&EPA
United States
Environmental Protection
Agency
Environmental Research
Laboratory
Athens GA 30605
EPA-600/3-80-022
January 1980
Research and Development
Sediment-Pollutant
Relationships in
Runoff from Selected
Agricultural,
Suburban, and Urban
Watersheds
A Statistical
Correlation Study
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are.
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5, Socioeconomic Environmental Studies
6, Scientific and Technical Assessment Reports (STAR)
7 Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the ECOLOGICAL RESEARCH series. This series
describes research on the effects of pollution on humans, plant and animal spe-
cies, and materials. Problems are assessed for their long- and short-term influ-
ences. Investigations include formation, transport, and pathway studies to deter-
mine the fate of pollutants and their effects. This work provides the technical basis
for setting standards to minimize undesirable changes in living organisms in the
aquatic, terrestrial, and atmospheric environments.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/3-80-022
January 1980
SEDIMENT-POLLUTANT RELATIONSHIPS IN RUNOFF
FROM SELECTED AGRICULTURAL, SUBURBAN,
AND URBAN WATERSHEDS
A Statistical Correlation Study
by
Stanley W. Zison
Tetra Tech, Inc.
Lafayette, California 94549
Contract No. 68-03-2611
Project Officer
Charles N. Smith
Technology Development and Applications Branch
Environmental Research Laboratory
Athens, Georgia 30605
ENVIRONMENTAL RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U. S. ENVIRONMENTAL PROTECTION AGENCY
ATHENS, GEORGIA 30605
-------
DISCLAIMER
This report has been reviewed by the Environmental Research Laboratory,
U.S. Environmental Protection Agency, Athens, Georgia, and approved for
publication. Approval does not signify that the contents necessarily re-
flect the views and policies of the U.S. Environmental Protection Agency,
nor does mention of trade names or commercial products constitute endorsement
or recommendation for use.
ii
-------
FOREWORD
As environmental controls become more costly to implement and the penal-
ties of judgment errors become more severe, environmental quality management
requires more efficient analytical tools based on greater knowledge of the
environmental phenomena to be managed. As part of this Laboratory's research
on the occurrence, movement, transformation, impact, and control of environ-
mental contaminants, the Technology Development and Applications Branch
develops management or engineering tools to help pollution control officials
achieve water quality goals through watershed management.
Essentially all control technology for nonpoint sources is related to
erosion or sediment control. Several techniques are available today that
estimate nonpoint source pollutant loadings based on the assumption that
a reasonable correlation exists between various pollutants and sediment.
For these tools to be useful, sediment-pollutant correlations (potency
factors) must be developed to provide essential inputs into lumped-parameter
runoff models such as the EPA's Nonpoint Source Model (NPS) and Storm Water
Management Model (SWMM). This report provides statistical correlation
values for the potency factor, based on estimates from runoff data represent-
ing agricultural, suburban, and urban watersheds.
David W. Duttweiler
Director
Environmental Research Laboratory
Athens, Georgia
tii
-------
ABSTRACT
Data from agricultural, suburban, and urban watersheds were subjected
to statistical correlation analysis to estimate potency factors. These
factors are coefficients that, when multiplied by sediment mass emission
rates (transported in runoff), provide estimates of mass emission rates for
other pollutants. The potency factors are required input for such lumped-
parameter runoff models as the Nonpoint Source (NFS) Model and the Storm-
water Management Model (SWMM).
The data were also subjected to multiple regression analysis to examine
the effect of storm parameters on runoff water quality and the interrelation-
ship among runoff water quality constituent concentrations themselves (other
than sediment load). The multiple regression analysis was primarily explora-
tory with the objectives of explaining variance of water quality and identi-
fying important independent or predictor variables rather than developing
predictive expressions.
This report was submitted in fulfillment of Contract No. 68-03-2611 by
Tetra Tech, Incorporated, under the sponsorship of the U.S. Environmental
Protection Agency. The report covers the period September 20, 1977, to
September 19, 1978, and work was completed as of September 19, 1978.
1v
-------
CONTENTS
FOREWORD Hi
ABSTRACT iv
FIGURES vi
TABLES vili
ACKNOWLEDGEMENTS xi
1. INTRODUCTION 1
2. CONCLUSIONS AND RECOMMENDATIONS 3
CONCLUSIONS 3
RECOMMENDATIONS 4
3, BACKGROUND AND TECHNICAL ISSUES 6
SUSPENDED SOLIDS AND WATER QUALITY 7
PAST STUDIES 10
4. METHODOLOGY 28
GENERAL ANALYSIS PROCEDURE 28
5. RESULTS AND DISCUSSION 51
SIMPLE LINEAR REGRESSION 51
MULTIPLE REGRESSION 75
REFERENCES i . 112
APPENDIXES
A. REGRESSION ANALYSIS THEORY 114
B, RUNOFF WATER QUALITY ANALYSIS METHODS 126
-------
FIGURES
Number Page
1 Nitrogen availability pattern in fresh urban runoff
sample(s) ............................. 11
2 Pollutant variations with Q and time for storm no. 13,
date 3/16/72 ........................... 14
3 Pollutant variations with Q and time for storm no. 13,
date 3/16/72 ........................... 15
4 Pollutant variations with Q and time for storm no. 20,
date 6/20/72 ........................... 16
5 Pollutant variations with Q and time for storm no. 20,
date 6/20/72 ........................... 17
6 General data analysis procedure .................. 29
7 Sample plot showing output of simple linear regression program
used in this study ........................ 31
8 Soils and topography, watershed P-01 ............... 40
9 Soils and terrace configurations, watershed P-03 ......... 41
10 Soils and terrace configurations, watershed P-04 ......... 42
11 Buffalo Bill Watershed configuration and location ......... 44
12 Location and configuration of Michigan State University
test plots ............................ 47
13 Location map for Seattle catchments ................ 49
14 A comparison of statistical results among Southcenter, South
Seattle, and Viewridge 1 ..................... 70
15 Scatter plot and regression of iron concentration on suspended
solids concentration ..... ............... ... 73
16 Scatter plot of log^Q iron concentration on logjQ suspended
solids concentrations ....................... 74
v
-------
FIGURES (continued)
Number Page
A-l Schematic showing an interpretation of R2 118
vii
-------
TABLES
Number Page
1 Pollutant Fractions Associated with Particle Sizes ... 13
2 Total Suspended Solids and Organics Data Averaged for 36 Storm
Events and Ratios of Organics to Total Suspended Solids 18
3 Total Phosphorus, Kjeldahl Nitrogen, and Fecal Coliforms in
Runoff Averaged for 36 Storm Events and Ratios of Each to
Total Suspended Solids 20
4 Metals Concentrations (Ca, Co, Cu, Cr, Fe) Averaged for 36
Storm Events and Ratios of Each to Total Suspended Solids 22
5 Metals Concentrations (Pb, Ni, Mg, Mn, Zn) Averaged for 36
Storm Events and Ratios of Each to Total Suspended Solids 24
6 Regression of Log [N03 + N021 33
7 Significance of Entries in Table 6 35
8 Summaries for Data Bases Used in this Study 39
9 Buffalo Bill Watershed Flow and Rainfall Data 45
10 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown From Watkinsville, Georgia - Data are
From Plot P-04 . . 52
11 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown From Watkinsville, Georgia Test Plot P-03. . 53
12 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown. Data are From Watkinsville, Georgia Test
Plot P-01 54
13 Regression of Dissolved Species on Concentration in Suspended
Solids Using Data From Watkinsville, Georgia 56
14 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown From the Buffalo Bill Watershed, Iowa ... 58
viii
-------
TABLES (continued)
Number Page
15 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown From Michigan State University Study .... 60
16 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown From Viewridge 1, Seattle Data 61
17 Statistics for Correlations Between Turbidity and Constituents
Shown From Viewridge 1, Seattle Data 63
18 Statistics for Correlations Between Antecedent Dry Days and
Constituents Shown From Viewridge 1, Seattle Data 64
19 Statistics for Correlations Between Suspended Solids Concentrations
and Constituents Shown From the South Seattle Data Base 66
20 Statistics for Correlations Between Antecedent Dry Days and
Constituents Shown From South Seattle Data Base 68
21 Statistics for Correlations Between Suspended Solids and
Constituents Shown From Seattle Southcenter Data Base 69
22 Statistics for Correlations Between Suspended Solids and
Constituents Shown From the Honey Creek Watershed 72
23 Summary Listing of Potency Factors Estimated From Agricultural,
Suburban, and Urban Watersheds Examined in This StudyA ...... 76
24 Variables Examined in Multiple Regression as Candidate
Predictors ofRunoff Water Quality 78
25 Simple Correlations (r) Between Dependent and Independent
Variables for Watkinsville Plot P-04 Data 79
26 Multiple Regression Statistics forWatkinsville Plot P-04 81
27 Regression Equations With Confidence Intervals for Slope and
Intercept From Watkinsville Plot P-04 82
28 Simple Correlations (r) Between Dependent Variables (Trifluralin
and Diphenamid) and Independent Variables Data are From
Watkinsville Test Plots 84
29 Multiple Regression Statistics for Watkinsville Plots P-01 and
P-04 Data 86
30 Regression Equations with Confidence Intervals for Slope and
Intercept From Watkinsville Plots P-01 and P-03 Herbicide Data . . 87
ix
-------
TABLES (continued)
Number Page
31 Simple Correlations (r) Between Dependent and Independent
Variables for the Buffalo Bill Watershed 89
32 Multiple Regression Statistics for the Buffalo Bill Watershed ... 90
33 Regression Equations with Confidence Intervals for Slope and
Intercept From the Buffalo Bill Watershed Data Base 91
34 Simple Correlations (r) Between Dependent and Independent
Variables for Southcenter, Seattle 94
35 Multiple Regression Statistics for Southcenter, Seattle 95
36 Results of Validation-Generalization Using Data From
Southcenter and Coefficients From Viewridge 1 97
37 Regression Equations with Confidence Intervals for Slope
and Intercept From Southcenter, Seattle Data 98
38 Simple Correlations (r) Between Dependent and Independent
Variables (South Seattle) 100
39 Multiple Regression (t) Statistics and R2 for South Seattle
Data 102
40 Results of Validation-Generalization Using Data From South
Seattle and Coefficients From Viewridge 1 103
41 Regression Equations with Confidence Intervals for Slope and
Intercept From South Seattle Data 104
42 Simple Correlations (r) Between Dependent and Independent
Variables (Viewridge 1) 107
43 Multiple Regression (t) Statistics and R2 for Viewridge 1
(Seattle) Data 109
44 Regression Equations, with Confidence Intervals for Slope and
Intercept From Viewridge 1 (Seattle) Data 110
-------
ACKNOWLEDGEMENTS
The author would like to express his gratitude to the Environmental
Research Laboratory at Athens, Georgia, for its support of this study (EPA
Contract Number 68-03-2611) and in particular, to Messrs. Charles N. Smith
and Rudolf Parrish for their suggestions and editorial conments. Thanks also
go to Messrs. William B. Mills, Paul Johanson, Larry Woods, and Mrs. Mei-Chi
Hua for assisting with analyses; to Mrs. Bernice Bujacich and Ms. Susie Madson
for typing; and to Mrs. Anna Zison for drafting.
Finally, the author would like to thank Dr. Dave Baker (Heidelberg
College, Tiffin, Ohio), Mr. David M. Cline (EPA-ERL, Athens, Georgia), Mr.
Roger Splinter (Iowa State Hygienic Laboratory, Iowa City), and Dr. Wayne
Huber (University of Florida, Gainesville) for providing data.
xi
-------
SECTION 1
INTRODUCTION
In recent years, much attention has been paid the problem of point dis-
charges, their effects on receiving waters, and the study and simulation of
resulting water quality interactions. Less work has been done on the problem
of nonpoint loadings, despite the fact that stormwater runoff is frequently
the most significant determinant of receiving water quality. Stormwater run-
off can be important, for example, in rural areas where point sources are
minor and where intensive farming and other agricultural activities (e.g.,
feed-lot operation) lead to heavy concentrations of potential pollutants
spread over the land. In large urban areas, where point sources are commonly
very Important, nonpoint loads may also be substantial as they contribute lit-
ter, heavy metals, gardening wastes, and construction debris through the storm
sewers to the receiving waters.
Emphasis has commonly been placed on the impacts of point sources rather
than on nonpoint sources because:
Point discharges tend to be easier to identify than nonpoint,
and therefore draw public attention. Further, this has
brought them to the fore among environmental researchers.
t Point discharge characteristics tend to be easier to quantify
and are commonly less variable over time than are nonpoint
characteristics.
Reliable mechanistic predictive tools have been easier to
develop where only point sources are considered, since in
the overall modeling process, the point source represents
a boundary condition, while nonpoint loads must themselves
be estimated. This means one more level of simulationand
one of substantial difficulty and uncertainty.
The combined impact of the importance of nonpoint waste!oads and the
difficulties inherent in modeling their effects has led to development of
various storm runoff models such as the Environmental Protection Agency's
(EPA) Stormwater Management Model (SWMM) (1), EPA.'s Nonpoint Source Pollutant
Loading Model (NPS) (2), STORM (Water Resources Engineers/U.S. Army Corps of
Engineers) (3), EPA's Agricultural Runoff Management (ARM) Model (4), and
EPA's HSP-F, (5).
With respect to runoff water quality, these models, and others of this
class, may be categorized into those which attempt to mechanistically simulate
-------
soil and runoff water chemistry, and those which assume the mass emission rate
of each pollutant to be a simple function of sediment mass emission rate in
runoff from the watershed. ARM falls into the first category (4). It in-
cludes algorithms for simulating pesticide decay on the watershed such that
the availability of these pollutants to be washed off during a storm is a
function of time elapsed since application. Analogously, for nutrients, the
ARM model simulates soil storage and decay between storms, thereby mechanisti-
cally estimating quantities of pollutants available for washing off in each
storm being modeled.
