Unrted States Office of Weter
Environmental Protection Regulations and Standards NoVSItlbef 1987
Afiencv C riteria and Standards Divitior. SCD# 14
Washington DC 20460
Water
SEPA
SEDIMENT QUALITY CRITERIA METHODOLOGY
VALIDATION: UNCERTAINTY ANALYSIS OF
SEDIMENT NORMALIZATION THEORY FOR
NONPOLAR ORGANIC CONTAMINANTS
-------
2-
SEDIMENT QUALITY CRITERIA METHODOLOGY VALIDATION:
UNCERTAINTY ANALYSIS OF SEDIMENT NORMALIZATION
THEORY FOR NONPOLAR ORGANIC CONTAMINANTS
Work Assignment 56, Task 3
June 1987
Prepared by:
S. Pavlou, R. Kadeg, A. Turner and M. Marchllk
Envlrosphere Company
10900 N.E. 8th Street
Bellevue, Washington 98004
for:
U.S. Environmental Protection Agency
Criteria and Standards Division
Washington, D.C.
Submitted by:
BATTELLE
Washington Environmental Program Office
Washington, D.C.
-------
ABSTRACT
The U.S. Environmental Protection Agency (EPA), under the Criteria and
Standards Division, is currently pursuing the development of numerical
sediment quality criteria. Initial activities supporting this effort involved
developing and evaluating sediment criteria methodologies. This study
continues earlier work in which equilibrium partitioning (EP) theory was used -
to estimate permissible sediment contamination concentrations (PCCs) for
nonpolar hydrophobic organic contaminants. The PCCs were computed from a
simple predictive equation which expresses the PCC value as the product of the
organic carbon-normalized sediment/water partition coefficient (KQC) and the
chronic water quality criterion (CWQC) for the compound in question. The
applicable chronic water quality criterion is established by the EPA. For the
development of sediment quality criteria, CWQC is applied to the interstitial
water. Because benthic organisms are exposed to sediment contaminants for
extended periods, chronic criteria are more appropriate than acute criteria
for protecting benthic biota from impacts induced by chemical-specific
exposure In their natural habitat.
In computing the PCCs, the following activities were performed: the
pertinent partitioning literature was updated; empirically based regression
equations which predict KQC values from octanol/water partition coefficients
(Kqw) were reexamined with refinement of the descriptive statistics; and the
environmental variables influencing partitioning were evaluated.
The current study is an extension of the previous investigation. The main
objective Is quantification of the uncertainty associated with and the
refinement of the initially predicted PCC values. This will allow the
uncertainty Inherent in any criterion value to be quantified, thereby
permitting better-Informed environmental and regulatory decisions. Two
approaches were used to Investigate uncertainties. The first approach
employed parametric statistics assuming a log normal distribution on measured
Kqc values. These KQC values (distributions) were then subjected to a
quantitative uncertainty analysis technique to define cumulative probability
distribution functions (probability of exceedance curves). From these curves
4468a
i
-------
one may determine the percentage of time (i.e., probability) that a given
Kqc value will be equaled or exceeded for a specific chemical. PCC values
are determined, based on EPA chronic water quality criteria, by multiplying
the chemical-specific K distribution function by the chronic criterion
value for that chemical, converting the curve to a PCC distribution. The
probability that any specific PCC value will be equaled or exceeded can then
be determined, allowing selection of PCC values at various levels of
protection. The second approach employed nonparametric statistics using Latin
Hypercube sampling and computer-intensive (bootstrap) method of analysis.
This second approach ultimately proved inconclusive due to the limited raw
K /K data pairs from which the data distributions were generated.
vv UW
The resulting probability distribution based on the first approach yields
the following PCC values for those compounds with CWQC (at the 95 percent
protection level, in micrograms contaminant per gram of sediment normalized to
organic carbon content (ug/goc)).
F1uoranthene
Chiordane
DDT
Dieldrin
Endri n
Heptachlor
310
0.06
0.06
0.002
0.002
0.006
In addition, PCC values based on acute-to-chronic toxicity ratio
(ACR) estimations were developed for 13 other compounds for which no
CUQC exist.
4468a
ii
-------
CONTENTS
Page
ABSTRACT . • i
1.0 INTRODUCTION
1.1 BACKGROUND AND RATIONALE
1.2 SUMMARY OF UNCERTAINTY ANALYSIS APPROACH 1-3
2.0 ANALYTICAL METHODS: DESCRIPTION/RATIONALE 2-1
2.1 GENERAL APPROACH 2-l
2.2 LATIN HYPERCUBE SAMPLING 2-2
2.3 CONFIDENCE INTERVALS ON PREDICTED VALUES OF KQC AND PCC . 2-4
3.0 AVAILABLE KqC-Kqw 0ATA
3.1 COMPOUNDS AND Koc-Kow DATA 3_!
3.2 KQC DISTRIBUTION
4,0 ?f,nE!;S2!:?!-r!£ uSi!S!fI^«RmssiBLE contamination concentrations
AND ASSOCIATED UNCERTAINTY 4-1
4.1 CWQC-VALUES 4.!
4.2 DISTRIBUTION OF PCC VALUES AND ASSOCIATED UNCERTAINTY . . 4-3
5.0 CONCLUSIONS
6.0 REFERENCES
APPENDIX A: ALTERNATIVE NONPARAMETRIC APPROACH TO DETERMINE
UNCERTAINTIES A-1
APPENDIX B: ANALYSIS OF AVAILABLE K0C-KQW DATA B-l
APPENDIX C: CUMULATIVE PROBABILITY DISTRIBUTION PLOTS C-l
4468a
111
-------
FIGURES
1.1 Analytical Sequence I-5
'2.1 Illustration of LHS. Sampling . 2-3
TABLES
3.1 Representative Hydrophobic Organic Chemical Compounds .... 3-2
3.2 Summary of Kow and Koc Data for Selected Hydrophobic Organic
Chemical Compounds 3-3
3.3 Log K„ Distribution 3-4
oc
4.1 Matrix of Available Water Quality Criteria Data 4-2
4.2 Specific Chronic Water Quality Criteria and Acute to Chronic
Criteria Ratios for Selected Compounds 4-4
4.3 Distribution of Permissible Sediment Contaminant Concentrations
Ug/9oc) Derived from Existing or Estimated Chronic Water
Quality Criteria 4-6
4.4 PCC Values for Tested Fractiles 4-7
5.1 PCC Values from Equilibrium Partitioning Methodology and Other
Preliminary Criteria Values 5-3
4463a
iv
-------
1.0 INTRODUCTION
The U.S. Environmental Protection Agency (EPA), under the Criteria and
Standards Division, is currently pursuing the development of numerical
sediment quality criteria. Initial activities supporting this effort involved
developing and evaluating sediment criteria methodologies. This study
continues earlier work in which equilibrium partitioning (EP) theory was used
to estimate permissible sediment contamination concentrations (PCCs) for
ncnpolar hydrophobic organic contaminants. The PCCs were computed from a
simple predictive equation which expresses the PCC value as the product of the
organic carbon-normalized sediment/water partition coefficient and the
-chronic water quality criterion (CWQC) for the compound in question. The
applicable chronic water quality criterion is established by the EPA. For the
development of sediment quality criteria, CWQC is applied to the interstitial
water. This application is appropriate because the benthic organisms are
subjected to a long-term exposure of sediment contaminants and therefore the
chronic values are more appropriate for providing criteria that can protect
benthic biota from any impacts Induced by chemical-specific exposure in their
natural habitat.
A brief background and rationale for the uncertainty analysis of the
sediment normalization equilibrium partitioning theory, together with a
summary of the uncertainty analysis approach, is presented in the following
section.
1.1 BACKGROUND AND RATIONALE
Preliminary estimates of permissible sediment contamination concentrations
(PCCs) based on equilibrium partitioning theory have been reported in earlier
Investigations (Kadeg et al. 1986; Pavlou 1984; Pavlou and Weston 1984) as
part of an overall effort by the U.S. Environmental Protection Agency to
evaluate and develop sediment criteria methodologies for the development of
sediment quality criteria. These studies investigated PPC values based on
both acute and chronic water quality criteria. The assumption of equilibrium
and the exposure of benthic organisms to the sediment dictates that PCC values
4468a
1-1
-------
based on CWQC are more appropriate than values based on acute criteria.
Therefore, no further investigation of acute-based PCC values has been
performed.
Work by Kadeg et al-. (1985) confirmed that percent organic content of the
sediments is the proper normalizing parameter for reporting equilibrium
partition coefficients (K) for nonpolar organic contaminants. In addition,
the study concluded that other parameters (salinity, temperature, dissolved
organic carbon, sediment particle size, and suspended particulate matter) have
a varying but limited influence on partitioning. There are insufficient data
to quantify the effects of these factors. For the organic carbon-normalized
sediment/water partition coefficient (K ) and the octanol/water partition
coefficient (Kow)» evidence was also provided that the use of chemical
class-specific K„/K regression equations improved the estimation of
QC UW
partition coefficients for chemicals which did not have measured Kqc
values. The recomputed PCC values were higher (less stringent) than those
previously calculated by Pavlou (1984).
The results of this investigation pointed to the Inherent limitations oT
using empirical regression equations to estimate PCCs and raised the question
of how reliable these numbers are as "first-cut" sediment criteria for
nonpolar hydrophobic chemicals without laboratory and field verification. To
address the question of reliability and usefulness of these numbers, it was
necessary to perform an uncertainty analysis on the mathematical expressions
used to derive the PCCs. This was done by refining the values as appropriate
and establishing a generic statistical methodology that could test the
reliability of any criterion value determined by other methods.
This report discusses the methodology for the quantitative treatment of
the data; the results of applying the methodology to the PCCs computed by the
equilibrium partitioning approach; recommendations for laboratory and field
studies to be performed to reduce the uncertainty inherent in the predictive
equations; and appendices describing additional statistical investigations.
4468a
1-2
-------
1.2 SUMMARY OF UNCERTAINTY ANALYSIS APPROACH
To quantify the uncertainty in the calculated PCCs for sediment, the
sources of uncertainty must first be identified. PCC values of chemicals for
which measured KQC values exist are calculated from
PCC = $ x CWQC (1.1)
oc
A
where = true K + bias + error
oc oc
CWQC a EPA-specified chronic water quality criterion.
Because K__ values are measured and therefore subject to error, any
OC
observation is composed of the true (actual) KQC value plus-or-nrlnus
measurement error. Because the available data reflect observations from
OC
numerous laboratories and, commonly, multiple observations from any single
laboratory, the error within any reported K • value is composed of the
OC
potential bias of the laboratory reporting the value and the within-1aboratory
variability (errors) of these measurements.
For chemicals for which no measured values exist but for which
oc
measured KQW values are found, the PCC value can be expressed as
PCC = £ x CWQC (1.2)
oc
where *oc " * B0 * B1 *ow * error (l,3)
nqw - true KQW ~ bias ~ error.
For any chemical without a measured KQC, the value can be estimated from
the linear regression model for that compound's chemical class. The model
describes the empirical relationship between measured K^. and KQW values
as observed from chemicals with both It and K data in chemically
ow oc
similar (in terms of structure and behavior) classes. Bg and are the
model constants determined for each of the chemical classes. The estimated
KqC carries error due to the imperfect prediction of from KQw values
for that chemical class plus the measurement error of K, which in turn
ow
contains both bias and precision errors (as KQC measurements, above).
4468a
1-3
-------
Two observations on Equations (1.1) and (1.2) are pertinent to the methods
applied in the current analysis. First, comparison of the equations indicates
that errors/uncertainty accrued to PCC values for chemicals requiring
prediction of Kqc wiTl be greater than the errors accrued to PCC values for
chemicals whose K values have been measured. This observation is true
oc
unless the measurement error on K is much smaller than measurement error
ow
on and/or K J functions as a perfect predictor of K„. Hence, the
oc ow oc
assumption that Equation (1.1) will yield a smaller error/uncertainty is
intuitively reasonable because confidence in a measured value is generally
greater than in a predicted value. Second, in both equations, the estimated
PCC value is found by multiplying the observed KQC or predicted KQC by the
CWQC for that compound. It is recognized that CWQCs, which are developed from
toxicological data with some known or unknown variance, carry error
themselves. This study focuses on the variances of Kqw and KQC as they
affect predicted PCC values. Therefore, it has initially been assumed that
the error associated with CWQC values is zero. The actual nonzero error
associated with CUQC values may require further evaluation in the future.
Figure 1.1 schematically describes the analytical sequence followed in the
current study. The first phase of the study involved extensive exploratory
analyses and resulted in distributions of KQW and KQC for each chemical
with any reported measurements. The representativeness of these distributions
was examined within the context of the extremely limited raw data and the
comparative data indicated by the dashed feed-back line from the paired
Koc-Kow box t0 the raw Koc"Kow data box'
Following the exploratory analyses, two approaches were used. The first
approach applied straightforward parametric statistics to the measured K
oc
distributions for those compounds for which measured K values exist.
Previous work (Kadeg et al. 1986) has verified the log normal distribution of
these data and appropriateness of using this approach. A decision-model!ing
system employing Latin Hypercube sampling was then applied to the KQC
distribution Equation (1.1) to define cumulative probability distribution
function (CDF) curves for the corresponding PCC values. The 0.01 and 0.05
probability fractiles, corresponding to the 99 and 95 percentile protection
levels, were then calculated, yielding PCC values for these levels of
protection. This procedure is discussed in Chapter 2.
4468a
1-4
-------
FIGURE 1.1 Analytical Sequence
1-5
-------
The second approach was to apply nonparametric statistics to those
conpounds without measured KQC values. The data were analyzed to determine
their unique distributions, from which data sets were sampled. These data
were used to develop multiple linear regression equations with associated
confidence intervals (CI) for compounds with no Kqc values. The Kqc
values were then used in equation (1.2) with the CI to define uncertainties on
PCC values. This work is detailed in Appendix A.
