PRINCIPAL COMPONENTS ANALYSIS OF DIOXINS FROM
THE ROCKY MOUNTAIN ARSENAL & DENVER CO.
Paul D. Jones1, John P. Giesy1, John L. Newsted2, John Slocomb2, Gerry
Henningsen3, Laura Williams3, Carol Bicher4
1
Michigan State University, 2Entrix Inc., 3USEPA, 4Gannett Fleming Inc.
Introduction
The statistical analysis of data for 17 PCDD/F congeners in ambient soil samples
collected from the Denver Front Range Dioxin Survey are reported. Principal
Components Analysis (PCA) was conducted to evaluate the utility of pattern
recognition techniques for identification of congener profiles that could
characterize typical land uses for risk assessment purposes. The five typical land
uses that were studied were: residential, commercial, open space, agricultural
and industrial. The results of these analyses are intended to be used to help
determine if suspected point sources from the Rocky Mountain Arsenal have
released TCDD-like congeners (i.e. “dioxins”) above ambient levels and/or in
different forms than those found in well characterized reference areas.
Materials and Methods
Samples were collected, analyzed, verified and reported by USEPA Region 8.
Sampling locations are provided (Figure 1). Soils were collected from 160
locations divided into 5 pre-designated off-post land-use types (open space,
agricultural, residential, commercial and industrial) as well as from 60 locations
at the Rocky Mountain Arsenal. The data were validated and provided in a
spreadsheet that contained 220 rows, each representing a single surface soil
sample location, with 24 columns of results and related sample information. The
results consist of 17 columns of data for 17 PCDD/F congeners, 2 columns for
G.P.S. data and 5 columns containing sample identifiers and location indicators.
Data reports: (http://www.epa.gov/region8/superfund/sites/rma/rmdioxrpt.html).
Assessment of Normality and Data Transformation
Data sets were next tested for normality using a one-sample KomolgorovSmirnoff test that was visually assisted by using probability plots. Normality is
not an absolute requirement for PCA, since PCA is not a strict statistical test (i.e.
it does not accept/reject hypotheses based on alpha and beta values). However
to minimize the influence of ‘outliers’ and highly non-normal distributions, data
transformations can be used to produce ‘nearly normal’ data. The issue of datanormality for PCA has been discussed extensively (Wilkinson et al 1996). After
log transformation, the distributions for the data were generally normal or near
normal. The exceptions to this rule were congeners where a significant number
of results below method detection limit (<MDL) (e.g. 2,3,7,8-TCDD).
Organohalogen Compounds, Volumes 60-65, Dioxin 2003 Boston, MA
Comparison of On- and Off-Post Samples
Without specifying a number of factors to be retained, the algorithm returned
two factors (PCs) that accounted for 84.5% of the variation, as compared to the
88.3% being explained by three factors. When the algorithm was forced to
retain four factors, 90.8 % of the variation in all the data was explained - only an
additional 2.7% of the total variation. In addition, the maximum loading for any
congener on the fourth Factor was 0.350, indicating that the fourth factor had
little power to segregate the samples. For these reasons the PCA that retained
three factors is presented here. The component loadings were similar to, but not
the same as, those observed for the On-Post or Off-Post PCAs (Table 1). The
first Factor was loaded heavily on most of the higher chlorinated PCDF
congeners, while the second Factor was more heavily weighted on the higher
chlorinated PCDDs. The third Factor was loaded the greatest for 2,3,7,8-TCDD,
with the loading for 2,3,7,8-TCDF being the second highest but not as large as it
was in the On-Post PCA (represented mostly by Factor 1).