In contrast, SWMM, STORM and NFS all simulate runoff water quality simply
as a linear function of total sediment transported, with appropriate multi-
pliers (potency factors) provided as input. The representations are of the
form:
C = P S (1)
where C = mass of pollutant washing off the watershed, any
mass units
P = potency factor converting S to C, units depending on
units of S and C. For use with the NPS model, P is
expressed as a percent. Therefore P, as presented
later in this report, must be multiplied by 100.
S = mass of sediment washing off the watershed, any
mass units
Recognizing that SWMM, NPS, and other similar models are popular and are
likely to continue in common use, EPA's Environmental Research Laboratory at
Athens, Georgia, funded a statistical study geared essentially toward defin-
ing the nature of relationships between runoff sediment loads and concentra-
tions of other water quality constituents. This report discusses results of
that statistical study and emphasizes observed correlations between runoff
water quality and suspended sediment. Secondarily, findings are also pre-
sented which relate runoff water quality to storm characteristics and certain
other variables. The former provides estimated potency factors for use with
such models as SWMM and NPS, and the latter can provide guidance in further
model and research study development.
-------
SECTION 2
CONCLUSIONS AND RECOMMENDATIONS
CONCLUSIONS
1. The temporal variance of suspended sediment concentration in storm runoff
can account for a relatively small proportion of the temporal variance of
nearly all other water quality constituents considered.
2. Watershed type appears to be very important in determining the reliabil-
ity of potency factors. Based upon the results of this study, potency
factors computed for urban runoff are more reliable than those developed
for suburban, rural, and agricultural areas.
3. There is a very substantial variability of potency factors among agricul-
tural watersheds representing diverse geographical and climatological re-
gions.
4. Within the single urban area examined (Seattle) the potency factors are
quite similar from sampling station to station for a number of pollu-
tants. This suggests possible similarities among pollutant sources, and
deposition and transport phenomena as well.
5. When factors in addition to suspended sediment load are taken into con-
sideration, explication of runoff water quality improves significantly.
Factors which proved important include number of dry days preceding
storms, time elapsed since the beginning of storms, cumulative rainfall
since the beginning of storms (the latter two sometimes simultaneously
important), time since application of chemical (agricultural watersheds),
and elapsed time since some reference date (suggesting longterm trends).
6. The fact that time since the beginning of storms is sometimes significant
even after cumulative storm rainfall is accounted for suggests that wet-
ting rates may be important in the temporal profile of runoff water qual-
ity from a watershed. Additionally, it may be that some pollutants do
not reside at the very surface, and some amount of surface material must
be washed off before those pollutants are exposed to the dislodging and
transporting action of raindrop impingement and overland flow.
7. In some cases, the correlation between suspended solids and dry days pre-
ceding, storms is relatively weak. In many of these same cases, the cor-
relation between suspended solids and certain water quality constituents
is also weak, while the direct correlation between dry days preceding
storms and runoff water quality concentrations is substantially stronger.
-------
Both SWMM and the NFS model simulate dust and dirt accumulation as a
function of dry days preceding storms, and compute water quality as a
function of sediment transported.
8. Correlations among water quality constituents are generally strong rela-
tive to other correlations examined. In some cases, however, correla-
tions are lower than might be expected (e.g., suspended solids concentra-
tion with turbidity).
9. Rainfall intensity, lagged 0, 1, and 2 hours prior to the time of water
quality sampling, is a consistently poor predictor of runoff water qual-
ity. This may be due simply to inadequate data, however, rather than to
a lack of real relationships between rainfall intensity and runoff water
quality. Data to examine temporally lagged rainfall intensity were se-
verely limited in this study,
10. In the development of the potency factors reported here, a simple linear
model was fitted to the available data. This model included a constant
(y-intercept) which was not forced to zero. Also, in many instances fit-
ting the linear model without transforming the data would have resulted
in clustering of points, unacceptable weighting of selected data, and
biasing of computed statistics. Transforming the data to a log scale was
generally found to be an acceptable solution. However, both the exis-
tence of a non-zero y-intercept and the use of a log transformation mean
that some of the potency factors reported here are only approximations to
those required by such models as SWMM and NPS.
RECOMMENDATIONS
Modeling nonpoint pollution transported in storm runoff is a difficult
task at best, based upon the findings of this and innumerable other studies.
The difficulties stem, in part, from the complexities of sediment and pollu-
tant transport phenomena. Perhaps even more important, however, is the dif-
ficulty of satisfactorily describing, in mathematical or modeling terms, any
real watershed of significant size. It would be easy to recommend, based upon
the results obtained in this study, that much further research be done to
characterize pollutant deposition and transport phenomena, especially in terms
of accounting for the spatially variable nature of potency factors. Yet, it
seems clear that the costs of an effective program would be very large, par-
ticularly for such studies of non-urban watersheds. In such watersheds, there
are almost certainly too many interdependent factors involved which influence
runoff chemistry. These have too broad a range of values from watershed to
watershed to make practical a few comprehensive pollutant washoff studies of
universal validity. Considering the large and small scale complexities of
real watersheds, the interaction of factors affecting runoff water quality,
and the fact that the impact of watershed topographic features is strongly
dependent upon the spatial orientation and positioning of those features, the
model resolution needed to realistically represent a watershed spatially seems
prohibitive.
-------
It appears, then, that modelers must be content to represent large water-
sheds in the grossest of terms using lumped-parameter models, and to consider
only spatially and temporally averaged phenomena. Given that such is the
case, funding agencies and the research community must decide just where ad-
vances in the state of the art can realistically be expected. The primary
recommendation of this study is that the real need for accuracy in nonpoint
waste load estimation be carefully assessed and that future directions in non-
point modeling be very carefully planned. As far as potency factors are con-
cerned, this would seem to emphasize urban watersheds, followed by suburban,
rural, and agricultural watersheds. Further, the; practicality, value, and
cost of developing improved models should be weighed against their potential
benefits, the costs of field sampling and direct watershed characterization on
an individual basis, and the acceptability of statistical, empirical, or sim-
plified computational procedures for nonpoint loading estimation.
Specific recommendations regarding potency factors as input to the NPS
model or SWMM are as follows:
1. In applying potency factors to non-urban watersheds, modelers should be
especially cognizant of the site-specific nature of the potency factors
as well as the large degree of uncertainty in the values. The latter is
indicated by various statistics presented in the text.
2. Where the y-intercept is non-zero, either the particular model should be
altered to accommodate this or at least the possibility of error intro-
duced by ignoring the constant should be recognized. Where the potency
factor has been developed using log-transformed data, the model might
also be modified, or a piecewise-linear approximation might be used.
3. In view of some of the findings outlined above under "Conclusions," item
7, the modeler might consider altering the code for selected constitu-
ents. In some cases (see pertinent parts of the text) it may be appro-
priate to model water quality directly as some function of antecedent dry
days. Although this is less appealing conceptually than predicating pol-
lutant concentrations on sediment loads, correlation results suggest it
may allow for greater accuracy.
4. Modelers should recognize that the fundamental assumption underlying the
concept of potency factors is some reasonably consistent relationship be-
tween quantity of sediment transported and concentration of other con-
stituents. Even if one ignores differences in particle sizes, suspended
sediment surface area-to-mass ratios, sediment composition, and the fact
that a given pollutant may not, in fact, be physically associated with
particulate, and even if one assumes such a relatively consistent rela-
tionship between suspended sediment and pollutant concentration, it is
still possible that there may be little or none of the pollutant avail-
able to wash off. At the other extreme, the pollutant may be present in
such amounts as to swamp available sediment sorption sites. The "result
is a large error in the potency factor even under such ideal assumptions.
The modeler, therefore, should recognize the conceptually tenuous nature
of the potency factor itself, and use it accordingly.
-------
SECTION 3
BACKGROUND AND TECHNICAL ISSUES
The chemical characteristics of stormwater runoff are influenced by many
different phenomena which are manifested from raindrop nucleation in the at-
mosphere to the point where the droplet ultimately intercepts a receiving
water body. The phenomena include:
Scrubbing of atmospheric pollutants,
Surface impingement and suspension of particulate
matter,
Dissolution of chemical species during and following
impingement on a surface, and
Chemical transformations in the aqueous phase.
Atmospheric scrubbing determines the chemical characteristics of the
raindrop to the point of impaction with the earth's surface. Pollutants
washed out and contained within the raindrop can include particulate matter,
sulfuric acid, nitrate and nitric acid, complex organic substances, traces/of
hydrocarbons, and other substances introduced by local atmospheric conditions.
Upon and following surface impingement, a vast array of substances is avail-
able for transport. Much of it is mineral (clays, silts), while a substantial
portion is organic. Depending upon the site of impaction and the path of run-
off, material transported may include significant amounts of animal wastes,
litter, chemicals of anthropogenic origin, natural atmospheric fallout, soil
particles, and plant debris. In transit to a receiving water channel, chemi-
cal and physical transformations also occur. Although biochemical decay is
generally considered unimportant given the time scale involved, physical
effects, such as disaggregation of clumps of matter, sorption, and chemical
phenomena (e.g., complexation, precipitation, and dissolution) can all be im-
portant determinants of runoff water quality.
This study, being a statistical rather than mechanistic investigation,
and using actual field data, directly or indirectly involves virtually all of
the phenomena which affect runoff water quality. The study specifically ex-
amines relationships between sediment loads and runoff water quality, which is
the major emphasis, and then examines certain other variables (e.g., time
since beginning of storm, cumulative rainfall) hypothesized to influence the
chemical composition of the runoff.
-------
SUSPENDED SOLIDS AND WATER QUALITY
Potency Factors
Potency factors are used in the NFS and SWMM models as described in
equation 1. In later sections of this report, P is denoted as m, which is a
computed approximation to the true value of P. It is based upon concentra-
tions of pollutant and sediment in storm runoff rather than actual mass
emission rates. The way in which "S" in equation 1 is defined is important
in interpreting the meaning of potency factors. Commonly, in applications of
the models to real watersheds, mass emission data are unavailable, and the
modeler finds himself calibrating with data representing total concentrations
(best), concentrations of dissolved species, or possibly even concentrations
in the suspended solids fraction. The significance of any potency factor
depends upon which of these is being used. If, for example, the calibration
is made using dissolved lead values, there are two major determinants of the
potency factor values. One is the partition coefficient for lead between
suspended solids and the aqueous phase. The other is the amount of lead
available to wash off the watershed relative to the amount of runoff trans-
porting the lead. Thus, where the partition coefficient favors the solid
phase and there is little lead to be washed off, the value of P in equation 1
for dissolved lead could be extremely low. It is clear that an increase in
the amount of available lead, a change in the partition coefficient favoring
lead in the aqueous phase, or both, would increase P. The same is true if the
suspended fraction lead concentration were of interest, only with the change
in the partition coefficient favoring the suspended sediment. In contrast,
however, if total lead is used to estimate P, the partition coefficient is
irrelevant, while the amount of lead available on the watershed to wash off
still influences P.
The fact that the amount of any substance available to wash off strongly
affects P means that P must inherently be only as consistent as is the rate of
pollutant accumulation on the watershed. Since it is clear that the rate of
accumulation of many pollutants on watersheds is quite variable (e.g., nu-
trients and pesticides in agricultural lands, salt on city streets) P for
those pollutants must similarly be quite variable.
With respect to the reliability of simulations of suspended or aqueous
fraction pollutant levels, the partition coefficient is important to the ex-
tent it is variable. As will be described later, studies have suggested that
the partition coefficient is quite variable for many kinds of pollutants in-
cluding nutrients.
Given these sources of variability, and where P is to be estimated from
field data, P for dissolved species might be considered conceptually as:
CB
PB = p . S . A or CB = PB . S . p A (2)
where Pg = potency factor for pollutant B assumed to be 50% in
aqueous phase and maximally available to wash off
-------
Cg = concentration of dissolved species B
S = concentration of suspended matter
p = partition coefficient (CSecn;ment/cB) for B distributed
between the suspended matter and the aqueous phase.
A = Availability factor, maximum value = 1 where the
greatest amount of B has accumulated to wash off.
Where none has accumulated, A approaches zero.
The fact that the potency factor is not ordinarily estimated as in equa-
tion 2that is, p and A are unavailable and represent random variablesimeans
that P determined from field data must be proportionately as variable as phe-
nomena represented by A and p. Since A is important regardless of whether
dissolved, suspended, or total pollutant is estimated, and since A is likely
to be quite variable, PB computed from the pollutant concentration and the
suspended sediment load must be subject to very substantial error as well.
Pollutant Transport
Relationships between suspended solids and water quality involve the
following.
Transport as particulate
0 Dissolution, precipitation, and volatilization
Sorption phenomena
/
Transport as Particulate
To varying degrees, water quality constituents of interest may exist as
colloidal or larger compositionally homogeneous masses and are then trans-
ported as a fraction of the suspended solids. Examples include coliforms, BOD
(biochemical oxygen demand), and naturally occurring mineral matter. Addi-
tionally, where surface dumping of waste chemicals occurs, virtually any low
solubility species may be transported long distances this way.
Dissolution, Precipitation, and Volatilization--
At any point in time, the aqueous portion of runoff tends to approach a
state of equilibrium among the rates of dissolution of each chemical species,
of removal to the solid phase, and of exchange with the atmosphere. Because
conditions are continually changing with location, equilibrium is not likely
to be attained for any significant period of time, and the process is one of
continuous changes in the rate and direction of approach to concentration
equilibrium.