The remainder of this report discusses the analytical methods and
interprets the results. Chapter 2 describes the basic analytical/statistical
¦ethods and confidence interval calculation for both measured KQ<; and
regression-predicted KQC values. The actual data analysis is presented in
Chapter 3 together with results from exploratory analyses and the K0C-KQw
distributions. Chapter 4 describes the development of PCC values and
associated uncertainty. Chapter 5 presents overall conclusions and
interpretation of the study results.
4468a
1-6
-------
2.0 ANALYTICAL METHODS: DESCRIPTION/RATIONALE
This chapter provides the description and rationale for the analytical
methods employed to define the uncertainty associated with the PCC values
derived from equilibrium partitioning theory.
2.1 GENERAL APPROACH
As noted in the introduction, initial exploratory analyses were performed
on the Kqc - KQW data to evaluate chemical specific distributions.
Details and results of these analyses are presented in Appendix B. Previous
work (Kadeg et al. 1986) on these data sets verified that the log-transformed
data for the group of chemicals were normally distributed. This was
established using Kolomogorov-Smirnov (K-S) goodness-of-fit tests. The
exploratory analyses indicated that chemical-specific data were normally or
nearly normally distributed for larger data sets, but not necessarily nonaal
for compounds with limited data. The trend was clear, however, that as the
number of samples Increased, the distributions tended more toward a normal or
near-normal curve. Therefore, standard parametric statistics were applied to
the chemical-specif1c measured K data to determine the median value (u),
wV
regardless of .the number of observations, and the standard deviation (j),
defined as the median value times the maximum coefficient of variation (Cv),
assuming normal distribution. This subsequently defined the characteristics/
range of KQC values to be employed in Equation (1.1). Latin Hypercube
Sampling (LHS) as described in Section 2.2 was then employed to define the
uncertainty of the PCC values associated with the corresponding measured K
values and a fixed CWQC value. Specific fractiles (0.01, 0.05) corresponding
to the 99 percentile and 95 percentile of the cumulative distribution
function, respectively, are the final PCC values.
For compounds for which EPA has not specified a CWQC value, a value was
estimated from EPA acute water quality criteria (AWQC) using acute-to-chronic
ratios (ACR) described by Kenaga (1982). If a compound-specific ACR value was
not available, a range of CWQC values was determined using a range of ACR
values from Kenaga (1982). In this case, the distribution of the range was
considered to be uniform (equally probable), and the corresponding uncertainty
4627a
2-1
-------
(i.e., CWQC uncertainty) was incorporated, using the LHS technique, into the
cumulative probability distributions. The fractiles and recommended PCC
values were then determined as described in the preceding paragraph.
To allow direct comparison of intermediate results (e.g., PCC
distributions) without introducing CWQC uncertainty in only select cases, PCC
values were also calculated assuming a fixed ACR value of 10 (order of
magnitude) for all compounds. These results, however, were not incorporated
into the final PCC values.
An alternative, nonparametric statistical approach was also investigated
as presented in Appendix A. This approach is summarized as fallows. The
distributions developed during exploratory analyses (Appendix B) were treated
as the raw data, although the relationship between the distributions for these
small data sets and the distribution of the true populations is not known.
Samples from these data were then used for all subsequent analyses:
*ow~*oc Pa^rs ^or chemicals (having both measures) within each of
three chemical classes (low-weight PAH, high-weight PAH and pesticides) were
randomly selected by a Latin Hypercube sampling method. Regression analysis,
using ordinary least squares, was performed on 100 samples for each chemical
class. Resulting linear equations [of the form: KQC = bq + Bi Kow^
were applied to randomly sampled Kqw values for compounds within that class
for which no K „ measurements are available to predict the K values
oc oc
needed to develop PCC values for those compounds. For each regression and
KQw value, the predicted KQC value and the values defining the limits of
the 95 percent confidence Interval were multiplied by the CWQC for that
compound to obtain a median value and confidence intervals for the PCC value.
2.2 LATIN HYPERCUBE SAMPLING
Latin Hypercube Sampling (LHS) is a constrained random-sampling technique
which is used to transform distribution functions (e.g., normal or uniform) to
cumulative probability functions or to select actual values for input
variables. As shown in Figure 2.1, this technique generates a sample from a
given distribution by first dividing the total distribution into a number of
equal segments, and then sampling within each segment. This forces sampling
of the tails of the distribution, resulting in a small sample being more
4627a
2-2
-------
INTERVAL ENOTOINTS USED WITH A LHS OF SIZE S (TOP) AND
SPECIFIC VALUES OF X SELECTED THROUGH THE INVERSE OF THE
DISTRIBUTION FUNCTION (BOTTOM)
FIGURE 2.1 Illustration of LHS Sampling
2-3
-------
representative of the distribution than unconstrained Monte Carlo sampling
which may require several thousand samples to adequately represent the data,
depending on the variance. Detailed description of the technique is found in
McKay, Conover, and Beckman (1979).
2.3 CONFIDENCE INTERVALS ON PREDICTED VALUES 0F.Kgc and PCC
For compounds with a measured K value, the only source of uncertainty
oc
in Equation (1.1) is the measured K . The resulting confidence interval on
oc
PCC will be the same width as that associated with the KQC value with the
distribution shifted by '"•jltiplying the scalar CWQC. The distribution on the
measured l( results from random error associated with the intra-laboratory
oc
variability and experimental bias resulting in inter-laboratory variability.
The combination of both sources of uncertainty leads to a final estimate of
the total uncertainty for measured K values for a given compound.
oc
For compounds without a measured K„ value (Appendix A), the use of
vC
Equation (1.3) introduces several sources of measurable variance. First, the
•
ability of regression equations to accurately predict values for KQC is
uncertain based on the ability of the regression to explain the variance in
Koc. If Kow is a perfect predictor of KQC, the variance introduced by
use of the regression is zero. The study by Kadeg et al. (1986) indicates
that there are other factors which probably affect KQC that are not used in
the regression due to lack of data (such as dissolved organic matter and
particulate matter). Thus, for this study, it can be assumed that Kqw is
not a perfect predictor for Kqc and that the error term will not be zero.
Since the error term and associated confidence interval is nonzero, any
value of K predicted from one of the regression equations will have some
Vv
uncertainty associated with it. This uncertainty is caused-purely by the use
of the ordinary least squares (OLS) estimators BO and Bl. From the use of
ordinary least squares criteria, it is possible to estimate the confidence
interval (which reflects the error) on the estimated K . This confidence
Uv
interval is used to quantitatively measure the variance in the final sediment
criteria values contributed by the use of a K predicted from the linear
oc
regression model.
4627a
2-4
-------
The error term in Equation (1.3) which defines the confidence interval is
calculated as
error » Tc x (2.1)
Tc is the critical value of the Student's t distribution (at a = 0.05,
95 percent confidence interval) with n-k-1 degrees of freedom where _n is the
number of sample data points and Jc is the number of regressors used in the
regression. S^ is the estimated standard error of the forecast, calculated
as the positive square root of
Sf2, Sn2 [1 + (1/n) + (Xf -X)2]/SSQX (2.2)
where
2
Sn = the estimated variance of the error term in the regression
ji * the number of data points in the regression
X^ » the measured KQW used to predict KQC
X * the average of the K values used in the regression
VH
SSQX « the sum of squares of the KQW values used for the regression.
From Equation (2.2), it follows that the magnitude of the error term in
Equation (1.2) increases as Sf increases due to a poor fit of the data, or
as ji decreases, or as the value of KQW used in Equation (2.2) moves away
from the mean of the regressed K... values, or as the value of SSQX decreases
ow
due to a bunching of the KQw values.
4627a
2-5
-------
3.0 AVAILABLE - K„tl DATA
oc ow
This chapter identifies the specific compounds under investigation,, the
associated available K_ - , summarized data, and the determined K
oc ow oc
distributions to be applied to Equation (1.1).
3.1 COMPOUNDS AND K - K_ DATA
oc ow
Table 3.1 identifies the representative hydrophobic organic chemical
compounds for'which K and data were collected. Compounds in capital
ow oc
letters have common matched Kni. - K„ data pairs. These data, taken from
OW oc
the available literature, were screened prior to analysis, limiting data to
observations exhibiting internal consistency (Kadeg et al. 1986). Detailed
evaluations and descriptions of these data are presented in Appendix B. Based
on these evaluations, summary statistics, using a normal distribution, were
developed for the compounds shown in Table 3.2.
3.2 Kgc DISTRIBUTION
The summary data presented in Table 3.2 were used to calculate KQC
values for the 5th, 25th, 50th, 75th, and 95th percentiles of the normal
distribution from which each chemical specific KQC is drawn. The
percentiles were calculated as follows:
5th percentile = K - 1.64 x SD
oc
25th percentile a K - 0.67 x SD
oc
50th percentile ¦ K
oc
75th percentile = + 0.67 x SD
oc
95th percentile * K „ + 1.64 x SD.
oc
These results are presented in Table 3.3 and indicate the uncertainty
(error) associated with the K „ terra in Equation (1.1). When the LHS
oc
technique is employed to develop the CDF curves, the information in Table 3.3
is internally generated from the data in Table 3.2 and by specification of a
normal distribution.
4628a
3-1
-------
a/
TABLE 3.1 Representative Hydrophobic Organic Chemical Compounds-'
Polycyclic Aromatic Hydrocarbons (PAHs)
Low Molecular Weight PAHs (2 and 3 rings)
FLUORENE
NAPHTHALENE
Acenaphthylene
Acenaphthylene
ANTHRACENE
PHENANTHRENE
High Molecular Weight or "Combustion" PAHs (3 to 6 rings)
FLUORANTHENE
PYRENE
CHRYSENE
BENZ0(A)ANTHRACENE
BENZO(A)PYRENE
Benzo(b)f1ouranthene
Benzo(k) f 1ouranthene
Indeno(1,2,3-cd)pyrene
0IBENZO(A,H)ANTHRACENE
Benzo(ghi)pery1ene
Pesticides
ODD
ODE
DDT
Acrolein
AL0RIN
CHLORDANE
DIELORIM
a-Endosulfan
s-Endosul fan
ENORIN
HEPTACHLOR
Heptachlor Epoxide
o-Hexach1orocyclohexane
B-Hexachlorocyclohexane
Y-Hexachlorocycl ohexane
6-Hexachlorocyclohexane
Isophorone
TOXAPHENE
Polychlorinated Biphenyls (PCBs)
Aroclor
Aroclor
Aroclor
Aroclor
Aroclor
Aroclor
Aroclor
1016
1221
1232
1242
1248
1254
1260
a/ Compounds in capital letters have common matched Ka«
pairs.
-------
table 3.2 Summary of KQW and KQC Data for Selected Hydrophobic
Organic Chemical Compounds
Log K
.ov
Log K(
C02E0und Mean ' _so_ Mean"' "°CJ£
High Weight PAHs
Benzo(a)anthracene 5.76 0.66 6 27 o 82
Benzo(a)pyrene 6.02 0.55 6 75 07!
Chrysene 5.70 0 65 "•'» 0.74
D1 benzol a,hlanthracene 6.25 0.71 6 36 n'«
Fluoranthene 5.26 0.60 si? 0*70
5-11 0.58 4.88 51 4 .
"UXT *¦" 0.49 Im S'S3
Napttialene 3.37 „.31 3
Phenanthrene 4.49 0.51 4^2 ass
Pesticides
Aldrin c e,
Chlordane 5'if
ODD I'H
DOE f*ff
DOT 5.69
Dieldrfn
Endrin J-Jf
Heptachlor J-J®
Toxaphene J'27
0.77
4.79
0.52
0.45
5.15
0.68
0.68
5.38
0.71
0.65
5.17
0.68
0.57
5.52
0.48
0.68
3.81
0.50
0.51
3.55
0.47
0.61
4.00
0.53
0.45
3.00
0.39
4628a
3-3
-------
TABLE 3.3 Log Kqc Distribution (1/kg)
Percentile
Compound .
5 25 50 75 95
High Weight PAHs
Benzo(a)anthracene J.925 5.720
Benzo(a)pyrene 5.536 6 254
Chrysene J-JjJ r985
Di benzo(a,h)anthracene 5.442 9.«
Fluoranthene
Pyrene
Low Weight PAHs
6.27
6.823
7.615
6.75
7.246
7.964
5.77
6.279
7.016
6.36
6.735
7.278
5.31
5.779
6.458
4.88
5.168
5.585
Anthracene 3.469 ^ ^ Ql 065 4>879
Fluorene J-JZJ 3*259 3.52 3.781 4.160
Naphthalene ^ ^ 4.588 5.122
Phenanthrene
Pesticides
3 937 4.442 4.79 5.138 5.643
Aldrin J.937 # 5AS S6Q6 6>265
Chlordane J-g* Jg04 5.38 5.856 6.544
ODD J-S| Jim 5.17 5.626 6.285
DOE T ,11 s 198 5.52 5.842 6.307
DOT HS 5:475 3.81 4.145 4.630
Dieldrin 9770 3 235 3.55 3.865 4.321
Endrin *"131 3.645 4.00 4.355 4.869
Heptachlor » 360 2.739 3.00 3.261 3.640
Toxaphene
4.031 4.42 4.809 5.371
4628a
3-4
-------
4.0 DEVELOPMENT QF SEDIMENT PERMISSIBLE CONTAMINATION
CONCENTRATIONS AND ASSOCIATED UNCERTAINTY
This chapter presents the results of applying the determined Kqc
distributions to Equation (1.1)," using either EPA provided chronic water
quality criteria (CWQC) or CWQC derived from acute-to-chronic ratios (ACR).
Subsequent cumulative probability distribution (CDF) curves are presented,
together with 0.01 and 0.05 probability fractiles which correspond to the 99th
and 95th percentile protection levels and form the basis for the PCC final
values.
