A
4 mm h
2
\ PR0^ [ELDS
Lange and Revere Street Canals
Sediment Sampling Report
Ten-Mile Drain Site
St. Clair Shores, MI
Prepared by:
FIELDS Program, US EPA, Region V
Linda Jacobson, Brennan Pierce, John Canar, and Chuck Roth
June 2012
1
-------
TABLE OF CONTENTS
List of Acronyms and Abbreviations 3
I. Introduction 4
II. Sampling Methods 4
A. Ponar - Shallow Sediment 4
B. Core - Below the Surface Sediment 5
III. Analysis Methods 5
A. Sediment Thickness and Water Depth 5
B. Core Depth Intervals and Sediment Type 6
C. PCB Results 6
D. Interpolations and Mass/Volume Estimates 6
E. Comparisons between Depth Intervals 7
F. Statistical Evaluation of PCB Results and Sediment Clay Content 7
G. Comparison between Lange and Revere Street Canals 7
IV. Results 7
A. Sediment Thickness and Water Depth 7
B. Core Depth Intervals and Sediment Type 7
C. PCB Results 8
D. Interpolations and Mass/Volume Estimates 8
E. Statistical Evaluation of PCB Results and Sediment Clay Content 9
F. Comparison between Lange and Revere Street Canals 9
V. Discussion 9
2
-------
List of Acronyms and Abbreviations
ANOVA Analysis of Variance
FAST FIELDS' Analysis and Sampling Tool
FIELDS Field Environmental Decision Support
FIELDS Tools Field Environmental Decision Support Software Tools for ArcGIS
GPS Global Positioning System
PCB Polychlorinated biphenyl
3
-------
I. Introduction
The FIELDS Group conducted sediment sampling in the Lange and Revere Street canals from
August 23 to September 1, 2011 as part of the Ten-Mile Drain, Superfund site, characterization
activities. The sampling was performed to assess the extent of PCB contamination in the
sediment of the Lange and Revere Street canals. This report summarizes the sampling and
analysis methods, and results of the sampling. Statistical methods used in this report are
described in detail in Appendix A and Appendix B.
The Ten-Mile Drain Site is located northeast of the City of Detroit on the western shores of Lake
St. Clair in St. Clair Shores, Macomb County, Michigan (see Figure 1). The site is located in a
mixed commercial/residential area near the intersection of Bon Brae Street and Harper Avenue.
It includes a portion of the Ten-Mile Drain storm sewer system, which consists of concrete sewer
pipes and soil surrounding the pipes in a utility corridor extending to approximately 15 feet
below ground surface. The site covers several blocks where polychlorinated biphenyls (PCBs)
have been found in the storm sewer system at levels as high as 200,000 milligrams per kilogram
(parts per million). The PCBs migrate into the storm sewer, which discharges into the Lange and
Revere Street canals which are connected to Lake St. Clair. The canals, which provide
recreational boating access to Lake St. Clair for approximately 125 homes, are private property
and are used for recreational boating, swimming, and fishing.
II. Sampling Methods
A. Ponar - Shallow Sediment
The first phase of sampling was the collection of surface sediment samples using a petite ponar
in the Lange and Revere Street canals. The locations of the Lange and Revere Street canals are
shown in Figure 2. A sample design of 100 sampling locations within the canals was created
using the FIELDS Tools software in ArcGIS (see Figure 3). The sample design was a systematic
aligned triangular grid based on a random start location with a distance of 46 feet between each
proposed location. During the sampling event, each sample was taken as close as possible to the
proposed sample locations. Some of the proposed locations were obstructed by boats or were
very close to the seawall. In those cases, the sample was taken as close as possible to the
proposed location without causing damage to other boats. The actual sample locations are also
depicted on Figure 3.
In addition to the 100 proposed sample locations, five additional surface sediment samples were
collected during the sample event. Two samples were located in the northwestern corner of the
Lange Street canal near the storm water drain outlet labeled P300 and P301 (see Figure 3).
Three sediment samples labeled P200, P201, and P202 were collected from the Lakecrest and
Rio Vista Street canals since the former Martin Drain used to discharge into these canals (see
Figure 4).
At each sample location the ponar was used to collect surficial sediment. The sediment from the
ponar was placed in a metal bowl, excess water was decanted, and the sample was mixed
thoroughly and placed in a Ziploc® bag. Each bag was labeled with the sample identification
4
-------
number (e.g., P001), date and time, and photographed. A probe marked at half foot intervals was
driven into the sediment at each location until "refusal" (a hard bottom was reached or the pipe
could not be pushed any farther). The water depth and total depth to refusal were recorded in
FAST, a data management software program developed by the FIELDS Group and used to
collect real time data continuously in the field. An estimate of sediment thickness was later
calculated by subtracting water depth from total depth to refusal. As each sample was taken, a
Trimble ProXR GPS unit connected to a laptop running the FAST software recorded the
geographic location. The sample identification number and description of each sediment sample
was entered into the FAST software program at each sample location.
B. Core - Below the Surface Sediment
The second phase of the sediment sampling consisted of collecting sediment cores below the
surface sediment in the Lange and Revere Street canals. The sediment core sampling was done
in two stages with 29 cores collected in the first stage and 12 collected in the second stage.
In stage one, a sample design was created based on the sediment thicknesses obtained during the
ponar shallow sampling event. Polygons were created in ArcGIS around areas having a
sediment thickness of two feet or more. The FIELDS Tools software in ArcGIS was then used to
create a systematic aligned triangular grid sampling design within each polygon (see Figure 5).
If a proposed sampling location was obstructed by a boat or the seawall, the actual sediment core
samples were taken as close as possible to the proposed locations.
Stage two core sample locations were determined in the field; therefore, a proposed sample
design was not created beforehand. Stage two consisted of collecting additional sediment cores
at locations to spatially fill in areas where cores had not been taken during stage one and in areas
with elevated surface PCB results from the shallow sediment sampling event. These locations
are also shown in Figure 5.
The sediment cores were collected with Lexan® tubes attached to a vibracore device. Each
Lexan® tube was cut at intervals of 0 to 6 inches, 6 to 12 inches, 12 to 24 inches, and subsequent
one-foot intervals to the bottom of the tube (five feet total length). The sediment sample for each
interval was placed in a metal bowl, mixed thoroughly, and placed in a Ziploc® bag. Each bag
was labeled with the sample identification number and depth interval (e.g., COO 10 6), date and
time, and photographed. The geographic location for each core was recorded with a Trimble
ProXR GPS unit connected to a laptop running the FAST software. The sample identification
number and a description of the sediment for each sample were also entered into FAST at each
sample location.
III. Analysis Methods
A. Sediment Thickness and Water Depth
Sediment thickness and water depth were estimated in the Lange and Revere Street canals using
both the inverse distance weighting and natural neighbor interpolation methods. The
interpolations were performed on depth measurements taken with each ponar sample. The
5
-------
inverse distance weighting takes into account the distance between points and assumes that the
influence of each point diminishes with increasing distance. With the natural neighbor
interpolation, nearby points are weighted based on a shared geometric boundary between points.
Since the interpolation results were very similar, the more conservative inverse distance
weighting interpolation results are shown in this document.
B. Core Depth Intervals and Sediment Type
A distribution and a map of core depth intervals were created to show how many cores were
sampled at each depth interval. The sediment type throughout the canals was displayed on cross
sections. Sediment type (based on visual observations) was classified as sand, silt, clay, silt and
clay in a mixture, and sand and silt or sand, silt, and clay in a mixture.
