An In Situ Laser-Induced Fluorescence System for
Polycylic Aromatic Hydrocarbon-Contaminated
Sediments
Prepared for:
Marc Tuchman, Project Officer
Proj ect Number GL975149- 01 - 0
Prepared by.
Joseph Aldstadt
Chemistry Department
University of Wisconsin - Milwaukee
Randy St. Germain
Dakota Technologies, Inc.
Timothy Grundl
Geosciences Department
Robert Schweitzer
Schweitzer Consulting
Submitted to:
United States Environmental Protection Agency
Great Lakes National Program Office
Chicago, Illinois
FINAL REPORT
May 2002
-------
TABLE OF CONTENTS
DISCLAIMER 5
INTRODUCTION 6
Statement of the Problem 6
"Direct Push Testing": A Brief History 9
Laser-Induced Fluorescence Spectroscopy 10
Understanding ROST Fluorescence Waveforms 12
PAH Time Decay Waveforms 13
Fluorescence Spectra of PAHs 14
PAH Multi-Wavelength Waveform (MWW) 15
FVD Colorization 18
PAHs in Harbor Sediments: "Ex Situ" LIF 20
PAHs in Harbor Sediments: Our Approach 20
In Situ LIF of PAHs in Harbor Sediments: Data Interpretation 21
EXPERIMENTAL 23
Reagents 23
Instrumentation & Methods for PAH Analysis 23
ROST System Description 25
Laser 26
Fiber Optic Cable 26
Shock-Protected Optical Compartment 26
Emission Detection System 26
Oscilloscope 27
Instrument Control & Data Acquisition Computer 27
Depth Control and Acquisition Module 28
Calibration and Normalization 28
Field Sampling Methodology 29
Shipboard Delivery Vehicle 29
In Situ Sediment Samplers 31
Field Testing Program 32
Data Analysis 36
RESULTS & DISCUSSION 38
Field Performance of the SDV/LIF System 38
PAH Quantification by Chemometric Data Analysis 38
-2-
-------
Source Apportionment 49
CONCLUSIONS 51
REFERENCES 52
APPENDICES 55
-3-
-------
FIGURES & TABLES
Figure 1. Spectral property of fluorescence 11
Figure 2. Temporal property of fluorescence 12
Figures. Examples of the temporalnature of fluorescence 13
Figure 4. Examples of spectral fluorescence 14
Figure 5. Example wavelength-time matrices of common contaminants on sand 16
Figure 6. Multi-wavelength waveform concept 17
Figure 7. Waveforms of common contaminants 18
Figure 8. An illustration of how color-coding is performed 19
Figure 9. Laboratory analysis scheme for PAHs in sediments 24
Figure 10. ROST system and key peripheral devices 25
Figure 11. SDV components and key peripheral devices 30
Figure 12. The R/V Mudpuppy in Milwaukee Harbor 31
Figure 13. Aerial photograph of study area 35
Figure 14. Depth profiles of total fluorescence 39
Figure 15. Depth profiles of total fluorescence for two replicate pushes taken at station 5.5 40
Figure 16. Predicted versus measured concentrations of representative 2-ring PAHS 42
Figure 17. Plots of field fluorescence measurement and chemometric prediction: Station 7 45
Figure 18. Plots of field fluorescence measurement and chemometric prediction: Station 1 46
Figure 19. Plots of field fluorescence measurement and chemometric prediction: Station 2 47
Figure 20. Depth profile of four individual PAH compounds at station 7 48
Figure 21. Source apportionment 50
Table 1. Ecological impacts potentially associated with contaminated sediments 7
Table II. Environmentally pertinent characteristics of several PAHs 8
Table III. Descriptions of the stations occupied during field testing 34
Table IV. Gross sediment characteristics 41
Table V. Relative errors between measured and predicted concentrations 43
-4-
-------
DISCLAIMER
The U.S. Environmental Protection Agency through its Great Lakes National Program Office
funded the research described here under Grant GL975149 to the University of Wisconsin-
Milwaukee. It has not been subjected to Agency review and therefore does not necessarily reflect
the views of the Agency, and no official endorsement should be inferred.
Reference herein to any specific commercial products, process, or service by trade name,
trademark, manufacturer, or otherwise, does not necessarily constitute or imply its endorsement,
recommendation, or favoring by the United States Government. The views and opinions of authors
expressed herein do not necessarily state or reflect those of the United States Government, and
shall not be used for advertising or product endorsement purposes.
-5-
-------
INTRODUCTION
Statement of the Problem
In the early 1970s the poor environmental health of the Great Lakes was recognized and
both the United States and Canada banded together to form the International Joint Committee (UC) in
an effort to reverse a long trend of toxic pollutant discharge to the Great Lakes. The original Great
Lakes Water Quality Agreement (GLWQA) between Canada and the Unites States was signed in
1972. Almost immediately, the UC realized the importance of sediment-bound contamination. In 1975,
the "International Working Group on the Abatement and Control of Pollution from Dredging Activities"
was instituted. By 1978, a "Contaminated Sediment Task Force" was established to look at the
assessment and remediation of contaminated sediments. In 1987, the United States demonstrated its
support of the GLQWA by establishing, under the Clean Water Act, the "Assessment and Remediation
of Contaminated Sediments" (ARCS) program. The ARCS program was specifically charged with
assessing the nature and extent of sediment contamination and to demonstrate the effectiveness of
selected remedial options at 5 different sites [1]. The ARCS final report was issued in 1994 and
immediately the UC identified "remediation and management of sediments contaminated with persistent
toxic substances" as one of its 1995-1997 program priorities [2].
This 20-year record of study has led to a consensus that contaminated sediments are a major
cause of environmental problems in the Great Lakes. All of the original Areas of Concern (AOC) that
were identified by the UC as problem sites have contaminated sediment [2]. As such, contaminated
sediments represent a universal obstacle to the environmental recovery of these areas. The GLWQA
defines fourteen types of ecological impacts arising from contamination. Of these fourteen impacts,
eleven are affected by the presence of sediment contamination. Table I lists these eleven impacts and the
number of AOCs that display each type of impact.
One of the primary organic contaminants of concern in Great Lakes sediments are poly cyclic
aromatic hydrocarbons (PAHs). Table U lists three chemical characteristics of representative PAH
compounds. For comparative purposes, the same characteristics are listed for the familiar compounds
benzene and water. These characteristics define the environmental behavior of these compounds.
Inspection of Table U reveals that PAH solubilities are very low and the hydrophobic sorptive capacity
(Kow) is correspondingly high. This, coupled with low volatilities (Henry's Law constant) and general
chemical stability mean that PAHs are environmentally persistent compounds that are strongly held to
solids, both suspended particles and bottom sediment. Relatively little transport via the dissolved phase
is seen. Furthermore, the high partitioning to organic carbon (as reflected by high KOW) is the root cause
behind the high rate of bioconcentration for these compounds and the ease with which they enter the
food web. This illuminates the current dilemma facing Great Lakes managers. In spite of the fact that
significant decreases in the discharge of these compounds to the Great Lakes has been achieved, a large
pool exists in the bottom sediments and this pool is easily re-mobilized into the base of the food web by
benthic organisms. Environmental managers cannot afford to ignore the existence of contaminated
sediments.
-------
Table I. Ecological impacts potentially associated with contaminated sediment.
The numbers of Areas of Concern with such impacts are also shown (modifiedfrom [2]).
ECOLOGICAL IMPACT
Restrictions on fish and wildlife
Degradation offish and wildlife
populations
Fish tumors or other
deformities
Bird or animal deformities or
reproduction problems
Degradation of benthos
Restrictions on dredging
activities
Eutrophication or undesirable
algae
Degradation of aesthetics
Added costs to agriculture or
industry
Degradation of phytoplankton
or zooplankton populations
Loss offish and wildlife habitat
HOW CONTAMINATED
SEDIMENT MAY AFFECT
ECOLOGICAL IMPACT
Contaminant uptake via contact with
sediment or through food web
Contaminant degradation of
Habitat; contaminant impact through
direct sediment contact; food web
uptake
Contaminant transfer via contact
with sediment or through food web;
possible metabolism to carcinogenic
or more carcinogenic compounds
Contaminant degradation of
habitat; contaminant impacts through
direct sediment contact; food web
uptake
Contact; ingestion of toxic
contaminants; nutrient enrichment
leading to a shift in species
composition and structure, due to
oxygen depletion
Restrictions on disposal in open
water due to contaminants and
nutrients, and their potential impacts
on biota
Nutrient recycling from temporary
sediment sink
Re-suspension of solids and
increased turbidity; odors associated
with anoxia
Re-suspended solids; presence of
toxic substances and nutrients
Toxic contaminant release;
Re-suspension of solids and adsorbed
contaminants, and subsequent
ingestion
Toxicity to critical life history stages;
degradation of spawning and nursery
grounds due to siltation
NUMBER OF AREAS OF
CONCERN WITH THE
ECOLOGICAL IMPAIRMENT
(n = 42, % in parentheses)
36 (86%)
30 (71%)
20 (48%)
14 (33%)
35 (83%)
36 (86%)
21 (50%)
25 (60%)
7 (17%)
10 (24%)
34 (81%)
-7-
-------
Table II. Environmentally pertinent characteristics of several PAHs [3].
Water
Benzene
Benzo(a)anthracene
Benzo(a)pyrene
Benzo(&)fluoranthene
Benzo(&)fluoranthene
Pyrene
Solubility
(ng/g)
—
1.75xlO+6
12
3.9
14
0.6
135
Henry's Law Coefficient
(atm-m3/mol)
6.6 x lO'7
5.4 x 10'3
2.3 x 10'6
2.4 x 10'6
1.2X10'5
l.OxlO'3
l.lxlO'5
Octanol-Water
Partition Coeffiiceint
(log KoW)
—
2.12
5.90
6.00
6.57
6.85
5.09
Ecosystem health is not the only problem associated with the presence of sediment
contamination. Large scale economic costs to the commercial and sport fishing industries arise from fish
consumption advisories, the viability of commercial ports are threatened by restriction on navigational
dredging, and municipalities have to employ more expensive water treatment techniques. Over half of
the eight million cubic meters of Great Lakes navigational dredging performed between 1985 and 1989
had to be placed in some form of containment facility due to high contaminant levels. Placement of
dredge spoils in a containment facility increases dredging costs three to five times [2].
Relatively slow progress has been made in the remediation of the AOCs (by 1996, only one
of 43 had been "de-listed") [2]. A more recent report indicates that within the Great Lakes watershed
as a whole, 33 sites are listed as past or current remediation projects and an additional 15 more are
listed as upcoming sites [4]. Many of these sites are on inland waterways and as a result are relatively
small. Even with these generally smaller and more accessible sites, the average cost of remediation is
close to $100 per cubic meter of contaminated sediment [4].
A complex set of political, socioeconomic and scientific obstacles are impeding progress in
the cleanup of these sites. One recognized scientific impediment to continued progress is the inability to
define the extent of the problem. One key area has been identified by the IJC that needs additional
research and development is the "accurate physical identification and differentiation of sediment deposits
which allow both cost-effective assessments and clean-ups" [2].