4.1 CWQC VALUES
To apply water quality criteria to Equation (1.1), the appropriate water
quality values must be selected. As noted in Section 1.0, chronic water
quality criteria are most appropriate because of the long-term
exposure of benthic organisms to sediment contaminants. This study estimates
PCCs for the marine environment. A parallel analysis employing freshwater
chronic criteria could be performed to develop PCC values for freshwater
environments. The effects of salinity on KQC values have been discussed
previously (Kadeg et al. 1986). The ACR values employed were developed
primarily from non-benthic, freshwater organisms. Due to the paucity of
saltwater CWQCs and the absence of saltwater ACR values, the freshwater data
were used. Furthermore, it is assumed that the freshwater species used in
developing both CWQCs and ACRs have the same sensitivity as marine benthic
organisms. The errors associated with these assumptions are mitigated through
the uncertainty analyses performed on the data and the use of a range of ACR
values.
For the compounds with sufficient data to develop KQC distributions
(Table 3.3), a matrix of available water quality data was developed (Table
4.1). Based on this matrix, three approaches were taken. For those compounds
with specified CWQC, the CWQC values were combined with the KQC
distributions directly in Equation (1.1), from which CDF curves were
developed. For those compounds without specified CWQC but with specific ACR
values (naphthalene and toxaphene), the ACR values were applied to acute
4498a
4-1
-------
TABLE 4.1 Matrix of Available Water Quality Criteria Data
Criteria, ug/1
Compound
Benzo(a)anthracene
Benzo(a)pyrene
Chrysene
01benzof a,h)anthracene
Fluoranthene
Pyrene
Anthracene
FT uorene
Naphthalene
Phenanthrene
Fresh
Acute
Fresh
Chronic
3,980£/
2,300£/
Mari ne
Acute
300^/
300£/
3001/
300j>/
40°/
300&/
300£/
3005/
620£/ 2,350°/
300^/
a/ As reported by Kenaga (1982).
b/ Insufficient data for criteria,
for compound class.
Marine Specific
Chronic ACR£/
lgc/
29
Aldrin
3.0
«¦»*
1.3
—
--
Chlordane
2.4
0.0043
0.09
0.004
9
000
¦ — _
DOE
1,050£/
• •
14£/
— .
DDT
1.1
0.001
0.13
0.001
53
Dteldr1n
2.5
0.0019
0.71
0.0019
- -
Endrfn
0.18
0.0023
0.037
0.0023
2
Heptaqhlor
0.52
0.0038
0.053
0.0036
4-
Toxaphene
1.6
0.013
0.07
—
100
Based on lowest-observed-effect level
£/ Insufficient data for criteria. Based on lowest-observed-effect level.
4498a
4-2
-------
criteria values to generate a calculated CWQC value, which was then treated as
the specified CWQC value above. For the balance of the compounds, for which
no specific CWQC or ACR values exist, a range of ACR values (3-29) was applied
to the CDF based on Kenaga (1982), assuming a uniform distribution. This
range portrays the effect of introducing uncertainty into the CWQC.
A1 ternatively, for comparison purposes in the PCC distributions, the general
assumption was made for this group of compounds that the chronic criteria
would be an order of magnitude less than the acute criteria (i.e., ACR equals
10). This assumption appears to be a reasonable first-order estimate based on
compound-specific ACR values presented by Kenaga (1982). The fixed ACR value
of 10 was then used as before to generate a calculated CWQC for which a second
set of CDF curves were generated. This alternate approach had the effect of
eliminating the uncertainty in these curves associated with water quality
criteria uncertainty. For calculation of the final PCC values, the curves
generated from the uniform distribution of ACR values were selected over the
curves generated from this alternate approach. Both sets of curves are
presented in Appendix C. Compounds with specific CWQC values, or specific ACR
values and associated calculated CWQC values, are presented in Table 4.2
together with the corresponding values. It is apparent that the chronic
criteria values estimated from ACR values are generally higher than the EPA
measured chronic criteria values.. This reflects the limitation of the ACR as
a good predictor of CWQC, but in view of the absence of measurements for
chemicals for which only acute data were available, the ACR was the only
method by which CWQC could be estimated. Furthermore, it should be recognized
that the similarity of a saltwater ACR to the freshwater ACR is also an
underlying assumption in these computations. This may warrant further
investigation; however, for the present study, an adjustment factor is not
recommended.
4.2 DISTRIBUTION OF PCC VALUES AND ASSOCIATED UNCERTAINTY
The CWQC values as determined above were combined with the K
distributions as described in Chapter 3 (Table 3.3) in Equation (1.1) to
develop PCC distributions and define the uncertainty range. Table 4.3
presents the resulting distributions for the compounds under investigation.
4498a
4-3
-------
TABLE 4.2 Specific Chronic Water Quality Criteria and Acute
to Chronic Criteria Ratios for Selected Compounds
Compound
EPA Marine Chronic ACR Calculated Marine Chronic
Criteria (ug/1) (species)-^ Criteria (ug/1)-^
Fluoranthene
Naphtha!ene
Chiordane
DDT
Die!drin
Endrfn
Heptachlor
Toxaphene
16
0.004
0.001
0.0019
0.0023
0.0036
3 (MS)
29 (CS)
2 (DM)
53 (FHM)
13.3
81.0
0.045
0.0025
2 (SHM)
4 (DM)
100 (DM)
0.0185
0.0133
0.0007
a[/ Based on Kenaga (1982). Species upon which ACR is based is as
follows:
MS a mysid shrimp
CS « coho salmon
DM a Daphnia magna
FHM a fathead minnow
SHM a sheepshead minnow
b/ Calculated marine chronic criteria were only used in subsequent
calculations for those compounds without specific EPA chronic
criteria (napthalene and toxaphene).
4498a
4-4
-------
For those compounds with no specific CWQC or ACR value, the assumption of ona
order of magnitude difference between acute and chronic criteria (i.e., ACR
equals 10) was used to permit comparison.
The LHS technique was applied to the PCC distributions (the
-------
TABLE 4.3 Distribution of Permissible Sediment Contaminants
Concentrations (u^/goc) Oerived from Existing or
Estimated Chronic Water Quality Criteria
Percentile
Compound
High Weight PAHs
Benzo{a)anthracene
Benzo(a)pyrene
Chrysene
D1benzo(a,h)-
anthracene
F1uoranthene
Pyrene
Low Ueight PAHs
Anthracene
F1 uorene
Naphthalene
Phenanthrene
Pesticides
Aldrin
Chlordane
000
ODE
00T
OieTdrin
Endri n
Heptachlor
Toxaphene
5
25
50
75
- 95
1.24E+03
7.83E+03
2.79E+04
9.96E+04
5.27E+05
5.18E+03
2.7QE+04
8.44E+04
2.64E+05
1.37E+06
5.05E+02
2.71E+03
8.83E+04
2.84E+04
1.55E+.05
4.19E+03
1.46E+04
3.44E+04
8.11E+04
2.82E+05
2.32E+02
1.11E+03
3.27E+03
9.S2E+03
4.89E+Q4
2.27E+02
5.8SE+02
1.14E+03
2.20E+03
5.71E+03
4.40E+01
1.61E+02
3.95E+02
9.67E+02
3.54E+03
2.10E+01
6.81E+01
1.53E+02
3.46E+02
1.12E+03
6.14E+01
1.47E+02
2.68E+02
4.89E+02
1.17E+03
3.06E+01
1.06E+02
2.49E+02
5.86E+02
2.02E+03
1.1
3.6
8.0
1.80E+01
5.82E+01
4.34E-02
1.98-01
5.65E-Q1
1.61
7.36
3.0
1.45E+01
4.32E+01
1.29E+02
6.23E+02
8.0
3.63E+01
1.04E+02
2.95E+02
1.34E+03
5.41E-02
1.58E-01
3.31E-01
6.95E-01
2.03
1.86E-03
5.67E-03
1.23E-02
2.65E-02
8.UE-02
1.38E-03
3.95E-03
8.16E-03
1.69E-02
4.82E-02
4.87E-03
1.59E-02
3.60E-02
8.15E-02
2.66E-01
1.60E-04
3.83E-04
7.00E-04
1.28E-03
3.06E-03
4498a
4-6
-------
TABLE 4.4 PCC Values for Tested Fractiles
PCC (ug/goc) for
Probability Tractile
Compound 0.01 0.05 Basi sj/
High Weight PAHS
Benzo(a)anthracene ISO 2,100 est.
Benzo(a)pyrene 1,800 4,500 est.
Chrysene 120 440 est.
Dibenzo(a,h)anthracene 1,200 3,500 est.
Fluoranthene 160 310 CVIQC
Pyrene 85 190 est.
Low Weight PAHs
Anthracene 19 38 est.
Fluorene 5.9 15 est.
Naphthalene 50 72 calc.
Phenanthrene 11 24 est.
Pesticides
A1 dri n
Chiordane
000
DOE
DDT
Oieldrin
Endri n
Heptachlor
Toxaphene
0.43
0.84
est.
0.03
0.06
CWQC
0.6
2.2
est.
2.7
6.0
est.
0.04
0.07
CWQC
0.001
0.002
CWQC
0.001
0.002
CWQC
0.004
0.006
CWQC
0.0001
0.0002
calc.
a/ CWQC « based on existing EPA marine chronic water quality criteria
value. J
Calc. a based on calculation from compound-specific ACR value.
Est. . based on an estimate using a uniform distribution of a
range of ACR values.
4498a
4-7
-------
5.0 SUMMARY AND CONCLUSIONS
This study has investigated the uncertainty associated with the
equilibrium partitioning (EP) approach to developing sediment criteria for
hydrophobic organic chemicals, thereby allowing the uncertainty associated
with a criterion value to play a role in the regulatory decision-making
process. This method assumes that the concentration of a contaminant at the
water/sediment interface is at equilibrium, that the bioavailable form is the
free concentration of the contaminant in the aqueous phase, and that the
sorption is controlled by the physical and chemical properties of the sorptive
matrix and the chemical constituency of the aqueous phase. Hence, proper
normalization of the contaminant concentration to these parameters is
required. Previous studies together with the analyses presented have
indicated that, for the hydrophobic organic chemicals, the key normalization
parameter is organic carbon.
Permissible sediment contamination concentrations (PCCs) were estimated by
multiplying measured or predicted organic carbon-normalized partitioning
coefficients (K ) by EPA specified chronic water quality criteria (CWQC)
oc
values. The uncertainty associated with the PCC values can be quantified by
use of parametric statistics on measured values or nonparame.tric
analyses (Appendix A) on KQc values determined from
regressions. These later evaluations confirmed the need to use the
experimental data rather than the regression-predicted data because of the
larger error associated with the predicted information. Furthermore,
employing a quantitative uncertainty analysis technique, probability of
exceedance curves are generated which measure the risk of exceeding an
acceptable level of a given contaminant 1n sediments as a fraction of sediment
contaminant concentration. PCC values for 10 PAHs and 9 chlorinated
pesticides were computed for two levels of exceedance probability, 1 percent
and 5 percent (which are within the range of current regulatory
stipulations). These analyses were performed using water quality criteria
values reflecting marine chronic effects.
4839a
5-1
-------
In the absence of chronic water quality criteria, estimates were based on
a uniform distribution on a range of*acute-to-chronic' ratios (ACR) or from a
compound-specific ACR value. Uncertainties associated with ACR estimates were
also incorporated into the probability of exceedance curves.
Table 5-1 summarizes the results of these analyses and compares the
computed PCC values to similar quantities estimated in other studies. A 5
percent increase in the probability of exceedance induces approximately a
two-fold increase in the estimated PCC value. It is apparent that in certain
cases there is consistency in the estimated values among the different
methods. For the majority of the pesticides, the values based on equilibrium
partitioning (EP) agree with screening level concentration (SLC) values
reported by Neff et al. (1986). The Apparent Effect Threshold (AET) values
for benthic population data are, in general, higher for most of the compounds,
but the amphipod equivalent AET appears to be lower. These discrepancies
cannot be rationalized at this point because of the databases selected and the
inherent assumptions and limitations of all three methods. However, this
range of values provides a useful guide for regulatory decisions and argues
for more field and laboratory experimentation to reduce the ranges. The"
EP-based criteria also demonstrate a variability between compounds that is
expected based on their diverse KQC values (e.g., the PAHs tend to have
higher criteria values than the pesticides).
In review of the above results, it is obvious that for the EP approach
there 1s need for field and laboratory measurements of values, both to
oc
verify and refine the PCC values. Also, marine-based chronic water quality
criteria should continue to be developed and additional ACR data compiled.
4839a
5-2
-------
TABLE 5-1 PCC Values Fro« Equilibrium Partitioning Methodology and Other
Preliminary Criteria Values
Coapound
EPA CWQA j/
PCC
(»g/goc) for
probability
fractlle
or
Other Studies
AET AET
e Fractlle Benthlc Aaphlpod
0.05 ratio i/ |ug/goc)jy (ng/goc)£'
SIC
ug/goc)£^
Fluoranthene
160
310
1.9
>6,300
160
43.2 ,
Chlordane
0.03
0.06
1.9
—
—
0.0981'
DDT
0.04
0.07
1.6
>3.7
>1.2
42.8 ,
Dleldrln
0.001
0.002
1.6
—
0.0211'
Endrin
0.001
0.002
1.6
—
Heptachlor
0.004
0.006
1.6
—
0.0081^
Est. CWQA 1/
Benzofa)anthracene
160 2,100
12.6
>4,500
110
26.1
Benzo(a)pyrene
1,800 4.500
2.4
>6,800
99
39.6
Chrysene
120
440
3.7
>6,700
110
38.4
Oibenzo(a.h)-
anthracene
1,200 3,500
2.8
—
—
Pyrene
85
190
2.2
>7,300
>210
43.4
Anthracene
19
38
2.0
--
—
—
Fluorene
5.9
16
2.6
—
__
Naphthalene.
50
72
1.4
>330
>200
36.7
Phenanthrene
11
24
2.2
>3,200
180
25.9
Aldrln
0.43
0.84
2.0
—•
--
--
ODD
0.6
2.2
3.5
DOE
2.7
6.0
2.2
»_
Toxapiiene
0.0001
0.0002
1.4
—
—
--
a/ Bete rained prior to rounding of fractlle values,
b/ Tetra Tech 1986.
c/ iteff et al. 1986.
d/ PCC values based on EPA CWQC or EPA suggested value,
e/ Freshwater value.
f/ PCC values based on chronic water quality criteria estimated froai acute criteria.