C. PCB Results
The sediment samples were analyzed onsite for PCBs by the US EPA Mobile Laboratory (ML)
on a wet weight basis in order to allow for a faster analytical turn-around. Additionally, each of
these samples was submitted to the Region 5 Central Regional Laboratory (CRL) for percent
solids measurement. The percent solids values were used to adjust the wet weight PCB values to
dry weight. The dry weight PCBs values are used for all maps and analyses in this document.
Ten percent of the sediment samples were also analyzed for PCBs on a dry weight basis by CRL
as a quality control on the ML results. Both laboratories analyzed for aroclors 1016, 1232, 1242,
1248, 1254, and 1260. The PCB results from the ML and CRL were compared using simple
linear regression. (See Appendix B for a more complete discussion.)
The maps and analyses in this report depict the PCB concentrations as the sum of all detected
aroclors in a sample from the ML results.
D. Interpolations and Mass/Volume Estimates
Inverse distance weighting and natural neighbor interpolations of PCB concentrations were
created for the 0 to 6 inch depth interval using ponar and core results combined. Inverse distance
weighting and natural neighbor interpolations of PCB concentrations were also created based on
core results for the depth intervals 6 to 12, 12 to 24, 24 to 36, 36 to 48, and 48 to 60 inches.
Since the inverse distance weighting and natural neighbor interpolations were similar, the inverse
distance weighting interpolations are displayed in the maps in this document to show the most
conservative results.
The mass of PCBs and volume of sediment containing PCBs were estimated from both the
inverse distance weighting and natural neighbor interpolations at each depth interval using the
FIELDS Tools for ArcGIS. For samples without any detected aroclor results, half the reporting
limit of one of the aroclors (all aroclors within a sample had the same reporting limit) was used
as the PCB concentration for the sample. A density value of 2500 lb/yd was used in estimating
the mass.
6
-------
E. Comparisons between Depth Intervals
PCB concentrations were compared by depth interval using a one-way ANOVA statistical test.
(See Appendix A for a more complete discussion.)
F. Statistical Evaluation of PCB Results and Sediment Clay Content
The one-way ANOVA test was used to determine if there was a difference between PCB
concentrations and the presence or absence of clay in the sediment. Sediment was classified as
either containing all clay, majority clay, any clay, or no clay. (See Appendix A for a more
complete discussion.)
G. Comparison between Lange and Revere Street Canals
PCB concentrations in the Lange and Revere Street canals were compared using a one-way
ANOVA test. (See Appendix A for a more complete discussion.)
IV. Results
A. Sediment Thickness and Water Depth
The inverse distance weighting interpolation resulted in a range of 0.3 to 9.5 feet of sediment
thickness (the measured sediment thicknesses ranged from 0.25 to 9.5 feet) with the thickest
sediment located in the central and eastern portions of both the Lange and Revere Street canals
(see Figure 6). The natural neighbor interpolation gave a similar distribution of sediment
thickness.
The inverse distance weighting interpolation showed a water depth range of 4.3 to 12.2 feet (the
measured water depth ranged from 4.25 to 12.25 feet) with the greatest water depths located
mostly in the Lange Street canal (see Figure 7). The natural neighbor interpolation gave a
similar spatial pattern of water depths.
B. Core Depth Intervals and Sediment Type
All 41 cores had sediment in the 0 to 6, 6 to 12, and 12 to 24 inch intervals, 24 cores had
sediment in the 24 to 36 inch interval, 15 cores had sediment in the 36 to 48 inch intervals, and 6
cores had sediment in the 48 to 60 inch interval with the deepest 3 cores reaching to 56 inches
(see Figure 8). The cores with the deepest sediment were located predominantly in the central
part of the Lange and Revere Street canals. The sample core depths corresponded with the
sediment thickness interpolations with deeper cores taken in areas with the larger sediment
thickness and shorter cores taken in areas with smaller sediment thickness.
Sediment types in the Lange and Revere Street canals are shown in Figures 9, 10, and 11. Silt
was the most common sediment type at the sediment surface. Below the silt, there was
predominantly a mixture of silt and clay followed by a layer of clay. In the eastern end of the
Lange and Revere Street canals, a mixture of sand and silt and sand, silt, and clay was present.
7
-------
In the Lange Street canal in the eastern end next to Lake St. Clair, the sediment was entirely
sand.
C. PCB Results
The confirmation analysis comparing the ML and CRL total PCB values showed the two
laboratories gave slightly different results. (See Appendix A for a more complete discussion of
the statistical methods used and the results found.) Simple linear regression was used to quantify
the relationship between the ML and CRL total PCB values. Using the best linear fit equation
for the relationship between ML and CRL total PCB values, a ML total PCB value of 2 ppm is
equivalent to an adjusted total PCB value of 3.44 ppm. For 20 ppm, the adjusted value would be
35.7 ppm, and for 100 ppm, the adjusted value would be 183.2 ppm. (See Appendix B for a
more complete discussion of the findings from linear regression.)
Of the 270 samples taken in the Lange and Revere Street canals (including all ponar and core
samples) and analyzed by the ML, aroclor 1016 was detected in 221 samples, aroclor 1248 was
detected in 5 samples, and aroclor 1260 was detected in 25 samples. Aroclors 1232, 1242, and
1254 were not detected in any of the samples (see Figure 12). Since the majority of the aroclors
detected were aroclor 1016 and the highest aroclor concentrations were aroclor 1016, maps
depicting PCB concentrations are presented as total PCBs rather than separated by aroclor.
Ponar and core PCB results representing the top 6 inches of sediment are depicted on Figure 13.
The highest PCB concentrations were located in the western portion of the Lange and Revere
Street canals. The PCBs were not detected in the three additional ponars located in the Lakecrest
and Rio Vista Street canals (see Figure 14).
The PCB concentrations for the cores from 6 to 12, 12 to 24, 24 to 36, 36 to 48, and 48 to 60
inches are shown in Figures 15 to 19. The PCB concentrations decreased with depth in most of
the cores.
Figure 20 shows the ponar and core PCB results in the dredged and undredged areas of the
Lange and Revere Street canals. (Dredging was performed in 2003.) The one-way ANOVA
statistical method found a significantly greater PCB concentration inside than outside the
previously dredged area indicating the PCBs have been re-deposited into the Lange and Revere
Street canals since the dredging occurred. (See Appendix A for a more complete discussion of
the statistical methods used and the results found.)
D. Interpolations and Mass/Volume Estimates
Within each depth interval, the inverse distance weighting and natural neighbor interpolations
gave very similar spatial distributions of PCBs throughout the canal. Figures 21 through 26
show the inverse distance weighting interpolation results. Overall, the concentration of PCBs
was highest in the western portion of the canal near the storm water drain outlet for each depth
interval.
8
-------
PCB mass was calculated for each depth interval using the inverse distance weighted and natural
-3
neighbor interpolations. A density value of 2500 lb/yd was used in estimating the mass. The
mass of PCBs was greatest in the top 6 inches of sediment (see Table 1). Using the inverse
distance weighting interpolation, the mass of PCBs in the top 6 inches of sediment was 62% of
the total PCB mass in the canal sediments. In the 6 to 12 inches depth interval, the mass of PCBs
was 29% of the total. Each of the depth intervals between 12 to 60 inches contained 7% or less
of the total PCB mass.
E. Statistical Evaluation of PCB Results and Sediment Clay Content
The one-way ANOVA statistical method showed there was a significantly lower PCB
concentration in sediment containing clay (any clay, majority clay, or all clay) than sediment
containing no clay. (See Appendix A for a more complete discussion of the statistical methods
used and the results found.)
F. Comparison between Lange and Revere Street Canals
The one-way ANOVA statistical method showed the Lange Street canal sediment had a
significantly greater PCB concentration than the Revere Street canal sediment. (See Appendix A
for a more complete discussion of the statistical methods used and the results found.)
