MPB-07853; No of Pages 8

Marine Pollution Bulletin xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Marine Pollution Bulletin

journal homepage: www.elsevier.com/locate/marpolbul

Chemometric techniques in oil classification from oil spill fingerprinting

Azimah Ismail a,b,e, Mohd Ekhwan Toriman e,⁎, Hafizan Juahir a, Azlina Md. Kassim c, Sharifuddin Md Zain d,
Wan Kamaruzaman Wan Ahmad c, WongKok Fah c, Ananthy Retnam c, Munirah Abdul Zali c,
Mazlin Mokhtar e, Mohd Ayub Yusri a
a
East Coast Environmental Research Institute (ESERI), Universiti Sultan Zainal Abidin, Gong Badak Campus, 21300 Kuala Terengganu, Malaysia
b
Faculty of Design and Engineering, Universiti Sultan Zainal Abidin, Gong Badak Campus, 21300 Kuala Terengganu, Malaysia
c
Environmental Health Division, Department of Chemistry Malaysia, Ministry of Science, Technology and Innovation, Jalan Sultan, Petaling Jaya, Selangor, Malaysia
d
Chemistry Department, University of Malaya, 50603 Kuala Lumpur, Malaysia
e
Universiti Kebangsaan Malaysia, Bangi, Selangor, Malaysia

a r t i c l e i n f o a b s t r a c t

Article history: Extended use of GC–FID and GC–MS in oil spill fingerprinting and matching is significantly important for oil clas-
Received 9 April 2016 sification from the oil spill sources collected from various areas of Peninsular Malaysia and Sabah (East Malaysia).
Received in revised form 23 June 2016 Oil spill fingerprinting from GC–FID and GC–MS coupled with chemometric techniques (discriminant analysis
Accepted 24 June 2016
and principal component analysis) is used as a diagnostic tool to classify the types of oil polluting the water. Clus-
Available online xxxx
tering and discrimination of oil spill compounds in the water from the actual site of oil spill events are divided
Keywords:
into four groups viz. diesel, Heavy Fuel Oil (HFO), Mixture Oil containing Light Fuel Oil (MOLFO) and Waste Oil
Oil spill (WO) according to the similarity of their intrinsic chemical properties. Principal component analysis (PCA) dem-
Peninsular Malaysia onstrates that diesel, HFO, MOLFO and WO are types of oil or oil products from complex oil mixtures with a total
Principal component analysis variance of 85.34% and are identified with various anthropogenic activities related to either intentional releasing
Discriminant analysis of oil or accidental discharge of oil into the environment. Our results show that the use of chemometric tech-
Chemometric niques is significant in providing independent validation for classifying the types of spilled oil in the investigation
Fingerprinting of oil spill pollution in Malaysia. This, in consequence would result in cost and time saving in identification of the
oil spill sources.
© 2016 Published by Elsevier Ltd.

1. Introduction communities and aggravate human health (Zakaria and Hideishige,

Oil spill pollution in the marine environment can aggravate human
health and disrupt sustainability of marine resources. The increase in ur-
banization has led to an increase in oil and gas exploration that causes
oil and gas elated substances to be released from oil wells into the
ocean water (Elijah et al., 2014). Near Malaysian coastlines, oil pollution
usually results from incidents such as collision of crude oil vessels, in the
course of transporting the crude oil from oil-rich nations of Africa and
Middle East to Asian countries like China (Zakaria and Hedeishige,
2007). Oil spill is also generated from refinery oil plants or crude oil plat-
forms, underwater pipelines as well as large-scale industrial-based ac-
tivities. Oil spill released into the marine ecosystems can be
intentional or due to accidents (Mishra and Kumar, 2015). Crude oil
spillage in the marine ecological environment can endanger marine
ecosystems if the substances persist in the waterways. The spillages pre-
vent sunlight from reaching seabed marine ecological systems and may
potentially be a threat to the natural environmental marine
⁎ Corresponding author.
E-mail addresses: ikhwan@ukm.edu.my (M.E. Toriman), hafizanjuahir@unisza.edu.my
(H. Juahir).
2007). According to the Maritime Agency of Malaysia, there were 154
cases of petroleum spillage incidents reported from 2007 until 2015
within the Malaysia coastlines (i.e. Straits of Malacca, South China Sea
and Johor Straits). Tar balls collected on the Peninsular Malaysia shore-
lines are known to originate from the Middle East countries based on
the biomarker signatures (Zakaria et al., 2000, 2001a, 2001b).
Petroleum compound consists of a complex mixture which consists
of aliphatic, polycyclic aromatic hydrocarbon and non-hydrocarbon
(Bayona et al., 2015). The unambiguous identification of source from
spilled oil in terms of properties and characteristics has become essen-
tially important to assess the fate and the environmental damage
caused by the spilled oil. The oil spill fingerprinting prominently used
recently allows extensive diagnostic capabilities to the complex mix-
tures of carbon compounds in oil samples for oil spill identification
and its environmental fate (Bayona et al., 2015). Moreover, oil spill fin-
gerprinting analysis methods require very low contaminant concentra-
tions (Guitart et al., 2012). Chemical fingerprinting (Stout and Wang,
2007) plays a main role in the investigation of pollution sources in ma-
rine environment towards to a known-source of anthropogenic activi-
ties which affect various humanitarian aspects viz. insurance, statutory
compensation (company's commitment), fatality, material loss and

http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
0025-326X/© 2016 Published by Elsevier Ltd.

Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
2 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx

other special loss (Chen et al., 2015). Assessment for oil spill restoration distinguish chemical fingerprints and analyze various types of oil sam-
is time consuming and a complete system for rehabilitation (i.e. com- ples from sources recovered from Peninsular Malaysia and Sabah (East
pensation, insurance and funding) of the marine water quality has Malaysia) waters and the application of chemometric techniques for
been implemented in the U.S, but the system is yet to be established oil type classification.
in many developing countries like China (Lai et al., 2015).
Deposition of various oil spills such as oil slicks and tar ball is among
2. Material and methods
the major concerns in our marine ecosystems. A fundamental knowl-
edge of how the concentration of weathered oil spill component evolves
2.1. Sample collection
over time along beaches and coastline areas (Mulabagal et al., 2013) is
of utmost importance for a better evaluation in assessing long-term ef-
Oil and oil-water samples were collected at thirty-two spilled oils at
fects of residues in the marine ecological systems. Oil type classification
location-selected water areas of Peninsular Malaysia and Sabah (East
obtained via intelligent laboratory analysis allows the study of physico-
Malaysia) (Fig. 1) based on in-situ observations by the Department of
chemical characteristics and background sources of the spilled oil. Gas
Environment, Malaysia (DOE) which were made between 2013 and
chromatography/flame ionization detector (GC–FID) and gas chroma-
2014. Description of the samples and the sampling locations are provid-
tography mass spectrometry (GC–MS) are the advanced analytical in-
ed in Table 1. Observational sources were not conducted in this study;
struments used to analyze weathered oil spill samples. This paper
only oil type identification was carried out.
briefly discusses oil spill fingerprinting and data interpretation through
Chemical fingerprinting of up to 32 samples of spilled oil at various
various techniques of chemometrics in ascertaining the source of spilled
locations were carried out in this study. Oil samples were obtained
oil origin. According to Satheeshkumar and Khan (2011), chemometric
from the Department of Environment (DOE), Malaysia covering the pe-
techniques are the most reliable techniques for a meaningful data re-
riod of 2013 until 2014. Upon arrival, the samples were registered and
duction and interpretation. Likewise, the principal component analysis
stored at 4 °C prior to analysis. In this study, oil spill samples were ran-
(PCA) and cluster analysis (CA) are the application of unbiased and pre-
domly collected from 32 distinctive locations (Table 1) and the samples
cise methods, while multilinear regression offers better understanding
range from C9 to 29bbs. All the organic solvents used were of analytical
and in-depth interpretation of oil spill fingerprinting in waters. Various
grade A.
applications of chemometric analyses allow reduction of the complexity
of dataset amounts for better data interpretation (Kannel et al., 2007).
According to Dominick et al. (2012), CA is the technique of unsuper- 2.2. Analytical laboratory method
vised pattern recognition functions to group variables and observations.
It provides the establishment of oil hydrocarbon signatures and evalu- Two types of solvents were used in this oil spill analysis viz. dichlo-
ates oil spill source identification of pollutant variables. Results from romethane (DCM) and hexane. For this study, specific internal reference
the clustering enable data interpretation and pattern recognition of standards commonly used in oil fingerprinting analysis were chosen,
the spilled oil. To support CA, the application of DA is important as a pat- namely, terphenyl-d14 for PAHs, C3017 β (H) for biomarker and 5α-
tern recognition methodology (Juahir et al., 2011a, 2011b; Ismail et al., androstane for 21β(H)-hopane and n-alkanes. The surrogates for this
2016). PCA application provides detection correlation between oil spill analysis consist of five mixtures of PAH compounds, namely, phenan-
variables for oil type classification and source identification in water. Ac- threne-d10, perylene-d12, acenapthene-d10, benz(a)anthracene-d12
cording to Zhou et al. (2007), large amounts of variables can be reduced and ο-terphenyl for n-alkanes. All standards and surrogates for PAHs
by PCA into smaller set of components that can be easily interpreted. and n-alkanes were purchased from AccuStandard Inc., but the internal
This study aims to establish the fingerprints via identification, standard and biomarker standard were purchased from Chiron.

Fig. 1. Map showing the various sampling locations of oil spills in Peninsular Malaysia and Sabah (East Malaysia).

Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx 3

Table 1
Oil spill samples and sampling locations.
Source from Department of Chemistry, Malaysia.