CI
(0.0-64.3)
(0.0-0.136)
(0.0-113.7)
(0.0-0.084)
(0.0-0.029)
(0.0-41.0)
(0.0-46.8)
(0.0-60.5)
(0.0-74.4)
(0.0-41.4)
(0.0-38.4)
4839a
-------
6.0 REFERENCES
Kadeg, R. 0., S. P. Pavlou and A. S. Duxbury. 1986. Elaboration of
sediment normalization theory for nonpolar hydrophobic organic
chemicals. Envirosphere Company, Bellevue, Washington.
Kenaga, E. E. 1982. Predictability of chronic toxicity from acute
toxicity of chemicals in fish and aquatic invertebrates.
Environmental Toxicology and Chemistry 1:347-358.
Mackay, M. D., W. J. Conover and R. J. Beckman. 1979. A comparison
of three methods for selecting values of input variables in the
analysis of output from a computer code. Technometrics 21:239-245.
Neff, J. M., D. J. Bean, B. W. Cornaby, R. M. Vaga, T. C. Gulbransen,
and J. A. Scanlon. 1986. Sediment quality criteria methodology
validation: calculation of screening level concentrations from
field data. U.S. Environmental Protection Agency, Criteria and
Standards Division, Washington, O.C.
Pavlou, S. P. The use of the equilibrium partitioning approach in
determining safe levels of contaminants in marine sediments. Paper
presented at the 6th Pellston Conference on the Role of Sediments
1n Regulating the Fate and Effects of Chemicals in Aquatic
Environments, August 12-17, 1984, Florissant, Colorado.
Pavlou, S. P., and D. P. Weston. 1984. Initial evaluation of
alternatives for development of sediment related criteria for toxic
contaminants in marine waters (Puget Sound) phase II: development
and testing of the sediment water equilibrium partitioning
approach. JRB Associates, Bellevue, Washington.
Tetra Tech. 1986. Tasks 4 and 5a - Application of selected sediment
quality value approaches to Puget Sound data. Draft report
prepared for Resource Planning Associates for U.S. Army Corps of
Engineers, Seattle 01 strict. Bellevue, Washington.
4965a
6-1
-------
APPENDIX A
ALTERNATIVE NONPARAMETRIC APPROACH
TO DETERMINE UNCERTAINTIES
4489a
-------
CONTENTS
Page
A.O INTRODUCTION A-l
A.l BACKGROUND. A-l
A.2 DEVELOPMENT OF Koc AND Kow DISTRIBUTION AND THE
BOOTSTRAP SAMPLING SPACE A-3
A.3 ANALYSIS AND RESULTS A-ll
A.4 CONCLUSIONS A-18
A. 5 REFERENCES A-22
4489a
1
-------
TABLES
Page
A.l Summary Statistics for Normal Distributions of Low/High
Weight PAH and Pesticides, Based on Median Observed
Kow/Koc and Calculated SD Given Fixed Maximum N . . . a-4
A.2 Normal Distribution Parameters for Kow and Koc:
Calculated and Observed High Weight Polynuclear Aromatic
Hydrocarbons A-7
A.3 Normal Distribution Parameters for Kow and Re-
calculated and Observed Low Weight Polynuclear
Aromatic Hydrocarbons A-8
A.4 Normal Distribution Parameters for Kow and KpC
Distributions Calculated and Observed Pesticides .... A-9
A.5 Summary of Calculated Ranges in PCC Values for
Compounds Without Measured Koc A-17
A.6 Regression Statistics Used in Parametric Sensitivity
Analysis ...... A-19
A.7 Permissible Sediment Contaminant Concentration Values
(ug/goc) by Kow Values A-20
FIGURES
Page
A.l Example Figure - Comparison of Simulated and Observed Oata
Pairs for Phenanthrene A-6
A.2 Example Linear Plot of Bootstrap Sample - Simulated
Kow vs. Koc for Pesticides A-13
A.3 Example Cross-Specific Plot - Predicted Koc vs.
Simulated Kow for Pesticides
A.4 Example Residual Plot - Residual Koc vs. Simulated
Koc for Pesticides A_15
4489a
ii
-------
A.O INTRODUCTION
As described in Chapters 1 and 2 of the main report, an alternative
approach using nonpararaetric statistics to determine uncertainties on K
and Kqc and subsequent PCC values was evaluated. This appendix provides
background information on the approach (bootstrap method), and summarizes the
K and K distributions and development of bootstrap method sampling
ow oc
space, the results, and conclusions of this evaluation.
A.l BACKGROUND
A common problem in applied statistics Involves estimating a statistic
based on a finite number of observations or a sample taken from an unknown
population of observations. Typically, the estimated statistic carries
meaning only within the context of an interval surrounding that estimate.
Classical point-interval estimation statistics handled this problem
satisfactorily by assuming that sample estimates, such as the mean and
variance reflected the true mean (u) and variance (o^) of the population
from which the sample was taken. Classical distribution theory, confidence
intervals and hypothesis testing tacitly assume that data follow a specific
distribution, most commonly the normal distribution; these are called
parametric methods. For the typical problem, applied data are rarely adequate
to define the distribution of available data. Therefore, unless data exhibit
completely anomalous behavior, 1t is commonly assumed that the data
approximate the normal distribution, and investigators proceed to apply
classical methods. Most classical parametric statistical methods are fairly
robust; i.e., Insensitive to bias, etc., when derived from nonnormal data.
However, many applied statisticians have become increasingly uncomfortable in
making this 'leap of faith1 and have developed alternative estimation methods.
In the mid-1970s, Increasing Interest resulted in development of what have
been designated as 'nonparametric' methods. These methods typically use
ranked or ordered data {as defined by the raw data), rather than the raw data
Itself for which no true distributional characteristics could be defined. The
methods make no assumptions about the distribution of the observations, and
4489a
A-l
-------
methodological development has relied on exhaustive specification of the
probability of the specific value found for a test, given a specific sample
size. Examples of nonparametric methods-and their corresponding parametric
methods include: Wilcoxon's test as an alternative to the classical T-test;
Kruskall-Wallis and Friedman tests as alternatives to classical analysis of
variance; and Kendall's 'tau' statistic as an alternative to the classical
correlation coefficient.
With decreasing costs for computer time and developments in exploratory
data analysis, the approach applied to develop nonparametric methods was
expanded. The conceptual approach of bootstrapping, developed Dy Efron and
Tibshlrani (1985), empirically defines the distributional attributes of the
estimated statistic by replacing the distribution assumptions with the
observed distribution of the estimated statistic when the sample is expanded
into a population. The actual method involves iteratively sampling the
observations, with replacement; each sample so derived is called a 'bootstrap'
sample. The statistic of interest is then calculated for each bootstrap
sample. The resulting empirical distribution of the statistic, from all
possible estimates of the statistic given sample size and noted observations,
provides both an estimate of the statistic and a context within which that
estimate can be evaluated. Briefly stated, the bootstrapping method replaces
the classical, assumed 'population' (from which the observed sample has been
collected) with the bootstrap 'universe' of samples of which the observed
sample is a single example. Bootstrap methods have been applied to define
standard error, bias and precision errors for the one-sample situation, as
well as to more complicated data structures such as time series and censored
data. Usefulness of the method 1s evidenced in its application by widely
divergent areas of research.
Given the limitations to the available data on which sediment PCC values
were to be based (detailed in Appendix B), it was felt that the bootstrapping
method could be an extremely powerful tool to quantify the variability of the
possible regression models defining the relationship between K and K
oc ow
values for each chemical class.
4489a
-------
A.2 DEVELOPMENT OF K and KgvJ DISTRIBUTIONS AMD THE BOOTSTRAP
SAMPLING SPACE
Given the results of the exploratory analysis (Appendix B), it was
concluded that although the data are very limited, the distribution of the
logarithmic transformation of both Kqw and Kqc values will be assumed to
follow a normal distribution. To develop a balanced set of Kow-Koc data
for the bootstrap regressions, normal distributions for the Kqw and KQC
values of each chemical were constructed. The mean of the normal distribution
was chosen as the median value of all observed values. The standard deviation
was slightly more difficult to determine because no estimates (parametric or
nonparametric) of dispersion (on the observed data) were available for sets in
which the number of observations was less than three. To determine the
standard deviation for the K -K sets, the maximum coefficient of
ow oc
variation (C ) observed in all the and the sets where the number
v ow oc
of observations was greater than 10 were examined (Tables B.2 through B.4,
Appendix B). The maximum Cy in the three sets with greater than 10
data points was 9.12 percent and in the four sets was 8.76 percent. The
close agreement 1n the Cy values was unexpected because of the sediment
variability Included in the measured variability of Koc.
To estimate the standard deviation, the maximum Cy was held constant for
all
-------
TABLE A.l Summary Statistics for Normal Distributions of Low/High
Weight PAH and Pesticides, Based on Median Observed
Kqw/Kqc and Calculated SD Given Fixed Maximum N
L°gK
,OW_
Compound
Low Weight PAH
Benzo(a)
, anthracene
Benzo(a)pyrene
Chrysene
Dibenzo(a,h)
anthracene
Fluoranthene
Pyrene
High Weight PAH
Anthracene
FT uori ne
Naphthalene
Phenanthrene
Mean
5.76
6.02
5.70
6.25
5.26
5.11
4.45
4.27
3.37
4.49
SD
0.53
0.55
0.52
0.57
0.48
0.47
0.41
0.39
0.31
0.41
Adjusted
SD
0.66
0.55
0.65
0.71
0.60
0.58
0.51
0.49
0.31
0.51
6.27
6.75
5.77
6.36
5.31
4.88
4.42
4.01
3.52
4.22
0.58
0.55
0.58
0.56
0.44
0.48
0.38
0.34
0.30
0.38
Pesticides
A1 dri n
5.66
0.52
0.77
4.79
0.45
Chiordane
3.32
0.31
0.45
5.15
0.40
DDO
5.99
0.55
0.68
5.38
0.42
ODE
5.6°
0.52
0.65
5.17'
0.48
DDT
6.13
0.56
0.56
5.52
0.42
Die!drin
4.95
0.45
0.68
3.81
0.34
Endrln
4.48
0.41
0.51
3.55
0.30
Heptachlor
4.48
0.41
0.61
4.00
0.30
Toxaphene
3.27
0.30
0.45
3.00
0.20
4489a
A-4
-------
generate 100 random observations (for both KQC and KQW) for each
chemical. These 100 values per partition coefficient and per chemical
represented the balanced data sets from which bootstrap samples were taken.
To assess the reliability of the generated data sets given the limited data
and the many assumptions that were made, random pairs of K and '<
wv OW
values generated for each chemical compound were plotted along with the 'true'
pairs available from the literature. A sample of these plots is presented in
Figure A.l. To further corroborate the realism of data pairs selected from
the hypothetical distributions, descriptive statistics were calculated from
the bootstrap samples. Tables A.2 through A.4 document the desired statistics
for K and K data for each chemical per compound (based on exploratory
OW oc
analyses) and the.actual statistics calculated from the randomly assigned
pairs. The agreement between the two sets of data is quite good.
The plots presenting the simulated data pairs and the observed data pairs
fell into three general cases as follows:
Case i: Many true pairs fell within the center of the simulated
distribution (10 plots).
Case ii: Occasional true pairs exhibited mildly anomalous behavior; i.e.,
occurred at the fringes of the simulated data (4 plots).
Case 111: On rare occasions, observed pairs were highly anomalous within
the simulated distributions (3 plots).
These results Indicate that while available data were inadequate to
accurately describe the actual distributions of the KQC and Kqw data, the
simulated distributions did not erroneously describe possible paired points.
The existence of cente^d as well as anomalous pair behavior indicates that,
overall, the simulated data is representative of the observed KQC and KQW
data.
4489a
-------
6.0+
I R
R R
R R R R 2 R
5.0+ 2 2RR R R RS
K - R 2 R R 2 R
2 RR R 2 R R
0 - R R R R ** RR
R R RR RR2R 22 RR
C 4.0+ R R RR R22 R R RRRR R
R F: 2R R R R
R R R 2 R
R 2 R R
R R R
3.0+ R
R
-+ + + + + + K 0 U
3.00 3.40 4.20 4.80 5.40 6.00
phenanthrene: 2 pairs
R * RANDOM BINORNAL KOC-KOW PAIRS
* * PAIRED KOC-KOU P0INTCS3 FROM LITERATURE
INTEGER * NUMBER OF COINCIDENT OBSERVATION^
FIGURE A.l Example Figure - Comparison of Simulated and Observed Data Pairs
for Phenanthrene.
A-6
-------
TABLE A. 2 Normal Distribution Parameters for Kow and Koc:
Calculatadi' and Observed zf High Weight Polynuclear
Aromatic Hydrocarbons
Compound
-ow-
Mean
SD
-oc-
Mean
SO
Pibenzo(a,h)anthracene
Observed 6.25 0.71 6.36 0.56
Calculated 6.26 0.71 6.45 0.60
Benzo(a)pyrene
Observed 6.02 0.55 6.75 0.74
Calculated 5.91 0.53 6.76 0.75
Benzo(a)anthracene
Observed 5.76 0.66 6.27 0.82
Calculated 5.73 0.76 6.09 0.7S
Chrysene
Observed 5.70 0.65 5.77 0.76
Calculated 5.78 0.76 5.63 0.74
H uoranthene
Observed 5.26 0.60 5.31 0.70
Calculated 5.22 0.64 5.28 0.75
Pyrene
Observed 5.11 0.58 4.88 0.43
Calculated 5.08 0.60 4.93 0.44
y Calculated distribution parameters (X, s) were calculated from the
random samples constrained to the expected distribution (a,-?).
b/ Observed distribution parameters (u»a) were determined from
~~ available data and defined as follows:
m > median value of observed Kq* or Koc data, regardless- of the
number of observations and
a » (median value) x (max coefficient of variation)
where max Cv (Kow) = 0.0912
max Cy (Koc) = 0.0876.