The rate of approach to equilibrium is ordinarily influenced by such
factors as the instantaneous concentration of the dissolved form, the avail-
ability and nature of the substance in the solid phase and the particle sizes
8
-------
involved, the volatility of the substance, the availability of other chemical
species and ions with which a precipitate may form, the turbulence and rate
of flow and resulting rates of mass advection-diffusion, and temperature. The
equilibrium concentration itself is mainly dependent upon the particular chem-
ical species involved, the presence of other species, and the pertinent solu-
bility products or stability coefficients. These, in turn, are a function of
temperature, but also .may be influenced by pH and redox potential.
Precipitation, it should be recognized, is not a simple process of solid
matter coming out of solution and settling or being transported in pure form.
In natural systems, several phenomena may be of importance with respect to
precipitation and transport in the solid phase. These include adsorption
(discussed below) and occlusion. Occlusion is the entrapment of water within
a precipitate or floe which may result in any substance transported being
associated with the suspended matter fraction.
Sorption Phenomena--
The term "sorption" refers to the physical association of a chemical
species with particulate matter. Where particulates are of small size and
concentrations are high (as in water turbid with clay) such substances as
pesticides and heavy metals may be associated with suspended solids because
of surface adsorption. Adsorption, it should be noted, may be due to elec-
tronic charge, as in ions adsorbed on the surface of a crystal lattice, or
due to other kinds of affinities, as in lipophilic substances adsorbing onto
organic floes. As is true for dissolution-precipitation, adsorption in run-
off represents a dynamic equilibrium.
Implications for This Study
This study is mainly concerned with the degree to which levels of chemi-
cal substances transported in runoff may be correlated with suspended solids
loads, and the development of potency factors for use in mathematical model-
ing. Potency factors, as used in such models as NPS, SWMM, and STORM, may be
interpreted as coefficients which when multiplied by the suspended solids con-
centration, yield concentrations of other species, such as coliforms, nitrate,
and BOD. Many studies have provided data with which potency factors can be
computed. The potential reliability of each factor, and as a consequence,
the reliability of runoff water quality simulations, however, does not depend
upon the average degree to which each constituent is associated with suspended
matter, but upon the amount of variation in the relationship. Coliforms, for
example, might be expected to associate with the suspended fraction. Coli-
forms themselves are particulate, of course, and if appropriate techniques are
used, can always be removed from the aqueous phase. Nevertheless, corre-
lations between conforms and suspended matter are far from perfect as will be
discussed in more detail below. This may reflect e/rors of measurement,
differences in assay technique, nonuniformity of coliform availability for
washoff, differences in die-off rates, and a host of other factors.
At the other end of the spectrum, nitrate can be expected to associate
with the aqueous phase rather than with the suspended matter, based upon con-
siderations such as the Peneth-Fajans-Hahn Law and the high solubility of
-------
virtually every inorganic nitrate species. The Law states that where at least
two types of ions are present, and other factors being equal, the ion forming
a compound of lowest solubility with the crystal lattice ions will be prefer-
entially adsorbed as will ions with highest charge and those closest in ionic
radius to the ion which would normally be at the particular lattice location.
With respect to nitrate, correlations may be expected to be poor and erratic,
and potency factors are likely to be very unreliable,
Finally, somewhere between the two extremes is BOD. A variable degree
of association of BOD with suspended matter can be expected based upon the
fact that much BOD can be present as organic particulate*whereas a substantial
component may also exist as dissolved organic matter (e.g., organic acids).
A number of studies have been performed in the past, in part geared to
determining degrees of association of various water quality constituents
(especially nutrients) with suspended matter, although these studies commonly
do not have as a specific objective the development of potency factors. In a
report on an application of the NPS Model to nutrients in runoff, Donigian and
Crawford (6) state that:
"In summary, the literature appears to indicate sediment can be
used as a reasonable indicator of nutrient loadings by surface
runoff from agricultural and urban lands. However, soluble
nutrient losses can be significant in watersheds where subsur-
face discharge is a major component of the runoff."
That the literature actually supports the contention that sediment loads are
a reasonable indicator of nutrient loads can certainly be debated. Further,
the literature suggests that sediment levels are, at best, only a fair index
of loads of pesticides, heavy metals, BOD, and other substances which migjrt
be expected to associate with particulate matter.
PAST STUDIES
Urban Storm Runoff
Cowen e£ al. (7) investigated nitrogen availability in urban runoff.
Findings showecTthat for organic nitrogen the relative portions present as
soluble and as particulate nitrogen varied substantially. Based upon the
data represented in Figure 1, from about 0 percent (B-8) to about 80 percent
(D-8) of total organic nitrogen is in the aqueous phase. Whereas the samples
within a particular letter designation represent different samples from a
single site (e.g., A-6, A-8, A-9, A-12 are from site A, a low-density residen-
tial area), it is also clear that variability was high within sites (see e.g.,
A-6 versus A-8 and D-8 versus D-10). Cowen ejt al_. (7) point out in general
terms, however, that of 13 samples, 10 had more particulate organic nitrogen
than soluble organic nitrogen.
It is important to note here that even if the relative proportions of
soluble and particulate organic nitrogen were essentially constant, this does
not provide information about the ratio of nitrogen to total sediment, which
10
-------
100
90
80
Z 70
_J
^f
£6°
|_
u.
O 50
1
H
^y
UJ 40
O
cr
UJ
0- 30
20
10
.
-
-
-
-
-
-
^
\
\
\
\
S
\
s
s
x
\
S
X
\
\
\
X
\
X
s
\
\
\
\
_
~- »
\
\
\
s
s
\
\
X
\
\
s
X
,\
s
\
X
s
\
1*
\
X
X
X
X
\
\
S
X
\
X
\
:
-r
\
X
X
\
v
X
S
X
X
s
\
S
X
X
X
"X
X
\
S
-,
::':
s
\
\
\
\
\
N
\
\
X
\
,x
\
\
X
X
X
\
s,
X,
.X
"x
'x
s
\
\
X
\
X
N
\
S
s
X
S
x
X
x,
X
x
x
X
'x
X
X
X
V
x,
\
x.
'.-
±
\'
'-\
£
'.'
*!
'[
__
-,
X
x,
x,
X
X
vv
x,
X
X
X
X
X
X
X
X
;
':
:
^
s
\
X
\
\
\
s
\
\
\
\
S
\
\
\
\
\
J
Y//////////////////////
I*'
'.
:'.,
:
j:
v
\
\
s
\
\
\
\
s
X
\
s\,
s
s
\
s>v
"^
71
T
T
\
S
s
s
x
X
'-.
".-
I;.
;.*
N
N
D
Q
A-6 A-8 A-9 A-12 B-6 B-8 B-9 B-12 D-6 D-8 D-IO D-12 F-9
RUNOFF SAMPLE NUMBER
Site A = low density residential area, B = medium density area,
D = University of Wisconsin-Madison campus, F = recently sodded
residential area previously under construction.
Figure 1. Nitrogen availability pattern in fresh urban runoff sample(s)
(Redrawn after Cowen et al. (7))
-------
is the basis for potency factors. Cowen ert al_. (7) had not intended to ad-
dress this issue. The results do suggest, however, that the nature of nitro-
gen species transported from urban sheds and/or the nature of the sediment
tend to be quite variable.
The study by Cowen ert al_. (7) suggests how variable the partition coef-
ficient can be, and thus has major implications for predicting dissolved or
suspended species concentrations. Cowen and Lee (8) have performed a parallel
study to that of Cowen e_t cil_. (7) examining the availability of phosphorus in
urban runoff-transported particulate. Results showed that in the 44 samples
taken, particulate phosphorus represented from 13 to 97 percent of total
phosphorus.
Sartor et al_. (9) have studied runoff quality in 12 U.S. cities having
populations ranging from about 13,000 to about 900,000. Table 1 shows data
from the study associating various pollutants with particle size ranges. Al-
though Sartor et al. (9) point out the importance of the fine silt-like par-
ticulate fractTorfTn that it represents a very small portion of the total
solids (TS) but accounts for large portions of other contaminants, they do
not discuss the variability of their data within or between the study sites.
Colston (10), in a report on urban land runoff characterization and
treatment, has presented data which more directly address the issue of potency
factors. Colston monitored storm activity and runoff for 36 storms on the
Third Fork Creek Watershed within the City of Durham, North Carolina. Figures
2 through 5 show some of his results. Note that within a given storm event,
the curve of total suspended solids (TSS), which is shown on all plots, cor-
relates, at least upon visual inspection, fairly well with certain other
variables (notably fecal coliforms, total phosphorus, iron, and lead), but
correlates very poorly with some others. For example, the substantial undu-
lations in the curves of TSS do not very well match those of TOC (total or-
ganic carbon), BOD (biochemical oxygen demand), or calcium.
In his report, Colston (10) also presents runoff quality for 36 storms
averaged by storm. These data, which represent flow at the downstream ter-
minus of the watershed, are presented in Tables 2 through 5. Also shown in
the tables are ratios of each pollutant to total suspended solids. The data
are representative of the total runoff from the watershed studied since gaug-
ing and sampling were done at the downstream watershed terminus. Tables 2
through 5 show the variability of the ratios of each water quality character-
istic to suspended sedimentthat is, the potency factors. Note in Table 2,
for example, that the ratios of 36 storm means of COD, TOC, and BOD to TSS
vary by a factor of about 20. Note, also, that concentrations of each of
these are substantial, suggesting such variability is not entirely the result
of errors of measurement. Similarly, Tables 3, 4, and 5 exhibit high vari-
abilities for ratios of means of the various water quality parameters to the
mean of total suspended solids.
With regard to this variability, three additional points should be con-
sidered. First, 521 separate samples were taken over the 36 storms, averaging
between 14 and 15 samples per storm. Because the ratios shown in Tables 2
through 5 represent ratios of means, variability among the ratios will probably
12
-------
TABLE 1. POLLUTANT FRACTIONS ASSOCIATED WITH PARTICLE SIZES (9)
Measured Pollutant
TS
BOD5
COD
VS
Phosphates
Nitrates
Kjeldahl nitrogen
All heavy metals
All pesticides
PCB
Fraction of Total (% by weight)
<43 urn
5.9
24.3
22.7
25.6
56.2
31.9
18.7
43 ym-246 urn
37.5
32.5
57.4
34.0
36.0
45.1
39.8
51.2
73
34
>246 ym
56.5
43.2
19.9
40.4
7.8
23.0
41.5
48.7
27
66
13
-------
KEY'
2200
o>
1800
Q 1400
tr 1000 -
UJ
O 600 F-
o
200 -
0
o
o
D
A
Total Solids
Total Susp. Solids
Total Volatile Solids
Volatile Susp. Solids
1700 1900 2100 2300
TIME (hrs)
oioo
700
600
6
^ 500
O
g* 400
2 300
cr
S 20°
o
o
° 100
KEY'
BOD
Fecal Coliform
COD
TOC
Total Suspended
Solids x 3
1700
J900
2100
TIME (hrs)
2300
OIOO
Figure 2. Pollutant variations with Q and time
for storm no. 13, date 3/16/72.
(Redrawn after Colston (10))
14
-------
KEY'
Manganese
K- Nitrogen
Zinc
Lead
Total - P
Total Suspended
Solids x I02
1700
1900
2100
TIME (hrs)
2300
0100
36
5 20
QL
Z 12
UJ '*
O
z
O 4
O
1700
KEY=
A
1
1900
Iron
O Magnesium
O Calcium
+ Total Suspended
Solids x 50
2100 2300
TIME (hrs)
0100
50
40
30
20
10
Figure 3. Pollutant variations with Q and time
for storm no. 13, date 3/16/72.
(Redrawn after Colston (10))
15
-------
2700
E
V.
6 2100
c
° 1800
,§ '500
z
O (200
(E 900
Z
UJ
2* 600
O
O
300
KEY'
Fecol Col if or m
Total Susp. Solids
Total Volatile Solids
Total Solids
Volatile Susp. Solids
i i i i i r
0500 0700 0900
i r
noo
1300
TIME (hrs)
KEY'
700 -
COD
BOO
O TOC
a Total Suspended
Solids x 4
0500
0700 0900 1100
TIME (hrs)
1300
Figure 4. Pollutant variations with Q and time
for storm no. 20, date 6/20/72.
(Redrawn after Colston (10))
16
-------
KEY'
D Iron
A Manganese
O Total P
O Total Suspended
Solids x I02
0500
T 1 1T I
0700 0900 1100
1300
TIME (hrs)
KEY'
D K- Nitrogen
A Lead
O Total Suspended
Solid* x I03
0500 0700 0900 1100
TIME (hrs)
1300
Figure 5. Pollutant variations with Q and time
for storm no. 20, date 6/20/72.
(Redrawn after Colston (10))
17
-------
TABLE 2. TOTAL SUSPENDED SOLIDS AND ORGANICS DATA AVERAGED FOR 36 STORM EVENTS
(DATA OF COLSTON (10)) AND RATIOS OF ORGANICS TO TOTAL SUSPENDED SOLIDS
Storm
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Total Suspended
Solids (TSS)
mg/1
Avg.
-------
TABLE 2 (continued)
Storm
Number
22
23
24
25
26
27
28
29
30
31*
32
33
34
35
36
Total Suspended
Solids (TSS)
mg/1
Avg. a
2332
S54
2889
3913
2522
1024
1326
1340
83
777
1246
1463
1029
643
1090
290
1266
2204
2434
376
624
1100
62
788
550
923
288
202
COO
rag/l
Avg. a
402
96
348
187
184
253
140
142
157
132
110
93
374
289
92
430
52
198
79
80
232
60
59
69
83
77
28
103
101
31
TOC
mg/1
Avg. a
165
26
94
48
50
51
21
38
44
49
34
38
105
99
31
148
9
41
14
18
41
11
16
13
15
10
14
35
19
14
BOO
mg/1
Avg. o
73
100
80
16
220
41
--
138
182
80
49
50
100
10
5
79
2
10
24
15
60
74
~
20
12
20
~
COD/TSS
.17
.17
.12
.05
.10
.14
.11
.12
1.59
.14
.07
.26
.28
.14
Range:
.05 + .95
.95/.05=19
TOC/TSS
.07
.05
.03
.01
.02
.02
.03
.03
.59
.04
.03
.07
.10
.05
.01 -i. .25
.25/.01=25
BOD/TSS
.03
.18
.03
__
.06
.02
.10
.14
.96
.04
.03
.10
<.01 + .20
.20/. 01=20
Mote that storm 31 has not been considered a* 1t nay represent erroneous values.