The current practice for delineating the extent of sediment contamination at any given site comes
from the installation of expensive networks of discrete sampling sites. Sediment is obtained from core
samples that are often times vertically blended before laboratory analysis. Blending decreases analysis
costs, but results in poor vertical discretization. Original identification of the AOCs was based on
sampling networks with spacings that were hundreds of meters apart. This is clearly insufficient coverage
upon which to base a decision as to the most suitable remediation technique. Active remediation efforts
require a more detailed understanding of site conditions and typically must install a more complete —
and more expensive — sampling network with spacings on the order of tens of meters. We report on
-------
the development of an inexpensive, rapid, vertically discrete, in situ technique for the measurement of
sediment contamination by PAHs. The technique is an adaptation of a proven terrestrial technique;
direct push testing (DPT). The technique will be invaluable in the following contexts.
The ability to clearly assess the areal extent and concentration levels of sediment
contamination. This background information is a prerequisite for any decisions on remediation
designs and remediation costs.
In-the-field guidance of dredging operations.
• Post- closure monitoring of remediated sites.
• A better understanding of bulk sedimentation rates as well as bioturbation or other re-
suspension processes through vertical discretization. These processes are especially important
to the understanding of contaminant flux from sediment to the overlying water column and
biota.
• To serve as a practical data collection technique for the long-term analysis of trends and
surveillance monitoring of sediment contamination into the future.
"Direct Push Testing": A Brief History
Direct Push Testing (DPT), also referred to as "Cone Penetrometer Testing" (CPT), is a
well-established technique for geophysical and chemical characterization of subsurface terrestrial
environments [5]. DPT is based on driving a conical steel tip (typically 60 apex and ~44-mm o.d. at
base) under hydraulic pressure at a constant rate. The tip is advanced by sequentially attaching a series
of cylindrical steel extensions or "push rods" to the assembly that are linked to one another by
threadings. The thick-walled DPT push rods, typically 1 m in length, are hollow (i.e., push "rods" can be
more accurately described as tubes) to accommodate sensor electronic cables, sample transfer lines,
optical fiber bundles, etc. Trucks for DPT vary in size on the basis of the requirements of the site
geology, from hand-operated rigs that are used to probe shallow formations of loose clay and sand to
large trucks in the 20- to 40-ton range that can push to depths of 50 m or more in loose sands, although
intervening layers containing coarse gravel and cobblestones will limit the maximum depths that can be
reached. The load from large hydraulic rams, typically in the 20,000-kg range, is transferred to the push
rods by using a hydraulic clamping system.
Direct push testing was developed as a geophysical tool early last century in the Netherlands
to gather civil engineering data on the load-bearing capacity of weak soils (i.e., to identify the depth of
firmer strata) [5]. In modern geophysical investigations based on DPT, changes in the force required to
advance the cone through the underlying strata are recorded. For the unconsolidated and semi-
consolidated geologic formations that it can successfully penetrate, DPT is advantageous compared with
conventional auger well drilling in that it generates a continuous subsurface profile, causes minimal
disturbance to the site, generates minimal waste at the surface, and is much faster (typically 2 cm/s).
Over the past 15 years or more, DPT has been recognized as a promising means to obtain
chemical as well as geophysical information about the subsurface. DPT has been used over the last
decade for a broad range of sampling and analytical techniques for the chemical characterization of the
subsurface [5]. In situ chemical samplers can be driven into the subsurface by DPT to collect soil gas
-9-
-------
[6,7], groundwater [8-18], or soil samples [19] at depth, thus providing a means of determining the
vertical and horizontal extent of contamination. Furthermore, DPT has also been used as a means for
delivering various in situ subsurface sensor probes, including laser-induced fluorescence (LIF)
spectroscopy for semi-volatile organic compounds [20-29], x-ray fluorescence [30, 31] and laser-
induced breakdown spectroscopies [32-33] for heavy metals; gamma ray spectroscopy for
radionuclides [34], and various optical and electrochemical sensors, including for chlorinated aliphatic
organics [35] and nitroaromatic explosives [36], respectively.
Laser-Induced Fluorescence Spectroscopy
The pioneering work in this area was done by the Naval Research, Development, Test, and
Evaluation Division (NRaD) in collaboration with the Army's Waterways Experimental Station (WES)
and by Dakota Technologies, Inc. (DTI). The U.S. Air Force-sponsored LIF work was carried out by
Gillispie and co-workers at North Dakota State University and Dakota Technologies, Inc [37]. This
work led to a commercial LIF system known as the "Rapid Optical Screening Tool" (ROST™). There
are currently seven ROST LIF-DPT systems in the United States, two of which belong to the Air
Force. Of the remaining five, only the one owned by Dakota Technologies is available for research and
development work. The ROST system is optimized to determine two-ringed aromatics.
Fluorescence spectroscopy is one of the most widely applied spectroscopic techniques in
use today. It is, by nature, a fast, sensitive and typically reversible process that makes it ideal for
incorporation into a continuous screening technique that uses an optically transparent window as the
conduit between the sensor and the analyte. Luminescence is the emission of light from any substance
that returns to the ground state after being excited into an electronically excited state. If the bulk of the
molecules emit their photons in less than a microsecond the emission is referred to as fluorescence.
Emission that takes longer than this is called phosphorescence.
Fluorescence is typically observed in molecules that have an aromatic structure. One class
of aromatics are the polycyclic aromatic hydrocarbons (PAHs) found in quantity in typical petroleum
products. The PAHs found in sediments are also fluorescent, but they fluoresce much less efficiently
than PAHs dissolved in solvent-rich environments, such as the aliphatic body that makes up the bulk of
petroleum fuels. The PAHs continue to absorb the excitation light, but there is a much higher likelihood
of the PAHs finding a non-radiative mechanism with which to shed the additional energy they picked up
during the absorption of the excitation photon(s). In spite of this, the PAHs in sediments can still be
coaxed into fluorescing well enough to allow in situ laser-induced fluorescence screening via a
sapphire-windowed probe.
Plots of the relative distribution of the different colors (or energies) of the photons being
emitted by an excited sample of PAHs are called the spectra. Figure 1 illustrates the concept of PAH
absorbance and fluorescence spectra. The spectra of individual PAH species can contain enough
structure (peaks and valleys) to be identified in simple mixtures in the lab. The fluorescence of PAHs in
sediments however, is originating from such a wide variety and concentrations of PAHs and differing
local environments (dissolved phase, sorbed to particles, microcrystals, etc.) that the resulting spectra
-10-
-------
are broad and contain relatively little "structure" that can be used to determine which individual PAHs
are responsible for the fluorescence.
Another property of fluorescence that can be measured is the varying amount of time it takes
for the molecules to emit the photons after exposure to a pulsed excitation source, such as a laser. This
is illustrated in Figure 2. An oscilloscope is used to observe the number of photons being emitted over
time, thereby deriving additional information about the nature of the fluorophores and their environment.
The different PAHs and the differing environments that exist in sediments all combine to change the
observed decay times. This information is readily obtained when using a pulsed source such as the laser
we used in this application. Our apparatus (described in "ROST System Description" below) allows
us to investigate not only what colors are being emitted, but also how long it takes for the excited
population of PAHs to emit the fluorescence photons. We use a patented method of combining the
photons from four regions of the emission spectrum optically collected over 20 nm wide sections of the
emission spectra at 340, 390, 440, and 490 nm.
General Fluorescence Properties
Spectrum (colof or energy distribution)
perylene
(nm, nano-metefs) •
Energy
LJHraA/iQlQl (UV)
Figure 1. Spectral property of fluorescence.
-11-
-------
General Fluorescence Properties
Temporal (time-decay waveform)
perytene
abscxpfion of
pulsed lose*
wovefofms
from
OSCtlOSCOpe
decaying emission o« energy
In 1he form of photons
[fluorescence]
Time (nono-seconds, ns)
Figure 2. Temporal property of fluorescence.
These four "channels" are delayed in time through successively longer fiber optic delay lines and
eventually arrive at a photomultiplier tube. The resulting waveform allows the simultaneous observation
of the spectral and temporal qualities of the fluorescence. It is these multi-wavelength waveforms,
measured continuously and stored vs. depth, that ultimately serve as our indicator of PAH concentration
vs. depth in the sediment.
In the natural environment, additional fluorescent materials in sediments will also absorb the
laser light and fluoresce intensely enough to complicate the measurement of the PAH fluorescence.
Example materials include minerals such as calcite and a variety of biological materials. Both living
organisms and their associated breakdown products fluoresce well enough to interfere with the
observation of the fluorescence of the target PAHs. This fluorescence, along with scattered excitation
laser light and Raman light generated throughout the optical train (fiber optics) will ultimately make it
back to the detector, mixed in with true PAH fluorescence, and must be accounted for in some fashion.
Throughout this document we will refer to all these sources of non-PAH emitted photons as
"background" fluorescence, even though the true source might well be non-fluorescent (scatter) in
nature.
Understanding ROST Fluorescence Waveforms
Spectroscopic techniques involve probing the target matrix with light and learning about the
contents of that matrix by analyzing the light that is emitted or absorbed by the target matrix. For
screening tools it is crucial to glean as much information from this light as possible in as little time as
possible. The fluorescence data from ROST can be gathered quickly, but is deceptively simple. It is
actually a two-dimensional data set that contains three-dimensional fluorescence information. A full
-12-
-------
description of the multi-wavelength waveform data follows to give the reader an understanding of the
data acquired during this study.
PAH Time Decay Waveforms
Each type of PAH molecule emits fluorescence over a unique time period after being excited
by a pulsed excitation source such as the laser used in ROST. The emission starts out at maximum
intensity, and then decays away at a rate unique to each type of PAH. The number of rings, the bonding
between them, the amount of substitution on the rings, and other structural features of the molecule
determine, to a great extent, the decay rate exhibited by a particular PAH. The environment in which the
PAH exists also has a substantial influence on the decay rate. Quenching, which refers to any process
that causes a decrease in the decay time (as well as the intensity) of the fluorescence, is dependent on a
myriad of matrix dependent conditions.
Figure 3 illustrates the differing decay times one might observe for four different PAHs, along with the
time profile of the laser pulse that excited them. There are large populations of PAHs being excited and
while some begin emitting immediately, other individual PAH molecules "wait" many nanoseconds
before emitting a photon. Figure 3 is a picture of the distribution of times that the PAHs are remaining in
the excited state before emitting photons. The bold curve in Figure 3 illustrates the fluorescence decay
profile that would result if all four PAHs fluoresced simultaneously and with the same intensity. This
same concept applies in harbor sediments however the fluorescence decay profile observed in
sediments is not made up of equal amounts of fluorescence from the various PAHs found in them. The
spectra emitted by and the relative fluorescence yields of the different PAHs are all quite different, but
the concept is still valid. The
Example PAH Waveforms
Laser pulse
PAH with 5 ns lifetime
PAH with 10 ns lifetime
PAH with 20 ns lifetime
PAH with 40 ns lifetime
Decay from all 4 PAHs
100
200
300 400
Time (ns)
500
600
700
-13-
-------
Figure 3. Examples of the temporal nature of fluorescence.
decay profile of the PAHs observed in the sediment results from the decay profiles of a mixture of
different PAHs, along with fluorescence from other materials in the matrix. It should be noted that there
is no predictable trend between decay rate and structure like the trend that exists between spectrum and
structure as described below.