V. Discussion
Based on the findings of the 2011 Lange and Revere Street canals sediment sampling event, the
highest PCB concentrations are located near the Ten-Mile Drain outfall. Overall, PCB
concentrations decrease with depth and distance from the outfall. PCB concentrations are
significantly lower in the deeper sediment, usually clay, than the surficial sediment, usually silt.
The highest PCB concentrations on the western ends of the canals during the 2011 sediment
sampling event likely indicates that PCBs have continued to discharge out of the Ten-Mile Drain
outfall into the canals.
9
-------
N
UTM Zone 17N, NAD 83
Map Author: Linda Jacobson
300,000 600,000 900,000 1,200,000
Feet
[fields
Figure 1: Location of St. Clair Shores in Macomb County, Michigan
10
-------
Lange Street Canal
Revere Street Canal
UTM Zone 17N, NAD 83
Aerial Source: Bing Maps
Map Author: Linda Jacobson
1.050
1.400
Feet
FIELDS
Figure 2: Location of the Lange and Revere Street canals and surrounding streets
11
-------
P300
P094
. P051¦P026
P066
P083
P079
P025
P081
P080
P036
P015
P003
Legend
Proposed Ponar Locations
Actual Ponar Locations
P100
P099
Figure 3: Proposed and actual ponar sample locations
12
-------
Lakecrest Street Canal
Rio Vista Street Canal
% mX*
FIELDS
1,425
N UTM Zone 17N, NAD 83
Data Source: USEPA
Aerial Source: Bing Maps
Map Author: Linda Jacobson
1,900 A
ZD Feet
Figure 4: Additional ponar sample locations
13
-------
C038
C015
C014
C024
Legend
Stage 1 Proposed Core Sample Locations
Stage 1 Actual Core Locations
o Stage 2 Actual Core Locations
Polygons of Areas with Sediment Thickness of 2 ft or More
N UTM Zone 17N, NAD 83
150 300 450 600 A Data Source USEPA
wj, Aerial Source: Bing Maps
Map Author: Linda Jacobson
FIELDS
Figure 5: Proposed and actual core sample locations
14
-------
Legend
o Ponar Sample Locations
Sediment Thickness (Feet)
0.3-0.5
(ft)
N UTM Zone 17N, NAD 83
150 300 450 600 A Data Source USEPA
Feet Aenal Source: Bing Maps
Map Author: Linda Jacobson
<3®
FIELDS
Figure 6: Inverse distance weighting interpolation of sediment thickness from measurement taken at ponar locations
15
-------
as?*
{¦#¦1
Legend
o Ponar Sample Locations
Water Depth (Feet)
4.3 - 5.0
5.0-7.0
7.0 - 9.0
9.0-11.0
11.0-12.2
q
Feet
N UTM Zone 17N, NAD 83
Data Source: USEPA
Aerial Source: Blng Maps
Map Author: Linda Jacobson
V fields
Figure 7: Inverse distance weighting interpolation of water depth from measurement taken at ponar locations
16
-------
Legend
Core Depth (inches)
O 0 to 6
N UTM Zone 17N, NAD 83
08z) ° ™™ £5—s?»_A ^Keld-
% PBo-rt" ^^^ ^ Map Author: Linda Jacobson \J ^ l^LDb
Figure 8: Core sample depth intervals
17
-------
Lange Street Canal
| C037 |
coiri\Roon
C015
C010
C016
345200
CO
-J—-'
CP
0.0 J.
-0.5
u
-1-0 J?
-1.5 3
—2.0 ^
-2.5 i?
-3.0 #
-3.5 $
-4-0-g
x:
-4—'
a.
o>
o
345000 345100
Easting (meters)
East End
34530C
¦ Sand
¦ Sand and Silt/Sand, Silt, and Clay
¦ Silt
J Silt and Clay
J Clay
¦ Water
Ul Core Sample Locations
Figure 9: Cross section of sediment type from core samples in the Lange Street canal
18
-------
Revere Street Canal
C028
C020
C029
C021
C007
East End
^ -0 5
—
20
=- -2.5
I- -3.0 §
r- -3
=~ —4 0 -r
345300
West End
30 345000 345100
Easting (meters)
345200
¦ Sand
¦ Sand and Silt/Sand, Silt, and Clay
¦ Silt
¦ Silt and Clay
J Clay
¦ Water
I Core Sample Locations
Figure 10: Cross section of sediment type from core samples in the Revere Street canal
19
-------
C032
C034
C004
North End
E ^°-°
|~ -0.5
E- -1.0
|—1.5
§- -2.0
=- -2.5
-3.0
-.3.5
"4.0 |
CD
-------
1016 1260 1248 1232 1242 1254
Aroclor
Figure 12: Frequency of detected and not detected aroclors for all samples
21
-------
IHKer, , .*
2/% fi
T^LtW
a
75
Legend
Ponar
0 to 6 inches
detected
UTM Zone 17N, NAD 83
Data Source: USEPA
Aerial Source: Bing Maps
Map Author: Linda Jacobson
FIELDS
Figure 13: Ponar and core PCB results for depth of 0 to 6 inches
22
-------
Lange and Revere Street Canals
i
i
¦/
Lakecrest and Rio Vista Street Canals
r^L " —1 ™
"T
n
Legend
Ponar PCBs (ppm)
^0 to 6 inches
°
Not detected
°
0.45 -1
°
1-10
•
10-50
°
50-100
| •
100-570 L
1,425 1,900
i Feet
UTM Zone 17N, NAD 83
Data Source: USEPA
Aerial Source: Bing Maps
Map Author: Linda Jacobson
FIELDS
Figure 14: Ponar PCB results in Lakecrest and Rio Vista Street canals for depth of 0 to 6 inches
23
-------
Legend
Core PCBs (ppm)
6 to 12 inches
O Not detected
N UTM Zone 17N, NAD 83
i JHL \ n icn ,nn c„„ A Data Source: USEPA
\ 30^ t A Aerial Source: Bing Maps
% pbot11 Map Author: Linda Jacobson
' -Tfi
FIELDS
Figure 15: Core PCB results for depth of 6 to 12 inches
24
-------
Legend
Core PCBs (ppm)
12 to 24 inches
O Not detected
N UTM Zone 17N, NAD 83
{A) ° —300 —
^^Map Author: Linda Jacobson VjiUhLDb
Figure 16: Core PCB results for depth of 12 to 24 inches
25
-------
Legend
Core PCBs (ppm)
24 to 36 inches
O Not detected
N UTM Zone 17N, NAD 83
lift) ° 300 ™L.k A,°?soo:SeBSp, ^Seld"
% ^^^ Map Author: Linda Jacobson \J * l^LDb
Figure 17: Core PCB results for depth of 24 to 36 inches
26
-------
Legend
Core PCBs (ppm)
36 to 48 inches
O Not detected
(A)
«( PROt*-
600
] Feet,
UTM Zone 17N, NAD 83
Data Source: USEPA
Aerial Source: Bing Maps no
Map Author: Linda Jacobson \j ^ liiLdJo
Figure 18: Core PCB results for depth of 36 to 48 inches
27
-------
HP*^
W* C i
;x&tn£u
Legend
Core PCBs (ppm)
L'
48 to 60 inches
O Not detected
0-1
1 -10
10-15
J*"**
UTM Zone 17N, NAD 83
Data Source: USEPA
Aerial Source: Bing Maps
Map Author: Linda Jacobson
&
FIELDS
Figure 19: Core PCB results for depth of 48 to 60 inches
28
-------
I
Legend
Ponar and Core PCBs (ppm)
0 to 6 inches
O Not detected
Previously Dredged Area
OS
FIELDS
Figure 20: Ponar and core PCB results for depth of 0 to 6 inches and previously dredged area
29
-------
N UTM Zone 17N, NAD 83
,,n „nn .,n Rnn A Data Source: USEPA
150 300 450 60° A Aerial Source: Bing Maps
Map Author: Linda Jacobson
"Tfields
Legend
Ponar and Core PCBs (ppm) 0 to 6 inches
j Inverse Distance Weighting Interpolation
| o-1
B1 -10
10-50
[ 50 -100
¦ 100-570
Mass of PCBs is approximately 327 pounds.
