Subscriber access provided by University of Winnipeg Library
Environmental Measurements Methods
Rapid Identification of Deepwater Horizon Oil Residues using X-Ray Fluorescence Anna P. M. Michel, Alexandra Morrison, Charles Marx, and Helen K. White Environ. Sci. Technol. Lett., Just Accepted Manuscript • DOI: 10.1021/acs.estlett.8b00589 • Publication Date (Web): 26 Nov 2018 Downloaded from http://pubs.acs.org on December 4, 2018
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 13
Environmental Science & Technology Letters
1
Rapid Identification of Deepwater Horizon Oil
2
Residues using X-Ray Fluorescence
3
Anna P.M. Michel†*, Alexandra E. Morrison†, Charles T. Marx‡, and Helen K. White‡
4 5
† Department of Applied Ocean Physics and Engineering, Woods Hole Oceanographic
6
Institution, 266 Woods Hole Rd, Woods Hole, MA, 02543, USA.
7
‡ Department of Chemistry, Haverford College, 370 Lancaster Avenue, Haverford PA, 19041,
8
USA
9
10
11
* E-mail:
[email protected] ABSTRACT
Oil residues are found on Gulf of Mexico (GoM) beaches due to the Deepwater Horizon
12
(DWH) incident.
Yet, natural seepage and other anthropogenic inputs also contribute oil
13
residues to these beaches. Therefore, in order to identify the sources of oil residues found on
14
beaches, especially after a spill, it is critical to have techniques that can be used in the field, can
15
provide rapid source identification, and can be used easily by response workers. Here we present
16
the utility of using handheld x-ray fluorescence (XRF) for rapid source identification of oil
17
residues. By coupling XRF data with a machine-learning model, DWH samples could be
18
distinguished from non-DWH samples with 95% accuracy. This approach allows us to analyze
19
bulk samples without sample preparation, paving the way for utilizing XRF in the field for
20
immediate oil residue source identification.
ACS Paragon Plus Environment
1
Environmental Science & Technology Letters
21
Page 2 of 13
INTRODUCTION
22
The presence of oil residues in the marine environment is of both environmental and public
23
concern.1,2 Local communities, especially those with fishing-based or recreation-based
24
economies, can experience significant financial losses as a result of oil contamination on public
25
beaches, particularly in the aftermath of an oil spill.3 Oil residues in the marine environment,
26
however, can originate from a variety of different sources, including anthropogenic inputs such
27
as spills, as well as natural oil seeps.4 Therefore, determining not only if persistent oil is present,
28
but also the source of the oil is fundamental to understanding oil residue presence on beaches.
29
Mixed inputs of oil from natural and anthropogenic sources are well known, particularly in the
30
Gulf of Mexico (GoM) following the Deepwater Horizon (DWH) incident5,6, an area which has
31
natural oil seeps7,8 and small inputs of unknown oil likely from anthropogenic sources.9 It is also
32
important to monitor these oil residues over time as they contain, among other chemicals,
33
persistent polycyclic aromatic hydrocarbons (PAHs), which can be toxic if exposed to humans
34
and ecosystems.10 Monitoring the persistence of a specific source of oil in the marine
35
environment in the field can be challenging, although is a tractable problem once samples are
36
collected and analyzed in the lab.