Table 1. Component loadings for PCA on the combined
(60 On-post and 160 Off-Post samples) data set.a
Factor 1
Factor 2 Factor 3
Congener
PCDF-12378
0.914
0.195
0.205
HxCDF-123789
0.888
0.222
0.13
0.882
0.351
0.072
HpCDF-1234789
HxCDF-123478
0.878
0.406
0.125
HxCDF-123678
0.86
0.385
0.137
TCDF-2378
0.758
0.298
0.37
OCDF
0.753
0.562
0.004
HxCDF-234678
0.701
0.518
0.232
PCDF-23478
0.69
0.557
0.251
HpCDF-1234678
0.604
0.668
0.119
HpCDD-1234678
0.283
0.927
0.174
HxCDD-123678
0.35
0.892
0.191
0.88
0.085
OCDD
0.246
HxCDD-123789
0.352
0.86
0.194
0.85
0.187
HxCDD-123478
0.37
PCDD-12378
0.367
0.786
0.262
0.847
TCDD-2378
0.284
0.367
a
Relatively large loadings (>0.7) are highlighted
When these results were displayed graphically, the clustering of the various
groups of samples could be readily seen (Figure 2). Samples that were collected
from the RMA were generally separated from the reference samples along the xaxis (Factor 1). Whereas, the reference sites generally are spread along the yaxis (Factor 2). From the component loadings (Table 1), Factor 1 represents
Organohalogen Compounds, Volumes 60-65, Dioxin 2003 Boston, MA
higher chlorinated PCDF congeners (found more predominantly On-site), while
Factor 2 represents the higher chlorinated PCDDs.
Conclusions
Many soil samples from the RMA have a PCDD/F signature profile containing
higher chlorinated PCDF congeners. In contrast, Off-Post samples are more
dominated by the higher chlorinated PCDDs typical of general human activities,
such as traffic emissions and light industrial activities. The presence of a unique
PCDD/F signature in samples from the RMA does not necessarily indicate a
major on-site source of these compounds. Indeed, the relatively diffuse nature of
the sample clustering here would argue against the presence of a single large
source. The predominance of the PCDD/F congeners in the On-Post samples
from the central region of the RMA reflects the historical mixed industrial
activities carried out on the site, while many of the ‘peripheral’ site-wide RMA
samples were indistinguishable from the Off-Post reference congener profiles.
This PCA analysis of trace-level concentrations of dioxins in surface soil
samples was a valuable and sensitive tool that contributed useful information for
the environmental risk assessment. The PCA results helped USEPA risk
assessors and managers to better distinguish the potential source contributions of
the site vs. ambient conditions. This led to improved science-based decisions for
remedial action based risk reduction.
References
Wilkinson, L., G. Blank & C. Gruber. 1996. Desktop Data Analysis with
SYSTAT. Prentice Hall, NJ, 798 pp.
Organohalogen Compounds, Volumes 60-65, Dioxin 2003 Boston, MA
3
In d u s t r ia l