Figure 21: Inverse distance weighting interpolation of ponar and core PCBs for depth of 0 to 6 inches
30
-------
&&*
Legend
Core PCBs (ppm) 6 to 12 inches
Inverse Distance Weighting Interpolation
Mass of PCBs is approximately 155 pounds
N UTM Zone 17N, NAD 83
i JBL,\ n . cn „„„ cri. A Data Source: USERA )
IK; 0_^50 300_^50 600 A Aeria| Source; Bjng Mgps ,
% ^^^ Map Author: Linda Jacobson \J rihLDb
Figure 22: Inverse distance weighting interpolation of core PCBs for depth of 6 to 12 inches
31
-------
3g
jLegend
Core PCBs (ppm) 12 to 24 inches
| Inverse Distance Weighting Interpolation
0-1
1-10
| 10-34
Mass of PCBs is approximately 37 pounds.
im
WJ
N UTM Zone 17N, NAD 83
150 300 450 600 A Data Source: USEPA
P , Mk Aenal Source: Bing Maps
Map Author Linda Jacobson
' 'Tn
FIELDS
Figure 23: Inverse distance weighting interpolation of core PCBs for depth of 12 to 24 inches
32
-------
N UTM Zone 17N, NAD 83
150 300 450 600 A Data Source: USEPA
P . mk Aenal Source: Bing Maps
Map Author: Linda Jacobson
FIELDS
Legend
Core PCBs (ppm) 24 to 36 inches
|Inverse Distance Weighting Interpolation!
| 0-1
I 1 -10
1 10-16
Mass of PCBs is approximately 11 pounds.
Figure 24: Inverse distance weighting interpolation of core PCBs for depth of 24 to 36 inches
33
-------
Mass of PCBs is approximately 1.08 pounds
N UTM Zone 17N, NAD 83
150 300 450 600 A Data Source: USEPA
P , Mk Aenal Source: Bing Maps
Map Author Linda Jacobson
FIELDS
Legend
Core PCBs (ppm) 36 to 48 inches
| Inverse Distance Weighting Interpolation
| 0-1
I 1 -10
10-13
Figure 25: Inverse distance weighting interpolation of core PCBs for depth of 36 to 48 inches
34
-------
USE
N UTM Zone 17N, NAD 83
150 300 450 600 A Data Source: USEPA
Feet JL Aenal Source: Bing Maps
Map Author Linda Jacobson
FIELDS
Legend
Core PCBs (ppm) 48 to 60 inches
|Inverse Distance Weighting Interpolation!
| 0-1
I 1 -10
jl I 10-15
Mass of PCBs is approximately 0.34 pounds.
Figure 26: Inverse distance weighting interpolation of core PCBs for depth of 48 to 60 inches
35
-------
Depth Interval
Sample
Volume
(yd3)
Mass
(lb)
Percent of Total
Mass of PCBs
0 to 6 inches
Ponars and Cores
3,855
327
62%
6 to 12 inches
Cores Only
3,653
155
29%
12 to 24 inches
Cores Only
4,369
37
7%
24 to 36 inches
Cores Only
1,930
11
2%
36 to 48 inches
Cores Only
963
1.08
0.2%
48 to 60 inches
Cores Only
396
0.34
0.06%
Totals
15,166
531
Table 1: Mass and volume estimates and percent of total mass of PCBs in each depth interval
36
-------
A
4 mm h
2
\ PR0^ [ELDS
Lange and Revere Street Canals
Appendix A of
Sediment Sampling Report
Ten-Mile Drain Site
St. Clair Shores, MI
Prepared by:
FIELDS Program, US EPA, Region V
Linda Jacobson, Brennan Pierce, John Canar, and Chuck Roth
June 2012
1
-------
TABLE OF CONTENTS
I. Introduction 3
II. Analysis Methods 3
A. Paired Comparisons of PCBs 3
B. Comparison between Dredged and Non-Dredged Areas 3
C. Comparisons between Depth Intervals 3
D. Statistical Evaluation of PCB Results and Sediment Clay Content 3
E. Cumulative Distribution Functions 4
F. Comparison between Lange and Revere Street Canals 4
III. Results and Discussion 4
A. Paired Comparisons of PCBs 4
B. Comparison between Dredged and Non-Dredged Areas 4
C. Comparisons between Depth Intervals 4
D. Statistical Evaluation of PCB Results and Sediment Clay Content 5
E. Cumulative Distribution Functions 5
F. Comparison between Lange and Revere Street Canals 5
IV. References 6
2
-------
I. Introduction
The FIELDS Group conducted sediment sampling in the Lange and Revere Street canals in St.
Clair Shores, Michigan from August 23 through September 1, 2011. The sampling was
performed to assess the extent of PCB contamination in the sediment of the Lange and Revere
Street canals. This appendix provides a more complete discussion of the statistical methods
employed and results found that were reported in "Lange and Revere Street Canals Sediment
Sampling Report".
II. Analysis Methods
A. Paired Comparisons of PCBs
The total PCBs measured in samples by the Mobile Laboratory (ML) and Central Regional
Laboratory (CRL) were compared using the Wilcoxon signed rank sum test, which is a non-
parametric test that compares the difference between two paired variables. The difference
between the CRL and ML in total PCBs for each sample was calculated, and the Wilcoxon
signed rank sum test was used because the differences were not normally distributed.
B. Comparison between Dredged and Non-Dredged Areas
One-way ANOVA was used on both the ponar and core rank-transformed PCB data to determine
a difference in PCB concentrations from 0 to 6 inches inside and outside the previously dredged
area of the Lange and Revere Street canals. For the statistical analysis, the ponar and core PCB
data were given ranked values since the untransformed and natural log transformed data did not
meet the assumption of normality. Ranks from 1 to 143 were assigned to each sample by sorting
the PCB values from low to high with the lowest PCB value receiving a 1 and the highest 143.
Welch's ANOVA was used since the assumption of homogeneity of variances was violated (a p-
value of less than 0.05 for Levene's test). The general linear model (GLM) procedure in the
statistical software SAS was used.
C. Comparisons between Depth Intervals
PCB concentrations were compared by depth interval using one-way ANOVA on both the ponar
and core rank-transformed PCB values, using the general linear model (GLM) procedure in SAS.
The Least Squares Means Tukey-Kramer Multiple Comparisons test was used to determine
differences in mean ranked PCB concentration between depth intervals. The Least Squares
Means Tukey-Kramer Multiple Comparisons test was selected because it accommodates unequal
sample sizes and is the most robust test for pairwise comparisons (SAS, 2011). The ANOVA
and multiple comparisons test were performed on ranked PCB data since the untransformed and
natural log transformed PCB data were not normally distributed.
