Modeled PCB Weathering Series in Principal Components Space

MODELED PCB WEATHERING SERIES IN
PRINCIPAL COMPONENTS SPACE:
CONSIDERATIONS FOR MULTIVARIATE
CHEMICAL FINGERPRINTING
Glenn W. Johnson
Energy and Geoscience Institute
Department of Civil & Environmental Engineering
University of Utah, Salt Lake City, UT
Michael J. Bock
ENVIRON International
Portland, ME
1
Introduction
• Well-recognized that PCBs can weather after
release to the ambient environment.
• Degree of weathering differs for different
congeners.
 Lighter congeners are more susceptible to
weathering
 Degree of weathering is dependent on the material
to which PCBs are adsorbed.
•We have seen this empirically, both in benchscale lab experiments, and field studies.
Johnson PCDD/F - INEF 2013 – Penn State University
Example: Volatilization Experiments
Chiarenzelli, et al., 1997. (ES&T, 31: 587). Reported results of a series of laboratory
volatilization experiments for four Aroclors.
Residual from Arolcor 1242 experiment looked more like A1248.
How do such alteration series manifest in multivariate space?
Lake Hartwell, South Carolina
4
Weathering of PCBs
• Just two examples
•A number of studies have demonstrated this.
•Such alteration is most dramatic when “lighter” Aroclors are
weathered (e.g. Aroclor 1242).
• Tremendous implications for PCB fingerprinting/forensics.
•In this paper we demonstrate the influence this has on a
common PCB fingerprinting tool: Principal Components Analysis
(PCA).
• We use fugacity to simulate the weathering of Aroclor 1242 in
different environments.
• We show the data from these weathering models, and how the
changes in congener profiles manifest in PCA score-plots.
Johnson PCDD/F - INEF 2013 – Penn State University
Principal Components Analysis (PCA)
• Widely used in Environmental Forensics and beyond.
• Can be a stand-alone Exploratory Data Analysis Method.
• Or as an intermediate step in receptor modeling
• Objective: In its most common form - A data visualization tool.
PCA scores plot
•I’ll give you two simple PCB examples
 Both are three source systems (i.e. 3 Aroclors)
 First Example is strongly clustered
 Second Example has the same three sources, but Mixtures
Johnson PCDD/F - INEF 2013 – Penn State University
PCBs: Aroclor Congener Compositions
Data from Frame, et al., 1996
Figure from Johnson, et al., 2006 (p. 182)
Data Set 1: A Stongly Clustered Data Set
From Johnson, et al. (2007)
Johnson PCB - INEF 2013 – Penn State University
Same 3 Sources, but Mixtures
From: Johnson, et al., 2007: Chapter 7 (p. 231)
9
Fugacity Modeling
•Fugacity is a quantitative measure of the
“escaping” tendency of a chemical from one phase
to another.
• Objective: Predict concentrations of a chemical in
multiple phases (e.g., air, soil, and water) based
on equilibrium partitioning equations.
•It has been applied to PCBs but rarely to full PCB
congener profiles.
•Bock and Johnson recently developed a fugacity
model based on Land Applied Biosolids model
(LABS) of Hughes and Mackay.
• Model used to estimate the mass loss over time
and alteration of congener profiles over time.
Fugacity Modeling: Formulation
•Starting Composition: Full 209 congenercomposition of Aroclor 1242 (Frame, et al.1996 – Spl A3).
•Fugacity equations used to model congener
specific weathering due to volatilization and
solubilization/leaching,
•Modified LABS (Hughes and Mackay 2011).
– Model simulated single release of PCBs to soil/sed
– Koc values based on Poly parameter linear-free energy
relationships (PP-LFER) of Hawthorne et al.(2011)
– Two types of organic matter substrates were modeled:
1. Natural Organic Mater (NOM) (surrogate for fresh)
2. Coal Tar (CT) (surrogate for decayed organic matter)
Model Formulation
•In the full paper (Bock and Johnson, in prep), we
will report model results for five separate
environments.
•For purposes of PCA demonstration, the type of
environment modeled was submerged sediment,
with parameters shown below.
Table 1. Model Parameters
Model Aroclor
Run
Source
Matrix
Porewater
Diffusion
Exchange/
Diffusivity Leaching Boundary Path Proportion Proportion
Depth in Water
Layer
Length of Volume of Volume Organic Model
Rate
(cm)
(m2/h)
(mm)
as Air
as Water Matter Duration
(mm)
(mm/day)
1
Aroclor Submerged
1242
Sediment
5
0.000002
500
4
4
0.01
0.49
CTb
50 years
2
Aroclor Submerged
1242
Sediment
5
0.000002
500
4
4
0.01
0.49
NOMa
50 years
a
b
NOM - Natural Organic Matter as surrogate for fresh organic material
CT - Coal Tar as a surrogate for decayed organic matter
PCA Applied to A1242 Volatilization Series
•
•
Starting composition: Frame (1996) Aroclor 1242
31 samples
–
–
•
27 weathering intervals and
4 Aroclors as references
Congeners filtered to exclude variables that never exceed
0.2% of total PCBs
–
Reduced data set from 209 congeners to 130 congeners.
•
We used pretty standard data pre-treatment for PCA
– Normalize samples to percent
– Normalize congeners by auto-scaling (Z-transform)
• 3 Principal Components account for 95% of variance…BUT
13
Look More Closely at Goodness of Fit
14
Let’s Look at Goodness of Fit
15
Zoom in on 3 Congeners
What do the degraded samples look like as compared to Aroclors?