Fluorescence Spectra of PAHs
Figure 4 depicts the fluorescence emission spectra of the same four PAHs used in the
temporal example in Figure 3. The fluorescence emission spectrum of a pure PAH is simply a graphical
representation of the energy distribution of photons that are emitted from a large population of the PAHs
as they release energy that was absorbed from the excitation laser beam. Spectra of pure PAHs such
as these are typically acquired by dissolving a sample of the pure PAH in a pure solvent that does not
fluoresce.
Example PAH Spectra
a
o
.5 -
2 -
.5 -
1 -
.5 -
0
290
excitation (laser) wavelength
•PAH#1
•PAH #2
•PAH #3
•PAH #4
•Total
340 390
Wavelength (nm)
440
Figure 4. Examples of spectral fluorescence.
The laser wavelength is also shown in Figure 4, demonstrating the principle that fluorescence
occurs at longer wavelength (lower energy) than the excitation wavelength. The basic trend is toward
longer wavelength emission as more rings are added or substitution increases. Naphthalene emits at
around 340 nm and the spectra "red-shift" as the number of rings increase. Another general property of
fluorescence is that for a pure PAH the emission spectrum remains the same regardless of the
-14-
-------
wavelength of the excitation light. This is not true for mixtures however, because changing the excitation
wavelength might well change which PAHs are being excited and to what degree. The bold spectrum in
Figure 4 is the combined spectra of all 4 PAHs. This is a simplified illustration of the total fluorescence
of a mixture of different PAHs. Any change in the relative amounts of the differing PAHs or changes in
the matrix in which they exist will cause a change in the spectrum of light actually emitted.
The fairly well defined structure (multiple peaks, valleys, and their various positions) of the
spectra in Figure 4 suggests that perhaps algorithms could be used to directly extract information about
the relative concentrations of the individual PAHs. While this is possible for very simple mixtures (two
to three PAHs) under controlled conditions, the algorithms quickly fail when many PAHs are present
and interference fluorescence from humics, fulvics, and minerals is introduced.
PAH Multi-Wavelength Waveform (MWW)
Real-world environmental samples typically contain multiple PAH compounds along with
other fluorophores, and the PAH fluorescence spectra overlap to form broad and fairly featureless
spectral and temporal emission (compared to pure PAH spectra). A complete record of the temporal
decay waveforms across the entire spectrum would describe the fluorescence emission completely.
Such records are called wavelength-time matrices (WTM). To create a WTM, the emission selection
monochromator is scanned from wavelength to wavelength, monitoring the pulsed emission vs. time at
each wavelength with an oscilloscope.
Figure 5 contains the WTMs of diesel, jet, creosote, and gasoline on sand at several
thousand |ig/g. The difference between the contaminants is clear and identification is straightforward.
WTMs are excellent for identifying/classifying the PAH fluorescence of environmental samples because
of the unique information that both dimensions of PAH fluorescence exhibit when acquired in unison.
While WTMs make different contaminants readily discernable from one another, they are three-
dimensional and large. Also, the screening tool must be held still while the measurement is being made.
All of these qualities make WTMs unwieldy for environmental screening tools that are designed to
continuously log (typically 1 Hz) the presence of PAHs vs. depth.
Because WTMs are so difficult to implement in screening mode, DTI developed a multiple-
wavelength waveform (MWW) technique that allows multi-dimensional PAH fluorescence
measurements to be acquired "on the fly". Figure 6 illustrates this concept. Select regions of the
spectrum are monitored for their temporal response. The responses are optically delayed and
recombined, and the resulting responses converge to form one two-dimensional waveform. There is
sometimes overlap between the "channels" with long decay times, and the spectral regions being
monitored are fewer and farther between than WTMs, but the resulting waveform still retains a unique
combination of spectral and temporal fluorescence information that makes speciation and identification
of PAH mixtures possible. Figure 7 illustrates the unique waveform produced by a variety of common
PAH-containing environmental contaminants.
-15-
-------
Creosote
Diesel
Figure 5. Example wavelength-time matrices of common contaminants on sand.
-16-
-------
ui 11 resolution WTM
select 4 key
capture fluorescence channels
with optical fibers
delayed fluorescence pulses arrive ai PMT
and a« recorded with oscilloscope
fluorescence pulses
multi-wavelength waveform
jr.
ttt fed
Ht HB
Figure 6. Multi-wavelength waveform concept.
-17-
-------
WTMs
Creosote
&,.
F
Diesel
Waveforms
, . .
Figure 7. Waveforms of common contaminants.
FVD Colorization
The waveforms that are continuously logged vs. depth with ROST contain a wealth of
information, but to make this information easily interpretable in fluorescence vs. depth (FVD) log format,
the data is further reduced to a one-dimensional data set that can be plotted vs. depth. DTI developed
a technique that effectively converts the shape of the waveforms into colors. These colors are then used
to fill in the area under the FVD that represents the total fluorescence measured at each point in the
-18-
-------
FVD. Figure 8, derived from data from a coal tar delineation project, illustrates the technique of
colorizing the FVD according to the shape of the waveforms.
now
Uteing channel 2. 3 and A
(channel i tgnaecf]
RGB = 1.OOR-1-0,506-l-0.12B
fill color used at 6.7 Teet on FVD
D0?1
i
gu
in- rHiiVr^-.r-nf*,
ft » 10 IB 2)
RGB = 0.41 R + 0.75G + 1.008
color used at 5,4 feet on FVD
Figure 8. An illustration of how color-coding is performed.
-19-
-------
The result is a data presentation technique that allows the user to assess similarities or changes in the
waveform shapes vs. depth by simply observing the colors that represent the shape of each and every
waveform in the data set. This technique was used on the sediment measurements made in this project,
both in the lab and in the field. It should be noted that black indicates that the algorithm for generating
color failed to de-convolve the waveform.
The colorization technique is limited to using three of the channels as a result of the red,
green, and blue (RGB) color definition which computer colorization systems typically implement. The
first three channels (340, 390, and 440 nm) were used to colorize the data in this study. The 490 nm
channel was used in a quantitative sense, but was ignored for the colorization. Note that Figure 8 is an
illustration from a previous study in which the 340 nm channel was ignored. It should be noted that a
strictly temporal change (where only the decay times change, not the spectrum) would not necessarily
result in a color change, since the ratios of the three channels used might remain constant even though
the area under the waveform itself will increase or decrease.
An added benefit of this technique is that it provides insight for situations where non-linear
response behavior is encountered. Many contaminants do not fluoresce with concentration in a linear
fashion. For instance, a 10-fold increase in PAH concentration might produce very little or no increase
in total fluorescence intensity. However, a spectral or temporal shift often does continue to occur with
changes in concentration due to energy transfer, photon cycling, and other phenomenon. The color of
the FVD fill continues to darken or shift in color, acting as an indicator of a change in the fluorescence of
the sample, alerting the analyst to a possible increase in concentration.
PAHs in Harbor Sediments: "£jc Situ" LIF
In addition to the use of LIF in terrestrial sediments, LIF spectroscopy has also been
successfully used in the detection of PAHs in sub-aqueous environments (i.e., laboratory analysis of the
sediment by LIF). One of the key obstacles to the discrimination of contaminants in aqueous
environments is the presence of high background levels of fluorescent humic substances. This problem
was successfully overcome in marine sediments by using time-resolved LIF spectroscopy [38,39]. This
differentiation is based on the fact that humic substances typically exhibit quite short-lived fluorescence
emission (fluorescence lifetimes of 4 ns) whereas PAH fluorescence is more long-lived (fluorescence
lifetimes as long as 128 ns).
PAHs in Harbor Sediments: Our Approach
In this report, we will describe the development of an adaptation of the proven terrestrial
DPT-LIF technique used for the detection of PAHs for the in situ characterization of PAHs in
freshwater sediments. This technique provides for the rapid, vertically discrete quantitation of PAH
compounds with the following specific objectives:
-20-
-------
• To adapt LIF-DPT techniques for the in situ detection of PAH contamination in submerged
harbor sediments.
• To develop an in situ sediment sampling system for the collection of submerged harbor
sediments that directly correspond to the LIF depth profile.
• To demonstrate the ability of this system to detect PAHs in sediment samples and its
applicability as a tool to accurately delineate the extent of sediment contamination in Milwaukee
Harbor.
Once the physical development of the delivery platform and sampler was in place, development shifted
to the acquisition and interpretation of LIF-DPT measurements taken in the field. There are several
important tasks that the data analysis focused upon. The first task was to identify the individual species
contributing to the overall fluorescence signal. A second task was to remove background signals. A
third task was to provide relative concentrations of the species detected and to relate these relative
concentrations to true concentrations in a quantitative manner.
In Situ LIF of PAHs in Harbor Sediments: Data Interpretation
The description of LIF waveforms (see "Understanding ROST waveforms" section above)
alludes to the complex interplay between the relative concentrations of individual PAHs, their
fluorescence yields, spectral and temporal behavior and the presence of a variety of matrix-dependent
background fluorescence signals. It is not possible to directly ascertain the contribution of a particular
PAH to the total fluorescence signature by a simple summation of standard PAH spectra. However the
wealth of information that is contained in each MWW matrix can be extracted by use of advanced
chemometric data analysis.
Chemometric analysis was used to both define background noise and to quantify species
contributing to overall fluorescence. Chemometrics is the application of mathematical and statistical
methods to the collection and analysis of data. It becomes important in such areas as signal processing,
design and optimization of experiments, development of calibration models, and classification of data
points via clustering and pattern recognition algorithms.
The chemometric analysis for the research described in this report falls into the category of
"mixture analysis". This is because the data collected in our work consist of fluorescence spectra that
are a summation of the fluorescence spectra of the individual PAH components. Classical chemical
analysis requires the use of a chemical separation step to provide selectivity for the individual
components, but ideally chemometrics allows the separation of the component signals and background
signals mathematically from the collected spectra. Most mixture analysis techniques use an approach in
which a new set of variables (factors) is created, and actual chemical components in the sample are
related to the factors in a way that allows the mathematical separation of the individual chemical
components.
Quantitative model building was chosen as the tool to allow the identification of PAHs in
sediment via LIF. Model building requires the analysis of multiple samples by both a primary method
(GC-MS) and a secondary method (LIF). The primary method establishes the "true" concentrations of
-21-
-------
the PAHs in the sediment samples, while the secondary method is the method by which one would like
to perform routine measurements of PAHs. The set of data resulting from the analyses of the samples
by both the primary and secondary methods is termed the calibration set. A mathematical tool is then
used to establish a correlation (model) between the concentration data and the LIF spectra. This model
can then be used to predict the concentrations of PAHs in sediment for samples for which only LIF
spectra are available. The most familiar models are univariate models - models in which only one
wavelength is used to build the relationship between a spectroscopic measurement and known
concentration. The technique used in this work is a multivariate method (Partial Least- Squares
Regression, PLSR) in which all spectroscopic information (all 351 points in a MWW matrix) is used
simultaneously to relate to concentration. Multivariate techniques have significant advantages over
univariate techniques. They provide a signal-to-noise improvement, greater robustness, greater
selectivity, and the ability to monitor a new sample's fit to the calibration set. PLSR is by far the most
commonly used quantitative multivariate model building technique in the chemometrics community. A
text by Marten & Naes provides an excellent description of multivariate methods in general and PLSR
in particular [40].