D. Statistical Evaluation of PCB Results and Sediment Clay Content
One-way ANOVA was used on both the ponar and core rank-transformed PCB data to determine
a difference between mean PCB concentrations and the presence or absence of clay in the
3
-------
sediment. Sediment was classified as either containing all clay, majority clay, any clay, or no
clay. Welch's ANOVA was used when the assumption of homogeneity of variances was
violated (a p-value of less than 0.05 for Levene's test). The ANOVA test was performed on
ranked PCB data since the untransformed and natural log transformed PCB data were not
normally distributed. The general linear model (GLM) procedure in SAS was used.
E. Cumulative Distribution Functions
Cumulative distribution functions were created for each depth interval and all depths to show the
percent of ponar and core samples that had a PCB concentration above 50 ppm, the limit set by
the Toxic Substance Control Act (TSCA) for the maximum PCB concentration of sediment.
F. Comparison between Lange and Revere Street Canals
Ponar and core mean rank-transformed PCB concentrations in the Lange and Revere Street
canals were compared using one-way ANOVA, and Welch's ANOVA was used since the
assumption of homogeneity of variances was violated (a p-value of less than 0.05 for Levene's
test). The ANOVA test was performed on ranked PCB data since the untransformed and natural
log transformed PCB data were not normally distributed. The general linear model (GLM)
procedure in SAS was used.
III. Results and Discussion
A. Paired Comparisons of PCBs
Boxplots and descriptive statistics of total PCB concentrations in the ML and CRL are shown in
Figures A-l and A-2. The Wilcoxon signed rank sum test indicates that there was a significant
difference in total PCB concentrations between the ML and CRL (see the p-value for the Signed
Rank Test, "S", in Figure A-3).
B. Comparison between Dredged and Non-Dredged Areas
Boxplots and descriptive statistics of PCB concentrations from 0 to 6 inches inside and outside
the previously dredged area are shown in Figures A-4 and A-5. Welch's ANOVA indicates that
there was a significant difference in ponar and core mean rank-transformed PCB concentrations
from 0 to 6 inches inside and outside the dredged area with the dredged area having a
significantly greater mean PCB concentration (see Figure A-6).
C. Comparisons between Depth Intervals
Boxplots and descriptive statistics of PCB concentrations in each depth interval are shown in
Figures A-7 and A-8. The one-way ANOVA comparing mean rank-transformed PCB
concentrations between depth intervals indicates that there was a significant difference in mean
PCB concentration between at least two depth intervals (see the F-value for the Type III SS, the
unbalanced case, in Figure A-9). According to the Least Squares Means Tukey-Kramer test of
differences (see Figure A-10), the mean PCB concentration in depth intervals 12 to 24, 24 to 36,
4
-------
36 to 48, and 48 to 60 were not significantly different from each other. The mean PCB
concentration in depth intervals 0 to 6 and 6 to 12 and 6 to 12, 12 to 24, and 48 to 60 also were
not significantly different from each other. Depth interval 0 to 6 had a significantly greater mean
PCB concentration than depth intervals 12 to 24, 24 to 36, 36 to 48, and 48 to 60. Depth interval
6 to 12 also had a significantly greater mean PCB concentration than depth interval 24 to 36 and
36 to 48.
D. Statistical Evaluation of PCB Results and Sediment Clay Content
Boxplots and descriptive statistics of PCB concentrations for the presence and absence of clay in
sediment classified as containing either all clay, majority clay, any clay, or no clay are shown in
Figures A-ll, A-12, A-14, A-15, A-17, and A-18. Welch's ANOVA indicates that there was a
significant difference in mean rank-transformed PCB concentration between sediment containing
all clay and no clay with sediment containing no clay having a significantly greater mean PCB
concentration than sediment containing all clay (see Figure A-13).
The one-way ANOVA shows there was a significant difference in mean rank-transformed PCB
concentrations between sediment containing majority clay and no clay with sediment containing
no clay having a significantly greater mean PCB concentration than sediment containing
majority clay (see Figure A-16).
The one-way ANOVA also shows there was a significant difference in mean rank-transformed
PCB concentrations between sediment containing any clay and no clay with sediment containing
no clay having a significantly greater mean PCB concentration than sediment containing any
clay (see Figure A-19).
E. Cumulative Distribution Functions
Cumulative distribution functions for all depths and depth intervals 0 to 6, 6 to 12, 12 to 24, 24 to
36, 36 to 48, and 48 to 60 inches are shown in Figures A-20 through A-26, respectively. Of all
the ponar and core samples taken at all depths, 10% were above 50 ppm. In the 0 to 6 inch depth
interval 16.08% were above 50 ppm, and in the 6 to 12 inch depth interval 9.16% were above 50
ppm. In the 12 to 24, 24 to 36, 36 to 48, and 48 to 60 inches depth intervals, no samples were
above 50 ppm.
F. Comparison between Lange and Revere Street Canals
Boxplots and descriptive statistics of PCB concentrations in the Lange and Revere Street canals
are shown in Figures A-27 and A-28. Welch's ANOVA indicates that there was a significant
difference in mean rank-transformed PCB concentrations between the Lange and Revere Street
canals with the Lange Street canal sediment having a significantly greater mean PCB
concentration than the Revere Street canal sediment (Figure A-29).
5
-------
IV. References
SAS Institute Inc., SAS/STAT® User's Guide. Version 9.2. Cary, NC: SAS Institute Inc., 2011.
(The GLM Procedure, Multiple Comparisons)
6
-------
Boxplots of Total PCBs (ppm) by Lab
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
600 -
400 "
1
•
s
ffi
£
£
200 ~
o -
CRL ML
Lab
Boxplots of Tolal PCBs Ranted by Lab
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
60 "
50 "
40 "
TJ
m 30 -
2
20 "
10 -
CRL ML
Lab
Figure A-l: Boxplots of total PCB results from the ML and CRL (untransformed PCB data top
graph, rank-transformed PCB data bottom graph; red diamonds are extreme values, blue circle is
the mean, and middle horizontal line is the median)
7
-------
Au.-.]y'i- Y.iu.U'lt-: Toc.tL PC ?._ppm Totr. L F'C B- -ppm
Lab
till!
01;-
NT
Mcin
Mc'l:r.u
Std Dev
Mmimiun
Maximum
Eli use
(I oeft ot'
V)ii> riott
CRL
27
27
4" -3
11.44
0.05
593.10
593.05
251.25
ML
2?
11
28.38
5.80
77.46
0.16
360.00
359.85
272.94
Figure A-2: Descriptive statistics of untransformed total PCB results from the ML and CRL
Figure A-3: Wilcoxon signed rank sum test on differences between the CRL and ML total PCB
results
8
-------
Boxplots of Ponar and Core PCBs from 0 to 6 Inches Inside and Outside Previously Dredged Area
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
600
400
E
&
ID
o
0.
200
No
Yfes
In Dredged Area
Boxplols of Ponar and Core PCBs from 0 to 6 Inches Inside and Outside Previously Dredged Area
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
150
100
i
n
oc
50
No
Yes
in Dredged Area
Figure A-4: Boxplots of ponar and core PCB results from 0 to 6 inches inside and outside the
previously dredged area (untransformed PCB data top graph, rank-transformed PCB data bottom
graph; red diamonds are extreme values, blue circle is the mean, and middle horizontal line is the
median)
9
-------
Ami; :i^ Vsinhlf FCB-:j)|)in FCB- -ppim
In
E'l f-'lscil
Ai e i
N
O'.i:
x
Mem
Mi limn in
Std Dev
Cocff Of
V.iyi.iTiou
No
IS
89
6.50
410
0.11
49.00
7.49
ir'i
r—1
Yes
54
54
r 2s
29,50
0.15
570.00
117,46
134.58
Figure A-5: Descriptive statistics of ponar and core untransformed PCB results inside and
outside the previously dredged area
V. elch : ANOVA f*i P F
I]j_Died;rd_Aif-i
1MQQ
E11
Figure A-6: Welch's ANOVA test of significant differences of ponar and core mean rank-
transformed PCB concentrations inside and outside the previously dredged area
10
-------
Boxplots of Ponar and Core PCBs by Depth Interval
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
600 -
•
400 "
•
f
a
,
»
m
o
~.