-------
TABLE 3. TOTAL PHOSPHORUS, KJELDAHL NITROGEN, AND FECAL COLIFORMS IN RUNOFF
AVERAGED FOR 36 STORM EVENTS (DATA OF COLSTON (10)) AND RATIOS
OF EACH TO TOTAL SUSPENDED SOLIDS
ro
o
Storm
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Total P
mg/1
Avg. a
0.28
0.6
1.03
0.47
1.05
0.75
1.07
0.58
0.5
0.57
1.05
0.81
1.98
--
1.03
0.56
0.56
0.62
1.09
1.17
0.04
0.21
0.35
0.13
0.86
0.31
1.05
0.24
0.43
0.26
0.36
0.22
4.65
0.79
0.23
0.29
0.34
0.93
0.55
K-Ni trogen
mg/1
Avg. a
0.91
9.52*
9.68*
1.30
1.94
0.67
0.55
0.56
0.59
0.76
0.65
0.6
~
0.67
0.44
0.31
0.82
0.88
0.49
0.09
4.01*
3.57*
0.17
0.49
0.59
0.31
0.46
0.12
0.30
0.15
0.21
--
0.58
0.2
0.03
0.12
0.29
0.25
Fecal Coli forms
#/ml
Avg . o
203
398
387
689
106
74
54
67
102
137
143
161
4
442
98
148
104
246
111
68
39
48
28
63
71
--
43
51
2
387
83
Total P/
TSS(xlO-3)
3.15
2.19
6.32
-.
3.03
1.58
0.73
0.47
0.29
1.00
1.06
5.55
2.88
0.40
0.37
0.66
0.69
1.22
0.43
Kjeldahl N/
TSS(xlO-3)
10.2
--
58.4*
3.76
4.09
0.46
0.45
0.32
1.03
0.77
4.45
0.87
0.26
0.29
0.37
0.91
0.98
0.18
Fecal
Coli forms/
TSS(xlO-3)
__
1245
1118
1454
73
60
31
117
699
199
_.
__
55
106
5
492
no
~
-------
TABLE 3 (continued)
ro
Storm
Number
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Total P
mg/1
Avg. o
1.13
0.44
1.42
0.73
0.85
0.36
0.71
.71
0.92
0.6
1.54
0.59
0.39
0.19
0.37
0.23
0.47
0.07
0.21
.24
0.62
0.18
0.51
0.13
K-N1trogen
mg/1
Avg. o
0.64
0.25
0.34
0.35
0.48
0.33
2.07
.43
0.34
0.2
0.70
0.48
0.33
0.08
0.25
0.11
0.19
0.14
1.28
.13
0.05
--
0.05
0.13
0.18
Fecal Coliforms
I/ml
Avg. o
175
280
549
~
172
248
94
273
44
258
94
125
363
--
221
65
31
168
21
~
105
Total P/
TSS(xlO-3)
0.48
0.79
0.49
~
0.34
0.35
0.54
0.53
11.08
~
0.48
1.05
0.92
Range:
0.29-* 11.08
11. 08/. 29= 38. 2
Kjeldahl N/
TSS(xlO-3)
0.27
0.45
0.12
0.19
0.32
1.56
0.32
4.10
--
0.16
0.48
--
0.75
0.12 -» 10.2
10.2/. 12=85.0
Fecal
CoHforms/
TSS(xlO-3)
75
505
190
68
242
70
3289
35
401
5 * 3289
3289/5=658
Questionable values.
-------
TABLE 4. METALS CONCENTRATIONS (Ca, Co, Cu, Cr, Fe) AVERAGED FOR 36 STORM
EVENTS (DATA OF COLSTON (10)) AND RATIOS OF EACH TO TOTAL SUSPENDED SOLIDS
Storm
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
31
Calcium
mg/1
Avg. o
~
~
7.0
2,5
2.7
5-5
4.1
14,3
6,2
24.7
5.3
4.2
5.2
~
~
"
~
~
7.8
0.5
1.6
7.9
1.3
14.2
3.5
4.9
1.3
1.4
3.7
--
~
--
~
Cobalt
mg/1
Avg. cr
0.36
0.13
0.08
0.14
~
0.1
0.09
"""
0.09
0.03
0.04
0.05
--
~
--
~
--
--
0.0
0.00
**"
Copper
mg/1
Avg. o
0.36
.14
0.10
0.13
-
--
..
0.15
0.10
0.10
""
0.10
.03
0.03
0.03
~
--
--
--
~
0.07
0.00
0.00
"**
Chromium
mg/1
Avg. a
.31
0.33
0.31
0.27
0.29
0.27
0.16
0.11
0.10
~
~~
--
.07
0.08
0.06
0.03
0.09
0.07
~
~
~
0.09
0.03
0.00
--
""*
Iron
mg/1
Avg. o
4.4
3.5
3.8
10.6
9.1
11.2
11.4
9.9
9.1
7.8
16.3
3.7
13.6
9.3
7.7
12.9
12.1
~~
1.3
.8
1.7
2.9
4.2
4.1
4.9
6.3
4.7
3.9
14.5
1.0
9.7
3.5
4.4
8.3
3.2
~~
Ca/TSS
(xlO-3)
--
~
20.2
5.27
1.85
4.46
2.34
25.00
6.26
16.9
36.3
3.86
6.17
""
Co/TSS
(xlO-3)
4,04
0.47
0.49
~
'
~
0.11
0.10
"-
Cu/TSS
(xlO-3)
4.04
0.51
0.61
~
~
~
0.06
0.07
0.12
~
~~
Cr/TSS
(xlO-3)
--
1.13
2.02
0.78
0.20
0.22
~
~
0.06
0.07
0.12
-
_.
Fe/TSS
(xlO-3)
49.4
12.8
23.3
26.3
23.6
7.81
8.03
5.19
13.6
16.5
25.3
93.2
8.56
9.13
14.4
13.5
ro
ro
-------
TABLE 4 (continued)
Storm
Number
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Calcium
mg/1
Avg. a
--
-»
2.3
--
2.1
--
..
--
0.8
--
--
1.3
~
Cobalt
mg/1
Avg. a
~
--
~
~
~
~
Copper
mg/1
Avg. a
--
0.14
0.12
~
--
0.12
0.12
0.13
0.02
--
0.01
0.02
0.02
0
Chromium
mg/1
Avg. a
0.10
0.15
--
0.16
0.11
__
~
0.04
--
0.04
~
~
0.03
0
..
Iron
mg/1
Avg. a
32.8
19.0
18.8
19.6
--
__
._
14.6
5.6
~
9.3
9.4
~
~
__
Ca/TSS
(xlO-3)
1.58
~
__
Range:
1.58* 169
169/1.59-107
Co/TSS
(xlO-3)
~
_.
0.1 "4.M
4. 04/. 1-40.4
Cu/TSS
(xlO-3)
~
0.09
--
~
0.10
0.08
0.13
_..
.06 > 4.04
4.0V .06.67. 3
Cr/TSS
(xlO-3)
0.11
0.11
0.11
_..
.06 - Z.02
S.02/.06-33.7
Fe/TSS
(xlO-3)
__
13.0
18.6
227.
25.2
__
5.19 -> ZZJ
227/5.19-41.7
ro
co
-------
TABLE 5. METALS CONCENTRATIONS (Pb, Ni, Mg, Mn, Zn) AVERAGED FOR 36 STORM
EVENTS (DATA OF COLSTON (10)) AND RATIOS OF EACH TO TOTAL SUSPENDED SOLIDS
Storm
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Lead
rag/1
Avg. a
.49
0.43
0.53
0.57
0.40
0.42
0.43
0.35
0.47
0.57
0.26
0.38
0.29
0.23
0.45
0.23
0.10
0.47
0.49
~~
.11
0.07
0.09
0.24
0.14
0.22
0.16
0.26
0.30
0.80
0.12
0.29
0.14
0.15
0.46
0.17
0.00
0.54
0.38
~~
Nickel
mg/1
Avg.
-------
TABLE 5 (continued)
Storm
Kunber
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Lead
"19/1
Avg. a
~
0.79
0.26
0.20
0.28
1.19
0.69
0.24
,
--
~
~
0.75
0.09
~
0.26
0.12
--
9.32
0.14
0.08
Nickel
ng/1
Avg. o
~
--
~
~
--
~
--
Magnesium
[»3/l
Avg. o
12.4
4.7
11.9
7.2
12.6
~
~
15.5
--
1.2
0.8
1.6
2.2
~
--
4.4
--
2.5
«
Manganese
mg/1
Avg. a
0.71
0.40
1.67
1.32
0.60
0.44
--"
0.17
0.25
0.24
0.91
0.16
«-
0.08
i
,_
Zinc
mg/1
Avg. a
0.42
0.22
0.53
0.34
~
0.33
0.28
0.28
--
0.06
0.10
0.19
0.11
0,23
0.14
0.05
«
^~
^10^)
0.31
0.25
2.41
0.36
0.81
0.67
0.37
Range:
0.12 -> 2.64
2.S4/. 12-22.0
Ni/TSS
(xlO-3)
--
0.10*0.98
.98/.1-9.8
Hg/TSS
(KlO-s3
5.32
8.48
4.12
--
9.50
~
10.6
--
4.12 * 64.4
64.4/4. IZ-1 5. 6
Hn/TSS
U10-3)
0.30
0.72
0.58
~
0.52
0.59
0.35
0.30 + 4.36
4. 36/. 3-14.5
Zn/TSS
(xlO-3)
0.18
0.40
0.18
--
~
3.98
0.36
0.22
--
--
0.1? » 3.98
3.98/.1Z-33.2
no
en
-------
be lower than for ratios of individual observations. Second, a similar moder-
ating effect on variability is probably present in the data because of the
large drainage area involved and the distance over which the runoff traveled
before being sampled. All of this strongly suggests that the ratios would be
substantially more variable if computed for individual samples taken at up-
shed locations. Finally, it should be noted that the data within each storm
are likely to be autocorrelated. This means that the data, not being indepen-
dent of time, may exhibit variable potency factors from storm to storm because
of the sampling schedule. Thus, in examining Colston's (10) data it would
have been much better had individual observations been available rather than
storm means.
Donigian and Crawford (6), in a study specifically designed to evaluate
the feasibility of simulating nutrient washoff using potency factors and the
NPS model, also looked at the Third Fork Creek (Durham, North Carolina)
Watershed (see Colston (10) and preceding discussion). Donigian and Crawford
simulated sediment loss, TKN, total phosphorus, and iron during five storms.
Their results showed that for the specific watershed, total phosphorus and
iron concentrations in the runoff could generally be simulated with reasonable
accuracy. TKN, on the other hand, proved difficult to predict, this being
attributable (according to the authors) to subsurface carriage of TKN and mis-
calibration of the potency factors based upon available data.
Non-Urban Watershed Studies
Donigian and Crawford (6) have applied the NPS model to data developed on
agricultural test plots located in Watkinsville, Georgia, and at East Lansing,
Michigan. Characteristics of these test plots have been described in detail
by Smith et al_. (11) (Watkinsville) and by Ellis et al. (12) (East Lapsing).
In general, the test plots considered by Donigian andTCrawford are the P-2
plot at Watkinsville and the P-6 plot at East Lansing. According to Smith
et al. (11), the P-2 plot is an area of 1.29 hectares representing common
PTecEont forms and management practices. Soil types are primarily Cecil sandy
loam with some Cecil sandy clay loam and loam. No soil or water conservation
measures were applied in the period during which the data used by Donigian and
Crawford were collected. Watershed P-6 is 0.8 hectares in area and consists
primarily of Spinks loamy fine sand with some Hillsdale fine sandy loam (12).
During the period considered by Donigian and Crawford, plot P-2 was in corn
and plot P-6 was in soybeans.
Results of simulation of storm events on the two watersheds with the NPS
model are characterized as reasonable (6), with discrepancies between observed
and simulated concentrations of total phosphorus, phosphate, nitrate, ammonia,
and total nitrogen being attributed primarily to the small size of the test
watersheds, the short duration thunder storms, and the inability of the NPS
model to accommodate tillage practices. Donigian and Crawford (6) conclude
that potency factors appear useful for simulating total nutrients. Total
phosphorus concentration closely mirrored those of sediment. Total nitrogen
also could be adequately represented through potency factors, although not as
well as total P; whereas simulating ammonia, nitrate, and phosphate as a func-
tion of sediment load was increasingly tenuous.
26
-------
The issue of effects of tillage, as noted by Donigian and Crawford (6),
was addressed some years earlier, at least for N and P in transported sedi-
ment, by Romkens et al. (13). In the study, which was conducted at Bedford,
Indiana, five different tillage-planting systems were used with simulated
rainfall providing runoff events. It was noted, first, that nitrogen and
phosphorus concentrations in the transported sediment were related to runoff
sediment (both expressed as mass per unit area of watershed) although the re-
lationship was not necessarily linear. Second, impacts of tillage methods on
nitrogen and phosphorus losses were found to be significant, and the tillage
method influenced the propensity of fine clay particles to wash off. Finally,
correlations between sediment nutrient (N and P) concentrations and sediment
clay concentrations were very high (.98 and .96, respectively).