There are a few simple principles that must be adhered to ensure successful model building.
The first is that a wide range of data samples must be selected for the calibration set. These samples
must cover a wide range of concentrations and include samples from a wide range of different sediment
matrices. It is helpful to think of each sample as a data point in an n- dimensional data space in which n
is the number of spectral points in the spectra collected for each sample. The second principle is that
predictions can only be made for unknowns that are well represented by the points in the calibration set.
The bigger the data space that is covered by the samples in the calibration set, the greater the number
and variety of unknowns whose concentration can be predicted. However, the bigger this data space,
the greater the number of calibration points that are needed to cover that data space completely and
evenly. The point to be made is that it is not sufficient to have a wide range of concentrations
represented in the calibration set. All spectral variations that might be encountered in the field must also
be represented in the calibration set. A reassuring note is that the degree to which an unknown sample
is represented by calibration points in the data space can be calculated, and the results of this calculation
can determine the confidence that one should place on the predicted value for that point.
-22-
-------
EXPERIMENTAL
Reagents
All chemical reagents used were analytical reagent (AR) grade or better. Reagent water (18
MQ-cm) was prepared using a NanoPure™ filtration system equipped with an ultraviolet lamp
(Barnstead-Thermolyne, Dubuque, IA, USA). PAH standards were obtained from Supelco
(Bellefonte, PA, USA). All glassware was acid-washed after use for at least 48 hours in 5% (v/v) AR-
grade nitric acid to remove background contamination. PAH-containing standards were prepared on
the day of use. All standards were stored at 4 C.
Instrumentation & Methods for PAH Analysis
PAH analyses were conducted according protocols based on U.S. EPA standard methods;
a flow-chart of the analytical process is shown in Figure 9. Samples were collected with a VibraCore
and sectioned (2-6"), dried, and stored. Pressurized Fluid Extraction (PFE) was then performed in
duplicate (1:1 acetone:hexane) based on EPA Method 3545 [41] implemented using a method that we
developed previously for SVOCs [42]. Our PFE method resembles commercial PFE approaches (e.g.,
Dionex Accelerated Solvent Extraction or ASE™), but a crucial difference is that the available
extraction cell that we use is inexpensive, making it more appealing to laboratories like ours that handle
relatively small numbers of samples. The stainless steel extraction cells (64-mm-long pipe (7 mm i.d.
and 12 mm o.d.), threaded at both ends and fitted with end caps) were obtained from Minnesota Valve
and Fitting (Eden Prairie, MN, USA). An Isotemp Model 301 laboratory oven (Fisher Scientific,
Pittsburgh, PA, USA.) or a Model 47900 furnace (Thermolyne, Dubuque, IA, USA) was used to heat
the extraction cells. With one end of the stainless steel cell capped, a 300-mg sample (500-mg for
SRM) was placed in the cell, followed by 3 mL of hexane. To ensure a good seal on the capped cell,
Teflon tape was placed on the threads on both ends of the cell (not in contact with the sample). The
loaded cell was then placed in an oven for a predetermined time (usually 60 min). The extraction cell
was cooled to room temperature under tap water. The cell was opened, and the contents were
transferred to a glass vial. A portion of the extract was transferred to a conical vial and used for GC-
MS injections without further cleanup.
PAHs were determined by Method 8270C [43] using a dedicated Gas Chromatograph with
Electron lonization Quadrupole Ion Trap Mass Spectrometry (GC-MS), performed in a Clean Room
environment on a Varian Saturn 2000 system (Walnut Creek, CA, USA). The GC-MS consisted of
the following components: Model 3800 capillary gas-liquid chromatograph (DB-5MS column, 30 m x
0.25 mm with 0.25 |im film, J&W Scientific, Folsom, CA, USA) with Model 1079 split/splitiess
injector; electron impact ionization source (70 eV); Saturn 2000 quadrupole ion trap MS (10-650 m/z
range, unit resolution). The automatic gain control (AGC) of the MS system was used throughout this
study; mass spectra were obtained by single ion monitoring at relevant retention times with a 0.7 s scan
time. Automated library searching was performed using the National Institute of Standards and
-23-
-------
Technology (NIST) Mass Spectral Database (version 3.0). The mobile phase was ultra-high purity
helium (Praxair, Milwaukee, WI, USA) at a constant linear velocity through electronic flow control.
Collect
Samples
i
Rotate for ^^^^^^ Field
24 hours ^^^^™ Blank
±
. . „ . Laboratory
Alr Drv for J Pnntrnl
48 hours Sample
I
Dried NIST Certified
Sample Reference Material
Matrix ^^^_ _^^^ Surrogate
Spike Mix ^^^^ "^^^™ Spike Mix
Pressurized Fluid Extraction 1
±
[Solvent Evaporation with
Kuderna Danish
•^^B^J Internal Standard Mix 1
^ r
PAH Standards _^ Determina^Hn by
for Calibration GC-MS
Figure 9. Laboratory analysis scheme for PAHs in sediments.
Data acquisition and reduction were accomplished using the Varian GC-MS software (vers.
5.21). GC instrument control and data acquisition were performed on a Pentium personal computer
(Dell, Optiplex GX1, Dallas, TX, USA). QA/QC measures included field blanks, solvent blanks,
method blanks, matrix spikes, and surrogates. Percent recovery was determined using three surrogate
compounds (Nitrobenzene-d5, Benzo(g/7/')perylene, and 4-terphenyl-d14) and matrix spikes
(Naphthalene, 2-Fluorobiphenyl, Pyrene); the recoveries ranged from 80-102%.
-24-
-------
Sixteen separate calibration models were built using Internal Standards (Naphthalene-d8,
12\
Acenaphthene-d10, Phenanthrene-d10, Chrysene-d12, 1,4-Dichlorobenzene-d4, and Perylene-d
validation was performed using a contaminated river sediment (SRM 1944) obtained from the NIST
(Gaithersburg, Maryland, USA); accuracy was < ±20%.
ROST System Description
The ROST system is contained in a ruggedized shipping container as shown in Figure 10.
The system actually consists of a variety of sub-systems that are described in detail in this section. The
numbers in each heading below refer to the equivalent numbers in Figure 10.
Figure 10. ROST system and key peripheral devices.
-25-
-------
Laser (3)
The ROST system employs a pulsed XeCl Excimer laser (MPB PSX-100) that generates
very fast pulses of 308 nm light (50 Hz). The width of each pulse is less than 10 billionths of a second
(i.e., < 10 ns) at half height. The energy of the source radiation efficiently excites the vast majority of
PAHs present within the sediments. A beamsplitter directs a fraction of the beam to an energy meter to
monitor excitation pulse energy. A photodiode is positioned near the beamsplitter and serves as the
trigger source for the time-resolved fluorescence measurement. A lens is used to launch the laser light
into an optical fiber for delivery a sapphire window at depth.
Fiber Optic Cable (6)
The fiber optic cable consists of two 40 m silica/silica optical fibers. One fiber delivers the
excitation pulse while the other is used to return a portion of the resulting fluorescence to the surface.
The diameter of the optical fiber core is 365 jim. The fibers terminate in a custom-made connector at
the sub-surface (see SPOC below). Because the fibers are readily broken if flexed or handled too
aggressively, they are housed in a flexible polyurethane-covered stainless steel sheath with a bend radius
limiting quality that effectively reduces fiber optic failure rates.
Shock-Protected Optical Compartment (SPOC) (5)
The optical fibers deliver and receive the light from a steel probe that consists of two
threaded Geoprobe rods (one inch diameter) terminated with a special optics module and tip called a
"SPOC". The SPOC employs proprietary elastomer supports, in combination with Swagelok™
fittings, to insure long term stability of the optical alignment along with protection against breakage. The
SPOC contains a parabolic mirror that acts to turn the excitation beam 90 degrees. The beam exits the
SPOC through a sapphire window that is flush-mounted on the side of the SPOC and strikes the
sediment that is pressed against the window as the probe advances through the sediment. Sapphire's
hardness (second only to diamond) allows it to resist scratching or breaking under all but the most
severe conditions. The laser beam illuminates several mm2 of the sediment that is exposed to the
window's surface and any resulting fluorescence (along with a substantial quantity of scattered laser
light) is emitted back into the window where a portion is reflected by the parabolic mirror into the return
fiber for transport to the surface.
The SPOC design is watertight; a necessary attribute when probing in saturated sediments.
Any leakage can result in evaporation of the intruding water into the air space within the SPOC. The
resulting water vapor can cause fogging on the interior surface of the sapphire window or corrosion of
the optics and hardware within the SPOC. An extra measure of humidity prevention is achieved by
purging the SPOC with an inert gas immediately prior to reassembly any time it is open to the
atmosphere for maintenance or adjustment.
Emission Detection System (hidden from view) (7 and 8)
The collection fiber returns the entire spectrum of light ("white light") that is collected from
the sediment. Since this is a multi-channel (multi-wavelength) detection system, we must disperse the
white light. The collection fiber is butt-coupled directly into an Acton SP150 imaging monochromator
that disperses the white light into a "rainbow" that can sampled at four regions (340, 390, 440, and 490
nm) for detection.
-26-
-------
Before the light is dispersed by the monochromator, the source light from the Excimer laser
(308 nm) must be removed and the amount of fluorescence light must be controlled. The detector does
not differentiate between laser light and fluorescence, so this laser light must be filtered out. To achieve
this, a long-pass cutoff filter (320 nm CFLP) is placed immediately inside the monochromator, rejecting
the vast majority of laser light, but passing the lower energy (longer wavelength) fluorescence. Butt-
coupling of the fiber to the monochromator eliminates the slits that are usually found on the entrance of a
monochromator. These slits are designed to control bandpass and the amount of light that enters the
monochromator to avoid saturating the detector. The ROST system employs a neutral density filter
wheel for controlling light levels instead. By selecting an appropriate optical density filter the light levels
can be controlled with precision. The reference emitter signal (Ml described later) was attenuated in
these studies while the PAH fluorescence was passed through without filtering due to its relatively low
intensity.
At this point, if all four fiber optics were of the same length and were directed into the
detector (PMT), all four channels would be combined into a single decay curve (waveform). To
achieve separation of the four wavelengths, the photons must be time-delayed so that they strike the
detector at different times. To achieve this the fiber optics are made in 10 m increments - 2, 12, 22,
and 32 m in length for the 340, 390, 440, and 490 nm wavelengths, respectively. Because of the finite
speed of light in the optical fiber core, these "delay lines" allow for the temporal resolution of each
channel by approximately 50 ns. These four fibers are then terminated in a single large-core SMA fitting
which couples to a large diameter (1500 micron) fiber optic that is 0.33 m in length. This large diameter
fiber is taken through a relatively sharp bend that serves to "mix" the four fiber optic beams into one
homogeneous beam. The large fiber is attached to a mount that directs the light at the photocathode of
the PMT detector.