37
To determine the source of oil to the marine environment, the chemical composition of oil,
38
specifically the characterization of oil-derived hydrocarbon compounds known as biomarkers,
39
are typically measured.11,12 Well-defined methods of oil and biomarker analysis for source
40
determination occur in specialized laboratories employing one-dimensional gas chromatography
41
(GC) coupled to flame ionization detection (FID) or mass spectrometry (MS) as well as
42
comprehensive two-dimensional gas chromatography (GC×GC) coupled to either FID or a time-
43
of-flight MS (TOF-MS).13 Bulk oil analysis methods such as Fourier transform infrared (FT-IR)
ACS Paragon Plus Environment
2
Page 3 of 13
Environmental Science & Technology Letters
44
spectroscopy and thin layer chromatography with flame ionization detection (TLC-FID) are also
45
employed to more generally characterize oil.5,9,14 The aforementioned methods are robust in their
46
applications, but are both time and labor intensive, use harmful solvents, are destructive to the
47
samples analyzed, require the use of a specialized laboratory, and a highly trained researcher. In
48
contrast to lab-based approaches, field protocols based on physical and visual characteristics of
49
oil residues have been developed. While useful, these protocols rely on human interpretation and
50
are not definitive in terms of distinguishing different sources of oil.15 In this study, we sought to
51
develop a rapid and portable approach that could be employed in the field to determine the
52
source of oil in different samples. A hand-held X-ray fluorescence (XRF) spectrometer was used
53
to analyze samples for their elemental content.16 A key benefit of this method is that it does not
54
require time consuming sample preparation steps, thus making it possible to analyze a large
55
number of samples in a relatively short period of time. The method itself only takes minutes to
56
complete from start to finish, where other analyzation techniques can take hours. A second
57
benefit is that this process is nondestructive, which allows for the preservation of samples.17
58
The XRF analytical approach detailed here focuses on the analysis of elemental
59
concentrations present in oil as a way to distinguish different sources of oil. Specific metals,
60
primarily nickel and vanadium, have previously been used to characterize oils from different
61
sources.18–20 This approach is successful in part because nickel and vanadium are two of the most
62
abundant elements in oil. However, a variety of other metals are also consistently found in oil
63
samples,19 but their utility with respect to determining oil source has not been extensively
64
explored. Typically, a broad range of metals beyond nickel and vanadium are examined with the
65
intent of understanding the toxicity of oil as well as how the concentrations of these elements
66
may be altered as oil is weathered in the environment.21,22 In studies related to the DWH incident,
ACS Paragon Plus Environment
3
Environmental Science & Technology Letters
Page 4 of 13
67
for example, chromium, aluminum, iron, magnesium, nickel, lead, thallium, vanadium, cobalt,
68
zinc, copper, and mercury were all examined21,23. Other studies have also measured the
69
abundance of calcium, titanium, and strontium in crude oils.24,25
70
For development of this XRF analytical approach, samples that had been collected from GoM
71
beaches in Florida, Alabama, Mississippi, and Louisiana between 2012 and 2017 were utilized.
72
Although we analyzed samples in the laboratory to develop the XRF approach, since the XRF
73
used is a handheld system, the same approach could be utilized directly in the field. The samples
74
analyzed included oil residues that originated from the 2010 DWH incident as well as oil from
75
natural seeps, other anthropogenic spills, road asphalt, and samples that visually resembled oil,
76
but did not contain oil-derived hydrocarbons. Subsequent analysis of the element content data by
77
machine learning approaches was then used to distinguish DWH oil residues from other sources
78
of oil and non-oil samples. The machine learning approaches employed included a rule-based
79
classifier, which uses IF-THEN rules for its predictions, as well as a decision tree classifier,
80
which is structurally organized as a flow-chart-like structure. These two approaches were chosen
81
based on their interpretability, which unlike traditional statistical approaches, produce outcomes
82
in the form of element content data, that can be deployed in the field when analyzing
83
environmental samples.
84 85
EXPERIMENTAL
86
Sample Collection. Between June 2012 and June 2017, 119 individual oil residue samples were
87
collected from GoM beaches as previously described.9,14 This sample set includes oil patties
88
originating from the DWH incident (62 samples), tar residues originating from natural seepage
89
offshore (27 samples), unknown oil residues (13 samples; 6 of which also contain sand), and
ACS Paragon Plus Environment
4
Page 5 of 13
Environmental Science & Technology Letters
90
non-oil samples that have the appearance of oil, but do not contain any oil residues (17 samples).
91
All samples were collected in pre-combusted glass jars (450 °C, 8 h), shipped to Haverford
92
College, Haverford, PA, and kept frozen until further analysis. The classification of these
93
samples as DWH oil patties, tar residues, unknown oil residues, and non-oil samples is described
94
in White et al., 2016 and Morrison et al., 2018.9,14
95
X-Ray Fluorescence of samples.