L it e
In d u s t r ia l
L it e
FACTOR(2)
2
1
0
-1
-2
-3
-2
5
LLititee
In d u s t r ia l
In d u s t r ia l
L it e
In d u s t r ia l H IS T
L it e
In d u s t r ia
l ral
In d u sRtur ia
In d u s t r ia l
L Litl it
e
In ed u s t r ia l
L it e
In d u s t r ia l
In d u s t r ia l
L it eL it e L it e
R u r aL litLeLit it
ee
L it e
In
d
u
s
t
r
ia
l
L
it
e
L it eL it e L it e R u In
r adl u s t r iaLl it e
L it e
L it e In d uInsdt ruiaslt r ia l
L it e
R u ra l
L dit e
itLel it e
uu rsatlLr ia
LIn
edeu sIn
t rR
itiaLelit e R u r a l
Lit it
L ituer L
L itIne d uR
In d u s t r ia l
it edTLuP
itset r ia l
a
lrlia l LW
In
s
t
r
ia
In
d
u
s
t
L it e
R u ra l
R uuL
rerLaLit
l el eeR u r a l
eRLMit
A aitit
SW
In
t rL
iaitrlaR
L itd eu sR
u
l
WRTT
uPP
rW
a l T P L it e R u r a l
L itLe itIn
LRit
dR
l IS T
RMA SW
L In
itRLeuduitrue
R
aslitlt ee
l H
ueuW
rusartlar ia
RrLa
u r raia
l lu W
R u ral
RPuMr aAl S W
L it eR uL ritael L R
e r a lTR
L it e
R M AL itSit
W
L
it
e
H IS T
W
T
u rIS
a lT
R u r a l RRMueIn
duPlS
asP
lPt r ia l L it eR H
A
W
rrTaaW
R
uR
l urT
LW
IS
rRaL
lMitTe
RR
uRrL
lit e
MauitlAreaRS
W R uH
s t r ia l
In
uR
LR
itdeu
WInATdPSuLW
rsatulr riaR
alluu rraall
it
e
r aal l S W R u r a lR M A S W
R u r a l RRuu
Ruur R
r aluR
lRruaurl raaRl l u r a l
L it e R uMrraA
l
R
a
R M A SHRW
R Lu itr ae l R M
R u r a l L it e R u
M
IS
SW
it eTA S W
RA
MA
S LW
R uSr W
al ral
R
M
A
R MHHAIS
In d uRsut R
rria
aull r a lR u r a l
R u ral
ISSTTW
RMA R
S uWr a l R u r a l
RM A SW
A SW
R Ru M
raA
l RSMW
RRuur raal l
RRMMAA SSWW
R u ra l
RMA SW
L it e
RM A SW
In dL uit e
s t r ia lR u r a l
H IS T
H IS T
W TP
H IS T
H IS T
H IS T
H IS T
RM A SW
H IS T
H IS T
R M AHSISWT
H IS T
H
H IS
T T
H IS
IS T
RMA SW
RMA SW
H IS T
R u ra l
R u ra l R M A S W
-1
0
1
2
3
F A C T O R (1 )
4
5
L it e
L it e
4
FACTOR(3)
3
2
1
0
-1
L it e
L it e
L it e
L it e
L it e
RMA SW
L it e
In d uInsdt ruiaslt r ia l
H IS T
RMA SW
L it e
R u ra l
it ed u s t r ia l
L it e In d u s t r ia lLIn
In d u R
s tur ia
r all
L it e
L it es t r ia l
I nLditIn
ue sdturRia
ul r a l In d u s t r ia l
R uInr ad lu s t r ia l
L it e
itWe L it e
R uLr it
a el R M A LSIn
R u ral
u sTtdrPia
L it e L itdeW
u sl t r iaul r a l
L it
H
PitIn
RLeit
u er aRl u
e T LR
it e
Inrda ul W
s tT
rLia
lIS
R
l l r aTl P
LRitudeRR
ruausulrtraraia
W
PR
l lR
RruauR
aluR
itlrTreaua
lR
rslit
aruLu
lritaueW
In
d ur aIn
sLl t it
r ia
l L
itruead
In
u
t
r
ia
l
L
e
R
a
l
e
L
Ru
H
IS
R
u
l
L
it
e
W
P uR
u lr a l TH IS TR u r a l L itR
InrLadit
lLl itTeR
l u sRtureia
ra
LR
it euR
e u ral
W
P
r ra
R
uareelraal l
ud ruaeslLtitr ia
l uuitrurra
R
uT
L
itit
R
u
L
it
e
R u rR
aInlM
RLR
a
l
e
L
In
d
u
R
a
r
a
l
l
A
S
W
R
u
l
R
M
A
S
W
R
u
r
a
l
dM
uAa
sltP
r ia
RL
Wlll
L itH
eL IS
W
P
it e Ts t r ia l H IS T