4489a
-------
TABLE A.3 Normal Distribution Parameters for Kow and Koc:
Calculated!/ and Observed**' Low Weight Polynuclear
Aromatic Hydrocarbons
Kow Koc
Compound Mean SD Mean SD
Phenanthrene
Observed 4.49 0.51 4.22 0.55
Calculated 4.55 0.54 4.31 0.59
Anthracene
Observed 4.45 0.51 4.42 0.58
Calculated 4.52 0.54 4.44 0.58
F1uorene
Observed 4.27 0.49 4.01 0.53
Calculated 4.24 0.48 4.00 0.51
Naphtha!ene
Observed 3.37 0.31 3.52 0.39
Calculated 3.29 0.33 3.49 0.38.
jj/ Calculated distribution parameters (X, s) were calculated from the
random samples constrained to the expected distribution (u,o).
b/ Observed distribution parameters (ji.cr) were determined from
available data and defined as follows:
ii = median value of observed Kow or Koc data, regardless of the
number of observations and
a a (median value) x (max coefficient of variation)
where max Cv (Kow) » .0912
max Cy (Kqq) a .0876.
4489a
-------
TABLE A.4 Normal Distribution Parameters for Kow and Koc
Distributions Calculated 1/ and Observed i>/ Pesticides
Compound
DDT
Observed
Calculated
DDD
Observed
Calculated
DDE
Observed
Calculated
Aldri n
Observed
Calculated
Dieldr-fn
Observed
Calculated
Heptachlor
Observed
Calculated
Endri n
Observed
Calculated
Chlordane
Observed
Calculated
4489a
vow
koc
Mean
SO
Mean
SO
5.13 0.57
6.10 0.60
5.52 0.48
5.50 0.44
5.99
6.20
0.68
0.74
5.38
5.38
0.71
0.67
5.69
5.57
0.55
0.67
5.17
5.08
0.68
0.70
5.66
5.67
0.77
0.82
4.79
4.76
0.52
0.50
4.95
5.02
0.68
0.67
3.81
3.80
0.50
0.46
4.48
4.41
0.61
0.61
4.00
4.03
0.53
0.53
4.48
4.39
0.51
0.55
3.55
3.61
0.47
0.47
3.32
3.34
0.45
0.43
5.15
5.11
0.68
0.60
A-9
-------
TABLE A.4 (Continued)
K0w Koc
Compound Mean SD Mean SD
Toxaphene
Observed 3.27 0.45 3.00 0.39
Calculated 3.24 0.92 2.95 0.41
£/ Calculated distribution parameters (X, s) were calculated from the
random samples constrained to the expected distribution
b/ Observed distribution parameters (n,a) were determined from
available data and defined as follows:
u = median value of observed Kow or Koc data, regardless of the
number of observations and
a = (median value) x (max coefficient of variation)
where max Cv (Kow) = .0912
max Cv (Koc) = .0876.
4489a
A-10
-------
Through the extensive exploratory analysis of measured KQC and KQW
values, the above distributions were derived which describe the observed
variability in the data sets. Although the data from which these
distributions were generated were very limited, simulated data sets generated
from these distributions using L?itin Hypercube sampling were generally
consistent with the observed data pairs. As a result, it appeared that such a
sampling method might be adequate to derive a sample space for estimating" tne
uncertainty in the PCC values.
A.3 ANALYSIS AND RESULTS
For compounds with a measured Kqw value but no measured value, the
uncertainty in the sediment criterion increases because of the need to use a
Kow~Koc re9ression relationship to calculate the KQc value. For each
class of compounds, the bootstrapping method was used to generate a sample of
100 possible regression equations. For each regression equation generated
using the bootstrap method, a random value of was selected from the
distribution based upon measured values. Then, using Equations
ow r ow
(1.3) (i.e., the regression equation), (2.1), and (2.2) as discussed in the
main report, a median PCC value and associated 95 percent confidence interval
values were determined. Combining these results for the 100 equations
permitted derivation of a distribution for the median and confidence intervals
for the PCC values.
To evaluate these distributions, linear plots of the bootstrap samples
were plotted which represent random and independent selections of KQW and
Kqc values (from the distributions described in Table A.l) for the chemical
classes. These plots were elliptical and the shape of the distribution is the
direct result of the uncertainty associated with K _-Knw values for the
oc ow
chemical within each chemical class as well as the result of the assumptions
made about the distribution of the data to compensate for the 'unpairedness'
of the available data.
Additional plots of the chemical class-specific 100 regressions lines
(predicted K vs. observed K ) resulted in plots approaching the
vv# OW
theoretical confidence intervals about the regression. The distributions were
4489a
A-ll
-------
generally parabolic, centered at the mean. These results suggest that 100
bootstrap samples will result in an adequate estimate of the variability in
the regression lines derived from the simulated data.
Residual plots were also constructed for the chemical class-specific
groups (i.e., high and low weight PAHs, pesticides). Residuals, which are tne
differences between simulated .< and predicted K (based on the observed
oc
K values) are the basis by which the success (or failure) of the linear
ow
model can be assessed. Plots of the residuals against the predictor or
carrier variable (here simulated K values) which exhibit random,
ow
horaoscedastic behavior (across the predictor) support the model. For example,
such a plot Indicates that the effect on the predicted variable (here, Kqc)
of the predictor has been accounted for by the model and no additional effect
2
(such as an X term) is required. The plots appeared to support such a
conclusion.
Plots of the residuals against the predicted values can also be used to
indicate the variability of the residuals with respect to the model
predictions. Systematic nonrandom behavior suggests that some systematic
variability 1n the predicted variable has not been accounted for by the linear
model. The plots exhibited no such cumulative behavior in the regressed
bootstrap samples, again supporting the model.
Finally, plots of the residuals against the observed values can and must
exhibit nonrandom behavior due to the correlation between the observed and
residual values. (Recall that the residual equals the observed minus the
predicted value.) The absolute value of the slope of the residual vs. the
observed Kqc plot equals zero only when the model is a perfect fit. The
plots Indicated that there is a less than perfect fit for each of the three
chemical classes.
Examples of the above plots are presented in Figures A.2 through A.4. The
results of this graphic evaluation suggested that, although the model appeared
to be meeting the general constraints for its application, the underlying
assumptions concerning the distribution of the data together with the limited
real (measured) data sets (Appendix B) would ultimately lead to
4489a
A-12
-------
4
Log Kow
¥
8
FIGURE A.2 Example Linear Plot of Bootstrap Sample - Simulated K VS K„
for Pesticides. ow oc
A-13
-------
;vJ_!
i
i
i
i
i
i r r ~~L P E~
12 4 6 5 5
Log Kow
FIGURE A.3 Example Cross - Specific Plot - Predicted VS Simulated K_.
for Pesticides. oc ow
A-14
-------
SIMULATED Log Koc
FIGURE A.4 Example Residual Plot - Residual KQC VS Simulated KQC for Pesticides.
-------
unrealistically high uncertainties due to the scatter in the simulated data
sets and resulting regression lines (as supported by the elliptically shaped
plots of the bootstrap samples). In simplest terms, it appeared that the
measured K and K data sets may have been too small to infer true "real
oc ow
world" distributions, as opposed to perceived, artificial distribution
patterns indicated by the simulated data sets.
Table A.5 summarizes the range in PCC values. The range is defined as the
difference between the log values of the PCC. The range in all values is. the
difference between the absolute minimum and maximum PCC values calculated in
100 samples including confidence intervals on the predicted Kqc. The
80 percent range is the difference between minimum and maximum values in the
range of the 10th to 90th percentiles on the median PCC. The minimum and
maximum confidence intervals are from those generated in the analysis for any
one sample observation. As these values indicate, there is substantial
variability across both the distribution of median-PCC values and of
confidence intervals for each median PCC.
From columns 3 and 4 of Table A.5, it can be seen that the 80 percent
confidence interval contributes from 40 to 48 percent of the range in the PCC
variability for the median values of the PCC. Columns 1 and 2 (which include
the confidence intervals) show a similar decrease in overall variability when
comparing the 80 percent range to the 100 percent range. This decrease in
range is probably due to the elimination of some poorly-fit regression lines
which were used to calculate the PCC values in the tails of the distribution
on the median PCC. This reduction in range would indicate that such 111-fit
regression lines ars not predominant, but do occur.
Columns 5 and 6 on Table A.5 show the range 1n the width of the confidence
Intervals. This variability in the width of the confidence intervals
corresponds to the wide range in slope and intercept terms reported in the
results of the bootstrapping method and shows the resulting Impact on the
range of PCC values (i.e., uncertainty). As discussed in Section 2.3 of the
main text, the width of the confidence interval is dependent on several
factors including the fit of the regression (as typically measured by the
R-square term), the number of data points, and the spread of the data used in
the regression.
4489a
A-16
-------
TABLE A.5 Summary o* Calculated Ranges in PCC Values for Compounds without
Measured Koc
Range of Values -
All Values^ Mediant Range in C.I.-^
Compound 100a 80a 100a 80a Max. Mi n.
High Weight PAHs
8enzo(b)fluoranthene
Benzo(g,h,i)pyrene
Benzo(k)f'1 uoranthene
Indeno{l,2,3,c,d)pyrene
Low Weight PAHs
Acenaphthene
Acenaphthylene
Pesticides
Endosulfan
Hexacyclochlorohexane
Isophorone
14.44
9.92
4.68
18.72
10.75
5.27
15.37
10.59
4.73
23.94
15.27
7.61
19.55 4.26 1.92
40.00 18.90 11.03
7.43 5.68 1.79
6.95 4.81 1.50
10.72 8.29 3.57
2.20
13.23
1.13
2.49
16.91
1.36
2.28
13.90
1.16
3.44
22.19
1.65
0.93 19.55 0.23
4.60 40.00 0.52
0.76 7.17 1.58
0.72 6.92 1.53
1.41 9.26 1.95
a/ Range is defined as the difference of the log values of the PCC.
]>/ The range in all values is the difference between the absolute minimum and
maximum values calculated 1n 100 samples. The 80 percent range is the
difference between minimum and maximum values in the 10th to 90th
percentiles on the median PCC.
c/ The range is calculated as the difference between median values of the
PCC, excluding confidence intervals on the PCC.
d/ Minimum and maximum confidence intervals from those generated in the
~~ analysis of 100 samples for any one sample observation.
4489a
A-17
-------
Tables A.6 and A.7 show the results of a parametric sensitivity analysis
using three regression equations, three-values for KQW> and the 80 percent
confidence interval on the predicted value of Kq(.. Table A.6 lists the
regression equations and Kqc values used. For each chemical class, three
regression equations were chosen based on the distribution of the calculated
slope term. From the values of B1 generated by the bootstrapping metnod,
regression equations were chosen for the median, upper, and lower quartile
values of Bl. For these analyses, the CWQC value in Equation (1.3) of the
main report was uniformly defined as an order of magnitude less than the acute
water quality criteria (i.e., acute to chronic ratio of 10). These results
show that there is a substantial range in the calculated PCC values when the
K -K regression relationship is used to calculate the PCC value. Much
oc ow
of this range is due to confidence intervals on the predicted value of K
with some contribution from variance in BO and Bl regression terms and the
value of KQw used in the regression.
The range in PCC values due to the confidence intervals on ic varies
oc
between compound classes. As would be expected, the confidence intervals for
the pesticides are more uniform because of the larger sample sizes used. The
largest width in the PCC confidence interval is seen for the low weight PAH
compounds, which have the fewest data points.
Due to the instability of the regression lines generated by the
bootstrapping method, it is not feasible to define a single probability
function for the PCC distribution for compounds without a measured K . It
oc
is certain, however, that a great deal of confidence can be gained by
measuring the Kqc value.
A.4 CONCLUSIONS
Based on the above analysis, the uncertainty associated with the PCC
values in Table A.6 is very broad. The application of the nonparametric
bootstrapping techniques, while technically correct, has resulted in an
overstatement of these uncertainties. This is directly attributed to the
assumptions required to generate a data set of suitable size to employ this
method. It is apparent that the selection of distributions (data patterns)
4489a
A-18
-------
TABLE A.6 Regression Statistics Used in Parametric Sensitivity Analyses
Regression Equation Parameters i/
Compound Class
BO
B1
S2
XBAR
SSQX
High Weight PAH
4.30
0.2400
0.54
5.90
9.62
3.50
0.4400
0.39
5.20
2.72
0.18
0.9600
0.15
5.70
0.94-
Low Weight PAH
3.00
0.1800
0.42
4.50
13.60
2.20
0.5100
0.95
3.80
6.54
0.57
0.7900
0.06
4.30
20.60
Pesticides
3.10
0.2800
0.54
4.80
3.11
2.30
0.3900
0.28
4.90
0.77
1.90
0.5400
0.62
4.70
1.43
Log Kqc
Compound
Q1
Median
Q3
H1ah Weight PAH
Benzo(b)fluoranthene
5.732
6.320
6.908
Benzolg,h,1)pyrene
6.503
7.050
7.597
Benzo(k)fluoranthene
5.850
6.450
7.050
Indeno(l,2,3,c,d)pyrene
6.986
7.700
8.414
Low Weight PAH
Acenaphthene
3.893
4.150
4.407
Acenaphthylene
3.761
4.010
4.259
Pesticides
Endosulfan
3.377
3.600
3.823
Hexacyclochlorohexane
3.752
4.000
4.248
Isphorene
1.595
1.700
1.805
£/ BO » regression equation intercept
B1 s regression equation slope
S2 » estimated variance of error term in regression
XBAR a average of log Kow values used in regression
SSQX » sum of the squares of the log Kqw value used for the
regression
Q1 a lower quartile value
Q3 a upper quartile value
4489a
A-19
-------
TABLE A.7 Remissible Sediment Contaninant Concentration Values (ug/goc) by Kow Values
Compound
25th Percentile
Predicted Confidence Interval
Regr Cow He? H
-------
inferred from the original raw data sets resulted in KQC ranges beyond those
which are chemically and physically feasible for a ciiaracteri stic marine
environment. Therefore, the PCC values and associated ranges reported in this
appendix are not recommended, but are rather included for illustrative
purposes. While the overall nonparametric approach is still suitable to
determine uncertainties, it is clear that more field K and K data are
oc ow
required.