96
Instruments X-MET 7500 x-ray fluorescence spectrometer, a portable and handheld device with
97
detection limits on scales of 1 ppm.16 No extraction steps were performed on the samples prior to
98
analysis. The minimum size of samples analyzed was approximately 2 cm x 3 cm x 1 cm. In
99
order to consistently analyze the interior of each sample, sand patties and oil residues were sliced
100
in half with a solvent rinsed razor blade prior to analysis. Samples were placed on a Mylar
101
circular window film (6 µ thick, 64 mm diameter) on the XRF x-ray detection site. The XRF
102
standard “soil” method was used, with a 60 second detection time.26 Samples collected are
103
heterogeneous mixtures of oil, sand, and other natural organic and inorganic material.14,15 To
104
capture the variation within individual samples, 5 data points were taken per sample, each the
105
average of 3 measurements. Each of the 5 data points was taken sequentially from a different
106
location on the inside of the sample, with approximately 2-5 millimeters distance between each
107
data point. Element concentration data were therefore collected in triplicate for a total of 595
108
sub-samples, representing 119 individual samples, each measured in 5 places. To evaluate the
109
heterogeneity of the samples further, eight samples representing different Gulf of Mexico coastal
110
beaches were solvent-extracted according to previously described methods (see White et al.,
111
2016) so that the resulting oil-residue extracts and remaining unextractable materials (primarily
XRF spectroscopy was performed using an Oxford
ACS Paragon Plus Environment
5
Environmental Science & Technology Letters
Page 6 of 13
112
sand and shells) could be analyzed separately from one another.14 The extracts and unextracted
113
residues were then dried overnight at room temperature prior to XRF analysis.
114
Machine Learning Analysis. The 595 XRF samples from 119 unique samples were used to
115
train interpretable machine learning models to infer the presence or absence of DWH oil from
116
XRF data. The element concentration data has been deposited with Figshare, with the DOI
117
10.6084/m9.figshare.7272644. All element concentrations were used as trainable features, with a
118
binary class variable (source of oil is DWH or non-DWH). The training set used to construct the
119
model was comprised of a random selection of 66% of the samples (393 sub-samples, 79 unique
120
samples). The remaining 33% of the samples were held out as a test set (202 sub-samples, 39
121
unique samples) so that the accuracy of the model with respect to unseen data could be
122
determined. Sub-samples from the same patty were grouped together into the training set or test
123
set to ensure the model did not have prior exposure to samples similar to those in the test set. The
124
PART rule-based classifier27 and C4.5 decision tree classifier28,25 were used for the binary
125
classification task of determining if a sample contained oil residues from the DWH oil spill.
126 127
RESULTS AND DISCUSSION
128
Elements detected by X-ray Fluorescence. A total of 25 elements were detected across the
129
range of samples. In order of typical average abundance across all samples collected the
130
elements are: calcium (Ca), iron (Fe), potassium (K), titanium (Ti), zirconium (Zr), strontium
131
(Sr), manganese (Mn), vanadium (V), molybdenum (Mo), thorium (Th), barium (Ba), rubidium
132
(Rb), nickel (Ni), cadmium (Cd), zinc (Zn), copper (Cu), lead (Pb), cobalt (Co), chromium (Cr),
133
mercury (Hg), thallium (Tl), antimony (Sb), tantalum (Ta), selenium (Se), and gold (Au).
ACS Paragon Plus Environment
6
Page 7 of 13
Environmental Science & Technology Letters
134
Minimum, median, maximum, and 1st and 3rd quartile concentrations of these elements for each
135
sample type are provided (Table S1; Supporting Information (SI)) along with the concentrations
136
of all 25 elements for each sample category (Figure S1-S4; SI). Often, the samples were found
137
mixed with sand. The most common elements detected in DWH extracted oil residues (without
138
the sand matrix) compared to the most common elements in the sand matrix highlight variations
139
in element concentrations within the extracted oil and sand even when separated from one
140
another (Figure S5; SI).
141
Element content can predict source of DWH oil. The rule-based and decision tree models
142
yielded 96% and 95% classification accuracy, respectively (detailed in Figure S6; SI). The two
143
models made decisions based on very similar features, primarily considering of mercury (Hg),
144
copper (Cu), calcium (Ca), zinc (Zn), potassium (K), strontium (Sr), zirconium (Zr) and iron (Fe)
145
(Table 1). The concentrations of these eight elements for all DWH and non-DWH samples are
146
shown in Figure 1.
147 148
Table 1. Rules and the number of each sample type classified as either DWH or non-DWH oil by each of these rules.
Rule 1 Rule 2 Rule 3
Rule a
Classification
If Hg > 7 and Cu 2869, Zn