No. Sample code Sample description Sampling location Source of spilled oil

1 BAS303A Blackish, spilled (oil + water) Selangor Source from a premise drain
2 BAS304B Blackish, source (oil + water) Selangor Waste oil from public drain
3 BAS497A Blackish, spilled (oil + water) Selangor Source from premise drain
4 BAS497B Blackish, spilled (oil + water) Selangor Source from premise drain
5 BAS497C Blackish, source oil Selangor Source from inside premise
6 BAS498B Blackish, spilled (oil + water) Selangor Source from river
7 BAS498C Blackish, source oil Selangor Source from inside premise
8 BAS498D Blackish, source (oil + water) Selangor Source from inside premise
9 BAS498E Blackish, source (oil + water) Selangor Source from inside premise
10 BAS498F Blackish, source (oil + water) Selangor Source from inside premise
11 BAS741A Yellowish, source oil Selangor Source from lorry tanker
12 BAS741B Yellowish, spilled oil Selangor Source from public drain
13 BAS1092-1 Blackish, oil Sabah Source from Sabah waters (illegal oil transferring from tankers)
14 BAS1092-2 Blackish, oil Sabah Source from Sabah waters (illegal oil transferring from tankers)
15 BAS1092A Blackish, oil Sabah Source from Sabah waters (illegal oil transferring from tankers)
16 BAS1092B Blackish, oil Sabah Source from Sabah waters (illegal oil transferring from tankers)
17 BAS1383E1 Blackish, oil Johor Sample from Johor Straits (illegal oil transferring from vessels)
18 BAS1383E2 Blackish, oil Johor Waste oil from lorry tanker
19 AAS106A Blackish, source oil Perak Waste oil collected from drain
20 AAS106B Blackish, spilled (oil + water) Perak Sample from factory drum container
21 BAS156A Blackish, source oil Selangor Source from factory drum container
22 BAS156B Blackish, spilled (oil + water) Selangor Source from public drain and river
23 BAS156C Blackish, spilled (oil + water) Selangor Source from public drain and river
24 JAS294A Yellowish, oil Johor Sample from Johor Straits (illegal oil transferring from vessels)
25 JAS294B Blackish, oil Johor Source spilled from Johor Straits (illegal oil transferring from vessels)
26 JAS294C Yellowish, oil Johor Source spilled from Johor Straits (illegal oil transferring from vessels)
27 JAS294D Blackish, oil Johor
28 MAS356A Blackish, spilled (oil sludge) Melaka Source spilled on earth
29 MAS356B Blackish, spilled (oil + water) Melaka Source from factory drainage
30 SAS141A Blackish, source oil Sabah Sample from lorry tanker
31 SAS141B Blackish, spilled (oil + water) Sabah Sample from public drain
32 SAS141C Blackish, spilled (oil + water) Sabah Sample from public drain

2.2.1. Sample preparation and column clean-up
All the oil samples were analyzed using GC–FID and GC–MS that
can target specific molecular features in petroleum. Two types of
oil samples were used viz. uncontaminated oil and oil-water to con-
firm the oil classification analysis and to generate quantitative values
for statistical analyses and the results obtained were used for the in-
terpretation of the different types of potential oils present in the
samples. Initially, the oil sample was prepared by approximately
weighing 0.8 g of oil sample and diluted with 10 ml hexane to obtain
a final concentration of 80 mg/ml. Then, contaminated water-oil
sample was transferred into a pre-cleaned separating funnel. The
sample bottles were thoroughly rinsed three times with 10 ml di-
chloromethane (DCM). To determine the type of oil which is present
in the sample, 60 ml of DCM was added repeatedly into the separat-
ing funnel, and subsequently was vigorously agitated. The extracts
were combined, dried and filtered by passing them through a pre-
cleaned sodium sulphate layer. The dried extracts were then concen-
trated to the appropriate volume and solvent-exchanged with hex-
ane by rotary evaporation prior to being concentrated to 10 ml.
Prior to analysis, total solvent extractable material (TSEM) was per-
formed in order to determine the concentration of oil that had been
extracted. For TSEM, the oil weight was set at 16.0 mg and the vol-
umes of extracted sample were quantified before promptly proceed-
ing with the column clean-up analysis.
The column clean-up analysis was carried out using a pre-cleaned
glass column (30 cm × 10.5 mm I.D.) containing approximately 6.0 g
pre-cleaned silica gel (100–200 mesh, Davisil grade 923) and topped
with 0.5 cm pre-cleaned sodium sulphate which was conditioned with
20 ml hexane. Prior to exposure of sodium sulphate, the extracts were
quantitatively transferred spiked with surrogates (4 mix PAH com-
pounds and ο-terphenyl) onto the column. About 3.0 ml of hexane
was added to complete the iterative process of transferring oil into the
column. Right before exposure of sodium sulphate, approximately
12.0 ml of hexane was added into the column to elute the aliphatic
compounds, which are composed of n-alkanes and biomarkers (ad-
dressed as F1). Then, approximately 15 ml of 50% DCM in hexane was
used to elute the aromatic compounds (addressed as F2). Both fractions
(F1 and F2) were concentrated under a gentle stream of N2 until the vol-
ume reached b 0.5 ml. Then, F1 was spiked with 1.0 ppm of C3017 β (H)
and 21β(H)-hopane and 20 ppm of 5α-androstane, while F2 was spiked
with 1 ppm of d14-terphenyl. All fractions were made up to 1.0 ml for in-
jection prior to analyses.
2.2.2. GC–FID and GC–MS analysis of oil spill fingerprints
An innovative use of GC–FID and GC–MS enables the determination
of oil spill-fingerprint which can be used as an environmental forensic
technique. Petroleum biomarkers viz. steranes, sesquiterpanes and
terpanes are oil components commonly used in oil spill fingerprinting,
as they are ubiquitous in most petroleum products and their presence
in the environment tends to be recalcitrant (Elijah et al., 2014). Accord-
ing to Wang and Fingas (2007), petroleum biomarker compounds,
steranes, sesquiterpanes and terpanes are unique to the oil's source.
An oil-specific fingerprint and distinctive compositional ratio are com-
monly generated by the unique compound distribution of petroleum
biomarkers. Furthermore, it can be used for comparative purposes to a
specific oil source in various environmental matrices. In this study,
sterane, sesquiterpane and terpane ratios were chosen and their reten-
tion times were fixed for the laboratory analysis. All the analysis were
performed under the determined temperature programmed condition;
a) n-alkane fragmentation (GC–FID) was operated at 50 °C for 2 min,
then 6 °C/min to 300 °C and held for 15.5 min, all of which the total run-
ning time was 60 min; b) biomarker (GC–MS) was set at 40 °C for 2 min
duration, then followed by 6 °C/min to 300 °C and held for 25 min with
the total running time of 70.3 min; and c) for PAH (GC–MS), the temper-
ature programmed condition was set at 70 °C for 2 min, then 30 °C/min
to 150 °C and held for 0 min, then 5 °C/min for 10 min, where total run-
time was 40.7 min. Results from laboratory analysis could generate
quantitative values so that statistical analyses can be performed and

Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
4 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx
compared using repeatability limits for further interpretation of oil type
classification of samples. Separation and analyses of Diesel, Heavy Fuel
Oil (HFO), Mixture Oil containing Light Fuel Oil (MOLFO) and Waste
Fuel Oil (WFO) in oil samples are accomplished using the methods re-
ported by Wang et al. (2005a, 2005b). Analyses for n-alkane distribution
and TPHs are performed on Perkin Elmer, Clarus 680 equipped with
flame ionization detector and PE AutoSystem GC with built-in
Autosampler. The column used for this analysis was 30 m × 0.25 mm
ID DB-5. While, analyses of target PAH compounds (including five
alkylated PAH homologous groups and other EPA priority PAHs) and
biomarkers such as sesquiterpanes, terpanes and steranes are per-
formed on Agilent, 7890A GC System equipped mass-selective detector
and CTC PAL ALS autosampler. The column used was 30 m × 0.25 mm
HP-5MS fused silica columns. For both instruments, all the samples
were splitless injection mode with helium as a carrier gas. Further, the
study to perform the quantitation of biomarkers and PAHs was under
selected ion monitoring (SIM) mode.
The validation was established based on five point calibrations for
PAHs, biomarkers and n-alkanes to enable the demonstration of a linear
range of standard. From the standard curve, relative response factor
(RRF) for each component was calculated based on the abundance of
known concentration of standard to the internal standard using the
equation below:
As Cis
RRF ¼
Ais Cs
where, As is denoted as a peak area for standard, Ais is a peak area for in-
ternal standard, Cs is a known concentration of standard and Cis is a con-
centration of internal standard. The concentration of each component is
further calculated using the following equation:


μg As Wis D
Concentration ¼
g Ais RRFWs
where, Ais and As represent the peak areas of internal standard and sam-
ple, Wis and Ws are the amounts of internal standard and weight of sam-
ple respectively, and D is a dilution factor. Method detection limit for
this analysis is 0.03 μg/ml of each component.
2.3. Statistical analysis: chemometric methods
The chromatographic data produced during the analysis were
imported into the computing environments for processing. The chro-
matographic data were produced during the GC based on the concentra-
tions of each compound from all the thirty-two (32) samples. Here, the
GC–MS data were exported to file formats which can be recognized by in-
tuitive commercial data analysis software such as Excel 2007, XLSTAT
2015 and JMP10 for Windows. Once the data are available in the desired
program, the first step is to process them prior to multivariate data anal-
ysis, to reduce data variability unrelated to the chemical composition (viz.
baseline, retention time shifts and concentration effects). Several chemo-
metric techniques were applied in this study for oil classification assess-
ment of oil spill samples. In this work, comprehensive statistical
analyses were performed using XLSTAT 2015, which is a Microsoft
EXCEL statistical add-in and a highly efficient statistical analysis software.
2.3.1. Discriminant analysis (DA)
Reliable classification of observations in CA analyses can be vali-
dated by the application of DA based on a set of variables known as
predictors. DA determines the oil spill variables and discriminates
them into two or more groups (clusters) (Zhou et al., 2007; Kannel
et al., 2007; Blundell et al., 2009; Juahir et al., 2011a, 2011b;
Retnam et al., 2013; Fermo et al., 2013; Ismail et al., 2016). According
to Kannel et al. (2007), the applied DA on the original datasetoffers
similar discriminant ability towards the original data set with or
Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
without a standardization in constructing the Discriminant Factor
(DF). Hence, the original data set is not necessarily significant for
standardization and the following function (Eq. (4)) (Kannel et al.,
2007):
Xn
f ðGiÞ ¼ ki þ j¼1
wij  pij ð4Þ
where i denotes the number of groups (G), ki is the constant inherent
to each group, n is the number of parameters used to classify a set of
data into given group and wij is the weight coefficient which is
assigned by DA to a given parameter (pij) (Zhou et al., 2007).
In this study, DA is used to construct Discriminant Factors (DFs)
based on three modes viz. standard mode, forward stepwise mode
and backward stepwise mode. In the forward stepwise mode, the
most significant oil spill variables are added one-by-one until no signif-
icant changes are observed. However, in the backward stepwise mode,
the variables of oil spill were removed one-by-one, starting with the
least significant variables until no significant changes are seen. DA was
performed on the oil spill raw data matrix with columns consisting of
all the oil spill compounds as independent variables and the rows com-
prising the dependent variables.
2.3.2. Principal component analysis (PCA)
According
to Kannel et al. (2007), the PCA technique is used to identify the most
significant variables that determine the spatial and temporal variations
of the dataset by excluding the less significant parameters with mini-
mum loss of important original information. PCA constructs latent vari-
ables (scores) which are the linear combinations of the original data
variables of oil spill compounds. In PC analyses for oil spill compounds,
latent variables with eigenvalue N1.0 are considered significant, and
PCA can be expressed with the following equation:
zij ¼ ail X lj þ ai2 X 2 j þ ai3 X 3 j þ aim X m j ð5Þ
where, z is denoted as the component score, a is the component load-
ing, X is the measured original variables, i is the component number
and m represents the number of the original variables. The loadings
ascertain the significance of the oil spill variables for each
component.
3. Results and discussion
3.1. Oil spill fingerprints
Oil spill fingerprinting analysis is applied in this study for classifica-
tion of spilled oil via the application of common petroleum biomarkers
and PAHs to track the spilled oil candidates. Chemometric techniques
are also applied for improving diagnostic techniques in the field of oil
spill-fingerprinting.
Research by Peters and Moldowan (1993) and Wang et al., (1997)
proved that the hydrocarbon structural assignment from the various
oil spill fingerprints was successfully accomplished through the mass
spectra pattern recognition, GC retention time data comparison with
reference standards and empirical data from previous research. The
source identification of spilled oil has been successfully conducted
through the use of the tiered methods (Wang and Fingas, 2003;
Wang et al., 2004; Stout and Wang, 2007). Fig. 2 reveals the GC–FID
chromatogram of selected oil samples which consists of diesel,
heavy fuel oil, lubricating oil, waste oil and waste oil containing mix-
ture of light fuel oil.
In this study, 94 compounds consist of n-alkanes (C10–C40), bio-
markers (sesquiterpanes, steranes and terpanes) and PAHs (alkylated
PAHs and EPA Priority PAHs) quantified and analysis of biomarkers
and PAHs performed via GC–MS run in selected ion monitoring (SIM)
mode, while n-alkane quantification was performed using GC–FID.
 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx 5