16
1242 Congener Profiles: 12 Hours
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 24 Hours
w/ Coal Tar as a surrogate for decayed organic matter
8
4
1242 Congener Profiles: 48 Hrs (2 Days)
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 4 Days
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 8 Days
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 30 Days
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 60 Days
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 90 Days
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 0.5 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 1 Year
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 2 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 3 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 4 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 5 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 10 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 15 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 20 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 25 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 30 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 35 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 40 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: 45 Years
w/ Coal Tar as a surrogate for decayed organic matter
1242 Congener Profiles: Years
w/ Coal Tar as a surrogate for decayed organic matter
Aroclor 1242 Fugacity Modeled Weathering Series
(CT Model) Plotted in 3 PC Space….
Unaltered 1242
Not a straight line between
A1242 and A1248.
A curvilinear trajectory
through multivariate space
NOM Model: 0.5 Years
w/ NOM as a surrogate for decayed organic matter
The model with coal tar as our organic
matter took 25 years to get this point
NOM Model: 4 Years
w/ NOM as a surrogate for decayed organic matter
Aroclor 1242 Fugacity Modeled Weathering Series
(NOM Model) Plotted in 3 PC Space….
Unaltered 1242
Not a straight line between
A1242 and A1248.
A curvilinear trajectory
through multivariate space
Lake Hartwell, South Carolina
44
Lake Hartwell PCBs – Multiple Source /
Multiple Alteration Conceptual Model
Dechlorination
Volatilization
45
4 Component Model
Conclusions
•Rate at which PCBs weather (and profiles change) depends
greatly on the type of organic matter present in
soil/sediment.
•In multivariate space, weathered patterns (dechlorination
and volatilization) do not manifest as simple linear mixing
of end members. Rather: curvilinear lines through multidimensional space.
•Linear Mixing Models (PCA-based or otherwise): Resolved
fingerprint patterns are interpretable in context of
weathering patterns published in the literature.
It may not be linear and it may not be mixing.
•But PCA & related methods still a good way to gain insight.
Just be aware of limitations of the conceptual model:
Conclusions (Cont.)
•A non-linear trend in PCA - may result in apparent need to
retain more principal components / sources to accurately
back-calculate data, and still there will be non-random lack
of fit
•If you suspect alteration, perhaps relax your goodness-of-fit
criterion.
•If alteration suspected - can’t blindly interpret end-members
as “sources” mixing proportions.
Receptor Modeling: 4 Patterns
• End-member 1 appeared to match a
relatively unaltered, homogenized
1248/1254 source
• Aroclor 1248 not a suspected source
• The true source was suspected to be
devolatilized 1242
Lake Hartwell PCBs – CMB Approach
50
Lake Hartwell PCBs – Receptor Modeling
EM-2: Devol 1242/1016
EM-4: Dechlorination 2 (H’)
EM-3: Dechlorination 1 (C)
51
EM-1: 1248/1254
Fugacity Modeling
Conceptual Model
Transport
Air
Volatilization
Aroclor in
Surface Soil
Surface Soil
Mineral
Grains
Pore Water
Leaching
Air
Model Formulation
•Starting Composition: Full 209 congener-composition
of Aroclor 1242 (Frame, et al., 1996).
•ackay Level III (non-equilibrium, non-steady state)
– Based on BASL model, single input of Arcolor
http://www.trentu.ca/academic/aminss/envmodel/models/BASL4110.html
– Congener specific
• Physical, chemical properties based on Oberg 2001 (QSAR
based) http://www.tomasoberg.com/qspr/papersource.html
– Currently models volatilization and dissolution/leaching,
modeling of dechlorination theoretically possible if rates and
pathways are completely described
– Can be adapted to model bioaccumulation
In Presence of Alteration
• Some of the corners are not source compositions.
• The are products of alteration.
• We carry on, treating degradation end-products as if
they behave (geometrically) as linearly mixing
sources.
• But if we look at experimental data with single source
and alteration - probably not that simple:
Aroclor 1248 Dechlorination
–
Next few slides from Johnson & Quensen (2000). Organohalogen
Compounds. 45: 280-283
– Aroclor 1248 dechlorinated in lab by anaerobic micro-organisms eluted
from Hudson River sediments.
– 20 week experiment, with 3 replicates collected at four week intervals.
– Congener specific PCB data were reported based on quantitation of 87
54 peaks.
In PCA: 3 Corners / 3 “Sources”
Unaltered Aroclor 1248
IUPAC Congener Number
Intermediate Dechlorination
IUPAC Congener Number
End-Member 3: End-Point Dechlorination
IUPAC Congener Number
Fugacity Modeling
•A framework for predictive modeling for movement of
chemicals from one phase to another
•Fugacity is a quantitative measure of “escapability”
•Objective: predict concentration in multiple phases
(air soil, water) based on equilibrium partitioning
equations.
•Has been applied to PCBs (Harner, 1995; Sweetman, 2009)
• Bock & Johnson recently developed fugacity model
based on LABS model of Hughes and Mackay.
– Used to estimate mass loss of PCBs over time, and
– Evaluate alteration of PCB congener profiles due to
differential weathering.
– Such data used here, plotted in PCA space to evaluate
degradation series.