Burwell et_ al_. (14) have noted that, although large amounts of nutrients
may be associated with runoff sediment, subsurface transport can also account
for substantial amounts of soluble nitrogen and phosphorus. This was also
cited by Donigian and Crawford (6) as one factor affecting their results in
applying the NPS model, particularly with respect to the poor nitrate and
phosphate simulations. Burwell et^ al_. (15) have also observed significant im-
pacts of soil cover and seasonal perfods on nutrient transport thus further
emphasizing the possible effects of land use and management practices on po-
tency ,factors.
27
-------
SECTION 4
METHODOLOGY
The study consisted initially of compilation of data bases representing
agricultural, silvicultural, urban, and rural watersheds, and subjecting the
data to regression analysis. Subsequent analysis was geared toward determin-
ing two things. First, the nature and quality (in terms of a simple mathe-
matical expression) of the relationship between suspended solids content of
runoff and other water quality characteristics was determined. Second, the
degree to which runoff water quality may be predicted using information on
storm characteristics and other variables in addition to suspended sediment
loads was evaluated.
GENERAL ANALYSIS PROCEDURE
Figure 6 shows the general procedure by which data were subjected to
analysis. The first step in the process consisted of fitting a simple linear
model of the form:
C = b + b S f^
o i 13)
where C = predicted concentration of water quality
constituent
S = concentration of suspended solids
b ,b = fitted coefficients
o i
The program for fitting the simple linear model provided several useful kinds
of output including a scatter plot of the data. Using the plots, it was pos-
sible to determine whether there was any obvious nonlinearity in the data as
well as to subjectively determine whether regression assumptions were met.
Where it was apparent that transforming the data might improve the quality of
relationships, appropriate transformations were used, and the plotting/linear
regression package was again applied. After it was clear for each data set
that further transformation was likely to be of little or no benefit in terms
of improvement of fit, processing moved on to multiple regression. In per-
forming the multiple regressions, other possible transformations were exam-
ined. Unless significantly better relationships were obtained by using other
transformations, the transformations used for simple regression were adopted
in multiple regression.
28
-------
Compile and visually
Inspect all data bases
Select Independent variables
for simple and multiple regression
r
Repeat
for each
data base
Select dependent variable within
data base to subject to analysis
Generate data scatter plots
Generate correlation statistics
Examine data and statistics
for deviations from assumptions
and need for transformations
Assumptions
cannot be met,
eliminate constituent
Perform multiple regression
analysis
Examine results to determine If
transformations are needed,1f assumptions
are met, or If run needs modification
L
Yes
ST
3P
1
Select next
dependent
variable to be
examined
Figure 6. General data analysis procedure,
29
-------
Because the details of statistical techniques in general and regression,
in particular, are not universally understood, some general theoretical con-
siderations are presented in Appendix A. The reader wishing to review the
materials, or requiring clarification of some point should refer to this Ap-
pendix.
Analytical Tools
Figure 6 shows the general procedure by which analyses were performed in
this study. The two distinct computer programs that were applied are de-
scribed below.
Simple Regression
The first program provided graphical output and certain basic statistics
using the simple linear model:
Yi = B0 + M, + EI (4)
and
?1 » + b1Xi («)
where Y^ = ith observed value of the dependent variable Y
X.j = ith observed value of the independent variable X
b = Y intercept of the fitted line, an estimator of 3
o o
b = slope of the fitted line, an estimator of 3
i i
3 = Y intercept of the true relationship
o
3 = slope of the true relationship
e. = error term correcting the value of the dependent variable,
for deviations from that described by 3 + 3 X.
o i i
s\
Y. = ith predicted value of the dependent variable
The output of the simple regression program appears as shown in Figure 7.
In the figure, and considering only the two variables (log^g l63^ concentra-
tion versus log zinc concentration), the following are provided.
Linear and nonlinear relationships (through appropriate data
transformations) of each bivariate population. The fitted
line is represented by "+" symbols, the number of observations
per plot coordinate by plotted numerals.
Confidence limits for prediction of the population mean and
individual values of the dependent (ordinate) variable for a
given value of the independent (abscissa) variable. The 95%
30
-------
Z.eStOM
2,«OtO«l
Log Pb Cone, (ppb)
l.«§50«»
i.saiees
4
»
1
*
*
»
i **
1 * 1 1
*
1 ** 2 2 1 t 1
» ..
...
» ., 1
1 * i.
* .,11
** .. 2 *
.. **
.. 2 1* 2
* ,., 2 »*«l
* .. 1* «
*** ., *« 11
** t. »»
* ., 1«* 11
.. *»» ..
... ** 2 71 .2 1
t. «» .t
t. * + ...
.. +» (.2
.. »* ..
** .. *
»* .. **
**« 1 » ,. » 2t bl* 1
»* .. «
** .. *
** t.
* .... **
**
** ..1 *
** ,, +t*
** t. **
** .. I**
* t., t **
»* .. *+
* ., H *1
** .. ** ..
* .. 1 ++ ,.
* .. ** ..
** .. »1* ...
* 1 ., +* i.
* 1 1. l»» .,
** .. ++ .. *
** 1. 1 t21 .. **
* 1 I.. *+ 1 J .. ***
til 1 »* ... **
., I 1 1 » +* 1 .. **
t. J 2 13 »»» 11.. **
1 1. 1 11 1 »* 1 .. **
J ..*! 1 2 l*+2 1 ,. *
,. 1 1 1 t*» 1 .. **
1 111 *! 3 1 1 ... **
»+ Z \ ,, **
2 Itl 1 11 .. **
*1 1 .. **
» *»t 1 1 1 .. **
1*1 12 1 1. **
.. ««
1 1... **
2 1 2. 1 **
1 1.* **
:i *
..
. **
*
i i
**
**
**i
»*
i
1,180 1,320 1,520 1.720 1,920 2,120 2,320 2.S20 2.720 2.910
LINtAK HtSHtSSIUN STATISTICS
Log Zn Cone, (ppb) ,975o LL A*-.Kit t(A)« ,11S ,<»TSO UL A« ,37J
,<>7b(v LL t>» ,*ZS l(o)s l,0e ,97bV UL o* 1,2U
N SQUARED*,Sb7189 f* 2i
-------
confidence limits are shown on the plot with the band for
prediction of the population mean shown as "" symbols
and the band for prediction of individual dependent
variable values shown as "*" symbols.
Estimates for the slope and intercept of the least
squares model (designated E (B) and E (A), respectively,
in Figure 7).
t Confidence limits for the slope and intercept. In Figure
7, these are designated as LL B, UL B, LL A, UL A for the
lower and upper limits of the slope and intercept,
respectively.
The r-squared and F values for the fitted relationship.
The former describes the proportion of the variance
shared by the ordinate and abscissa variables (given a
linear relationship), while the latter, given that
regression assumptions are met, indicates whether the
value of r-squared is significant at the selected a level.
N is the number of observations.
Multiple Regression
The multiple regression program uses a stepwise algorithm in fitting an
equation of the form:
Y = b + b X +bX +...+ b X (5)
01:22 mm
s\
where Y = the predicted value of the dependent variable
X = the observed values of the independent variables
b = the fitted coefficients
In this type of program, the independent variables are introduced into the
regression one at a time (one per step), and for each step, all statistics are
generated. The criterion for selecting variables is the increase in R^
(squared multiple correlation coefficient). Thus the first step introduces
the independent variable having the highest correlation with the dependent
variable. The second step introduces into the regression that independent
variable which, in conjunction with the first variable, gives the greatest R^.
The algorithm can also remove a variable which, although introduced in an
earlier step, no longer contributes significantly to R2.
Table 6 shows an example multiple regression. Specifically, a typical
output table from the regression package used in this study is shown. Table 7
provides an explanation of entries in Table 6.
32
-------
TABLE 6.
REGRESSION OF LOG [NO, + N09]
(SEATTLE DATA)* * L
STEP NO. 6
VARIABLE ENTERING
1
FLEVEL 6.034608
STANDARD ERROR OF Y
CONSTANT 2.459798
VARIABLE
X 1
X 5
X 6
X 9
X 13
X 15
COEFFICIENT
-.55947
-.09976
-1.67359
.20843
.00416
-.13748
STD ERROR OF COEFF
.22775
.02747
.36013
.02662
.00034
.04574
T-STATIST1C
-2.4565
-3.6314
-4.6471
7.8283
12.2785
-3.0054
PREDICTED VS. ACTUAL RESULTS
OB. NO.
1
2
3
4
5
6
7
8
9
10
181
182
183.
184
185
186
187
188
ACTUAL
-.2076
-.2147
-.1675
-.1739
-.1739
-.2676
-.3468
-.2596
.4472
.3424
PREDICTED
-.2886
-.2203
-.1562
-.1554
-.2095
-.2724
-.2939
-.2112
.0083
.0532
DEVIATION
.1055
.0079
-.0179
-.0292
.0527
.0060
-.0583
-.0650
1.7808
1.0696
.7447
.6198
.6383
.6778
.1487
.0128
.1612
.1461
.7084
.6144
.6207
.6259
.0031
.0032
.2354
.2420
-.0157
-.0030
-.0095
-.0267
-.2973
.0225
.1085
-.3458
PERCENT DEV,
17.01
1.29
-2.63
-4.36
7.87
1.10
-12.95
-11.81
63.60
48.62
-8.72
-1.24
-4.13
-12.70
-41.87
2.18
15.72
-24.70
NO. INDEPENDENT VARIABLE CODES
1 (LOG OF) DAYS SINCE 1/1/73
2 MONTH
3 HOURS SINCE STORM START
33
-------
TABLE 6 (continued)
4 (LOG OF)
5 (LOG OF)
6 (LOG OF)
7 (LOG OF)
8 (LOG OF)
9 (LOG OF)
10
11
12
13
LOG OF)
LOG OF)
LOG OF)
14 (LOG OF)
15 (LOG OF)
16
DRY DAYS BEFORE STORM
SUSPENDED SOLIDS, MG/L
CD, UG/L
PB, UG/L
ZN, UG/L
NH4, MG/L-N
ORGANIC N, MG/L-N
TOTAL P, MG/L-P
OP04-P, MG/L-P
CONDUCTIVITY, U-MHOS/CM
TURBIDITY, JTU
FLOW, CFS
STORM RAINFALL, INCHES
R-SQUARED FOR THE REGRESSION LINE IS .753564
CORRECTED R-SQUARED IS .745395
F FOR THE REGRESSION IS 92.25 F CORRECTED FOR AN EQUIVALENT OF
188 OBSERVS.= 92.25
AUTOREGRESSION COEFFICIENTS FOR THE FIRST TEN DIFFERENCES ARE-
.111693 .093102 .077329 .010425 -.003380 .000156 .001426
.000575 .002910 -.001002
transcribed from computer output, observations 11 through 180 not shown.
34
-------
TABLE 7. SIGNIFICANCE OF ENTRIES IN TABLE 6
Item
Interpretation
Step No. 6
Variable entering 1
F level
Standard error of Y
Constant
Variable
Coefficient
The program used provides stepwise multiple
regression. During each step, either a new
variable is added or a previously included
one is removed. Table 6 contains the sixth
step in the analysis.
During this step, variable 1 has been in-
troduced as a predictor variable.
This is the value of F for the current re-
gression step. It reveals whether the step
has significantly improved prediction over
the preceding step.
This is the standard error of the predicted
values, also called the "standard error of
estimate." It may be used in computing a
confidence interval for each predicted
value. Although the latter, also called
the "standard error of forecast," is not
shown in Table 6, it is provided in later
runs. The standard error of forecast is
not of much use here, where the main ob-
jective is explication. It is important,
however, where the regression equation Is
to be used as a predictive model.
The predictive model is of the general form
Y = b + b X + b X +...+ b Xn
01122 n n
The constant is bo, the intercept for the
regression equation.
The independent variable(s) used in the
model. In this case, variables 1, 5, 6, 9,
13, and 15 have been used.
These are the
equation.
"b" values in the regression
35
-------
TABLE 7 (continued)
Item
Interpretation
Standard error of
coefficient
t-statistic
Observation number
Actual
Predicted
Deviation
Percent deviation
Independent variable codes
R-squared for the
regression
The standard errors of the coefficients are
printed out here. These are used in com-
puting the t-statistic.
This is the ratio of the estimated coeffi-
cient to the standard error of the coeffi-
cient. In general, where the absolute
value exceeds 2, the probability that the
coefficient is non-zero is at least p= .95.
Thus, where Itl >_ 2, use of the variable is
justified on a basis of statistically non-
zero coefficient. If |t| < 2, there is no
justification in saying that the variable
predicts NOs + N02 for any reason other
than chance (at p = .95).
The number of the observation.
Observed concentration of N0£ + N03
transformation). The first observation was
10-. 2076 = .620 ppm.
Logged concentration of nitrate as7 pre-
dicted by the model 1Q--2886 = .515.
Untransformed difference between observed
and predicted, = .620 - .515 = .105.
The deviation as percent of the observed
untransformed value (.105/.620) x 100% =
17%.
Shows precise identity of each variable
used in regression.
Proportion (0 < R2 < 1) of the variance of
the dependent variable (log [NOs + NOg]
in this case) accounted for by the indepen-
dent variables (the regression model).
36
-------
TABLE 7 (continued)
Item
Interpretation
Corrected R-squared
F for this regression
Autoregression coefficients
for the first ten differences
Von Neumann ratio for first
ten differences
F corrected observs. =
R-squared adjusted for the number of vari-
ables used in the model. The more vari-
ables used, the more R-squared is likely to
be inflated, and the greater the correction
needed. This value is a more accurate re-
flection of the R2 likely to be obtained if
a new data set were collected under identi-
cal circumstances.