1
200 ~
o -
S
s
^ „
0-6
06-12 12-24 24-36 36-48
Depth Interval
48-60
Boxplots of Ponar and Core PCBs by Depth Interval
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
300 "
230 "
200 "
a
DC 150 -
100 "
50 "
—i 1 1 1
0-6 06-12 12-24 24-36
Depth Interval
36-48
48-60
Figure A-7: Boxplots of ponar and core PCB results by depth interval (untransformed PCB data
top graph, rank-transformed PCB data bottom graph; red diamonds are extreme values, blue
circle is the mean, and middle horizontal line is the median)
11
-------
An.ily-i? Variable : PCB;_ppm PCB- {ppwii
De-pfli
Iiu-:n.
ms
Ob;
V
Meati
Medina
Minimum
M.l'ilLf.UIlT
S rrl Dc'i
C ortt o:
Vrur.Uvlj
0-6
143
143
37.00
6.5C
ft. 11
:
82.0?
221.®
OS-12
41
41
22.00
4.20
0.1?
360 CO
v:"\4~
2 "0,33
41
41
5.24
1.6C
?::
34 CO
8.3-
If 3.55
24-36
24
24
2.98
1.51
C 12
16 00
~.OS
13\?8
3 5-48
15
15
1.47
o.:i
C 14
13 C>0
3.33
227.33
48-60
6
6
2.67
0.21
0.18
15.00
6.04
226.53
Figure A-8: Descriptive statistics of ponar and core untransformed PCB results by depth interval
Soul or
DF
Tvpc III IS
Me.iu Square
FViIuf
Fl -I
Dcjj:h_IiLrrn i]
3?"2Si.:S"*
_
14
..1
Figure A-9: One-way ANOVA test of significant differences of ponar and core mean rank-
transformed PCB concentrations between depth intervals
The GLM Procedure
Least Sq meres Means
Adjustment for Multiple Comparisons; Tuktp-Kramer
Beprh Inrei vn!
PCB' ranked
L«IEA>"
1. ;\l; V\
N n mb c r
0-6
154.J2
Oti-12
12-24
132 v::
;
85.395133
4
36-48
S3.1S666?
5
4S-60
59.08J333
§
Lfr-iTt Sqn ,t{ fov HO: LSMenn(U=LSMMn(j>
Dependent Variable: FCBi_i:mke
-------
Boxplots of Ponar and Core PCBs by Clay Presence
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clair Shores, Michigan
600
400
E
&
ID
o
0.
200
Absent
Present
All Clay
Boxplots of Ponar and Core PCBs by Clay Presence
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clar Shores, Michigan
E 150
Absent
Present
All Clay
Figure A-l 1: Boxplots of ponar and core PCB results by presence and absence of clay for
sediment containing all clay (untransformed PCB data top graph, rank-transformed PCB data
bottom graph; red diamonds are extreme values, blue circle is the mean, and middle horizontal
line is the median)
13
-------
V.iii.iblf- : Pi!3:_ppm PC3- ¦
Qny
N
Ob 5
_\"
Mean
Mlhiuulil:
Mmmiun
Sttl Dr'-
i! Orff of
Absent
221
221
28.90
530
0.11
570.00
"1 "2
14S r
Present
49
49
2.66
1.3-0
0.15
15.00
3..7S
141.95
Figure A-12: Descriptive statistics of ponar and core untransformed PCB results by presence
and absence of clay for sediment containing all clay
Welch'- ANDY A :oi
PC
DF
F V,: hie
Pr > F
L.-.v
1.0000
4LS2
<.0001
E l l
19.2021
Figure A-13: Welch's ANOVA test of significant differences of ponar and core mean rank-
transformed PCB concentrations between presence and absence of clay for sediment containing
all clay
14
-------
Boxplots of Ponar and Core PCBs by Clay Presence
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clar Shores, Michigan
600
400
E
&
m
o
a.
200
Majority Clay
Boxplots of Ponar and Core PCBs by Clay Presence
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clar Shores, Michigan
300
250
200
T3
I
(0
1 150
£
100
50
Absent
Majority Clay
Figure A-14: Boxplots of ponar and core PCB results by presence and absence of clay for
sediment containing majority clay (untransformed PCB data top graph, rank-transformed PCB
data bottom graph; red diamonds are extreme values, blue circle is the mean, and middle
horizontal line is the median)
15
-------
Ail,Y.iri.iWe : P
-------
Boxplots of Ponar and Core PCBs by Clay Presence
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clar Shores, Michigan
600
400
E
&
m
o
a.
200
Any Clay
Boxplots of Ponar and Core PCBs by Clay Presence
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clar Shores, Michigan
300
250
200
T3
I
(0
1 150
£
100
50
Absent
Any Clay
Figure A-17: Boxplots of ponar and core PCB results by presence and absence of clay for
sediment containing any clay (untransformed PCB data top graph, rank-transformed PCB data
bottom graph; red diamonds are extreme values, blue circle is the mean, and middle horizontal
line is the median)
17
-------
Aur.ly'r Y."ii],:l."'le-: Pi!3:_ppm Pi! 3: 'ppm:-
¦Clay
X
Ob-.