Fig. 2. GC–FID chromatogram profiling for various types of oils distinguishes by its distribution of n-alkanes a) HFO, b) MOLFO, c) lube oil, d) WO and e) diesel.

Each sample reveals different patterns of chromatogram (Fig. 2), de-
pending on the source background. Broader unresolved complex mix-
tures (UCMs) were observed in Fig. 2a and d, whereas the narrow
UCM pattern was detected on lube oil and MOLFO, but no UCM has
been identified in diesel.
3.2. Chemometric oil spill pattern recognition
Classification of spilled oil in the diagnostic process of oil spill finger-
printing is done through the oil classification analysis. Chemometric
techniques are used in analyzing the hydrocarbon spill source and the

Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
6 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx

Fig. 3. Dendrogram showing the clustering of hydrocarbon spill source into four clusters/groups.

nature of its background in addition to intelligent lab analysis for classi-
fying oil spill signatures (Christensen and Tomasi, 2007a, 2007b).
Hence, these methods offer a better explanation of the data saves time
in achieving the objective of identifying types of oil spills from various
sources compared to other common methods (Christensen and Tomas
2007a, 2007b).
To explore the similarity or differences of the hydrocarbon source oil
at the actual spilled source sites and its background sources, classifica-
tion and clustering viz. hierarchical agglomerative cluster analysis
(HACA) were performed on the oil spill data. According to the HACA
output, oil spill variables that have similar properties or similar-looking
fingerprints are clustered together as illustrated in the dendrogram
(Table 1 and Fig. 1). Oil spill classification, done through HACA, was
based on standardized data using Ward's method with squared Euclid-
ean distances. These data show that Cluster 3 is abundant in the sample
of oil spill sources, dominating the characteristic fingerprint in spill
source classification of CA (Fingas et al., 2002; Rosenbauer et al., 2010).
The level of environmental pollution in the waters is based on the
level of unique source-specific nature of fingerprints. The relative distri-
bution of various hydrocarbon spill sources that demonstrate the same
identity proves that these samples originated from the same source or
bears similar oil. Clearly, Cluster 3 dominates the spilled oil and this is
most probably due to illegal transferring of oil between vessels both
from the areas of Sabah and Johor waters, as well as waste oil collected
from the factory drainages. There are fourteen oil spilled samples in
Cluster 3 (BAS 497C, BAS497B, BAS498C, BAS498F, BAS1092-2,
BAS1383-E1, BAS1383-E2, BAS1092-A, BAS1092-B, JAS294A, JAS294C,
SAS141A, SAS141B, SAS141C).
DA analysis is a statistical validation process of the CA output. To fur-
ther assess the dataset, DA was applied after the grouping of the candi-
date samples of oil spill. The clusters were treated as dependent
variables and the measured oil spill compounds as the independent var-
iables. Results obtained from DA show that the sample oil sources are
discriminated into four significant groups of spilled sources in this
study. In the standard mode DA, the oil samples are discriminated to
99.51% classification matrix accuracy, performed by discriminant func-
tion (DF) (Fig. 3). Output from Wilk's Lambda test gives Lambda value
of 0.00 (p b 0.0001), thus the alternative hypothesis (Ha) is acceptable.
Further, forward stepwise mode DA provides a hundred percent
(100%) classification matrix accuracy separation of oil spill into four
clusters/categories. Forward stepwise and backward stepwise of DA
show that Diesel, HFO, MOLFO and WO (Fig. 4 and Table 2) are the dis-
criminant variables (p b 0.0001). The clustering for oil spill samples
from CA analysis is further confirmed based on this analysis. The cluster-
ing of the hydrocarbon-oil apportionment into four groups; diesel,
MOLFO, HFO and WO is based on the intrinsic chemical properties in
each oil spill sample.
Thus, DA outputs suggest that forward stepwise/backward stepwise
modes of DA with 93 variables are able to discriminate the hydrocarbon
oil spill, performed by discriminant function (DF) with four discriminat-
ing variables on the waters of Peninsular Malaysia and Sabah (East Ma-
laysia). Output from DA forward stepwise/backward stepwise modes

Table 2
Classification matrix of DA for all data measurement in oil classification.