F statistic showing whether or not R2 for
the model is significant. That is, it
shows how likely it is that the results
have occurred by chance alone.
This statistic examines the residuals (de-
viations) to determine whether there is
reason to suspect a lack of independence of
observations or residuals (a regression as-
sumption).
Corroborative statistic for the autoregres-
sion coefficient. Printed out in other
tables but not shown in Table 6.
F-statistic corrected for the number of
equivalent independent observations (here
equal to 188) computed from the number of
data points (in this case 188) and the
autoregression coefficients.
37
-------
Data Bases
Data for this study represent six different geographical areas and ten
discrete watersheds. Five of the watersheds are agricultural, one is silvi-
cultural, one is rural, and three are urban. Table 8 provides a summary
description of the data sets used. Appendix B describes water quality anal-
ysis methods as described in available reports for the data bases.
Watkinsville
The Watkinsville data sets were derived from agricultural test plots
maintained jointly by the U.S. EPA Environmental Research Laboratory, Athens,
Georgia, and the Agricultural Chemical Transport and Modeling Unit of the
Southern Piedmont Conservation Research Center, United States Department of
Agriculture. The locations and configurations of the test plots are as shown
in Figures 8, 9, and 10, and are described in detail in Smith et_ al_. (11).
Smith and coworkers also describe the experimental procedure and results of
several years of runoff sampling and analyses from the test watersheds.
All test plots were constructed to have a drainage channel consisting of
a stainless steel flume. Continuous runoff sampling for sediment and pollu-
tants was provided by use of a motorized sampling slot traversing back and
forth through the discharge. The slot was tapered so that a larger proportion
of low flows than high flows would be collected. Stationary slots below the
flume subdivided the sample, and a sequential sampler delivered samples to a
refrigerated compartment.
Data from test plot P-01 consisted of runoff concentrations of the pesti-
cides trifluralin, paraquat, and diphenamid (1972, 1973); paraquat and diph-
enamid (1974); and propazine and paraquat (1975). The data also included
rainfall and runoff information with measurements of flume stage height, sedi-
ment concentration, and sampling time provided. These data were also^rovided
for plots P-03 and P-04.
Test plot P-02 data were not used in the present study, and this plot
will not be discussed here. Test plot P-03 was used for runoff pesticide data
with analyses of trifluralin, paraquat, and diphenamid (1972, 1973); and para-
quat and diphenamid (1974) being available.
Test plot P-04 was used to develop pesticide data (paraquat and atrazine,
1973 and 1974; and atrazine, cyanizine, paraquat, and 2,4-D, 1975) and nutri-
ent runoff data as well. Nutrient data generated include nitrate, ammonia,
TKN (total kjeldahl nitrogen), phosphate, available phosphorus, total phos-
phorus, and chloride. With respect to chloride, phosphate, and nitrate, only
aqueous phase concentrations were measured. Only sediment concentrations of
available phosphorus were provided while the remaining constituents were mea-
sured both in the sediment and in the aqueous phase.
Storms were monitored from the time of cropping for varying durations on
the order of 6 months to 1 year, depending on plot and stage in the study.
Within storms, the number of observations varied with up to about 35 taken in
some individual storms.
38
-------
TABLE 8. SUMMARIES FOR DATA BASES USED IN THIS STUDY
Data Base
Uatklnsvllle
Test plot P-l
Test plot P-3
Test plot P-4
Buffalo Bill
Michigan State
University Farms,
Test plot P-6
Redwood National
Park
Seattle
Vlewrldge 1
South Seattle
Southcenter
Honey Creek
Location
Oconee County,
Georgia
Near Eldrldge, Scott
County, Iowa
East Lansing,
Ing ham County,
Michigan
Humboldt County,
California
Washington
Mel more, Ohio
Type
Agricultural test
plots
Agricultural
Watershed
Agricultural test
plot
SI 1v1 cultural
High density, old resi-
dential, urban, storm
sewer
Industrial, urban,
storm sewer
New shopping center,
urban, storm sewer
Predominantly rural
Area
(ha)
2.7
1.26
1.4
1,417
0.8
*
255
11.1
9.8
38,600
Number
of Con-
stituents
Examined
2
2
6
10
5
6
8
8
8
11
Number
3f Obser-
vations
Used
(range)
134-260
85-123
21-108
42-89
73-148
7-17
184-199
188-199
123-636
Approx-
imate
Number of
Storms**
13
7
24
30
31
30
Period of
Record Used
in the Study
7/72-5/74
7/72-10/75
5/74-5/75
8/73-12/73
1/75-4/75
9/73-9/75
2/73-9/73.
10/74-12/75
5/73-9/73.
10/74-12/75
2/73-9/73.
10/74-12/75
1/76-1/77
CO
vo
Several stations considered, precise drainage areas not available
Storms having data in the data base for runoff water quality and used in analysis.
Only approximate because some storms merged and the data were often unclear on this
Issue. In some cases (Michigan) some storms were snow storms.
-------
PRAINGAUGE
PACOLET GRAVELLY
SANDY LOAM
CECIL GRAVELLY SANDY LOAM
Location
AREA: 2.70 ha
CONTOUR INTERVALS-' 0.5 M
WATERSHED POI
SCALE:
20m
Figure 8. Soils and topography, watershed P-01. (Redrawn after Smith e_t al_. (11))
-------
TERRACE CHANNEL
GRASSED WATERWAY
RAINGAUGE o
FLUME
Location
AREA- 1.26 ha
SCALE-'
20m
WATERSHED P03
Figure 9. Soils and terrace configurations, watershed P-03.
(Redrawn after Smith e_t al_. (11))
-------
Location
KEY'
TERRACE CHANNEL
GRASSED WATERWAY
FLUME
AREA-' 1.40 ha
SCALE: I 20m I
WATERSHED P04
Figure 10. Soils and terrace configurations, watershed P-04,
(Redrawn after Smith et al_. (11))
42
-------
Buffalo Bill Watershed
The Buffalo Bill Watershed is a 1,417 hectare agricultural basin located
in eastern Iowa. During the period in which the data used in this study were
compiled, it was primarily in corn, beans, and pasturage, with very small
areas in hay and oats. The study from which the data were drawn was conducted
by the Iowa State Hygienic Laboratory (associated with the University of Iowa,
Iowa City) during the last five months of 1973.
Figure 11 shows the configuration and location of the drainage basin.
The nine sampling stations shown are the locations where runoff was sampled.
These were pooled for purposes of statistical analysis because of the high
quality of the data and relatively few observations available at each station.
Table 9 shows storm data for the Buffalo Bill data base (16). As shown,
there were seven storms, with the storms of August 13 and October 10 being the
most substantial.
Runoff quality variables included within the data base are as follows:
Dissolved Oxygen (DO) 0 DDT
Biochemical Oxygen Demand (BOD) DDE
0 Fecal Coliforms 0 Dieldrin
0 Total Kjeldahl Nitrogen (TKN) 0 Aldrin
0 Ammonia N 0 Heptachlor
0 Nitrite N 0 Heptachlor epoxide
0 Nitrate N 0 Gammachlordane
0 Phosphate 0 Organophosphate
0 pH 0 Turbidity
0 Suspended Solids (SS) 0 Dissolved Solids
Data were generally available for the non-pesticide parameters and for
DDT, DDE* and dieldrin, but not for the remaining pesticides. Accordingly,
the data on these pesticides (DDT, DDE, dieldrin) and BOD, fecal coliforms,
TKN, NH4, NOo, ^3, and phosphate were subjected to analysis. In their dis-
cussion of the study, Morris and Johnson (16) do not describe the way in which
phosphate,, TKN, Wty, and BOD were measured. Accordingly, it is not known
whether these represent dissolved, suspended, or total values. Nor is BOD
specified as representing five day, ultimate, or other basis.
Michigan State University Farms--
Data were drawn from a 27-month study of nutrient and pesticide losses
from great lakes watersheds, conducted by members of the Departments of Crop
and Soil Sciences and of Entomology of the Michigan State University at East
43
-------
Buffalo Bill Watershed
Scott County, Iowa
Figure 11. Buffalo Bill Watershed configuration and location,
(Redrawn after Morris and Johnson (16))
44
-------
TABLE 9. BUFFALO BILL WATERSHED FLOW AND RAINFALL DATA (16)
Date
8/13/73
8/13/73
8/13/73
9/04/73
9/17/73
9/26/73
9/26/73
10/10/73
10/10/73
10/29/73
12/04/73
12/04/73
12/12/73
Time
6:22 pm
7:10 pm
10:15 pm
1:00 pm
1:10 pm
8:00 pm
11:20 pm
6:15 pm
8:55 pm
11:15 pm
12:25 pm
3:15 pm
1:45 pm
Estimated Flow
(cu ft/sec)
2100
1300
200
70
50
40
110
380
70
5
45
35
3
Rainfall
(inches)
2.9
0.1
0.1
0.5
0.85
0.15
0.7
0.65
0.1
0.1
1.5
0.15
0.0
Data are representative of Station Number 9
(see Figure 11).
45
-------
Lansing. The study was supported by the EPA Environmental Research Labora-
tory, Athens, and was a parallel study, both in time (5/73-8/75) and in the
sense that it was in support of the Watkinsville study. Ellis ejt al_. (12),
(17) describe the study in detail in their project reports.
The Michigan study employed two watersheds of 0.8 hectare (east water-
shed, designated P-06) and of 0.55 hectare (west watershed, designated P-07).
Figure 12 shows the location and configuration of the watersheds. During the
10-year period prior to the study, they were in corn, but during the study
period, the watersheds were in soybeans.
Pollutants measured in the runoff from the watersheds included paraquat,
trifluralin, diphenamid, atrazine, nitrate, total kjeldahl nitrogen, ammonia,
total phosphorus, available phosphorus, phosphate, and chloride. With the
exception of nitrate, phosphate, and chloride, which were measured only in the
water fraction, and suspended solids concentrations of available phosphorus,
all pollutants were measured in both the water fraction and in the suspended
solids fraction.
Redwood National Park Studies
The United States Geological Survey (USGS) has conducted a study of sedi-
mentation and erosion processes in Redwood National Park in Humboldt County,
California, with details provided in Data Releases by Iwatsubo et al_. (18) and
by Iwatsubo et al. (19). The study involved data collection at 53 sampling
stations in tfie Redwood Creek and Mill Creek drainage basins. Although the
stations were located in-stream, water quality probably represents runoff
since many of the streams have very high slopes (typically up to about 25-30
percent) and have rocky, stable bottoms.
Water quality and hydro!ogic data collected include:
Stream stage Major dissolved solids components
Stream discharge Selected trace elements
t Turbidity Nitrogen
Sediment characteristics Phosphorus
Temperature* t Organic carbon
pH* Fish
Total alkalinity* t Periphyton
Specific conductance* t Phytoplankton
Dissolved oxygen* 0 Seston
*Measured in the field.
46
-------
Cultivated Watersheds
Michigan State University Campus
Erosion
Soil Legend
510 Hillsdale f.sa.l.
511 Tuscola f.sa.l.
515 Traverse f.sa.l.
819 Spinks l.f.sa.
1 0-25%
2 25-75*
+ Surface
deposition
Location
06
07
Figure 12. Location and configuration of Michigan State University test plots,
(Redrawn from Ellis (12))
-------
Organic carbon* Pesticides*
Bacteria Benthic Invertebrates*
The study also examined changes in geometry at 10 channel cross sections
on Mill Creek and the erosional landform distribution in that basin, quantity
and chemical quality of rainwater, and streambed characteristics at certain
Redwood Creek drainage basin stations.
For purposes of the present study, although the Redwood National Park
data base is extensive, problems of analysis arose because suspended sediment
analyses were not available for the samples for which chemical water quality
data were developed. Although this is no criticism of the USGS studies and
the situation probably arose because of the use of specialized sampling equip-
ment, it did severely restrict the number of observations where the chemical
water quality measurements and suspended solids concentrations could be con-
sidered to represent a single observation. In general, if the times of sam-
pling were sufficiently close (within about 10 minutes), conditions were con-
sidered constant, and the data for suspended solids were combined with water
quality data. Because of these limitations, however, there were few accept-
able observations for analysis (see Table 8).
Seattle Data Base--
Runoff data for the City of Seattle, Washington, were obtained from a
study performed for the Municipal Environmental Research Laboratory, U.S.
Environmental Protection Agency. A brief description of the data base is pro-
vided in a document prepared by Huber and Heaney (20). Runoff water quality
parameters available for analysis were:
t Orthophosphate P Nitrate plus nitrite N
Total phosphorus Organic nitrogen
Conductivity Lead
Ammonia N Zinc
0 Suspended solids Turbidity
The data base includes data for six different sampling sites in or near
Seattle. In the present study, data from three sampling sites were subjected
to analysis. These were Viewridge 1, South Seattle, and Southcenter. Figure
13 shows the location of the three sampling sites.
Honey Creek Data Base
Runoff data for Honey Creek at Melmore, Ohio, were extracted from a U.S.
Army Corps of Engineers report (21), which describes the sampling site as fol-
lows:
''Benthic analyses.
48
-------
PUGET / *!>
SOUND
0123 4 Miles
\/ Viewridge
Viewridge
Central
iiip Business
District
PUGET
SOUND
Figure 13. Location map for Seattle catchments.
(Redrawn from Huber and Heaney (20))
49
-------
"04197100 HONEY CREEK AT MELMORE, OH
Lat 4r01'20", long 83°06'35", Seneca County, Hydrologic Unit
04100011, at bridge on State Highways 67 and 100 at Mel more, 1.5 mi
(2.4 km) upstream from Buckeye Creek."