x
Mf Til
Mel:,-.it
M:]];iniur,
Muimiun
Std Dtv
(I t'rff of
V.irintiou
Absent
1-2
142
35.54
r
570.00
s: 02
224 iS
Present
128
121
10 38
2.05
0.12
3-60.00
36.08
347.51
Figure A-18: Descriptive statistics of ponar and core untransformed PCB results by presence
and absence of clay for sediment containing any clay
^OLUCr
I)F
Tvpe III
Mciui
FV^
pi =-r
Chy
1
1 j JtSj1 r .60 ji
i 5 j»i j 1 .iilj-Sii
£{. iv
--.wUl
Figure A-19: One-way ANOVA test of significant differences of ponar and core mean rank-
transformed PCB concentrations between presence and absence of clay for sediment containing
any clay
18
-------
Cumulative Distribution Function
Ponar and Core PCBs All Depths
Sediment Samping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
100
I
a
t
v
e
P
e
r
c
e
n
t
60
40
20
Summary Statetbs
N
270.0
Mean
24.14
Median
3,95
Std Deviation
65,67
1st Feroeritile
0.12
5th Percentile
0.17
10th Fteroemtile
0.20
90th Percentile
51.00
95th Percentile
131.3
99th Percentile
360.0
Upper Spes Limit
50.00
| Percent Ofcs > USL
10,00 |
200 300
PCB (ppm)
Specification: Upper= 50
-| i i i | i i i r
400 500 600
Figure A-20: Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for all depths
19
-------
Cumulative Distribution Function
Ponar and Core PCBs 0 to 6 Inches
Sediment Samping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
100
50 ppm
—r*
I
a
t
v
e
P
e
r
c
e
n
t
60
40
20
Summary Statetbs
N
143.0
Mean
37.00
Median
6,50
Std Deviation
82.03
1st Feroeritile
0.15
5th Percentile
0.45
10th Rsroemtile
1.00
90th Percentile
120.0
95th Percentile
180.0
99th Percentile
440.0
Upper Spes Limit
50.00
| Percent Ofcs > USL
1S.08 |
200 300
PCB (ppm)
Specification: Upper= 50
-| i i i | i i i r
400 500 600
Figure A-21:
inches
Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for depth interval 0 to 6
20
-------
Cumulative Distribution Function
Ponar and Core PCBs 6 to 12 Inches
Sediment Stamping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
I
a
t
v
e
P
e
r
c
e
n
t
100
80
60
40
50 ppm
100 200
PCB (ppm)
Specification: Upper= 50
Summary Statetbs
N
41.00
Mean
22.00
Median
420
Std Deviation
5947
1st Feroeritile
0.17
5th Percentile
0.22
10th Rsrcentile
0.32
90th Percentile
38.00
95th Percentile
£6.00
99th Percentile
360.0
Upper Spes Limit
50.00
Percent Ofcs > USL
9,76
300
400
Figure A-22:
inches
Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for depth interval 6 to 12
21
-------
Cumulative Distribution Function
Ponar and Core PCBs 12 to 24 Inches
Sediment Samping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
100
I
a
t
v
e
P
e
r
c
e
n
t
60
40
20
10
20 30
PCB (ppm)
Specification: Upper= 50
Summary Statetbs
N
41.00
Mean
5.24
Median
1.60
Std Deviation
8,04
1st Feroeritile
0.12
5th Percentile
0.16
10th Fteroemtile
0.17
90th Percentile
18,00
95th Percentile
21.00
99th Percentile
34,00
Upper Spes Limit
50.00
| Percent Ofcs > USL
0.00 |
40
50
Figure A-23:
inches
Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for depth interval 12 to 24
22
-------
Cumulative Distribution Function
Ponar and Core PCBs 24 to 36 Inches
Sediment Samping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
100
I
a
t
v
e
P
e
r
c
e
n
t
60
40
20
10
20 30
PCB (ppm)
Specification: Upper= 50
Summary Statetbs
N
24.00
Mean
2,98
Median
1.01
Std Deviation
4.06
1st Feroeritile
0.12
5th Percentile
0.15
10th Rsrcerntile
0.16
90th Percentile
7.70
95th Percentile
11.00
99th Percentile
16.00
Upper Spes Limit
50.00
Percent Ofcs > USL
0.00
50
Figure A-24: Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for depth interval 24 to 36
inches
23
-------
Cumulative Distribution Function
Ponar and Core PCBs 36 to 43 Inches
Sediment Samping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
I
a
t
v
e
P
e
r
c
e
n
t
100
80
40
20
10
20 30
PCB (ppm)
Specification: Upper= 50
Summary Statetbs
N
15.00
Mean
1,47
Median
0.21
Std Deviation
3.33
1st Feroeritile
0.14
5th Percentile
0.14
10th Rsrcentile
0.17
90th Percentile
3.10
95th Percentile
13.00
99th Percentile
13.00
Upper Spes Limit
50.00
Percent Ofcs > USL
0.00
50
Figure A-25:
inches
Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for depth interval 36 to 48
24
-------
Cumulative Distribution Function
Ponar and Core PCBs 43 to 60 Inches
Sediment Samping of Lange and Revere Street Canab Summer 20t1
St. Clair Shores, Mchigan
100
I
a
t
v
e
P
e
r
c
e
n
t
80
60
20
10
20 30
PCB (ppm)
Specification: Upper= 50
Summary Statetbs
N
6.00
Mean
2.67
Median
0.21
Std Deviation
6,04
1st Feroeritile
0.18
5th Percentile
0.18
10th Rsrcentile
0.18
90th Percentile
15,00
95th Percentile
15.00
99th Percentile
15.00
Upper Spes Limit
50.00
Percent Ofcs > USL
0.00
50
Figure A-26:
inches
Cumulative distribution function of ponar and core untransformed PCB values above 50 ppm for depth interval 48 to 60
25
-------
Boxplots of Ponar and Core PCBs by Canal
Sediment Sampling of Lange and Revere Street Canals Summer 2011
St. Clar Shores, Michigan
600
400
E
&
m
o
a.
200
Lange
Revere
Canal
Boxplols of Ponar and Core PCBs by Canal
Sediment Sampling of Lange and Revere Street Canals Summer 20TI
St. Clar Shores, Michigan
300 "
250 "
200 "
«
DC 150 -
100 "
50 "
Lange Revere
Canal
Figure A-27: Boxplots of ponar and core PCB results by canal (untransformed PCB data top
graph, rank-transformed PCB data bottom graph; red diamonds are extreme values, blue circle is
the mean, and middle horizontal line is the median)
26
-------
Ail.-]-.--it Ynti,-.Me- : PC3:_ppm PC3- 'ppin•
811111
Olr
V
Menu
Medirm
Minimum
Mnximum
Sr-1 Dev
octf o:
V.-1: i ci-iij
L-n-;
14-
14-
* C; ¦ C;
f.4C
J j .
-4: c j
-f
2."
'Revere
126
: :6
18.14
2.65
0.15
5To no
60.87
335.63
Figure A-28: Descriptive statistics of ponar and core untransformed PCB results by canal
Welch ¦ ANOVA fry
P<_ B i ulie-l
DF
F Value
Pj - F
;•. !'G
! y. -;
En-:.]
7'
Figure A-29: Welch's ANOVA test of significant differences of ponar and core mean rank-
transformed PCB concentrations between canals
27
-------
A
4 mm h
2
\ PR0^ [ELDS
Lange and Revere Street Canals
Appendix B of
Sediment Sampling Report
Ten-Mile Drain Site
St. Clair Shores, MI
Prepared by:
FIELDS Program, US EPA, Region V
Linda Jacobson, Brennan Pierce, John Canar, and Chuck Roth
June 2012
1
-------
TABLE OF CONTENTS
I. Introduction 3
II. Data and Data Handling 3
III. Results and Discussion 3
IV. References 4
2
-------
I. Introduction
Simple linear regression and regression diagnostics were used to find the "best fitting" linear
relationship between the total PCBs measured in samples by the Mobile Laboratory (ML) and
Central Regional Laboratory (CRL) using the SAS® software. This relationship is quantified
into a model (equation) of ML measurements of total PCBs and its corresponding CRL
measurement. The statistical methods employed were drawn from SAS® literature and three
regression texts: Statistical Methods in Water Resources. 1992; and Applied Regression Analysis
and Other Multivariate Methods. 1978 and 1988. (See "References" section for a complete list
of regression resources.)
The steps used to perform simple linear regression were:
1. Plot the data;
2. Compute the least squares regression statistics;
3. Examine adherence to the assumptions of regression using residual plots; and
4. Employ regression diagnostics (Helsel and Hirsch, 1992).
II. Data and Data Handling
All 270 sediment samples were analyzed onsite for PCBs by the ML. Ten percent of the
sediment samples were also analyzed for PCBs by CRL. A total of 27 sediment samples were
analyzed by both the ML and CRL and were used in the below regression. Both laboratories
analyzed for aroclors 1016, 1232, 1242, 1248, 1254, and 1260, and a total PCB value was
calculated for each sample. For samples without any detected aroclor results, half the reporting
limit of one of the aroclors (all aroclors within a sample had the same reporting limit) was used
as the PCB concentration for the sample.