Types of oil Predicted types of oil by DA Total % correct

DIESEL HFO MOLFO WO

Standard mode of DA
DIESEL 2 0 0 0 2 100.00%
HFO 0 12 0 0 12 100.00%
MOLFO 0 0 9 0 9 100.00%
WO 0 0 0 7 7 100.00%
Total 2 12 9 7 30 100.00%

Stepwise forward of DA
DIESEL 2 0 0 0 2 100.00%
HFO 0 12 0 0 12 100.00%
MOLFO 0 0 9 0 9 100.00%
WO 0 0 0 7 7 100.00%

Stepwise backward of DA
DIESEL 2 0 0 0 2 100.00%
HFO 0 12 0 0 12 100.00%
MOLFO 0 0 9 0 9 100.00%
WO 0 0 0 7 7 100.00%
Fig. 4. Plot of discriminant functions for hydrocarbon oil from the actual sites of oil spill
Total 2 12 9 7 30 100.00%
event based on the intrinsic chemical properties of DA forward stepwise.

Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx 7

Table 3
Factor loadings PCA after varimax rotation for oil spill chemical compound.

Statistics VF1(FL) VF2(FL) VF3(FL) VF4(FL)

Chemical C33(0.703); C39(0.741); C2-N(0.754); C23Tr(0.944); C24Tr(0.924); C10(0.720); C11(0.952); C12(0.989); C24(0.900); C25(0.970);
compound C3-N(0.891); C4-N(0.834); C0-P(0.906); 27Ts(0.943); 27Tm(0.978); 29ab(0.889); C13(0.985); C14(0.897); C15(0.965): C26(0.985); C27(0.989);
C1-P(0.932); C2-P(0.949); C3-P(0.939); 30ab(0.989); 31s(0.870); 32r(0.952); C16(0.970); C17(0.988); C18(0.982); C28(0.992); C29(0.992);
C4-P(0.949); C1-D(0.972); C2-D(0.981); 33s(0.988); 33r(0.987); 34s(0.984); C19(0.982); C20(0.952); C21(0.898); C30(0.992); C31(0.990);
C3-D(0.984); C4-D(0.983); C1-F(0.718); 34r(0.977); 35s(0.984); 35r(0.786); C22(0.738); CO-F (0.737); BpH(0.735) C32(0.989)
C2-F(0.716); C3-F(0.805); C1-C(0.982); 27bbR(0.975); 27bbS(0.972);
C2-C(0.981); C3-C(0.971); An(0.945); 28bbR(0.965); 28bbS(0.958);
FI(0.825); Py(0.949); BaA(0.942); 29bbR(0.979); 29bbS(0.977)
Pe(0.907); BeP(0.982); BP(0.923);
BbF(0.983); BkF (0.721); DA(0.937);
IP(0.713)
Eigenvalue 34.462 22.071 13.926 8.909
Variance (%) 37.056 23.732 14.974 9.580
Cumulative 37.056 60.788 75.762 85.342
variance
(%)

Note: Eigenvalue N 5; strong factor loadings N 0.7.

clearly shows that this technique is a useful tool to identify the discrim-
inating variables in the spatial variance of the oil spill samples.
3.3. Identification of oil spill sources of variation based on GC–FID/GC–MS
and PCA
PCA was applied in this study in order to reduce the complex mix-
ture oil data and to obtain a useful data interpretation of the most signif-
icant discriminating variables. PCA allows the complex mixture data
extracted from GC–FID and GC–MS to be classified and this is not usually
achievable through conventional fingerprints. PCA Varimax Factors
(VFs) (with mean data) yielded four variance factors comprising
85.34% of the total variance for complex mixture of oil spills (Table 3).
Varimax Factor (VF1) is responsible for 37.06% of the total oil spill com-
pounds (Fig. 5). VF2 is responsible for 23.73% of the total variance of oil
spilled in Peninsular Malaysia and Sabah (East Malaysia). Eigenvalues
greater than five are the main criterion considered in the varimax rota-
tion. Varimax Factor (VF1) with the highest loading factor recorded cor-
related with the individual saturated hydrocarbon oil variables of n-
alkanes (C10–C40), isoprenoids pristane, phytane, naphthalene (C0N–
C4N), phenantherene (C0P–C4P), dibenzothiophene (C0D–C3D),
flourene (C0F–C3F), chrysene (C0C–C3C) and other EPA priority PAH pol-
lutants (Acl, Ace, BpH, An, FI, Py, BaA, Bph, Pe. BeP, BP, BbF, BkF, BaP, DA
and IP) (Table 3). By contrast, VF1 is negatively correlated with other
compounds of EP priority PAHs such as C23Tr–29bbS. The main source
of n-alkanes (C10–C40), isoprenoids pristane, phytane, naphthalene
(C0-N–C4-N), phenantherene (C0P–C4P), dibenzothiophene (C0D–
C3D), flourenes (C0F–C3F), chrysene (C0C–C3C) and other EPA priority
PAH pollutants is contributed mainly from illegal oil transferring from
vessels (within the water areas of Johor and Sabah) and discharge
from the factories (producing oil from used tyres).
VF2 is positively correlated with other EPA priority PAH pollutants
contributed from the illegal or unintentional discharge of waste oil
and petroleum product related from the factories. According to Stuart
et al. (2004), the PAH with higher-molecular-weight from four to six
rings dominated the hydrocarbon compounds in the oil complex mix-
ture. VF3 and VF4 reveal high correlations with hydrocarbon oil com-
pound of n-alkanes reflecting major contribution from waste oil from
lubricant oil and diesel. Research by Bayona et al. (2015) states that n-
alkane has a prevalent molecular marker in the environment and
through the specific characteristic fragmentation, n-alkane identifica-
tion can be easily conducted.
The score plot of PCA reveals that the hydrocarbon oil from the ap-
portionment of Diesel, HFO, MOLFO and WO was detected based on
the observations as shown in Fig. 5. Two samples out of 93 observations
viz. 24 and 12 are seen as outliers as it lies far away from the centre of oil
samples in the score plot. Sample outliers in the oil spill fingerprinting
could be due to the type of spilled oil discharged intentionally or acci-
dentally from the industrial activities, anthropogenic activities and con-
tamination of the substance by the external factors after they were
released into the environment. This information can thus be applied
as evidence in the validation of oil classification - from the low variabil-
ity of scores for oil spill in the oil process validation compared to the
total variability of oil samples in the dataset - that the use of statistical
techniques (PCA) for oil types distinguishable is useful. Consequently,
PCA can be applied to separate the oil types based on GC–FID and GC–
MS fingerprints and provide an alternative methodology from the nor-
mal standard practice or conventional methods.