Oscilloscope (2)
The entire waveform arrives in less than 250 ns; therefore a very fast device is required to
accurately record the data. The ROST system employs a 100 MHz Tektronix® TDS 220 digital
storage oscilloscope capable of 1 billion samples per second (1 Gs) to record the waveforms. A 50-
ohm terminator at the input of the fluorescence channel converts the current to a voltage, allowing
measurement of a voltage vs. time waveform that represents the arrival of the photons at the PMT. A
second channel of the oscilloscope is used to monitor an energy meter (a much slower measurement)
before each test, to log the laser energy performance for maintenance/service tracking purposes. The
fluorescence waveforms are displayed on the oscilloscope in real time and are retrieved from the
oscilloscope via a general-purpose interface bus (GPffi) for storage and analysis. Approximately 50
laser shots are averaged for each sampling point along the test, which ends up being equivalent to a 1
Hz waveform storage rate. At the probe advancement speed used in this study, the vertical data density
averaged 0.3 to 0.5 inches (0.8 to 1.0 cm).
Instrument Control & Data Acquisition Computer (1)
A rack-mounted industrial computer is used to control the ROST system and log the data to
the hard drive. The computer controls the monochromator, the oscilloscope, a differential GPS beacon,
and the depth control and acquisition module (DCAM). The host computer program was written in
Visual Basic 5. The software provides a real time display of the test results while the test is in progress
and generates a full color picture of the log at the end of the test. The waveforms are continuously
-27-
-------
logged to the hard drive while a total FVD log is created by integrating the entire fluorescence waveform
and plotting its intensity vs. depth. The final data analysis and display was done on a separate
workstation.
Depth Control and Acquisition Module (DCAM) (4)
To accurately log PAH fluorescence vs. depth, it was crucial to accurately and dependably
monitor the position of the probe as it was being advanced or retrieved from the sediment. Since the
probe was positively engaged to a gearbox via a chain drive mechanism (direct drive), a depth sensing
system was designed based on monitoring the rotation of the gearbox. A waterproof proximity sensor
was mounted in a position that allowed the proximity sensor to sense the passing of the gear teeth as the
gear rotated. The proximity switch is phase sensitive, allowing determination of both the speed and
direction of the gearbox and ultimately the probe speed and direction.
A custom DCAM was built for this project. The DCAM controller for this project was
custom programmed to correctly sense the proximity sensor and calibrated to convert the proximity
sensor counts to distance (measured in feet). The DCAM's sole function is to monitor the proximity
sensor, calculate the speed, direction, and distance traveled, display the result on a digital display, and
make the results available to the ROST computer via RS-232 protocol.
Calibration and Normalization
The ROST system response depends on a host of factors. These include laser energy, fiber
termination quality, neutral density filter selection, parabolic mirror efficiency, fiber length, and others.
To account for changes over time and location, a single point calibration and system check is
performed. A reference emitter (coined Ml) is placed on the sapphire window and the response is
measured. The Ml solution is permanently stored in a quartz cuvette for convenience and the
measurement takes place through the wall of the cuvette. This proprietary mix of hydrocarbons
fluoresces efficiently across the entire system and serves as both an indicator of system function and as a
data normalization benchmark.
The total fluorescence intensity (area under the waveform) of Ml serves to normalize the
data from the push that immediately follows the reference emitter measurement. All the FVD logs are
presented as a percentage of the signal achieved with Ml. The area under every waveform in the data
set is integrated, resulting in a picovolt-seconds unit (pVs). These values are divided by the pVs
measured for Ml, and the result is multiplied by 100. The result is a log with x-axis units of percent of
Ml. This creates a normalized data set that takes into account the entire system performance, from
laser to oscilloscope. The shape of the Ml waveform acts to guide the operator in assessing proper
alignment of the detection system. The relative contribution for each channel and the shape of Ml
waveform is monitored for consistency to insure that the waveforms remain consistent from day to day.
-28-
-------
Field Sampling Methodology
A key part of this project was to develop a means to deliver the LIF probe into submerged
harbor sediments and to develop a means to collect sediment samples at depth from within the DPT rod
co-linear with the optical window. These samples were needed to directly calibrate LIF response to
measured PAH concentrations in the same sediment. Both the delivery vehicle and sediment sampler
were designed to operate from a surface vessel in depths as great as 45 feet.
Shipboard Delivery Vehicle
The shipboard delivery vehicle (SDV) consists of a steel framework supporting a 9-foot tall
DPT-like pushrod assembly. This framework has a rectangular base consisting of a 3.5 x 5 foot steel
plate with a hole in the center through which the DPT rod is advanced. Total dry weight of the SDV is
780 pounds. The pushrod assembly is advanced via a sliding trolley attached to a chain drive. Motive
force is provided by a 1/3 horsepower, variable speed, reversible electric motor and gear reduction
drive. Maximum insertion speed is 2 cm/sec.
The SDV is fitted with flotation tanks that can be evacuated and filled with air via electrically
driven solenoid cut-off valves. By emptying or filling the tanks with air, the SDV can be automatically set
for positive buoyancy (floating), nearly neutral buoyancy (a submerged weight of-50 pounds) or fully
weighted (flotation tanks full of water). Requisite air control valving, interior water level sensors, and
electrical harness are all housed on the SDV.
Ancillary equipment housed on the SDV includes automatic limit switches to avoid over
advancement of the pushrod assembly, video camera and lights, level indicators, an "on-bottom"
indicator switch and a gauge to measure depth of pushrod penetration. The penetration gauge is capable
of continuously monitoring at mm resolution, in both up and down directions, the location of the LIF
probe beneath the sediment-water interface.
The SDV is completely submersible and when in use is attached to the surface vessel only by
a lifting cable at the top and an umbilical cord. The umbilical contains a Kevlar strength member, a multi-
conductor electrical cable for power (120 VAC and 240 VAC) as well as valve control and switch
cabling, a fiber optic cable (with two optical lines), a video feed cable and compressed air. All controls
are lead to a weatherproof control box on the surface vessel. The physical detachment from the surface
vessel allows field work to be performed in other than flat calm conditions as would be necessary if the
probe was being advanced directly from the surface vessel. The maximum depth of operation is limited
only by the length of the umbilical.
The SDV was specifically adapted for use with the research vessel maintained by the EPA
Great Lakes National Program Office (GLNPO) (the R/V Mudpuppy) although it will work with any
vessel that has AC power and a large enough lifting capacity to handle the SDV. For use with the R/V
Mudpuppy, the floating SDV was cinched to the square bow of the vessel, and pushed to the site of
interest. After the vessel was anchored on-site, the SDV was released and the flotation tanks partially
filled with water to the preset negative buoyancy of 50 pounds. The SDV was gently lowered to the
bottom by the lifting cable, the flotation tanks completely filled with water and the pushrod assembly
advanced into the sediment. As the LIF probe, located near the tip of the pushrod assembly, advanced
-29-
-------
into the sediment a real-time fluorescence signal was collected and a paper copy was immediately
printed. Upon completion of the push, the sequence was reversed: the pushrod was retracted, the SDV
re-floated, cinched to the vessel and moved to the next site. The entire sequence, including anchoring
and transit time, took approximately 30 minutes. Figure 11 shows the SDV with the components
labeled. Figure 12 shows the SDV in transit attached to the .R/FMudpuppy.
Figure 11. SD V components and key peripheral devices.
Umbilical (1), DC AMsensor (2), drive motor (3), push rod (4), video camera (5), video light (6),
"over-penetration " limit switch (7), flotation tanks (8), air valving manifold (9), push rod drive
head (10), sediment sampler valving manifold (11).
-30-
-------
Figure 12. The R/VMudpuppy in Milwaukee Harbor.
The SDVis cinched to the bow of the vessel.
In Situ Sediment Samplers
Two sediment sampler designs were studied. Both samplers operate from within the pushrod
assembly above the LIF probe. Both collect sediment from a position approximately six inches above
and directly in line with the LIF window. The pushrod assembly can be stopped at any point, either
during insertion or retraction, to take a sample.
The first design was based on retracting a sleeve in the pushrod thereby exposing a spring-
loaded door that opened into the sediment. Closing the sleeve back down over the door pushed the
door shut and served to collect sediment. Retracting and closing the sleeve was accomplished by
reversing the direction of the entire pushrod assembly. The sleeve was held closed during normal
pushrod insertion/retraction by an electrically actuated, pneumatic latch. The latch was actuated from the
main control box on the surface vessel. This design never became fully operational and was not used
during field-testing.
The second sampler was in essence a small horizontally-oriented core barrel that extended
and retracted laterally out of the pushrod assembly. The core barrel was 1 cm in diameter by 3 cm in
length and could collect ~ 1 g of dry sediment. This amount was sufficient for GC-MS analysis
preceded by pressurized fluid extraction (as described in the Experimental section). The core barrel
extruded through a perforated septum that was designed to seal the end of the core barrel before and
after sample collection. The core extruded over a sliding base plate that provided sufficient suction (like
a syringe) to keep the sediment in the core barrel as it retracted into the pushrod assembly. An
electrically actuated, 2-way pneumatic cylinder provided the motive force for core barrel extension.
-31-
-------
Retraction was accomplished by a coil spring inside the pushrod assembly. The core barrel was
actuated from the main control box on the surface vessel. The requisite pneumatic valve manifold was
mounted in a watertight compartment housed on the SDV. Operation in the field consisted of removing
the pushrod assembly from the re-floated SDV (via the removal of one drift pin), removing the septa
and transferring the sediment into a sample container with a spatula. The core barrel was then cleaned,
a new septum was installed, and the pushrod assembly returned to the SDV.
This design did reach the field-operational stage and was used to collect several confirmatory
samples during field-testing. The sampler was damaged midway during field-testing and design
refinements are envisioned.
Field Testing Program
Development of the SDV and integration with the LIF spectroscopic system included
significant field-testing from a crane at the dock of the Great Lakes WATER Institute, the WATER
Institute vessel (^FNeeskay) and the EPA vessel (R/FMudpuppy). A similar testing program was
employed for the sediment samplers. Only a description of the final field demonstration (on the RV
Mudpuppy) is included in this report.
The final field demonstration concentrated on the lowermost reaches of the Kinnickinnic
River in Milwaukee, WI. During the course of 2.5 days, ten sampling stations were occupied along the
Kinnickinnic River, from the mouth (near the Jones Island sewage treatment plant) upstream to a point
halfway between the 1st Avenue and Becher Street bridges. Figure 13 is an annotated aerial
photograph showing the sampling locations and pertinent landmarks. Table HI lists the station numbers,
latitude-longitude locations and other pertinent data for each location.
At least one LIF push was performed at each site. At station 6.5, a replicate push was done
after re-positioning the SDV about one foot away from the original location. This was done to check on
the reproducibility of the LIF signal. At station 5.5 a replicate push was performed at a faster
penetration rate. This was done to examine how much of the vertical resolution was lost by a faster
insertion speed. The sediment sampler was used to collect samples at stations 1, 3, 5, and 7. On a
subsequent push the probe hit refusal and the SDV tipped over with the pushrod assembly extended in
the sediment. This damaged the sampler and no more samples were collected in this manner.