III. Results and Discussion
There was a statistically significant linear regression relationship between ML total PCB values
and their corresponding CRL values (results not shown). However, regression diagnostics found
that some of the assumptions of regression were violated. These violations included the lack of
extreme residuals and normality for the residuals (see Figures B-l and B-2, respectively). (The
null hypothesis of each of these four tests in Figure B-2 is that the residuals are from a normal
distribution. If using an alpha value of 0.05, one would reject the null hypothesis.) To overcome
these violations, two observations with Studentized residual values greater than 2.5, a value used
as a rule of thumb for extreme values, were removed from the data set. The new data set was
regressed and the linear regression was significant (results not shown). Again, some of the
assumptions of regression were violated: the lack of extreme residuals and normality of residuals
(results not shown). One additional observation with extreme residuals was removed from the
data set and the new data set was regressed. The regression results were significant but the
assumptions of regression were violated (results not shown). Another two extreme residuals
were removed from the data set. Although the regression results for the new data set were
significant, the assumptions of regression were not met (results not shown). This process was
repeated four additional times removing one, two, and one observation with extreme residuals,
3
-------
respectively, until the model had no observations with extreme residuals. However, the
assumption of normality of the residuals was violated in the final model. Given the inability to
achieve these assumptions, the natural log of the ML total PCB values and their corresponding
CRL values was taken.
The natural log of the ML and CRL total PCB values showed a statistically significant linear
regression relationship (see Figure B-3). All assumptions of linear regression were also met in
this model. Figure B-4 shows that the residuals were homoscedastic and none of the Studentized
residuals were greater than 2.5. The White test also found that the variance of the residuals were
homogenous (results not shown). Figure B-5 shows that the residuals are normally distributed.
Normality of residuals is required in order to test the hypothesis that "the slope coefficient (Pi) is
significantly different from zero" (Helsel and Hirsch, 1992). In other words, in order to
demonstrate a linear relationship between the two variables, ML and CRL total PCB values, the
slope coefficient must be significant. A visualization of the linear relationship between the
natural log of ML total PCB and CRL values is shown in Figure B-6.
The parameters of the best linear fit equation for the relationship of natural log of ML total PCB
and CRL values are:
Adjusted LN total PCB = 0.53136 + (1.01606)*(LN ML total PCB value)
However, as this equation is in natural log space, the antilog of the adjusted total PCB value
must be taken. For example, for a ML total PCB value of 2 ppm (0.693 ppm in natural log
space), the Adjusted LN ML total PCB value is 1.24 ppm. The antilog of this value is 3.44 ppm.
Hence, a ML total PCB value of 2 ppm is equivalent to an adjusted ML total PCB value of 3.44
ppm. For 20 ppm, the adjusted value is 35.7 ppm; for 100 ppm, the adjusted value is 183.2 ppm.
IV. References
Chen, X., Ender, P., Mitchell, M. and Wells, C. (2003). Regression with SAS, from
http://www.ats.ucla.edu/stat/sas/webbooks/reg/default.htm
Helsel, D.R. and Hirsch R.M., Statistical Methods in Water Resources. Elsevier, Amsterdam,
1992.
Kleinbaum, D.G. and Kupper, L.L., Applied Regression Analysis and Other Multivariate
Methods. Duxbury Press, Boston, Massachusetts, 1978.
Kleinbaum, D.G., Kupper, L.L., and Muller, K.E., Applied Regression Analysis and Other
Multivariate Methods. Second Edition. PWS-Kent Publishing Company, Boston, Massachusetts,
1988.
SAS Institute Inc., SAS/STAT® User's Guide. Version 8. Cary, NC: SAS Institute Inc., 1999.
(Chapter 55, The REG Procedure)
4
-------
SAS Institute Inc., SAS ® System for Regression. Second Edition. Cary, NC: SAS Institute Inc.,
1991. 210pp.
5
-------
St. Clair Shores Lange/Revere Canal
5 e d i re n i Saipling Auq, 11 - Sept. 1h 2D11
RegreEE: ian of Nabile Lab "and CRL PCB Results (dry v e i ¦] h t j
No Trantfdrnat i on
DRL_L0O*thDd_;_ppm = 14.406 + 1.183E HL_LQOelhDd_Z_ppm
w
150 2 n-n- 250 m
Fred i cted Value
Figure B-l: Residual plot from the SAS software for total PCB ML and CRL values
6
-------
ierr foy XviimiHy
Te-r
Statistic
[) Yr. hit-
SLi.-pji -Vi'iLl:
W
j V: :¦¦¦:
Pi - V."
C .0001
Kvlmo- - Mini nov
D
Pi -D
' i j ]
<¦'. LTi]iicL-vo]j Mi'e--
V'-M]
¦ nf.g"0"
Pi -V.o.:j
¦' .00 vC
Aitdei'ou-D.u-m;
A-Sq
V ;4v::'-y
Pi -A-Sli
00 v 0
Figure B-2: Tests of Normality from the SAS software for residuals from total PCB ML and
CRL values
7
-------
Si, Clair Shores Lange Revere Canal
Sediment Sampling Aug. 22 - Sept. 2, 2011
Regression of Mobile Lab and CRL PCS Results {dry weight)
Xantral Log Transformation
The REG Procedure
Model; MODELI
Dependent Variable; CSL_LOD_Method_2_ppm_ln
Nuinbei of OWt v.irioir- Read
27
Numbci -'A Ot'ci'MTijU" V't-il
27
Ajiniy-]- of Variance
Soitice
DF
Sum of
Squares
Me.m
Square
F Value
Fj --F
Model
1
.;jrs-:
. -X- J
Error
:c.4~isf
C on eeted Total
Root MSE
0.85974
JtC-S-JU.il c
0.8465
Dependent Menn
2.05168
Aflj R-S.,
0.8404
C oeff Vai
41.90424
P.itnmetei Estimates
Variable
DF
P.ii ,i mete i'
E'fim.itc
Standard
En >jl-
r Vi'.iu-
P:- ¦ t
I]irf] ."rpr
¦
c.-"ii:-f.
;•
"i :
i .'.1 i -
ML_ LOE> ".il
1
1.01606
G CS652
11.74
<.0001
Figure B-3: Simple linear regression output from the SAS software for the natural log of total
PCB ML and CRL values
8
-------
St. Clair Shores Lange/Revere Canal
5 e d i re n i Saipling Auq, 11 - Sept. 1h 2D11
RegreEE: ian of Nabile Lab "and CRL PCB Results (dry v e i ¦] h t j
Natural Log TransformatIon
1 "
-1 "
OD.Nethcd.
,2_ppm_ 1 n =
0,5314 41.D1&1 ML_L0D_Method_^_ppn_ 1 n
+
+
+
4
+
+
+ +
+++
+
-H-
* + +
+
-
+
+ +
4
+
+
0 —
T
-1
"1
1 2 3 4
Predicted Value
N
17
Rag
[;¦. S46-5
Ad jRsq
D. B.404
RMS E
P.&597
Figure B-4: Residual plot from the SAS software for the natural log of total PCB ML and CRL
values
-------
Ji\t" for Not tnnlity
SrnrHric
|) V,;hie
Siispko-Wilk
v.-
Pi ¦ V.'
o'-Simi u-.-'.
D
j ^ 5 •" "*
Pi -D
..."
Cramer-von Mises
Pi -V. o-j
; ¦ j|;.
Audei'-ou-D.it lius
A--1
"4
Pi -A-Si'i
•. ii
Figure B-5: Test of Normality from the SAS software for residuals from the natural log of total
PCB ML and CRL values
10
-------
Regression of the Natural Log of ML arid CRL Total PUB V&lues
Regrassion a q u a 1 i d n (j-intercapt)
St. Clair Shares Lonq«/Ravera Canal
SHirenl Scrip ling Aug. 12 - Sept. 2, 2D11
LN of DRL Total PCE (ppm)
0-
1"
fi-
ll-
4-
3-
1 -
1 -
o-
-1 -
-2-
R-saunretM.64652&&221
Figure B-6: Best-fit linear regression line from the SAS software for the natural log of total PCB
ML and CRL values
O /
o
0 /
o
<^o />
<*>A °
<0
/o
------- |