4489a
A-21
-------
A. 5 References
Efron, B., and R. Tibshirani. 1985. The bootstrap method for assessing
statistical accuracy, tech. Rept. 101, Division of Biostatistics,
Stanford University, Palo Alto, California.
4489a
A-22
-------
APPENDIX B
ANALYSIS OF AVAILABLE K - K DATA
oc ow
4484a
-------
CONTENTS
fifli
B.O DESCRIPTION/LIMITATIONS 8-1
B.l DISTRIBUTION OF KQC AND KQW B-5
*ow 8-10
Koc B-15
B.2 REFERENCES - R-'i
4484a
1
-------
TABLES
Page
B.l Kow and Koc Data: Summary by Chemical Class and Compound
Number of Observations, Labs, and Common Pairs B-2
B.2 Summary Statistics of Kow and Koc Distributions:
High Weight PAH Compounds . B-o
B.3 Summary Statistics of Kow and Koc Distributions:
Low Weight PAH Compounds ' B-7
B.4 Summary Statistics of Kow and Koc Distributions:
Pesticide Compounds 8-8
FIGURES
Page
B.l Benzo(a)pyrene Kow Data: Frequency Distribution B-j.2
B.2 Napthalene Kow Data: Frequency Distribution B-13
B.3 DDT Kow Data: Frequency Distribution B-14
B.4 Dibenzo(a,h)anthracene Koc Data: Frequency Distribution. . B-16
B.5 Pyrene Koc Data: Frequency Distribution B-17
B.6 Arochlor 1254 Koc Data: Frequency Distribution B-18
B.7 DDT Koc Data: Frequency Distribution B-19
4484a
3-ii
-------
APPENDIX B
AVAILABLE K -K DATA
oc ow
This appendix describes the detailed review and analysis of the Kqw and
Kqc data previously collected (Kadeg et al. 1986) and used for the
uncertainty analysis of sediment normalization theory.
B.O DESCRIPTION/LIMITATIONS
K and K data taken from available literature were screened prior
wn UW
to analysis, limiting data to observations exhibiting internal consistency
(Kadeg et al. 1986). Internally consistent, acceptable data are summarized in
Table B.l. The table indicates (by chemical compound within the four chemical
classes) the following: the number of Kqw and Kqc measurements; the
number of Individual laboratories which made the measurements; and the number
of true common pairs, i.e., paired Kow-Koc values for a specific compound,
generated by the same research laboratory. Compounds are ordered within
chemical class in the table, on the basis of the number of observed measures
on Kqw and KQC. Compounds ac the top of the class influenced all
subsequent analyses/conclusions by virtue of the relatively greater
information available on their partitioning behavior.
Several features of the available chemical partitioning data are
immediately apparent:
o Available data are unbalanced in terms of Information on partitioning
behavior. This imbalance is Indicated by the different number of
both measurements (ranging from 1 to 24 for KQw coefficients and 1
to 31 for K coefficients) and researchers (ranging from a single
oc
laboratory to 15 laboratories in and a single laboratory to 8
laboratories 1n KQC).
o Detailed Information indicative of the distribution of
measurements is limited to three compounds:
4484a
B-l
-------
TABLE 8.1 Kow and Koc Data: Summary by Chemical Class and
Compound Number of Observations^/, Labs, and Common Pairs
Compound
low.
N No. Labs
loc_
N No. Labs
Common
Pairs
High Weight PAH
Benzo(a)pyrsne 24
Pyrene 6
Dibenzo( a, h) anthracene *
Chrysene 5
Fluoranthracene 6
Benzo(a)anthracene 6
Benzolg,h,i)perylene 5
Benzo(b)fluoranthene 2
Benzo(k)fluoranthene 2
Indeno(l,2,3,c,d)pyrene 1
Low Weight PAH
7
6
4
6
6
6
5
2
2
1
Naphthalene
20
11
F1 uorene
7
7
Anthracene
7
5
Phenanthrene
6
5
Acenaphthy1ene
2
2
Polychlorinated Biphenyl Compounds
Arochlor 1254
2
2
Arochlor 1248
1
1
Arochlor 1260
1
1
Pesticides
DDT
18
15
DDE
7
7
Endri n
5
6
ODD
5
5
A1dri n
3
3
Chiordane
3
3
Heptachlor
3
3
Di el drin
2
2
Toxaphene
2
2
Acrolein
1
1
Isophorone
3
2
Endosulfan
1
1
Hexacyclochlorohexane
1
1
TCDD
1
1
4
31
15
1
1
1
0
0
0
0
8
1
3
2
0
14
0
0
0
0
0
0
1
3
2
1
1
1
0
0
0
0
6
1
2
1
0
1
0
0
8
1
2
1
4
1
1
2
1
1
0
0
0
0
0
2
2
1
1
0
0
0
0
0
4
1
2
2
0
0
0
0
0
0
0
0
0
4484a
3-2
-------
1) Benzo(a)pyrene (n=24)
2) Naphthalene (n=20)
3) DDT (n=18).
o Distributional attributes of K measures are limited to four
oc
compounds:
1) Pyrene (n=31)
2) Dibenzo(a,h)anthracene (n=15)
3) Arochlor 1254 (n=14)
4) DOT (n=14).
o The number of true paired measurements is severely limited. Of the
ten high weight PAHs, four chemicals have one to two sets of paired
observations. Four of the six low weight PAHs are paired, as are 9
of the 14 pesticides. No pairs are available for any of the PCS
compounds, thereby eliminating this class from analysis.
The inherent structure of available K^K data sets presents several
ow oc
analytical difficulties due to the assumptions associated with statistical
methods'used to predict values of a response variable (K ) on the basis of
Ov
observed values of a predictor variable (KQW). Ordinary least squares (OLS)
regression assumes that response and predictor variables are multivariate
normal 1n distribution. It further assumes that the predictor variable is
measured without error. Finally, and most importantly, OLS regression tacitly
assumes that the measures on predictor and response are paired observations.
The K -K__ data sets met few if any of the above criteria. Deficien-
ow oc *
cies are due, primarily, to the observational nature of the study and the
uncontrolled nature of expert mentation. Several of the limitations can be
compensated for by variations to OLS. Imbalance 1n the number of K and
ow
K values can be compensated by analyzing summary statistics, i.e., mean
OC
and variance of the K and K values. The variance associated with the
OW oc
estimated mean must be Included in the overall error estimate to ensure that
the sum of squares is not artificially deflated, resulting in an overly
optimistic estimate of the fit of the data. When the number of observations
4484a
3-3
-------
representing each mean value varies or the variances associated with the mean
values differ, weighted least squares (WLS) regression which 'weight'
observations on the basis of data 'reliability' is typically used. When the
measurement error of the predictor variable is much greater than that of the
response variable, Draper and Smith (1981) indicate that error in the
predictors may be effectively ignored without serious implications to the OLS
regression results. When the measurement errors (in predictor and response
variables) are of the same order of magnitude, Action (1966) has presented
analytical solutions for correcting the variance estimates.
None of the variations to the OLS regression solution are adequate to
address all the difficulties posed by the K -K data, and none of the
OW OC '
methods described above can deal with the minimal number of pairings in the
available data. Because the assumption that the predictor and response
variables are paired is basic to regression analysis and because the available
data are very limited in paired observations, the bootstrapping method was
applied (Appendix A).
Application of the bootstrapping method typically involves
o Randomly sampling reported KQw and KQC values for each chemical
for data from each chemical class.
o Determining the slope and intercept estimates of the KQw and KQC
relationship (for each chemical class) from this random sampling
using the least squares regression algorithm.
o Using the distribution of the slope and intercept estimates to define
the distribution of these estimates (e.g., the median or average
slope and intercept; and the first and third quartlles of each
estimate) and to define a chemical class-specific distribution of the
resulting KQC values.
However, given the available data, this classical approach for applying
the bootstrapping method would have given greater influence to chemicals for
which the least amount of information is available. If only a single K
4484a
8-4
-------
value is available for a specific chemical (as occurred for many of the
chemicals, Table B.l), that value would necessarily be included in each
bootstrap sample and in each regression solution because randomly selecting
one from a single valued distribution implies selection of that value. This
situation is different from the situation in which multiple observations or
measurements coincide. In this case, the phenomenon is considered to be
known. To eliminate the sample-biasing due to 'lack of information' in the
data included in the bootstrap samples, exploratory analyses of the available
data were made to
o Define the distributional structure of available data (within and
across chemical classes).
o Evaluate the laboratory factor as It affects apparent distributions
(defining Inter- and intra-laboratory variability).
o Develop a balanced set of observations which would be iteratively
sampled by the bootstrapping algorithm.
The results from these exploratory analyses are discussed below.
B. 1 Distributions of KQ(. and KQW
Tables B.2 through B.4 describe all available data per individual chemical
compound within the high weight PAH, low weight PAH, and pesticide classes.
The tables present the median value, i.e., the natural log-transformed value
of the K or K which is in the center of the range; N, the number of
ow oc
measured values; a brief descriptive summary of the type of frequency
distribution of the available data (approximately normal, bimodal, uniform,
etc.); and the coefficient of variation (Cy). The Cy Is the standard
deviation normalized by the mean value multiplied by 100 and gives a
percentage ratio of data variability to an average value. The median is the
most appropriate measure of central tendency for these data because it truly
reflects the center of the observed values. In normally distributed data, the
median and mean values will coincide. Given the sensitivity of the mean to
4484a
B-5
-------
TABLE B.2 Summary Statistics of Kow and K^c Distributions: Hijn
Weignt PAH Compounds
Compound
Log K
.ow
Log K
.oc
Pa i rs
Phenanthrene
Median
N
Hi stogram
Cv (%)
Anthracene
Median
N
Histogram
Cy(%)
F1 uorene
Median
N
Histogram
CVU)
Naphtha!ene
Median
N
Histogram
Cv(%)
4.49
6
Right skew
5.8
4.45
7
Sparse normal
4.7
4.27
7.
Uni form
2.6
3.37
20
Normal
1.8
4.22
2
Uni form
4.7
4.42
3
Uniform
17.1
4.01
1
Undef1ned
3.52
8
Uni form
15.7
N is the number of observations.
Cv is the coefficient of variance.
4484a
8-6
-------
TABLE B.3 Summary Statistics of Kow and Koc Distributions: Low
Weight PAH Compounds
Compound Log Knw Log Knr Pairs
Dibenzo(a,h)anthracene
Median
N
Histogram
CVU)
62.5
4
Sparse normal
8.9
6.36
15
Bimodal
4.1
2
Benzo(a)pyrene
Median
N
Histogram
cv(*)
6.02
24
Left skew
3.1
6.75
4
Unimodal
5.6
0
Benzo(a)anthracene
Median
N
H1stogram
CVU)
5.76
6
Approx normal
5.0
6.27
1
Undef1ned
1
Chrysene
Median
N
Histogram
CV(S)
5.70
6
Sparse normal
4.4
5.77
1
Undefined
1
Fluoranthene
Median
N
H1stogram
Cy(%)
5.26
6
Sparse normal
3.9
5.31
1
Undefined
1
Pyrene
Median
N
Hi stogram
CV(S)
5.11
6
Sparse normal
3.0
4.88
31
BImodal
4.6
2
N is the number of observations.
Cv 1s the coefficient of variance.
4484a
3-7
-------
TABLE B.4. Summary Statistics of Kow and Koc Distributions
Pesticide Compounds
Compound Log Knw Log Koc
DDT
Median 6.13 5.52
N 18 14
Histogram Normal Normal
CV(S) 9.1 8.8
ODD
Median 5.99 5.38
N 5 1
Histogram Right skew Undefined
Cv(*) 11.0
DDE
Median 5.69 5.17
N 7 1
Histogram Sparse normal Undefined
CV(X) 6.0
AT dri n
Median 5.66 4.79
N 34
Histogram Uniform Uniform
CVU) 18.3 26.3
Dieldrin
Median 4.95 3.81
N 2 2
Histogram Undefined Undefined
Cy(*) 35.9 10.9
Heptachlor
Median 4.48 4.00
N 3 ]
Histogram Uniform Undefined
CVU) 15.5
4484a
8-8
-------
TABLE B.4 (Continued)
Compound Log KQW Log Knc P^irs
Endri n
Median
N
Histogram
CVU)
4.48
6
Sparse normal
21.4
3.55
2
Undefi ned
13.0
Chlordane
Median
N
Hi stogram
Cv(%)
3.32
3
Uni form
37.0
5.15
1
Undef1ned
Toxaphene
Median
N
Hi stogram
CVU)
3.27
2
Undefined
1.52
3.00
1
Undef1ned
N is the number of observations.
Cv is the coefficient of variance.
4484a
B-9
-------
extreme values and the limited number of observations available for most of
the chemicals, the median, above and below which half of the observed values
lie, is a better summary statistic than the mean.
The Cy uses two statistics that are only relevant in the context of a
normal distribution. For purposes of comparison, an alternative,
nonparametric analog to the Cy was also examined. This alternative
nonparametric Cy value is the difference between the first and third
quartile value divided by the median value, multiplied by 100. The
denominator represents the nonparametric analog ta the mean. The numerator
represents the absolute range in variable values which includes the mid-lying
50 percent of the observations and serves as the nonparametric analog to the
dispersion of a normal distribution, i.e., standard deviation. Differences
between the parametric and nonparametric estimates of the coefficient of
variation were found to be negligible. The parametric estimate is presented
in the tables because it is the more commonly used statistic.
The limited number of K and K measures constrained the
ow oc
investigation of the distributional structures to three and four chemicals,
respectively. The chemicals are benzo(a)pyrene, napthalene, and DOT for
'
-------
o Twenty four measurements, determined by seven different analytical
laboratories, are available for the high weight PAH benzo(a)pyrene.