LR
it eu r a lL itR
uW
uIn
rIn
lT
RR
usuSrrtra
a
it
R u r aRlIn
derR
uTu
rRaia
l l
u
aaM
uRitsS
reM
lia
RReeS
uR
rW
MdRAuR
R
A
SerW
tarW
rLa
LruaA
itulrLelLa
erS
utu
l eW
uR
a l u r a Ll W
La
it
M
S
aAitlllRlR
leW
itA
H IS T
In d LuLR
se
r ria
itR
it
it e L it e
e
In
dLR
ia
lMMAA SSW
SM
W
R
RitR
u rMa LlAit
R
u sr at rl R
LR
it
eu
M
SS W
M
ue
l MAA
M
Ait
W
A
SAW
it eu r a lR u
rR
au
L it eL R
R
SW
W
LrRa
itSM
ea
lrR
LlR
l d u s t r ia lR M A S W In d uRsMt rA
L it e R uLritaIn
WSW
ia M
lS A
R
In d Lueitset r ia l
it M
e A SW
InLdR
RuMs tAr iaSl W
L it e
R u ral
In d u s t r ia l
H IS T
H IS T
H IS T
HH
ISIS
TT
H IS T
HTIS
IS H
H ISH
TT IS T
H IS T
H IS T
H IS T
R M A SRW
MA SW
RM A SW
W TP
In d u s t r ia l
-2
-3
-2
H IS T
H IS T
L it e
-1
0
RMA SW
1
2
3
F A C T O R (1 )
5
4
5
L it e
L it e
4
L it e
L it e
FACTOR(3)
3
2
L it e
L it e
L it e
RMA SW
H IS TL it e
In d u s t r ia l
In d u s t r ia l
R M LAitS
eW
In d u s t r ia l
L itLeit e
In dIn
ud
s tur siat lr ia l
L it e
L uit e
d
s
t
r
ia
l
In d u In
s td
r ia
ls t r iaHl IS T R In
In
d
u s t r ia l
u
u ra l
R u ral
L it e
L it e
RMA R
S uWr a l
H IS T
L
it
e
L
it
e
RLuitrIn
ea ld u s t r ia l
L it e
In d u s t r ia l
L it e
R TuT rPa l W T P W T P
W
H IS LTit e
H IS
L it e
d u ls t r iaHl IS T
RIn
el
LuRitrTua
eP
it
eP u r a
W
Lr aitle
RlLLuititrea
In d u s t r ia R
l u r a l In d u s t r iaRl u
R ruaW
rlLa T
lR
L it e R u r a l
RRu ur arHa
l lIS T
R uTrR
L lit e
RauluLrritaaell In d u s t r ia
Riau
rPraal l
H IS
u
R ruarl aHl IS
RRIn
u
Tuitu
drreau
tW
ul a
L it e R u r a l
alalsR
L
LW
itR
e
LruR
itrTera
TruParla l
H
T SR
R
r
l
l
R IS
MA
W RRuuHr a
R
u
l
R
u
L
it
e
IS
T
r
a
l
In
d
u
s
t
r
ia
l
L it e
r In
a l d u sR
Lr u
iturelraal l
u rlaR
lRuur a
R Lu itr ae l
tR
ia
LllR
itueurS
R M A S W HRIS
WL
TiteePW
a alW
LInit d
e u s t rLiaitle
LLit
it
RrW
aR
R
A
TS
lu
uL
aelrM
itu
IS
e T PR u r a lIn d u s t r ia l
L it e
RMMAAR
RHrR
uT
r it
aRrTelaul r a l
LW
d u s t rit
iael
R uRr aMlHAISSH
itRerIn
RT uT r aSl W
R M AH H
SRIS
W
Te
RL u
ua lr a l LL it
Tu rraall
LIS
In d u s t r ia l
Leit e InL ditHue
R uSrW
aIS
lRM
L it e
sitLteit
r ia l
R M A IS
SW
RHMISA TSRWMRA
R
uW
rS
a lW
u r a l RRA
R u ral
Lu
itSeA
RMA S
RRMuR
ArituaS
R u ral
RW
u r aRl M A RSMW
L Litite e
RMA SW
elr aWl
RM A SW
LLitLit
In d u sIn
t r ia
Leeit e
RMA SW
dr ia
ul sl t r ia l
RMA SW
RM A SW
In
d
u
s
t
L
it
e
L it e
In d u s t r ia l
RMA SW
RMA SW
R u r a l L it e
RRMMA ASSWW
W TP
In d u s t r ia l
H IS T
R u ra l
1
0
-1
H IS T
R u ra l
L it e
In d u s t r ia l
-2
H IS T
RM A SW
-3
-3
-2
-1
0
1
F A C T O R (2 )
L it e
2
3
Figure 2 Overall PCA comparing 17 PCDD/F congeners in
60 On-Post and 160 Off-Post soil samples
Organohalogen Compounds, Volumes 60-65, Dioxin 2003 Boston, MA