and distributed in a borderline zone based on PAHs and terpanes and steranes. However, the HFOs and OILs could not be discriminated based only on one type of hydrocarbon. 3.2. Screening the Sensitivity Index Based on Diagnostic Ratios. Principal components were analyzed based on 43 diagnostic ratios, including ratios for n-alkanes, PAHs, and terpanes and steranes (Table 2). The first four PCs (PC1, 23%; PC2, 13%; PC3, 8%; PC4, 7%), which accounted for 50% of the variance, were used to construct biplots (PC1/ PC2, PC1/PC3, PC1/PC4, PC2/PC3, and PC2/PC4) in Figures 3a−e. LFOs, HFOs, and OILs were distinguished well in the PC1/PC2 biplot (Figure 3a), except for several oil samples that were distributed in the border area. The WOILs were gathered and distributed near the zero point of the axis, and the WFOs were gathered and distributed far from the zero point of axis. In addition, various sensitive diagnostic ratios for each type of oil were relatively obvious. 3.3. Screening the Identification Indexes for OILs, HFOs, and LFOs, Based on Diagnostic Ratios. From the
respectively. FOs were mainly in the PAHs area, the LFOs were in the upper region, while OILs had a wider distribution and WOILs were located mainly in the sesquiterpene compounds area. As we know, sesquiterpene compounds are relatively volatile15 and easily weathered.16 From the PC2/PC3 biplot (Figure 1c), the three types of hydrocarbons were mainly distributed into two regions. One was the n-alkanes and the other was the PAHs and steroid terpenoids. Because PAHs and steroid terpenoids were not well separated, they did not allow for the distinction between OILs and HFOs. However, compared to the above-mentioned two biplots, the LFOs were distinguished better from the OILs and HFOs in Figure 1c, and the LFOs distribution was relatively more concentrated. PCA was also evaluated for the 164 samples based on the nalkanes, PAHs (including five series and individual aromatics) and terpanes and steranes. The corresponding biplots of the first two PCs are shown in Figures 2a−c. From these biplots, it was clear that LFOs could be clearly distinguished from HFOs F
DOI: 10.1021/acs.energyfuels.5b00443 Energy Fuels XXXX, XXX, XXX−XXX
Article
Energy & Fuels
ratios were chosen for the classification of the three types of oils, and good results were obtained (see Figure 4). 3.4. Screening the Identification Indexes for OILs and FOs (including LFOs and HFOs) Based on Diagnostic Ratios. The PC1/PC3 biplot (Figure 3b) and PC1/PC4 biplot (Figure 3c) could both better distinguish OILs and FOs (including LFOs and HFOs). From these two figures, it was clear that there were 19 diagnostic ratios (C23TER/C24TER, Ts/Tm, C29αβHOP/C30αβHOP, C27STERαββ/(αββ+ααα), C28STERαββ/(αββ+ααα), C29STERαββ/(αββ+ααα), C28STERαββ/(C27−C29)STERαββ, GAM/H31, C2-D/C2−P, C3-D/C3−P, ∑P/∑D, retene/C4−P, B(a)F/ 4-Mpy, B(b+c)F/4-Mpy, 2Mpy/4-Mpy, 1Mpy/4-Mpy, P/An, BaA/C, and pyrogenic index) that had an effect on classifying these two types of oils, of which OILs and FOs (including LFOs and HFOs) could be better discriminated (see Figure 5). 3.5. Screening the Identification Indexes for OILs and HFOs, Based on Diagnostic Ratios. From Figure 5 (based on 38 diagnostic ratios), considering the influences of the diagnostic ratios (a larger vector size represents more influence, and vice versa), 21 sensitive diagnostic ratios (i.e., C23TER/ C24TER, Ts/Tm, C33αβ(S/(S+R)), C35αβ(S/(S+R)), C28αββ/(C27−C29)αββ, GAM/H31, DH30/H30, C2-D/ C2−P, C3-D/C3−P, ∑P/∑D, 2-MP/1-MP, retene/C4−P, B(a)F/4-Mpy, B(b+c)F/4-Mpy, 2Mpy/4-Mpy, 1Mpy/4-Mpy, 2-MA/ΣC1P, P/An, BaA/C, (3 + 2)MP/(9/4 + 1)MP, and pyrogenic index) were selected for the discrimination of OILs and HFOs (including WOILs and WHFOs). In fact, for a single diagnostic ratio, there are some differences between these two types of oils. For example, in terms of P/An, although for most OILs, P/An is bigger (>10) and that of HFOs is smaller (