Figure B.l shows tne frequency distribution (histogram) for the
available data. Mai Ion and Harrison (1984) determined 18 of the data
entries; the remaining six were results from six other laboratories.
In Figure B.l the Mallon results are coded with an 1M1 and the other
laboratories' results are coded with an '0'. Numerical values
indicate multiple data points, the value corresponding to the number
of points. The plot suggests that anomalously high K values were
UW
reported by some of the laboratories; specifically, MacKay et al.
(1980), Rapport and Eisenreich (1984), and Rapport and Eisenreich
(op. cit., 1984). Data from the other three laboratories fall well
within the range of benzo(a)pyrene measures reported by MacKay et al.
(1980).
o The frequency distribution of the 20 measured KQw values for
naphthalene 1s presented in Figure B.2. While the data are sparse,
there appear to be two peaks in the distribution (at log.K of
OW
3.36 and 3.44). The figure also indicates that nine of the values
are from a single laboratory, Garst and Wilson (1984), coded with a
H(j" while the remaining 11 values are from 11 individual laboratories
coded with an "0." The Garst results span the entire range of
observed naphthalene KQw values, and the peak in the distribution
at 3.44 is due primarily to Garst's observations while the peak at
3.36 represents the data from the other laboratories.
o The 17 measurements of DDT KQW values represent results from 15
individual laboratories. Figure B.3 which exhibits the frequency
distribution of DDT K measurements indicates that tne data follow
ow
a normal distribution.
Examination of the l< data for the three chemicals for which an
ow
adequate number of KQW measurements are available indicates that
inter-laboratory variability is substantial. Although the distribution of
observations is relatively uniform among analytical laboratories, the
distribution of KQW values appears to approximate a normal distribution.
4484a
B-ll
-------
a:
Histisrss sf KOU N 5
I n t • r v r i
Midpoint Count
5,.50t °
4.25+
£.00+
5.7?+
H
"*
7
4
«
liALLQN l+HARRISOH +
n
0
+ + LABORATC
OTHER LAPS
FIGURE B.l Benzo(a)pyrene K_.. Data: Frequency Distribution.
u W
B-12
-------
ft:
Histos" sz.
of
KDw
Interval
Midpoint
Cc
3.26
1
*
:*
3.28
r.
3.30
n
¦If -If
3.32
1
3.34
0
7 TJ.
W • WW
c;
****!!:
3.38
*
m
3.40
2
*
3.42
0
3.44
4
3.4c
")
ft
3.48
1
*
b: frequency distributio?* py laboratory
r. - «AP" X 1IT! SON I"1C0AT
0 - INDIVirUAL~OTHER*LA?S
KQW
u
n
- - ">
3.430+ 2
3.360+ 3
_ o
+ + + + + + LABORATORY
GARS7 t UILSOM OTHER
FIGURE B.2 Napthalene K Data: Frequency Distribution.
Wn
B-13
-------
Hists2r3B of KQW N 5 --
Interval
Midpoint Count
*
mm**
m
*
5 > 2
1
*
5.6
4
6. 0
8
6.4
3
6,8
1
4
7
/ t m
0
1 L
¦ * w
1
FIGURE B.3 DDT KQW Data: Frequency Distribution.
Q_1 A
-------
These results are based on a very limited number of data points (maximum 24)
and chemicals (3) and are therefore tentative.. Further data are required to
validate these results for all the chemicals evaluated in this report.
K0c
As discussed in Section 1.2 of the main report, KQC measurements include
analytical error and potential bias; additionally, they reflect the role of
inter-laboratory variability in K measurements for chemicals. The effect
vv
of inter-laboratory variability will be evaluated for the four chemicals with
greater than ten measurements of KQC.
o Of the 15 K values reported for d1benzo(a,h)anthracene, 14 of the
measures are from Means et al. (1980) and one value is from Versar
(1984). The frequency distribution for the data is shown in
Figure B.4. The data do not appear to follow a normal
distribution—rather, they exhibit either a bimodal distribution or a
tendency to be skewed to the right. The figure also indicates that
Versar's single result lie* at the low end of the 14 measures made by
Means et al.
o The 31 reported values for the KQC of pyrene represent 17 values
from Karickhoff and coworkers (1979, 1981, 1985) coded with a "K" and
14 values from Means et al. (1980) coded with an "M." The frequency
distribution shown in Figure B.5 exhibits a highly right-skewed
and/or bimodal distribution of the observed values.
o The measurements of Arochlor 1254 KQC represent 14 observations
from a single laboratory, Weber et al. (1983). These values indicate
only replication (or analytical) error and cannot exhibit any effects
due to inter-laboratory differences. The data as presented in Figure
B.6 exhibit a near-normal distribution.
o The 14 values for the 00T Kqc represent data from eight
laboratories: six values from Gerstl and Mingelgrin (1984), and
with the remaining eight values from the seven other laboratories.
4484a
B-15
-------
p ' Hl2*. k G" N r i-
Irtt P "'-'ii
Count
5.9
1
*
5.?
**«
6.0
A
*
4.1
1
*
6.2
n
*•
n
6.3
A
6.4
5
*****
6.5
*
m
b: frequency distribution ^oratory^ ElM„
V » I'EP.SAR C19S42
6.50+ 2
K0C : I
6.25+ 2
- H
6.00+
n
3.75+ H
--+ + + + + + LABOs-*-'r
MEANS ET AL VEF3Ar
FIGURE B.4 D1benzo(a,h)anthracene Kqq Data: Frequency Distribution.
B-16
-------
A J Histesrsfc of r.OC N = 31
Ir.tsrvil
hidf-oint Count
4." 1 t
4.1 0
4.2 *
4.3 0
4.1 0
4.5 1 *
4.6 1 *
4.7 I ***
4.2 3 ********
4.? Z *****
1.0 2 *7
5.1 10 **********
b: frequency distribution by lab
K » KARICKHOFF ET AL [1979,1981>19853
M - MEANS ET At n®«02
KOC
"7
3
3
4,90+ K
4,55+
4.20+
K 4
3
3
K
K
K
+ + + +, + lABC-.i*l;
KARICKHQFF ET AL MEA*S ET AL
FISURE B.5 Pyrene KQC Oata: Frequency Distribution.
B-17
-------
Histosra® a? AROCHLQR
sGC [N s : - ¦
Interval
Midpoint Count
5.0
1
i.
*
»r n
0
5^4
0
5.6
0
5.9
3.0
**
T
t
6.2
5
*****
6.4
n
**
6.6
0
*
6.9
1
7.0
H
7.2
c*
FIGURE B.6 Arochlor 1254 K Data: Frequency Distribution.
Uw
B-18
-------
A*. HistoSr3# of f.OC N = ' 4
Interval
Midpoint
Count
4,6
1
*
•1,3
0
5.0
0
5.2
n
n
3.4.
4
nn
5.6
1
%
5.8
3
m
6.0
1
*
6.2
1
*
6.4
0
6 > 6
1
%
b: frequency distribution by laboratory
G ¦ GERSTL I MNGEL3RIN C19843
0 = OTHER IMDI'vIDUAL LAB
6.60+
KOC
6,00+ G
G
G
G
G
5.40+
4.30+
0
+ +- + a j. -+ LAPQRATOS:'?
GRIN
GES?HTfeEL! ' ' °™E*
FIGURE B.7 DDT KQC Data: Frequency Distribution.
B-19
-------
The data exhibit a fairly normal distribution (Figure 8.7) and, as
expected, the range of Kqc values from the seven other laboratories
is wider than that exhibited by the six observations from Gerstl and
Mingelgrin (1984).
The K „ data indicate that different laboratories give different
oc
estimates of the K for a given compound. Whether these differences can be
Vv
attributed to inter-laboratory differences or natural variability of the
sediments sampled cannot be determined from the available data. However, the
distribution of values tends to a more normal distribution as the number
oc
of laboratories that measured the K increases and/or balances in
oc
contribution. As with the K data, these results must be considered
ow
tentative because they are based on a limited number of data points. Further
data are required to validate these results for all the chemicals in this
report.
4484a
B-20
-------
3.2 REFERENCES
Action, F. S. 1966. Analysis of straight line data. John Wiley and
Sons, Inc. Mew York, Mew York.
Draper, N. R., and H. Smith. 1981. Applied regression analysis. John
Wiley and Sons, Inc. Mew York, New York.
Garst, J. E., and W. C. Wilson. 1984. Acute, wide-range, automated,
high performance liquid chromatographic method for the estimation
of octanol/water partition coefficients I: Effect of
chromatographic conditions and procedure variables on accuracy and
reproducibility of the method. J. Pharm. Sci. 73(11):1616-1623.
Gerstl, L., and U. Minglegrin. 1984. Sorption of organic substances
by soils and sediments. J. Environ. Sci. Health 319(3):297-312.
Kadeg, R. D., S. P. Pavlou and A. S. Duxbury. 1986. Elaboration of
sediment normalization theory for nonpolar hydrophobic organic
chemicals. Envirosphere Company, Bellevue, Washington.
Karickhoff, S. W. 1981. Semi-empirical estimation of sorption of
hydrophobic pollutants oh natural sediments and soils.
Chemosphere 10(8):333-846.
Karickhoff, S. W., and D. S. Brown. 1979. Determination of
octanol/water distribution coefficients, water solubilities, and
sediment/water partition coefficients for hydrophobic organic
pollutants. EPA-600/4-79-032, National Technical Information
Service, Springfield, Virginia.
Karickhoff, S. W., D. S. Brown and T. A. Scott. 1979. Sorption of
hydrophobic pollutants on natural sediments. Water Res. 13:241-248.
4963a
B-21
-------
Karickhoff, S. W., and K. R. Morris. 1985. Sorption dynamics of
hydrophobic pollutants in sediment suspensions. Environ. Tox.
Chem. 4:469-479.
Mackay, 0., A. Bobra and W. Y. Shiu. 1980. Relationships between
aqueous solubility and octanol-water partition coefficients.
Chemosphere 9:701-711.
Ma Hon, S. J. and F. L. Harrison. 1984. Octanol-water partition
coefficient of benzo(a)pyrene: measurement, calculation and
environmental implications. Bull. Environ. Contam. Toxicol.
32:316-323.
Means, J. C., S. G. Wood, J.J. Hassett and U. L. Banwart. 1980.
Sorption of polynuclear aromatic hydrocarbons by sediments and
soils. Environ. Sci. Technol. 14(2):1524-1528;
Rapport, R. A. and S. J. Eisenreich. 1984. Chromatographic
Determination of Octanol-Water Partition Coefficients for 58
Polychlorfnated Blphenyl Congeners. Environ. Sci. Technol.
18(3):163-170.
Versar, Inc. 1984. Chemical and toxicological review of priority
contaminants in nearshore tideflats and deepwater of Commencement
Bay, Washington. Oraft final report: Versar Inc., Springfield,
Virginia.
Weber, W. J., Jr., T. C. Voice, Massoud-Pirbazari, G. E. Hunt and D. H
Ulanoff. 1983. Sorption of hydrophobic compounds by sediments,
soils and suspended soils - II. Water Res. 17(10):1443-1452.
4963a
B-22
-------
APPENDIX C
CUMULATIVE PROBABILITY DISTRIBUTION PLOTS
4861a
-------
FIGURES
C.1A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Benzo(a)anthracene for: (A) Fixed Acute to Chronic
C.1B Toxicity Ratio of 10 and (8) Uniform Distribution of Acute to
Chronic Toxicity Ratio from 3 to 29 C-l
C.2A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Benzo(a)pyrene for: (A) Fixed Acute to Chronic
C.2B Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to
Chronic Toxicity Ratio from 3 to 29 C-2
C.3A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Chrysene for: (A) Fixed Acute to Chronic Toxicity
C.3B Ratio of 10 and (B) Uniform Distrioution of Acute to Chronic
Toxicity Ratio from 3 to 29 C-3
C.4A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Dibenzo(a,h)antiiracene for: (A) Fixed Acute to
C.4B Chronic Toxicity Ratio of 10 and (B) Uniform Distribution of
Acute to Chronic Toxicity Ratio from 3 to 29 . , C-4
C.5 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Fluoranthene C-5
C.6A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Pyrene for: (A) Fixed Acute to Chronic Toxicity
C.6B Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29 C-6
C.7A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Anthracene for: (A) Fixed Acute to Chronic
C.7B Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to
Chronic Toxicity Ratio from 3 to 29 . . . C-7
C.8A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Fluorene for: (A) Fixed Acute to Chronic Toxicity
C.8B Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29 C-8
C.9 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Naphthalene for Fixed Acute to Chronic Toxicity
Ratio of 29 C-9
4861a
1
-------
FIGURES (Continued)
C.lOA Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Phenanthrene for: (A) Fixed Acute to Chronic Toxicity
C.10B Ratio of 10 and (3) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29 C-10
C.11A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for Aldrin for: (A) Fixed Acute to Chronic Toxicity
C.11B Ratio of 10 and (8) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29 C-U
C.12 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Chlordane C-12
C.13A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for ODD for: (A) Fixed Acute to Chronic Toxicity Ratio
C.13B of 10 and (B) Uniform Distribution of Acute to Chronic Toxicity
Ratio from 3 to 29 C-13
C.14A Cumulative Probability Distribution Plots of Chronic Criteria
and Values for DDE for: (A) Fixed Acute to Chronic Toxicity Ratio
C.14B of 10 and (B) Uniform Distribution of Acute to Chronic Toxicity
Ratio from 3 to 29 c-14
C.15 Cumulative Probability Distribution Plot of Chronic Criteria
Values for DDT C-15
C.16 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Dieldrin C-16
C.17 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Endrin
C.18 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Heptachlor ... C-18
C.19 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Toxaphene for Fixed Acute to Chronic Toxicity Ratio
of 100. . . . . . . . , , . . . , . , , # . . C"19
4861a
ii
-------
BENZO(A)ANTHRACENE
I + + + + + + + + +*—+ + + +
*
*
*
*
0.500 + + + + +*+ + + + + + + +
*
g + +—*-+ + + + + + + * + + +
2 3 4 5 6 7
(log ug/goc)
Probability
Fractile
Chronic Criteria Chronic Criteria
(log ug/gQC)
(ug/gQC)
0.01
0.05
0.5
2.898
3.234
4.446
791
1,710
27,900
B
0.500 + +
*
*
*
*
+ + + * + + + +
*
*
*
*
3 4 5 6
(log ug/gQC)
—+—*-+——+
+ + +
+7
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria
(log ug/goc)
2.213
3.314
4.461
Chronic Criteria
(U9/90C)
163
2.060
28,900
FIGURES C.1A and C.IB Cumulative Probability Distribution Plots of Chroni
Criteria Values for Benzo(a) anthracene for: (A) Fixed Acute to Chronic
Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29.