Available water quality parameters were:
Total phosphorus P Silicate
t Orthophosphate P Suspended solids
Nitrate plus nitrite N Chloride
Ammonia N 0 Conductivity
Total kjeldahl nitrogen Iron
50
-------
SECTION 5
RESULTS AND DISCUSSION
Results of individual analyses will be presented for each data base.
irst, correlations between suspended sediment and each water quality constit-
uent will be discussed, followed by results of multiple regression analysis.
fn the discussion of simple regression results, tables of statistics are pro-
vided, with results being presented separately for each data base. A summary
table is also presented showing estimated potency factors by watershed. In
the discussion of multiple regression analyses results are organized around
location, and specific water quality relationships with variables other than
suspended solids are emphasized.
SIMPLE LINEAR REGRESSION
In all tables regarding simple linear regression, the notation described
below is used. Assuming the equation (same form as equation 3):
C = b + mS
columns in each table include the slope of the line (m) , the intercept (b),
and the 95% confidence limits for the slope and intercept. The limits are de-
noted as 1% (lower limit for the slope), mu (upper limit for the slope) and
corresponding b£ and bu. Note that where "m*" is to be used as a potency
factor in the NFS model, it must be multiplied by 100%. See note on Table 23.
The value for r2, the F statistic, and the number of observations are also pro-
vided. The columns headed Sm, S^, and Sp contain either blank or "X" entries.
Where "X" is shown, the particular statistic (e.g., F for Sc) is significant
at the a = 0.05 level. That is, the slope, intercept, and/or correlation
(Sm, Sb, and/or SF) are non-zero. The tables also show approximate ranges
for the dependent and independent variables as Xu, X^, Yu, Yo (for the upper
and lower limits of X and Y, respectively), and the number of observations, n.
Watkinsville
As discussed earl ier, three separate data bases from Watkinsville,
Georgia, were subjected to analysis. Results of simple linear regression for
the relationship between suspended solids and various other runoff water qual-
ity constituents are' shown in Tables 10 through 12.
51
-------
TABLE 10. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATIONS
(IN g/A) AND CONSTITUENTS SHOWN (CONCENTRATIONS OF DISSOLVED SUBSTANCES IN
RUNOFF) FROM WATKINSVILLE, GEORGIA. DATA ARE FROM PLOT P-04
Constituent.
Units
"Atrizine. ppb
TWI. ppm
NH4-1t. ppm
"NOj-N, pp«i
PO^-P. ppm
"tl, ppm
"u
.744
1.73
1.19
.514
.119
.443
m*
.105
1.30
.850
.541
.062
.247
^^^i
t
-.535
.878
.506
.268
.005
.051
MBHHta
Sffl
X
X
X
X
X
»u
.1.82
1.57
.186
-.601
.164
.323
b
1.02
1.23
-.085
-.821
.118
.168
"t
.227
.866
-.356
-1.04
.073
.014
Sb
X
X
X
X
X
rZ
.006
.37)
.185
.197
.069
.056
F
.117
37.8
24.0
15.7
4.78
6.25
V
X
X
X
X
X
*u
2.17
2.17
2.17
2.17
2.17 '
2.17
*l
.01
.01
.01
.01
.01
.01
\
2.57
4.31
3.00
0.82
.500
1.59
Tl
.27
.74
0.0
-1.02
.005
-1.04
n
21
66
108
66
66
108
cn
ro
Kate that m Is the potency factor for ta not significantly different from zero it a * .05
(Sj, blank) and the dependent variable not transformed. If b Is significantly non-zero,
then U should be recognized that C - mS was not a very acceptable linear relationship
for the particular constituent given the available data. Where ^ Is blink (in not signif-
icant) the potency factor a Is very tenuous. The value of r' represents the proportion
of the variance of the data for each constituent which may lie accounted for by suspended
sediment. For use with the NFS Model, n must be multiplied by 1001. Because of g/l
units for suspended solids at Hatklnsvllle and Michigan, n for these sites should
only be divided by 10.
~109IO transformation. That Is.
103]0 C b * BS
Hote that a very small Incremental constant (on the order of the seallest observed value)
MS added to the data where logs were used to prevent log 0.
Set text for Interpretation of colum headings.
-------
TABLE 11. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATIONS
(IN g/i) AND CONSTITUENTS SHOWN (CONCENTRATIONS OF DISSOLVED SUBSTANCES
IN RUNOFF)- DATA ARE FROM WATKINSVILLE, GEORGIA TEST PLOT P-03
GQMlttuent.
Units
TMfluralin, ppb
TrlfluraUn. ppb
DtpteunU. ppb
toiphen»1d, ppb
11
.512
.062
82.1
2.70
»*
.6«
.050
62.7
Z.30
M^^^^H^^V
»Jt
.475
.038
41.3
1.91
S«
X
X
X
X
>^MW0M^^
",
4.30
.580
42.2
.391
^M«M«^^^BIB
b
3.04
.511
-32.7
.ne
^^^^^
\
1.79
.442
-IDS.
-.154
Sb
X
X
r2
.191
.290
.15S
.371
f
39.7
68.5
40.6
134.
SF
X
X
X
X
*
17.7
17.7
17.7
1.264
't
0.0
0.0
0.0
0.0
Yu
36.30
1.42
33B4.
3.515
\
0.0
0.0
0.0
0.0
^.^^H
n
170
170
113
223
01
co
Kote: See ccmnents at foot of Tjble 10 and text for Interpretation of colum headings.
bath dependent and Independent variables had been log1Q transformed.
-------
TABLE 12. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATIONS (IN g/Jt)
AND CONSTITUENTS SHOWN (CONCENTRATIONS OF DISSOLVED SUBSTANCES IN RUNOFF)
DATA ARE FROM WATKINSVILLE, GEORGIA TEST PLOT P-01.
Constituent.
Units
"Diphenamid, ppb
"THfluralln, ppb
m
u
.0020
.031
m*
.0016
.022
"»
.0012
.016
S
X
X
bu
.772
.392
b
.579
.277
bl
.387
.161
Sb
X
X
r*
.499
.505
F
63.7
43.9
5F
X
X
Xu
2046.
58.4
h
0.0
.08
yu
3.0
1.45
Yl
0.0
0.0
n
66
45
cn
Nate: See comnents
-------
does not affect total water column loads of pollutants as they relate to sus-
pended solids. Table 13 shows correlations between dissolved pollutants and
pollutants in the suspended solids fraction for Watkinsville data.
Atrazine was not found to have a statistically significant relationship
to suspended sediment concentration at a = .05 (see Table 10). The slope was
not statistically different from zero, and r2 was low, so F was similarly low
"(below critical F for the degrees of freedom). The amount of data available
from this set was small for atrazine, and this may account for not detecting a
relationship. The fact that the intercept was significantly non-zero suggests
that some of the atrazine may move in the dissolved state so that the water
concentration is non-zero when suspended sediment is not present.
From Table 13, it is 'evident that despite the lack of a significant rela-
tionship between suspended solids and dissolved atrazine concentration (see
Table 10), there is a fairly consistent partition coefficient relating dis-
solved and suspended atrazine. These outcomes may seem to conflict, initial-
ly, but they may be easily reconciled. The significant partition coefficient
simply states that the water column atrazine concentration is related to the
concentration of atrazine in the suspended fraction, but not significantly to
the concentration of suspended matter itself. The relationship between sus-
pended fraction atrazine concentration (i.e., dissolved atrazine) and suspend-
ed solids load also depends upon the relative availability of both atrazine
and solids to wash off. If the ratio of mass of atrazine washing off to mass
of solids washing off were constant, then the relationship between dissolved
atrazine and suspended solids would undoubtedly be better. Other results of
the analysis for atrazine showed that atrazine runoff was very dependent on
time from cropping (r2 = 0.96, F = 453). Solids runoff data, although not
regressed on time, showed no such clearcut relationship over time upon visual
inspection.
All of the remaining regressions of dissolved species concentrations on
suspended solids concentrations were significant. This means that the trends
apparent in the data were not the result of chance alone and that "m," appro-
priately corrected to account for transformations used, and taking non-zero
Intercepts into consideration, may be used as estimates of potency factors.
Where the data were transformed, it may be desirable to estimate a series of
potency factors by a piecewise-li near approximation to the curve, and to sel-
ect those which are appropriate. Similarly, where "b" is significant, it
might be desirable to add a constant to estimates in NPS or SWMM in the cali-
bration and verification. This can easily be done directly in -the computer
code. I.
Regression results for TKN, Nfy, 1%, P04, and Cl are as would be ex-
pected, with TKN giving the highest r*. The highly soluble species (%, NO
P(>4, and Cl) may only be' related to suspended solids to the extent that both
are washed off simultaneously. Examining the relationships in Table 13 sug-
gests very strongly that this is so. There are no significant relationships
(essentially inverse partition coefficients) for TKN, NH/u or ROA. The sig-
nificant partition coefficients for atrazine, diphenanrid, and trifluralin in-
dicate that these are probably associated with particulate matter and that
sorption phenomena are probably involved. The idea of simulating these
55
-------
TABLE 13. REGRESSION OF DISSOLVED SPECIES ON CONCENTRATION IN SUSPENDED SOLIDS USING
DATA FROM WATKINSVILLE, GEORGIA. FORM OF RELATIONSHIP: Cn = b + mC_
WHERE Cn = DISSOLVED SPECIES CONCENTRATION, Cs = CONCENTRATION OF
SPECIES IN SUSPENDED FRACTION. UNITS ARE CONSISTENT BETWEEN
CD AND Cs
Constituent.
Units
Atnzlne. ppfa+
Atrulne, ppb
TIM. pp."
TW. PPB
tannla N. pp»
M4-P, PP-1
P04-P. pp!*'"
Dlphenaald. ppb
Dlphenaald, ppb**
Trlfluralln, ppb
Trinuralln. ppbf
Dlphenultf. ppb
DiphwMld, ppb**
Trinuralln. ppb**
Bu
3.10
.425
.157
5x1 (T6
6xlO'6
7x1 O'6
.017
.346
9.6X10"4
.034
;374
.308
.002
.0026
n'
2.44
.369
-.350
-3x1 O'5
-6x1 O"5
-2x1 O"5
-.065
.229
7.8X10"4
.027
.315
.192
.0016
.0004
"i
1.77
.313
-.857
-7x1 O"5
-2x1 O'5
-4xlO"5
-.147
.113
6.1x10-*
.020
.256
.075
.0012
-.002
S»
X
X
X
X
X
X
X
X
"u
-3.07
-5.95
4.91
2.48
.665
.212
.587
128.
1.20
4.43
.301
75.7
.772
.681
b
-4.67
-20.8
3.24
2.21
.508
.176
.348
61.4
1.03
3.41
.192
14.6
.579
.543
"l
-6.27
-35.6
1.56
1.93
.350
.140
.108
-5.28
.866
2.39
.082
-46.5
.387
.406
Sb
X
X
X
X
X
X
X
X
X
X
X
X
r2
.754
.815
.029
.044
.010
.028
.038
.064
.263
.266
.400
.076
.499
.004
F
58.3
177.
1.91
2.93
1.06
1.87
2.50
15.1
79.0
60.9
112.
10.7
63.7
.173
SF
X
X
X
X
X
X
X
X
Xu
3.00
1000.
4.5
35365.
96800.
8587.
3.93
2726.
2726.
780.
2.89
2024.
2024.
257.
h
1.84
70.0
2.1
120.
0.0
.33
1.55
0.0
0.0
0.0
0.0
0.0
0.0
0.0
\
2.57
370.
4.3
4.3
3.0
.50
.50
3384.
3.5
26.3
1.4
2867.
3.00
1.45
\
.30
2.0
.74
.74
0.0
.005
.005
0.0
0.0
0.0
0.0
0.0
0.0
0.0
n
21
42
66
66
108
£6
66
223
223
170
170
131
66
45
CJ1
CD " b * " lO $
"C,, - b * a log,,, Cs
*C0 P04, Cs Total P
Nate: See caments at foot of Table 10 and see text for explanation of column headings.
Note that m In this table does not represent potency factors. Note also that the
first seven lines In the table pertain to plot P-04. lines 8-11 pertalrf to plot
P-03, and lines 12-14 peruin to plot P-01.
-------
species as a function of suspended sediment is not entirely unrealistic. That
the r2 values are not nearly unity suggests that the nature of the sorption
equilibrium is variable, probably a result of changes in the nature of the
sediment and physical and chemical conditions.
Regressions of dissolved diphenamid on suspended solids were not only
significant, but also gave similar results for the partition coefficient for
vboth plots P-01 and P-03. Note in Table 13 that for untransformed variables
("Diphenamid, ppb") the slope and intercept ranges overlap for the two plots
and the r2 values are similar. The r2 values were not rigorously tested to
show a statistical difference, however. Where the diphenamid concentration
was transformed ("diphenamid, ppb**"), confidence limits on the slope and in-
tercept do not overlap but are close. Here, the r2 values are different and
are both substantially higher than for the untransformed diphenamid concentra-
tion.
The regressions of the log of dissolved trifluralin ("trifluralin,
ppb**") on suspended solids gave r2 values of 0.290 (plot P-03) and 0.505
(plot P-01). Note that for trifluralin, as well as diphenamid and other con-
stituents shown in Table 10 (except Nfy) the intercept is significantly non-
zero. Since the relationship in NPS and SWMM between suspended solids and
other pollutants does not include a constant, this cannot be accounted for
without changing the code. However, the desirability of adding a constant as
opposed to simply assuming a zero intercept, and potency factor m, is not
clear.
Buffalo Bill Watershed
Table 14 shows analytical results obtained with the Buffalo Bill Water-
shed data. The tabulated information is for regressions of various runoff
water quality constituent concentrations on suspended solids concentrations.