Immediately following completion of the LIF fieldwork, 8 stations were re-occupied and
VibraCore samples collected. The original LIF sites were re-located for coring by use of the shipboard
GPS system. The accuracy of this system is such that re-location was within 15 feet. Core samples
were collected in 3-inch clear plastic tubes. Core tubes were immediately sealed in the field and were
stored frozen in the dark to await sectioning and GC-MS analysis. A total of 44 individual sediment
samples were analyzed by GC-MS, of which 36 came from these core samples. The remaining eight
were collected from the sediment surface via Ponar ("clam-shell") sampler in other parts of Milwaukee
Harbor. Aliquots of all 44 samples collected for GC-MS analysis were directly exposed, in the
laboratory, to the LIF window and a LIF spectrum obtained. These samples were used as the
calibration set to build the chemometric model that related LIF spectra to the known concentration of
-32-
-------
each PAH compound. Note that in this calibration set GC-MS and LIF measurement pairs were
performed on exactly the same sediment thereby eliminating any bias from differences in sample location
or stratigraphy.
-33-
-------
Table III. Descriptions of the stations occupied during field testing.
Station
Number
1
2
2.5
3
4
5
5.5
5. 5 push 2
6
6 push 2
6.5
6.5 push 2
7
Latitude
(north)
43° 01.3674'
43° 01.1133'
43° 00.8970'
43° 00.7100'
43° 00.6957'
43° 00.5195'
43° 00.4940'
43° 00.4940'
43° 00.4972'
43° 00.4972'
43° 00.4957'
43° 00.4957'
43° 00.4713'
Longitude
(west)
87° 54.2120'
87° 54.2216'
87° 54.2414'
87° 54.3387'
87° 54.3491'
87° 54.4458'
87° 54.6223'
87° 54.6223'
87° 54.7009'
87° 54.7009'
87° 54.7765'
87° 54.7765'
87° 53.8263'
Water
Depth
(feet)
29.0
15.5
24*
14.0
16.2
21.3
5.6
5.6
5.6
5.6
7.9
7.9
8.0
LIFPush
Depth
(feet)
9.37
9.60
9.54
9.56
9.54
9.60
9.54
9.55
9.54
4.85
9.57
9.55
9.54
* depth varied from 23 to 25 feet across the bottom of the SDV
VibraCore Depth
(feet)
2.58
2.17
—
—
9.00
3.00
Description
near Jones Island treatment plant
at asphalt plant loading dock
at the coal loading dock
at wall just downstream of coke plant inlet
right in the middle of the coke plant inlet
75 meters east of KK Ave. bridge
5 -42 Lst off the fishing boat dock
penetration rate is 2.5 times faster
2-5° Hust upstream from 1st Street bridge
pit refusal and tipped sled over
° • palfway from station 6 to bend in river
pplicate push
9.33
on the corner where river turns to the south
-------
-.™y.. _^ Jones Island
Sewage Treatment Plant
asphalt loading dock
WATER Institute
abandoned coke plant
Figure 13. Aerial photograph of study area.
LIF test stations denoted by station number. North is to the top of the photograph.
-35-
-------
Data Analysis
The calibration set was used to build a PLSR model relating the LIF spectra to PAH
concentration for each of the 16 PAHs. The LIF spectra were collected by smearing a sediment
sample that had been collected by the VibraCore on the optical window of the LIF probe. These
spectra were then related to a GC-MS analysis (using EPA 8270C protocol) of the same sediment.
These data are referred to as "uphole" data. "Downhole" data refers to spectra collected from in-the-
field LIF pushes. Twenty replicate spectra were collected and an average spectrum was generated after
removing the first and last two spectra in a set of twenty. This was repeated for each sample in the
calibration set. Some preliminary PLSR models were built to test three common preprocessing
strategies. For this data it was found that auto-scaling gave the best results. For each of the 351 points
in the MWW matrix, the mean and norm was calculated over each sample. Each sample was then
subtracted by the mean and divided by the norm. This operation gives each of the 351 data points an
equal weighting. The PLSR algorithm can still give more weight to spectral points that provide
important correlations with concentration as opposed to those that contain little useful information. While
applying the models to the downhole data, it was found that one further preprocessing step was needed.
The LIF intensities of the downhole data are significantly different from the intensities of the uphole data.
Thus, it was necessary to scale each MWW between 0 and 1 for both the calibration set and the
downhole data before the auto-scaling over variables was performed. Even though absolute intensity
information is lost, PLSR still successfully built quantitative models by looking at the relative intensities of
the 351 spectral points for each MWW.
PLSR includes a data dimensionality reduction aspect analogous to principal components
analysis. Associated with this is the question of the number of latent variables to use for a given model.
As the number of latent variables is increased, the percentage of variance explained by the model for
both the concentration and spectral domains generally increase to a maximum of 100%. In addition, the
ability of the model to predict the concentrations of the data samples in the calibration set also increases.
This predictive ability is generally quantified by a root mean square error of calibration calculation
(RMSEC) where:
(actual — predicted )
RMSEC = -v/^-1 '• —
n
where n = total number of samples.
After a certain number of latent variables, however, the ability of the model to predict the concentration
of PAHs for spectra not in the calibration set begins to decrease. The use of a prediction set, in which
the samples are not used for calibration, allows the generation of a root mean square error of prediction
(RMSEP) that can be used to find an appropriate number of latent variables. For sets of data with small
numbers of samples it is possible to use another measure of merit called the root mean square error of
cross validation (RMSECV). With this method, a separate model is built for each sample in the
calibration set. Each model is built using all samples in the calibration set except for the sample whose
value is being predicted. The result is a predicted value for each sample where the predicted value for a
given sample is achieved with a model in which that sample is not used in the building of the model. The
RMSECV serves the same purpose as the RMSEP, but allows the model development to proceed
-36-
-------
without needing to split the available set of data into a calibration and prediction set. RMSECV analysis
was used for the research described in this report. After determining the appropriate number of latent
variables for each PAH model, the full data set was used to rebuild the model with all samples used in
the calibration set. The resulting model was used to predict downhole PAH values.
It is important to be able to catch the cases in which a spectrum will not result in a valid
prediction. In the current work, a very simple test has been implemented. The cosine of the angle
between the unknown MWW and the closest MWW in the calibration set is calculated for each
unknown MWW. By normalizing all vectors before doing the vector comparison, the maximum value
will be 1.0 (for a perfect match). If the cosine of the angle value is low, however, the ability of the
model to make an accurate prediction for that unknown cannot be guaranteed. An added benefit of this
outlier check is that sediment samples for outliers could be taken and the resulting data points (after LIF
and GC-MS analysis) could be added to the model, thereby increasing the predictive ability of the
model. It is also important to establish concentration limits below which the models cannot give
reasonable predictions (a defacto limit of detection based on the models). The calculation of an average
relative error for the entire sample set versus subsets in which the predicted PAH concentration is above
some threshold limit is one attempt to achieve such a defacto limit of detection.
-37-
-------
RESULTS & DISCUSSION
Field Performance of the SDV/LIF System
The field-testing program successfully demonstrated the ability of the SDV and the
associated LIF system to rapidly collect fluorescence-depth profiles in the field. Within a 2.5 day time
frame, 10 profiles were collected, 2 of which were replicated. A full 9-foot penetration was achieved at
all 10 profile sites. In only one case did the LIF probe hit refusal (at a depth of 5 feet) and that was in
an attempt to replicate a third site (station 6). The ability to penetrate these sediments is considerably
better than the equivalent ability of the VibraCore system. This is evident by inspection of Table HI that
shows only two VibraCores penetrating to a depth of 9 feet.
Printed results were obtained within minutes of completing a LIF push. An example of total
fluorescence-depth profiles obtained in the field is shown in Figure 14. These plots contain not only total
intensity, but also spectral information. The spectral information is shown at four specific wavelengths in
the waveform plots as well as by the color-coding of the main profile. The color-coding scheme is
explained in the "FVD colorization" section.
Figure 14 shows the two replicate pushes performed at station 6.5 and serves to illustrate the
reproducibility of the LIF signal. These two pushes were performed within 1 foot of each other and
show essentially identical spectral information (color changes) and very similar total fluorescence
response. Figure 15 shows the loss in vertical resolution when the penetration rate is increased from 1.0
to 2.5 cm/sec. It was concluded that the minimal time savings realized by a faster penetration rate was
not worth the loss in vertical resolution. The standard penetration rate for all stations was kept at 1
cm/sec. Appendix A contains all of the total fluorescence-depth profiles collected during the course of
this demonstration.
PAH Quantification by Chemometric Data Analysis
Total fluorescence in any given sediment reflects not only the presence of PAH compounds,
but also any other fluorophore that may be present in the sediment. To extract PAH fluorescence from
background fluorescence, a chemometric model based on PLSR analysis was performed as described
in the "Data analysis" section. Initial PLSR analysis was performed on a
set of 44 individual sediment samples for which PAH concentration and the corresponding LIF
response was known. The initial PLSR analysis identified one anomalous sample (station 6.5, depth
4.58-4.75 feet) that could not be modeled. The exact cause of this anomalous behavior is unknown, but
could be from an unidentified pollutant, or an unusual background fluorescence signal. This sample was
excluded from further analysis and the final chemometric model was built using the remaining 43
individual samples. The resulting chemometric model discriminated fluorescence from each individual
PAH compound and allowed the prediction of individual PAH concentrations from total LIF signal.
-38-
-------
LIFPII09
Figure 14. Depth profiles of total fluorescence.
Two replicate pushes taken at station 6.5 are shown. Replicate pushes were made within one
foot of each other. Spectral plots to the right of the profiles indicate the relative intensity of
fluorescence signal at 340, 390, 400 and 490 nm.
-39-
-------
a
:; r-,
06
Q.TS
1
1 .">-.
I i
1 75
1
•i :•:.
it
S75
a
i ;•:,
34
.-. :•-,
535
•j j
575
a
:, ?',
8.S
675
7
725
U
775
a
e.75
LIFPII10
UFPII11
90 100 160
Figure 15. Depth profiles of total fluorescence for two replicate pushes taken at station 5.5.
Replicate pushes were taken within one foot of each other. Profile on the right was taken at a
penetration rate 2.5 times faster than the profile on the left. Spectral plots to the right of both
profiles indicate the relative intensity of fluorescence signal at 340, 390, 400 and 490 nm.
Total PAH concentration in the calibration set ranged from 10 to 650 |ig/g dry weight. Basic
sediment characteristics of a subset of these 43 samples were measured. The subset was chosen to
span the widest range in characteristics observed in the field. Results are shown in Table IV. A wide
range in grain size distribution (% sand, silt, clay and carbonate) is evident. As would be expected in
harbor sediments, percent organic carbon is relatively high (1.2 to 5.4 %). Our ability to quantify PAH
levels and to discriminate between background and PAH fluorescence is applicable to sediments that
fall within the limits seen in Table IV.
-40-
-------
Table IV. Gross sediment characteristics.
Data shown is for a subset of the 43 samples used in the calibration set. Samples were chosen to
span the widest range in sediment characteristics observed in the field.