-------
BENZO(A)PYRENE
Probability
Fractile
(log ug/gQC)
Chronic Criteria
(log ug/gQC)
Chronic Criteria
(ug/goc)
0.01
0.05
0.5
3.539
3.840
4.926
3,460
6.920
84,300
B
1 H +-
.500 + + +
*
+ +
+ +
.+— +
+ + +
0 + +—
2 3
—+—
4 5 6
(log ug/gQC)
7 + +8
Probability
Fractile
Chronic Criteria
(log ug/gQC)
Chronic Criteria
(ug/goc)
0.01
0.05
0.5
3.259
3.648
4.996
1,820
4,450
99,100
FIGURES 4.2A and 4.2B Cumulative Probability Distribution Plots of Chronic
Criteria Values for Benzo(a)pyrene for: (A) Fixed Acute to Chronic Toxicity
Ratio of 10 and (B) Uniform Distribution of Acute to Chronic Toxicity Ratio
from 3 to 29.
C-2
-------
CHRYSENE
1 + + + + +- + + *+ + + + + +
*
*
*
*
0.500 + + + +*+ + + + + + + + +
*
*
*
*
0 + *+.
2
Probability
Fractile
—+ + + +-
4 5
(log ug/goc)
Chronic Criteria
(log ug/gQC)
.+ + + + +
6 7 8
Chronic Criteria
("9/9oc)
0.01
0.05
0.5
2.525
2.834
3.946
335
662
8,830
B
0.500 +
.+ + + + + +-
*
*
*
*
+ + +*+ + +
*
*
*
+
0 ~-—*+ + + + + + +——+ + -+ -+
Probability
Fractile
0.01
0.05
0.5
(log ug/goc)
Chronic Criteria
(log ug/gQC)
2.084
2.647
3.975
Chronic Criteria
(u9/goc)
121
444
9,440
FIGURES C.3A and C.3B Cumulative Probability Distribution Plots of Chronic
Criteria Values for Chrysene for: (A) Fixed Acute to Chronic Toxicity
Ratio of 10 and (B) Uniform Distribution of Acute to Chronic Toxicity Ratio
from 3 to 29.
C-3
-------
DIBENZO(A,H)ANTHRACENE
1 +-
.+—*«+ +
0.500 + + + + + + +*+ + + + + +
*
0 + + *+-.
3 3.500
Probability
Fractile
0.01
0.05
0.5
4 4.500
(log ug/gQC)
Chronic Criteria
(log ug/gQC)
3.494
3.720
4.537
.+ + + + +
5 5.500 6
Chronic Criteria
(u9/goc)
3,120
5*250
34,400
B
1 +-—+--
0.500 + + + + + +*+ + +
*
+ +
3 4 5 6
(log ug/gQC)
7
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria
(log ug/goc)
3.090
3.543
4.573
Chronic Criteria
1,230
3,490
37,400
FIGURES C.4A and^C.4B Cumulative Probability Distribution Plots of Chronic
Criteria Values foruibenzo(a,h)anthracene for: (A) Fixed Acute to Chronic
Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29.
C-4
-------
FLUORANTHENE
500
O
£. 500
3. 500
(log ug/gQC)
4. 500
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria
(log ug/g QC)
2.201
2.486
3.514
Chronic Criteria
(ug/gQC)
159
306
3,270
FIGURE C.5 Cumulative Probability Distribution Plots of Chronic
Criteria Values for Fluoranthene.
C-5
-------
PYRENE
2 2.500
3 3.500
(log ug/goc)
4.500
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria
(log ug/90C)
2.257
2.431
3.057
Chronic Criteria
181
270
1,140
8
l +-"
0
> + + 4 +«
3 4 5
(log ug/goc)
0.500 + + + + +* + + + + *-*» + +
*
*
*
Probability
FractHe
0.01
0.05
0.5
Chronic Criteria Chronic Criteria
(log ug/gQC) (»9/goc)
1.928
2.272
3.078
8A.7
187
1,200
FIGURES C.6A and C.6B Cumulative Probability Distribution Plots of Chronic
criteria Values for Pyrene for: (A) Fixed Acute to Chronic Toxicity Ratio
SV M Uniform Distribution of Acute to Chronic Toxicity Ratio from
« to "
C-6
-------
ANTHRACENE
I ^ + + + + + + + + + + * +
*
*
*
*
0.500 + + + + + + +* + + + + + +
*
+ *H—
1 1.500
Probability
Fractile
-+ + + +-
2 2.500
(log ug/gQC)
Chronic Criteria
(log ug/gQC)
—+ + + +
3 3.500 4
Chronic Criteria
(ug/goc)
a. oi
0.05
0.5
1.508
1.744
2.597
32.2
55.5
395
B
—+ +——+ + . ~+ + -+-*—4 + +——+
*
*
*
*
1.500 + + + + + +* + + + + + + +
*
*
0 <4 +
0 1
-+- +-
2
.* +-
3
(log ug/gQC)
.+ +—-+ + +
4 5 6
Probability
Fractile
Chronic Criteria Chronic Criteria
(log ug/gQC)
(ug/goc)
0.01
0.05
0.5
1.277
1.579
2.645
18.9
37.9
442
FIGURES C.7A and C.7B Cumulative Probability Distribution Plots of Chronic
Criteria Values for Anthracene for: (A) Fixed Acute to Chronic Toxicity
Ratio of 10 and (B) Uniform Distribution of Acute to Chronic Toxicity Ratio
from 3 to 29.
C-7
-------
FLCJORQJE
-+ * + + +
0.500 + + + + +*+ + + + + + + +
*
*
*
*
0 +
1 +-•
0.500 +
1 1.500
2 2.500
(log ug/gQC)
3 3.500
Probability
Fractile
Chronic Criteria
(log ug/goc)
Chronic Criteria
(ug/90C)
0.01
0.05
0.5
1.192
1.408
2.185
15.6
25.6
153
+ + +
*
*
*
*
+ * + + +
*
*
+ + + ~ 4
*
*
1
2 3
(log ug/goc)
4 5
Probability
Fractile
Chronic Criteria
(log ug/goc)
Chronic Criteria
(ug/90C)
0.01
0.05
0.5
0.773
1.195
2.234
5.93
15.7
171
FIGURES C.8A and C.8B Cumulative Probability Distribution Plots of Chronic
Criteria Values for Fluorene for: (A) Fixed Acute to Chronic
Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29.
C-8
-------
NAPTHALENE
o.500 *
*
o +—
i. soo
2. 300
(log ug/goc)
3. 300
4. 30O
Probability Chronic Criteria Chronic Criteria
Fractile
0.01
0.05
0.5
(log ug/g QC)
1.697
1.856
2.429
(ug/goc)
49.7
71.7
268
Figure C.9 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Naphthalene for Fixed Acute to Chronic Toxicity Ratio of 29.
C-9
-------
PHE24AWTHRE2JE
1 H—
.+ + * +.
*
0.500 + + + +
0 +-
1.500
+ + + + + + + + +
*
*
.+ +- + + +
2 2.500
(log ug/gQC)
3.500
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria
(log "9/90C)
1.355
1.581
2.396
Chronic Criteria
("9/9,„.)
22.6
38.1
249
B
i —+—+—+—+—+—+—+—+—
+—+—+—
*
0.500 + + + + + * + + + + + + +
*
0 +——+ + + + -+ + -+ + + + J
(log ug/gQC)
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria Chronic Criteria
(log ug/goc)
1.042
1.387
2.435
(ug/90C)
11.0
24.4
272
FIGURES C.10A and C.10B Cumulative Probability Distribution Plots of Chron
Criteria Values for Phenanthrene for: (A) Fixed Acute to Chronic
Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
_ j .ji., n - -i - # j- -
-------
ALDRIN
0.500
1.500
2.500
(log ug/gQC)
Probability
Fractile
0.01
0.05
0.5
Chronic Criteria Chronic Criteria
(log ug/goc)
•0.0832
0.131
0.903
(ug/goc)
0.826
1.35
8.0
B
1 H -+ +——+ +——+ + -+-*—+ + + +~ +
*
0.500 + + *+~+ + + •*¦ + + + + +
*
*
*
*
——+———+-—+——+——+ +—-+ +-—+ -+—-
0
0 +-
-1
.+——+.
1
,+ +-
3
.+ + +
4 5
(log ug/gQC)
Probability
Fractile
Chronic Criteria Chronic Criteria
(log ug/goc)
(ug/goc)
0.01
0.05
0.5
-0.368
-0.074
0.932
0.429
0.843
8.55
FIGURES C.11A and C.11B Cumulative Probability Distribution Plots of Chronic
Criteria values for Aldin for: (A) Fixed Acute to Chronic
Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29.
C-ll
-------
CHLORDANE
(log ug/g0(.)
Probability
Fractile
Chronic Criteria
(log ug/g QC)
Chronic Criteria
(ug/goc)
0.01
0.05
0.5
-1.524
-1.247
-0.248
0.0299
0.0566
0.565
FIGURES C.12 Cumulative Probability Distribution Plots of Chronic
Criteria Values for Chlordane.
C-12
-------
DDD
I A + + + + + * + + + + + +
*
*
*
0.500 + + + +*+ + + + + + + + +
*
*
*
*
0 +-*--+—-+ + + + + + + + + + +
0
(log ug/gQC)
Probability
Fractile
Chronic Criteria
(log ug/gQC)
Chronic Criteria
(ug/goc}
0.01
0.05
0.5
0.305
0.594
1.636
2.02
3.93
43.3
B
i —+—+—+——+—+——*+_—+——+
*
*
*
*
.500 + + + + + +*~ + + + + + +
0 +—*+—-+——+ +-—+_.—
-10 12
(log ug/gQC)
.+ + +
4 5
Probability
Fractile
Chronic Criteria
(log ug/goc)
Chronic Criteria
(U9/90C)
0.01
0.05
0.5
¦0.207
0.336
1.690
0.621
2.17
49.0
FIGURES C.13A and C.13B Cumulative Probability Distribution Plots of Chronic
Criteria Values for 000 for: (A) Fixed Acute to Chronic
Toxicity Ratio of 10 and (B) Uniform Distribution of Acute to Chronic
Toxicity Ratio from 3 to 29.
C—13
-------
DDE
1 H + + + + + + + + + + +-*--+
*
*
*
.500 + + + + + + * + + + + + +
*
*
*
*
0 + *h + + + + + + + + + + +
0.500 1 1.500 2 2.500 3 3.500
(log ug/gQC)
Probability
Fractile
Chronic Criteria Chronic Criteria
(log ug/gQC)
(ug/goc)
0.01
0.05
0.5
0.749
1.024
2.017
5.61
10.6
104
B
(log ug/gQC)
Probability
Fractile
Chronic Criteria
(log ug/gQC)
Chronic Criteria
<«9/9oc)
0.01
0.05
0.5
0.431
0.777
2.061
2.70
5.98
115
FIGURES C.14A and C.14B Cumulative Probability Distribution Plots of Chronic
in116!!1/^ for DDE for: Fixed Acute t0 Chronic Toxicity Ratio of
io and (B) Uniform Distribution of Acute to Chronic Toxicity Ratio from 3 to
C-14
-------
DDT
0. SCO + +
#
*
*
*
0 + » — — — + ——— «4~~
-i.so -o.so
(log ug/gQC)
0. 500
1. 500
Probability
Fractile
Chronic Criteria
(log ug/g QC)
Chronic Criteria
("9/9oc)
0.01
0.05
0.5
-1.381
-1.185
-0.480
0.0416
0.0653
0.331
FIGURE C.15 Cumulative Probability Distribution Plot of Chronic Criteria
Values for DDT.
C-15
-------
DIELDRIN
l +-
0. 500
-t- * +¦
0
-1. SO
(log ug/gQC)
-l
-0. so
0
Probability
Fractile
Chronic Criteria
(log ug/g QC)
Chronic Criteria
(ug/goc)
0.01
0.05
0.5
-2.849
-2.646
-1.911
0.00142
0.00226
0.0123
FIGURE C.16 Cumulative Probability Distribution Plot of Chronic Criteria
values tor Dieldrin.
C-16
-------
ENDRIN
0. 300
-2. 50
-1. 50
-0. 50
(log ug/gQC)
Probability Chronic Criteria Chronic Criteria
Fractile (log ug/g QC) (U9/S0C)
0.01 -2.970 0.00107
0.05 -2.779 0.00166
0.5 -2.088 0.00817
FIGURE C.17 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Endrin.
C^17
-------
HEPTACHLOR
o. soo
++
o *—
-£. 50
-d
-1.30
-1
(log ug/goc)
-o. so
o
0. soo
Probability
Fractile
Chronic Criteria
(log ug/g QC)
Chronic Criteria
(u9/90c)
0.01
0.05
0.5
¦2.438
•2.222
•1.444
0.00365
0.00600
0.0360
FIGURE C.18 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Heptachlor.
C-18
-------
1 +
0. 500
TOXAPHENE
(log ug/gQC)
Probability
Fractile
Chronic Criteria
(log ug/g QC)
Chronic Criteria
(ug/gQC)
0.01
0.05
0.5
-3.887
-3.728
-3.155
0.000130
0.000187
0.000700
FIGURE C.19 Cumulative Probability Distribution Plot of Chronic Criteria
Values for Toxaphene for Fixed Acute to Chronic Toxicity Ratio of 100.
C-19
------- |