4. Conclusion

Special attention to oil spill fingerprinting through GC–FID and GC–

Fig. 5. Score plot of PCA, the oil samples in the analysis replicates the Diesel, HFO, MOLFO
and WO.
MS in Malaysia is significantly advanced for oil type classification of oil
spill samples collected from various sources in Peninsular Malaysia
and Sabah (East Malaysia). DA reveals that there are four (4) types of
oils found in the oil samples viz. Diesel, HFO, MOLFO and WO. PCA val-
idated and confirmed that there are four (4) types of oil in the sample
that are responsible for 85.34% of the total spilled oil variation in Penin-
sular Malaysia and Sabah. Further analyses reveal that C10–C29ββs, pris-
tane and phytane are mainly responsible for the Varimax Factor total
variance of 85.34%. These oil spill pollutants can be attributed to various
anthropogenic activities related to the industries as well as port and
shipping activities. Nevertheless, natural weathering process may also
play a role as a source of these oil spills. From the application of various

Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
8 A. Ismail et al. / Marine Pollution Bulletin xxx (2016) xxx–xxx
chemometric tools, it is observed that the oil spill fingerprinting can be
optimally performed and it would save time and reduce the cost of oil
classification from the conventional practice. Importantly, the applica-
tion of chemometric techniques to oil spill fingerprinting from the
GC–FID and GC–MS is a complementary lead to oil type classification
from different sources to increase the objectivity of analysis, saves
time and cost effective based on the finding of this study. This will be
valuable to the relevant authorities.
Acknowledgements
The authors are thankful to East Coast Environmental Research Insti-
tute (ESERI), UNISZA, Department of Chemistry Malaysia, Department
of Chemistry, University of Malaya, Universiti Kebangsaan Malaysia
and Integrated Envirotech Sdn. Bhd. for their assistance, valuable com-
ments and support.
References
Bayona, J.M., Dominguez, C., Albaiges, J., 2015. Analytical developments for oil spill finger-
printing. Trends Environ. Anal. Chem. 5, 26–34.
Blundell, a., Hannam, J.a., Dearing, J.a., Boyle, J.F., 2009. Detecting atmospheric pollution in
surface soils using magnetic measurements: a reappraisal using an England and
Wales database. Environ. Pollut. 157 (10), 2878–2890.
Chen, X., Yu, F., Zao, C., Liu, X., Liu, G., 2015. Research of 3D oil spill response drill system.
Aquat. Procedia 3, 15–30.
Christensen, J.H., Tomasi, G., 2007a. A multivariate approach to oil hydrocarbon finger-
printing and spill source identification. Oil Spill Environ. Forensics XII, 293–325.
Christensen, J.H., Tomasi, G., 2007b. Practical aspects of chemometrics for oil spill finger-
printing. J. Chromatogr. A 1169 (1–2), 1–22.
Dominick, D., Juahir, H., Latif, M.T., Zain, S.M., Aris, A.Z., 2012. Spatial assessment of air
quality patterns in Malaysia using multivariate analysis. Atmos. Environ. 60, 172–181.
Elijah III, R., Buffy, M.M., Aminah, R., Edward, A., 2014. Oil source- fingerprinting in sup-
port of polarimetric radar mapping of macondo-252 oil in Gulf Coast marshes. Mar.
Pollut. Bull. 2014 (85), 85–95.
Fermo, P., Beretta, G., Facino, R.M., Gelmini, F., Piazzalunga, A., 2013. Ionic profile of honey
as a potential indicator of botanical origin and global environmental pollution. Envi-
ron. Pollut. 178, 173–181.
Guitart, C., Readman, J.W., Bayona, J.M., 2012. Comprehensive Sampling and Sample Prep-
aration. pp. 297–316.
Ismail, A., Toriman, M.E., Juahir, H., Sharifudin, M.Z., Abdul Habir, N.L., Retnam, A.,
Kamarudin, M.K.A., Omar, R., Azid, A., 2016. Spatial assessment and source identifica-
tion of heavy metals pollution in surface water using several chemometric tech-
niques. Mar. Pollut. Bull. 106 (1–2), 292–300.
Juahir, H., Zain, S.M., Mokhtar, M.B., 2011a. Spatial assessment of Langat river water qual-
ity using chemometrics. J. Environ. Monit.
Juahir, H., Zain, S.M., Yusoff, M.K., Hanidza, T.I.T., Armi, A.S.M., Toriman, M.E., Mokhtar, M.,
2011b. Spatial water quality assessment of Langat River Basin (Malaysia) using