Station
number
1
2
4
4
5.5
5.5
6
6.5
6.5
6.5
6.5
6.5
6.5
maximum
minimum
Depth
(feet)
1.75-2.25
0.75-1.25
2.75-3.25
6.75-7.25
1.92-2.08
4.92-5.08
1.75-2.25
1.50-1.67
2.50-2.67
3.00-3.17
3.58-3.75
4.58-4.75
8.08-8.25
% Sand
26.0
48.0
22.1
42.9
35.2
22.1
69.9
92.5
42.0
31.9
27.8
17.9
16.0
92.5
16.0
% Silt
65.0
42.4
65.3
39.3
57.8
66.3
24.2
5.0
51.0
60.0
62.2
62.7
68.7
68.7
5.0
% Clay
9.0
9.6
12.7
17.8
7.0
11.7
6.0
2.5
7.0
8.1
10.0
19.5
15.3
19.5
2.5
% Carbonate
39.3
46.2
38.8
68.3
39.3
39.7
32.5
34.6
38.9
37.4
39.1
39.4
38.4
68.3
32.5
% Organic
Carbon
3.7
3.4
2.6
3.4
2.2
3.3
3.4
1.2
4.1
5.4
4.1
2.0
3.7
5.4
1.2
Figure 16 shows the match between measured and predicted concentrations for representative 2, 3, 4,
and 5-ring PAH compounds. GC-MS concentrations for all 43 samples are given in Appendix B.
Structural formulas for all 16 PAHs are also provided in Appendix B. Plots of the match between
measured and predicted concentrations, along with the associated data tables, are given in Appendix C
for all 16 PAH compounds. The match between predicted and measured concentrations, as seen in
Figure 16 (and Appendix C), is sufficiently close for the technique to be used as a screening tool for all
compounds except acenapthylene, dibenz(ah)anthracene and naphthalene. The measured concentrations
of these three compounds in our calibration set were too low (<3 jig/g in most of the 43 samples) to
allow the building of a robust chemometric model.
-41-
-------
fluorene
CJ (N S 0->
sample site
- Actual
Pr&ineteij
anthracene
sample site
-Actual —o—Predicted
chrysene
-Actual —a— Predicted
benzo(a)pyrene
•ample cite
Actual — a — Predicted
Figure 16. Predicted vs. measured concentrations of representative 2-ringPAHS.
These are: 2-ring (fluorene), 3-ring (anthracene), 4-ring (chrysene) and 5-ring (benzo(a)pyrene) PAHs. Individual samples come from
station 1 (sample 1,2); station 2 (sample 3); station 4 (samples 4, 5, 6); station 5 (sample 7); station 5.5 (samples 8-18); station 6
(samples 19, 20); station 6.5 (samples 21-35); various surface samples collected in clean portions of the harbor (samples 36-43).
-42-
-------
Several techniques were used to quantify the extent to which the chemometric model can be
used as a predictive tool. The harshest measure of predictive ability is the root mean square error of
cross validation (RMSECV). In essence, RMSECV analysis uses all but one of the data points to
predict the concentration of the remaining point. This process is repeated until all 43 data points have
been predicted from models in which each data point was not included. Table V lists, for each PAH,
the relative errors between predicted and measured concentrations for the 43 samples in the calibrations
set using RMSECV analysis. Also included is the maximum and minimum concentration observed in the
calibration set. This maximum and minimum are the effective concentration range over which the
predictive model is valid. Complete data tables for the RMSECV analysis are given in Appendix D.
Table V. Relative errors between measured and predicted concentrations.
Error is shown for each compound as determined by root mean squared error of cross
validation. Compounds in italics contain measured concentrations that were too low to make a
robust prediction. RE=average relative error; TL=threshold limit in jug/g; [m]=measured
concentration (minimum, maximum) in jug/g.
Compound
Naphthalene
Acenapthylene
Acenapthalene
Fluorene
Phenanthrene
Anthracene
Fluoranthene
Pyrene
Benz(a)anthracene
Chrysene
Benzo(k)fluoranthene
Benzo(b)fluoranthene
benzo(a)pyrene
Indeno( 1 23 - cd)pyrene
dibenz(ah)anthracene
Benzo(ghi)perylene
overall average
average without low concentration
samples
RE overall
(%)
312.9
28.9
71.7
45.6
141.0
41.8
123.3
86.7
32.2
59.3
36.9
30.7
53.0
90.1
100.9
42.1
81.1
65.7
RE above TL
(%)
114.5
12.8
24.9
27.2
35.7
28.7
41.9
42.4
22.7
24.9
29.1
20.7
19.3
33.6
100.9
21.9
37.4
28.5
TL
1
1
2
2
10
2
10
10
10
10
10
10
10
10
2
10
[m] min
0.0
0.0
0.0
0.0
0.0
0.7
1.2
0.0
0.0
1.0
0.0
0.0
0.0
0.0
0.0
0.0
[m] max
12.5
2.7
10.0
12.1
95.8
16.1
155.4
131.2
42.3
47.2
83.4
34.0
44.2
48.6
11.9
30.9
For the chemometric model to discriminate background fluorescence from PAH fluorescence, a certain
level of PAH signal is required. The three compounds that were not found in sufficient quantity
(acenapthylene, dibenz(a/z)anthracene and naphthalene) were eliminated from further analysis because
-43-
-------
the PAH signals were too low to model. For the remaining 13 compounds, the average relative error is
65.7%. As the concentration of a particular PAH diminishes, its signal becomes difficult to distinguish
from background. There exists a threshold concentration below which relative errors become
increasingly large. This threshold limit has been set at 10 |ig/g except for compounds in which 10 |ig/g is
at or near the maximum level measured. In these cases, the threshold limit was set to 1 or 2 jig/g. Using
the threshold limit approach, the average relative error diminishes significantly — from 65.7% to 28.5%.
Although similar in concept to a "detection limit", the threshold limit is not a detection limit. Detection
limits have a well-defined meaning based on replicate analyses of a sample and the random nature of
indeterminate errors. The threshold limit is qualitatively defined as the level below which the match
between measured and modeled concentrations begins to diverge.
The chemometric models developed from this calibration set were applied to the LIF pushrod data
collected at all 10 field stations, including replicate pushes. Figures 17, 18, and 19 show the depth
profiles for stations 7 (farthest upstream site), 1 and 2 (farthest downstream sites), respectively. Depth
profiles of all the stations are given in Appendix E. Figures 17 through 19 show both the measured total
fluorescence intensity profile and a profile of the modeled concentrations of total PAHs. The former is
generated directly in the field, the latter is a summation the modeled concentrations for each of the 16
individual PAHs taken at each depth. The total PAH concentration line is the 5-point moving average of
modeled concentrations taken at approximately 1-cm intervals. Because LIF spectral data yields a
measurement of each individual PAH, depth profiles for individual PAHs can also be generated.
Example profiles of individual 2, 3, 4, and 5-ring PAH compounds are shown in Figure 20 for station 7.
Complete concentration-depth data tables (available in digital format) are a part of Appendix E.
-44-
-------
50 100 150
Total Fluorescence Intensity
Station 7
800
10
depth (feet)
Figure 17. Plots of field fluorescence measurement and chemometric prediction: Station 7.
The total fluorescence intensity profile as obtained in the field for station 7 is shown on the left.
On the right is the concentration profile of total PAHs obtained after application of chemometric
analysis. The concentration profile is a 5-point moving average of individual measurements.
-45-
-------
0
05
1
125
15
175
2.25
25
2.75
3
•^^* *T
£41
9)4.75
Q 5
55
6
65
6,75
7
7.25
75
7.75
8,75
*!
:
Station 1
800
, 700
E
£ 600
r 500
o
t! 400
300
200
100
456
depth (feet)
10
50
13C
150
Figure 18. Plots of field fluorescence measurement and chemometric prediction: Station 1.
The total fluorescence intensity profile as obtained in the field for station 1 is shown on the left.
On the right is the concentration profile of total PAHs obtained after application of chemometric
analysis. The concentration profile is a 5-point moving average of individual measurements.
-46-
-------
J "
0.25
r\ c
075
1,25
1.5
1.75
2.25
3 75
3
3.25
3.75
fj
•+-J if?
Q "
5.5
c 7-c
6
6.25
fi c
6.75
7
7.5
7 7J-i
ft 'W
O.-lO
ft A
O-3
ft 75
n
f
)
\
*
|
(
i
I
.
i
I
:
i
-
~
..
-
:
-
;
•
: -
'-
:
•
_
• -
_
r
;
r
-
.
50 100 150
Total Fluwesoence Insenaily
Station 2
800
456
depth (feet)
Figure 19. Plots of field fluorescence measurement and chemometric prediction: Station 2.
The total fluorescence intensity profile as obtained in the field for station 2 is shown on the left.
On the right is the concentration profile of total PAHs obtained after application of chemometric
analysis. The concentration profile is a 5-point moving average of individual measurements.
-47-
-------
Station 7
456
depth (feet)
10
Figure 20. Depth profile of four individual PAH compounds at station 7.
Representative multi-ringed aromatic compounds are shown. Blue =fluorene [two aromatic
rings]; green = anthracene [three aromatic rings]; red = chrysene [four aromatic rings]; black
= benzo(a)pyrene [five aromatic rings]. Trendlines represent the five-point moving average of
individual measurements.
A general agreement between total fluorescence intensity and total PAH concentration can
be seen in Figures 17 through 19, thereby indicating that a relatively large percentage of total
fluorescence is due to the presence of PAH compounds. This generally low background fluorescence is
a key feature of these sediments that has allowed for the successful discrimination of PAHs by their
fluorescence signals. However, this general agreement does not mean that raw fluorescence intensity
data can be used (i.e., without resorting to chemometric analysis). This is exemplified by comparing
stations 1 and 2. Total fluorescence intensity below 2.5 feet is low at both stations, with station 2
exhibiting less intensity (typically less than 10%) than station 1 (generally between 10 and 20%).
Chemometric data analysis shows the reverse, with considerably higher PAH concentrations in station 2
(200-300 |ig/g) than in station 1 (50-100 |ig/g). The difference lies in the spectral signature at the two
sites. The chemometric analysis uses both spectral and intensity data in the discrimination of PAH from
background fluorescence. This can be qualitatively seen by comparing the green color of the total
fluorescence plot of station 1 with the predominately yellow color of station 2.
-48-
-------
Source Apportionment
The rapidity with which LIF data can be collected, in combination with the ability to identify
individual PAH compounds, allows our technique to be used to deduce the sources of contamination in
a given stretch of harbor sediment. These source apportionment calculations can be very useful in
gaining an overall understanding of general contamination patterns. A detailed investigation into source
apportionment was not part of this project, but an example calculation is provided to illustrate the power
of this technique.
Previous workers have shown that the relative proportion of certain marker PAHs is
indicative of the industrial process that generated the PAHs. Li et al. [44] used seven marker
compounds to differentiate between three contaminant sources: wood burning, coke oven emissions and
highway dust. The primary assumption about the environmental behavior of PAHs that underlies this
analysis is that there is no preferential degradation of one PAH over another [44]. Wood burning
preferentially produces naphthalene, while coke ovens produce a preponderance of the 5-ring
compounds benz(a)anthracene and benzo(a)pyrene. Highway dust generates a relatively large
proportion of pyrene and phenanthrene. These three signatures are shown in Figure 21.
Figure 21 also shows data taken from two sites (station 1 and 7) superimposed on the
source signals. A strong highway dust signature is evident in station 7 extending to a depth of at least 8.0
feet. A similar signature is seen in all the other sites upstream of station 5.0 to depths as great as 9.0
feet. These upstream stations are in a portion of the Kinnickinnic River that is not currently dredged
(note the depth of water values in Table HI). PAH contamination in this stretch of river appears to be
primarily caused by runoff of generic highway dust throughout the Milwaukee metropolitan area. The
source signatures seen in station 1 are very different. Deep in the core (6.5 feet) the signature has a
coke oven component, whereas in the shallower section (1.3 feet) there seems to be a wood-burning
component.
Highway dust contamination extending to at least 9.0 feet necessitates a very high
sedimentation rate in the upstream portion of the Kinnickinnic River. If it is assumed that large amounts
of automobile traffic arose in the Milwaukee area in the mid 1930s, sedimentation rates would be
approximately 1.6 inches/year. Although this is surprisingly high, sedimentation rates of between 1.1 and
3.8 inches/year have been reported in the Kinnickinnic River [45]. Furthermore, it is known that during
World War n the Kinnickinnic River was dredged for large vessels as far upstream as the Becher Street
Bridge and that the sediments have filled in to the current depth of 5.6-8.0 feet since that time.
The enhanced understanding of contaminant sources and sediment dynamics that was gained
from source apportionment calculations is evident. This understanding was gained even though the LIF
survey was not designed for this purpose and generic literature values were used for source signatures.
A thorough attempt at source apportionment would necessitate a multi-variate analysis of the source
signatures as well as the sampling of possible source areas and generation of source signals specific to
the local area.
-49-
-------
source signals
co — we
station 1
- - «- - 1.3 feel - - •- - 6.5 fee*
£fl _
d/t -
1 30-
if
1fl
n
t
station 7
P
'* ,
.:- : • :
'-V 'ff
/«• * ^* a? ^4 ^» ^
- - •- - 4.5 feet - - •- - 8.0 feet
Figure 21. Source apportionment.
Source signals for highway dust (HWY), coke oven emissions (CO) and wood burning (WB) are
shown in the top panel. Data from station 7 and station 1 superimposed on source signals are
shown in the middle and bottom panels respectively. Source profiles are from [44].
-50-
-------
CONCLUSIONS
The results of this study lead directly to the following conclusions:
• The SDV is an effective means for rapidly deploying an LIF probe into submerged harbor
sediments. Penetration depths of nine feet can be routinely achieved. Station-to-station
turnaround time is -30 minutes. The SDV has a flexible design that can accommodate any
probe that can be fitted to the end of a DPT rod.
• LIF is a viable technique for the in situ detection of PAH compounds in harbor sediments.
Tracking of total fluorescence alone is not sufficient. A suitable chemometric model must be
developed to separate background fluorescence from PAH fluorescence.
• Field testing along the Kinnickinnic River as it enters Milwaukee Harbor shows that LIF can
predict individual PAH compounds to within an overall average relative error of 81%. For the
13 PAHs that were found in sufficient quantity to model, the average relative error drops to
28.5% if the sediments that contained less than a minimum threshold limit concentration
(between 2 and 10 |ig/g) were eliminated. Data repeatability was excellent and the method was
shown to apply to sediments that contained a total PAH concentration between 10 and 650
|ig/g. The type of sediment for which this technique was tested contained between 1.2% and
5.4% organic carbon and a wide range in sand-silt-clay content.
• The technique provides a rapid, cost-effective, in situ means to detect PAH contamination with
centimeter-scale vertical resolution. Total fluorescence plots are obtained in real-time, thereby
enabling the direction of sampling efforts to be accomplished in the field. The technique is
appropriate for screening level assessments of harbor contamination, the direction of dredging
efforts, and post-remediation analysis of contaminated sites.
-51 -
-------
REFERENCES
1 Assessment and Remediation of Contaminated Sediment (ARCS) Program, Final Summary
Report. EPA Report EPA-905-S94-001 (1994), 48 pp.
2 Great Lakes Water Quality Board of the International Joint Commission. Position Statement on
the Future of Great Lakes Remedial Action Plans (1996), 13 pp.
3 R.C. Knox, D.A. Sabatini, L.W. Cantner. Subsurface Transport and Fate Processes, Lewis
Publishers, Boca Raton, FL, 1993.
4 Realizing Remediation, A Summary of Contaminated Sediment Remediation in the Great Lakes
Basin, GLNPO Report (1998), 82 pp.
5 A.C. Meigh. Cone Penetration Testing — Methods and Interpretation, Butterworth, London,
1987.
6 D. Lucero. Hazardous Materials Control 1990, 3, 36.
7 R.V. Overby, M.E. Best. Proceedings of the Air & Waste Management Association 1993, 15,
133.
8 B.A. Torstensson. Ground Water Monitoring Review 1984, 4, 131.
9 M. Smolley, J.C. Kappmeyer. Ground Water Monitoring Review 1991, 11, 101.
10 B. Manchon. Journal of Soil Contamination 1992, 1, 3 21.
11 P.O. Kuhlmeier, T.E. Sturdivant. ASTM Special Technical Publication 1992, 1118,183.
12 C.Y. Chiang,K.R Loos, RA. Klopp. Ground Water 1992, 30, 428.
13 DA.Zemo, T.A. Delfino, J.D. Gallinatti, V.A. Baker, L.R. Hilpert. Ground Water Management
1992, 11,341.
14 C. Chiang, P. Petkovsky, M. Beltz, S. Rouse, T. Boyd, C. Newell, T. McHugh. Ground Water
Management 1993, 17, 661.
15 B.S. Mines, J.L. Davidson, D. Bloomqvist, T.B. Stauffer. Ground Water Monitoring.
Remediation 1993, 13, 115.
16 D.A. Zemo, Y.G. Pierce, J.D. Gallinatti. Ground Water Monitoring Review 1994, 14, 176.
-52-
-------
17 J.H. Aldstadt, C.H. Batson, A.F. Martin, M.D. Erickson, M.S. Costanza, M.B.J. Foster, T.R.
James, P.V. Doskey. Field Analytical Chemistry and Technology 1997, 1,239.
18 P.V. Doskey, J.H. Aldstadt, J.M. Kuo, M.S. Costanza. Journal of the Air & Waste
Management Association 1996, 46, 2.
19 K.F. Myers, J.M. Brannon, R.A. Karn, C.B. Price, D.Y. Eng, A.B. Strong, S.S. Cooper.
Fourth AWMA/EPA Field Screening Methods for Hazardous Wastes and Toxic Chemicals
Symposium, Las Vegas, NV, 1995, p. 177.
20 S.E. Apitz, L.M. Borbridge, G.A. Theriault, S.H. Lieberman. Analysis 1992, 20, 461.
21 S.E. Apitz, G.A. Theriault, S.H. Lieberman. Proceedings of the International Society of Optical
Engineering (SPIE) 1992, 1637, 241.
22 S.S. Cooper. Device for Measuring Reflectance and Fluorescence of In Situ Soil, U.S. Patent
#5,128,882(1992).
23 S.H. Lieberman, S.E. Apitz, L.M. Borbridge, G.A. Theriault. Proceedings of the International
Society of Optical Engineering (SPIE) 1993, 1716, 393.
24 A. Henderson-Kinney, J.E. Kenny. Spectroscopy 1995,10, 32.
25 W.C. McGinnis, M. Davey, K.D. Wu, S.H. Lieberman. Proceedings of the International
Society of Optical Engineering (SPIE) 1995, 2367, 2.
26 D.S. Knowles, S.H. Lieberman. Proceedings of the International Society of Optical Engineering
(SPIE) 1995, 2504, 297.
27 J. Lin, S.J. Hart, T.A. Taylor, J.E. Kenny. Proceedings of the International Society of Optical
Engineering (SPIE) 1995, 2367, 70.
28 J. Lin, S.J. Hart, W. Wang, D. Namytchkine, J.E. Kenny. Proceedings of the International
Society of Optical Engineering (SPIE) 1995, 2504, 59.
29 B.J. Nielsen, G Gillispie, D.A. Bohne, D.R Lindstrom. Proceedings of the International
Society of Optical Engineering (SPIE) 1995, 2504, 278.
30 W.T. Elam, J.V Gilfrich. Advances in X-Ray Analysis 1995, 38, 699.
31 W.T. Elam, J.V. Gilfrich. Proceedings of the International Society of Optical Engineering
(SPIE) 1995, 2367, 59.
-53 -
-------
32 R. Wisbrun, I. Schechter, R. Niessner, H. Schroder. Proceedings of the International Society
of Optical Engineering (SPIE) 1992, 1716, 2.
33 G.A. Theriault, S.H. Lieberman. Proceedings of the International Society of Optical
Engineering (SPIE) 1995, 2504, 75.
34 I.E. Meisner, W.F. Nicaise, D.C. Stromswold. IEEE Transactions in Nuclear Science 1995,
42, 288.
35 P.P. Milanovich, S.B. Brown, B.W. Colston. Penetrometer Compatible, Fiber-Optic Sensor
for Continuous Monitoring of Chlorinated Hydrocarbons — Field Test Results, UCRL-JC-
112182, Lawrence Livermore National Laboratory, Livermore, CA, 1993.
36 E.R. Cespedes,S.S. Cooper, W.M. Davis, WJ. Buttner, W.C. Vickers. Proceedings of the
International Society of Optical Engineering (SPIE) 1995, 2367, 33.
37 ASTM test Method for Deep, Quasi-Static, Cone and Friction Penetration Tests for Soils,
ASTM D3441, American Society for Testing and Materials, Philadelphia, PA.
38 S.M. Rudnick, RF. Chen. Talanta 1998, 47, 907.
39 RF. Chen, D.B. Chadwick, S.H. Lieberman. Organic Geochemistry 1997, 26, 67.
40 H. Marten, T. Naes. Multivariate Calibration, John Wiley and Sons, Chi Chester, 1996.
41 U.S. Environmental Protection Agency, Office of Solid Waste and Emergency Response, Test
Methods for Evaluating Solid Wastes (SW-846), Method 3 545 A: Pressurized Fluid Extraction,
SW-846, Chapter 4.3.1, 2001, Washington, D.C.
42 B. Szostek, J.A. Tinklenberg, J.H. Aldstadt. Chemosphere 1999, 38, 3131.
43 U.S. Environmental Protection Agency, Office of Solid Waste and Emergency Response, Test
Methods for Evaluating Solid Wastes (SW-846), Method 8270C: Determination of
Semivolatile Organic Compounds by Capillary GC-MS, SW-846 Chapter 4.3.2, 1997,
Washington, D.C.
44 K. Li, E. Christensen, R. Van Camp, I. Imamoglu. Environmental Science & Technology 2001,
35, 2896.
45 E. Christensen, A. Li, I. Ab Razak, P. Rachdawong, J. Karls. Journal of Great Lakes Research
1997,23,61.
-54-
------- |