environmetric techniques. Environ. Monit. Assess. 173, 625–641.
Please cite this article as: Ismail, A., et al., Chemometric techniques in oil classification from oil spill fingerprinting, Marine Pollution Bulletin
(2016), http://dx.doi.org/10.1016/j.marpolbul.2016.06.089
Kannel, P.R., Lee, S., Kanel, S.R., Khan, S.P., 2007. Chemometric application in classification
and assessment of monitoring locations of an urban river system. Anal. Chim. Acta
582, 390–399.
Lai, W., Hu, Z.W., Lin, D., Zhou, W., 2015. A damage assessment model of oil spill accident
combining historical data and satellite remote sensing information: a case study in
Penglai 19-3 oil spill accident of China. Mar. Pollut. Bull. 258–271.
Mishra, A.K., Kumar, G.S., 2015. Weathering of oil spill: modeling and analysis. Aquat.
Procedia 4, 435–442.
Mulabagal, V., Yin, F., John, G.F., Hayworth, J.S., Clement, T.P., 2013. Chemical fingerprint-
ing of petroleum biomarkers in Deepwater Horizons oil spill samples collected from
Alabama shoreline. Mar. Pollut. Bull. 70 (1–2), 147–154.
Peters, K.E., Moldowan, J.M., 1993. The Biomarker Guide. Interpreting Molecular Fossils in
Petroleum and Ancient Sediments. Prentice Hall, Englewood Cliffs, N.J., p. 363.
Retnam, A., Zakaria, M.P., Juahir, H., Aris, A.Z., Zali, M.A., Kasim, M.F., 2013. Chemometric
techniques in distribution, characterisation and source apportionment of polycyclic
aromatic hydrocarbons (PAHS) in aquaculture sediments in Malaysia. Mar. Pollut.
Bull. 69 (1–2), 55–66.
Rosenbauer, R.J.C., P.L., Angela, L., Lorenson, T.D., Hostettler, F.D., Thomas, B., Wong, F.L.,
2010. Reconnaissance of Macondo-1 well oil in sediment and tar balls from the
Northern Gulf of Mexico Shoreline, Texas to Florida. U.S. Geol. Surv. Open File Rep.
2010–1290, 22.
Satheeshkumar, P., Khan, A.B., 2011. Identification of mangrove water quality by multi-
variate statistical analysis methods in Pondicherry coast, India. Environ. Monit. As-
sess. 184 (6), 3761–3774.
Stout, S.A., Wang, Z., 2007. 1-Chemical fingerprinting of spilled or discharged petroleum -
methods and factors affecting the petroleum in the environment. Oil Spill Environ-
mental Forensics. 1–53.
Wang, Z., Fingas, M., Lambert, P., Zeng, G., Yang, C., Hollebone, B., 2004. Characterization
and identification of the Detroit River mystery oil spill (2002). J. Chromatogr. A
1038, 201–214.
Wang, Z.D., Fingas, M., Landriault, M., Sigouin, L., Feng, Y.P., Mullin, J., 1997.
Usingsystematic and comparative analytical data to identify the source of
anunknown oil on contaminated birds. J. Chromatogr. A 775, 251–265.
Wang, Z.D., Fingas, M.F., 2003. Development of oil hydrocarbon fingerprinting
andidentification techniques. Mar. Pollut. Bull. 47, 423–452.
Wang, Z.D., Yang, C., Fingas, M., Hollebone, B., Landriault, M., Hansen, A., Christensen,
2005a. J. Environ. Sci. Technol 39, 8700.
Wang, Z., Yang, C., Fingas, M., Hollebone, B., Peng, X., Hansen, A., Christensen, J., 2005b.
Characterization, weathering, and application of sesquiterpanes to source identifica-
tion of spilled lighter petroleum products. Environ. Sci. Technol. 39, 8700–8707.
Zakaria, M.P., Hideishige, 2007. 16 - case study: oil spills in the Straits of Malacca, Malay-
sia. Oil Spill Environ. Forensics 489–504.
Zakaria, M.P., Horinouchi, A., Tsutsumi, S., Takada, H., Tanabe, S., Ismail, A., 2000. Oil pol-
lution in the Straits of Malacca, Malaysia: application of molecular markers for source
identification. Environ. Sci. Technol. 34 (7), 1189–1196.
Zakaria, M.P., Takuda, T., Takada, H., 2001a. Polycyclic aromatic hydrocarbon (PAHs) and
hopane in stranded tar-balls on the Coasts of Peninsular Malaysia: application of bio-
markers for identifying sources of oil pollution (2001). Mar. Pollut. Bull. 42 (12),
1357–1366.
Zakaria, Takuda, T., Takada, H., 2001b. Polycyclic aromatic hydrocarbon (PAHs) and
hopane in stranded tar-balls on the Coasts of Peninsular Malaysia: application of bio-
markers for identifying sources of oil pollution (2001). Mar. Pollut. Bull. 42 (12) (pp.).
Zhou, F., Guo, H., Liu, Y., Jiang, Y., 2007. Chemometrics data analysis of marine water qual-
ity and source identification in Southern Hong Kong. Mar. Pollut. Bull. 54, 745–756.