In this data base, there were several different sampling locations. Be-
cause the main objective of this study is to estimate potency factors, because
the data appeared fairly homogeneous spatially, and because there were limited
numbers of observations at each sampling site, the data were pooled spatially.
The net effect of this, however, should be to reduce the significance of re-
lationships since it introduces another unaccountable source of variance into
the data. Thus true correlations are likely to be at least as good as sug-
gested by the results in Table 14.
As shown in Table 14, DDE, DDT (both examined in case other pesticides
prove to have similar behavior), and nitrate were not correlated with sus-
pended solids. The best correlations were with dieldrin, BOD,-and fecal coli-
forms, although r2 values for these were not high. It is instructive to com-
pare results for the Buffalo Bill and Watkinsville watersheds. In general,
results for constituents in common for the two data bases are very dissimilar.
It should be noted, however, that those for which data were available in both
data bases are also those for which correlation of concentrations between dis-
solved and suspended solids phases (Table 13) were weak. These are TKN, NH4,
N03, and P04. Clearly, though, even were data available for diphenamid,
57
-------
TABLE 14. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATION (IN mg/fc)
AND CONSTITUENTS SHOWN (CONCENTRATIONS IN RUNOFF) FROM THE BUFFALO BILL WATERSHED
(SCOTT COUNTY, IOWA)
Constituent.
Units
DUldrin, ppt*
Dleldrln, ppt**
ODE. ppt
DOT, ppt
DDT. ppt**
Phosphate, ppa
Phosphate, pp»**
NUrate-N, ppa
ftUrlte-M. ppa
N1tr1te-N, ppa**
AmonU N, ppn
TW. pp»
Fecal Colls. MPH/
100 air*
BOD. PP»
"u
.047
1.5x10"*
5.3x10"*
6.9x10"*
3.7X10"5
-2.7x10"*
-l.SxlO"5
7.U10"5
2.4xlO"5
7.9X10"5
9.6xlO"5
5.8x10-*
2.3x10"*
2.2x1 O"3
n*
.040
1.3x10"*
2.1x10"*
3.2x10"*
1.4xlO"S
-2.1xlO"5
-4.9xlO"5
-4.7xlO'6
l.SxlO"5
+.9xlO-5
6.3xlO"5
3.6x10"*
1.7x10"*
1.7xlCT3
.033
9.«x10"5
-1.1x10"*
-S.lxlO"5
S.SxlO"5
-4.0xlO"5
-8.4x10"5
-a.ixio'5
7.2xlO"6
1.9xlO"5
3.1X10'5
1.4x10"*
9.8x1 O'5
1.2x10"*
SB
X
X
X
X
X
X
X
X
X
X
bu
9.37
1.05
7. 05
5.72
.712
.314
-.608
1.52
.079
-1.23
.421
3.40
4.34
8.67
b
4.59
.968
6.10
4.33
.627
.261
-.707
1.31
.055
-1.32
.329
2.77
4.15
7.18
bi
-.200
.883
5.15
2.94
.541
.208
-.805
1.09
.032
-1.40
.237
2.15
3.95
5.69
56
X
X
X
X
X
X
X
X
X
X
X
X
X
rz
.597
.469
.020
.071
.038
.056
.084
1.7x10"*
.138
.105
.150
.111
.214
.320
F
123.
73.2
1.69
3.04
1.59
5.21
7.98
.015
13.9
10.2
15.3
10.8
23.7
40.9
SF
X
X
X
X
X
X
X
X
X
Xu
8563.
8563.
8563.
8563.
8563.
8560.
8560.
8560.
8560.
8560.
8560.
8560.
8560.
8560.
h
13.0
13.0
13.0
13.0
13.0
5.6
5.6
5.6
5.6
5.6
5.6
5.6
5.6
5.6
YU
130.0
2.11
24.8
17.9
1.25
1.09
.04
4.07
.466
-.332
1.59
9.47
5.96
29. S
Yt
2.7
.43
1.8
2.7
0.44
.010
-2.0
.17
.015
-1.77
.037
.27
2.44
.94
n
85
85
83
42
42
89
89
89
89
89
89
89
89
89
cn
cx>
*ppt * parts p«r trill ton
**Lo8,0 C b + -S
Mote: SM inmnli *t foot of Table 10 and text for explanation of coluon headings.
-------
atrazine, and trifluralin on the Buffalo Bill Watershed, results might not
compare well because of the availability factor discussed earlier. Thus fur-
ther emphasizes the highly site-specific nature of the potency factor.
Michigan State Farms
In a study paralleling that at Watkinsville, Georgia, runoff data were
collected from test plots near East Lansing, Michigan. Table 15 shows results
of regression analyses using the data from test plot P-06 from the Michigan
State University study.
Regression results with the Michigan State University data showed very
weak relationships between suspended solids and all other runoff quality con-
stituents, although statistically significant relationships were obtained for
all except nitrate. The best relationships were with atrazine in a log-log
regression (r2 = 0.138) and total phosphorus (r2 = 0.187). With respect to
TKN and ammonia, relationships are significant but borderline.
Comparing these results with those from the Watkinsville data (Table 10),
the atrazine confidence limits for slopes and intercepts are similar. In con-
trast, however, those for TKN and ammonia have none of their ranges in common.
Some comments should be made here regarding the Michigan State data base.
First, regarding comparison with the Watkinsville data, climatic factors
should be considered. During part of the year, precipitation at the Michigan
site is in the form of snow, and the ground is frozen. Secondly, several
large-scale errors were noted in the data base such as mispunched values and
off-column data. The fact that these errors were found leads to some suspi-
cion about the veracity of the remaining data, and although every effort was
made to purge errors from the data base in the present study, results pre-
sented in Table 15 should be used with appropriate caution.
Redwood National Park
As discussed earlier, the data base for Redwood National Park is very
large, but data for suspended solids were not obtained for the same samples as
were water quality data. Accordingly, observations from which correlations
could be computed were inadequate. An attempt was made to pool data both
spatially and temporally, justifying this by the relatively uniform geology,
land-use, and soil type and requiring suspended sediment to have been observed
within 10 minutes of other water quality parameters. Nonetheless, it was pos-
sible to obtain only from 7 to 17 observations, and regressions were either
non-significant or borderline-significant. In view of the data limitations,
non-significant fitted lines, and likely misleading results of analyses, the
statistics will not be presented.
Seattle; Viewridge 1
Data for three distinct urban watersheds in Seattle, Washington, were
subjected to regression analysis. The first to be discussed here is Viewridge
1. As described in Table 8, this watershed represents a high density urban-
ized area containing many old residences. Table 16 shows results of regres-
sions of runoff water quality on runoff suspended solids concentration.
59
-------
TABLE 15. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATION {IN g/£)
AND CONSTITUENTS SHOWN {CONCENTRATIONS OF SUBSTANCES IN RUNOFF) FROM
MICHIGAN STATE UNIVERSITY STUDY
Constituent,
Units
Atruine. pf*"
Atrazlne, ppb1"
TKN. ppn
1KH. p?Bt
Hirnonia-N, ppm
Aroonii-N, ppm"
t ^
Atnonid-N, ppm
ToUl P. ppm
ToUl P. ppm**
ToUl P, ppn1"
Kitrate-N. ppm
HHr«te-N. ppn**
Nitrite, ppm*
,g
,107
.431
.653
.321
.299
.032
.316
.076
.077
.284
.157
.081
.346
rf
.060
.298
.271
.387
.164
.OK
.193
.053
.051
.191
-.020
.028
.154
"t
.013
.165
-.H2
.054
.029
.014
.070
.030
.026
.099
-.198
-.025
-.037
i§
R
X
X
X
X
X
X
X
X
"u
1.52
1.6Z
5.33
.674
2.08
.253
.299
.476
-.351
-.294
1.92
.117
.120
b
1.44
1.56
4.63
.623
l.BQ
.163
.253
.428
-.404
-.330
1.55
.006
.046
bl
1.35
1.49
3.74
.572
1.53
.114
.206
.360
-.458
-.366
1.18
-.104
-.028
5b
3.
X
X
I
X
X
K
X
X
X
X
r'
.049
.138
.021
.080
.058
.075
.092
.187
.148
.157
.0006
.012
.028
F
£.35
19.8
1.58
4.23
4.65
6.19
7.74
16.6
12.5
13.4
.043
.873
2.04
5f
j
X
1
a
X
X
X
X
X
Iu
B.06
.895
8.07
.910
3.07
8.07
.855
8.07
8.07
.895
8.07
8.07
.895
Xi
.
-------
TABLE 16. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATION (IN mg/£)
AND CONSTITUENTS SHOWN (CONCENTRATIONS OF DISSOLVED AND SUSPENDED SUBSTANCES IN RUNOFF)
FROM VIEWRIDGE 1, SEATTLE, DATA
Constituent.
Units
Ortho K)4-P, pp>f
Total P. pp»f
Inunla N, ppn
NOZ + HOj-N. ppaf
Orj «, ppi*
Conductivity,
lotos/catr
Turbidity, JTUt
Lead. ppbf
Hue. ppb*
"u
.255
.311
.481
-.047
.348
-11. 1
.432
.572
.347
m*
.172
.237
.336
-.140
.280
-24.6
.358
.490
.283
"t
.089
.162
.191
-.233
.212
-38.2
.284
.408
.218
S»
X
X
X
X
X
X
X
X
X
"u
-1.31
-.807
-1.83
.030
-.346
164.
.646
1.54
1.57
b
-1.43
-.917
-2.05
-.108
-.447
144.
.537
1.42
1.47
bl
-1.55
-1.03
-2.26
-.246
-.547
124.
.428
1.30
1.38
^
X
X
X
X
X
X
X
X
r2
.083
.175
.102
.046
.263
.065
.332
.430
.289
F
16.9
39.5
21.0
8.92
66.5
13.0
92.5
140.1
75,5
Sf
X
X
X
X
X
X
X
X
X
"u
3.66
3.56
3.66
3.66
3.66
3.66
3.66
3.66
3.66
h
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Yu
.012
.616
.108
.63
.798
246.8
2.00
3.22
2.77
Y
-2.01
-1.23
-2.0
-1.49
-2.04
26.0
-.968
1.28
.979
n
188
188
188
188
188
188
188
188
188
iC-b + Blo^s
^C b + 1og1(J S
Note: See coBKnts it foot of Table 10 and text for explanation of colum headings.
-------
All runoff water quality constituents were significantly correlated with
suspended solids. In some cases, notably lead, zinc, turbidity, and organic
nitrogen, values of r2 were substantial with r* being as high as 0.43 for
lead. Thus 43% of the variance of runoff lead concentration can be explained
by the suspended solids concentration. Again, it must be emphasized that this
may be in part due to association of lead with particulate matter, but it may
also be due at least in part, to some concomitant factor. In reality, the
strong correlation probably represents both physical association and concomi-
tant variables causing both lead and suspended matter to be flushed from the
watershed.
Zinc was also found to be strongly correlated with suspended solids
(r2 = 0.289) and as well, with lead in the runoff (r2 = 0.557 for logio [Zn]
= b + m log [Pfc,]). As might be expected, total phosphorus was correlated more
strongly with suspended solids than was orthophosphate, which is highly solu-
ble, and would be expected to be less associated with suspended solids. Even
the sum of nitrate and nitrite (as a single parameter), which is very soluble
and generally considered to be unassociated with particulate matter, was sig-
nificantly correlated with suspended solids. Perhaps it is reasonable to as-
sume that this correlation (r2 = 0.046) represents that purely due to concomi-
tant factors such as the propensity of overland flow to carry both pollutants
and suspended solids when they are not associated.
It is of interest to compare correlations shown in Table 16 with regres-
sions on turbidity. These are shown in Table 17. Despite the fact that the
correlation between turbidity and suspended solids is not very strong (r2 =
0.332), the values of r2 in Table 17 correspond very well in all cases with
those in Table 16. Apparently, the variance shared in common between turbid-
ity and suspended solids is also essentially variance shared with other pollu-
tants.
It is also of interest to examine the correlations between runoff water
quality parameters and the number of antecedent dry days. This is the case
because some of the runoff models (e.g., SWMM) predicate quality on "dust and
dirt" accumulation, which is, in turn, a function of antecedent dry days.
Table 18 shows these correlations.
The correlation between suspended solids concentrations and number of
antecedent dry days is significant (r2 = 0.114, F = 23.9). However, with only
three exceptions (conductivity, N02 + N0s» and organic nitrogen), direct cor-
relations between pollutants in the runoff and number of antecedent dry days
are higher than between suspended solids and antecedent dry days for this data
base. This suggests that it might be better to estimate many pollutants in
urban runoff directly as a function of dry days without introducing the sus-
pended solids intermediate. This is especially true for orthophosphate, total
phosphorus, and ammonia which, while correlating well with number of antece-
dent dry days, correlated very poorly with suspended solids. Since they are
poorly correlated with suspended solids, and because suspended solids is, in
turn, poorly correlated with number of antecedent dry days, this means the al-
gorithm as used in SWMM can be especially poor for simulating these pollutants
exported from urban sheds. It must be emphasized, however, that this is true
for this data base and other watersheds do not suggest this as strongly.
62
-------
TABLE 17. STATISTICS FOR CORRELATIONS BETWEEN TURBIDITY (IN JACKSON TURBIDITY UNITS) AND
CONSTITUENTS SHOWN (CONCENTRATIONS OF SUBSTANCES IN RUNOFF) FROM
VIEWRIDGE 1, SEATTLE, DATA
Constituent,
Units
Ortho P04-P, DOB*
Total f. pp»f
Annofila N, ppnt
NOj + M03-N. ppnf
Org-H. ppmt
Conductivity!
Mhos/cn'n-
Lead, ppb+
line. ppb+
"u
.420
.531
.700
.047
.571
-.034
.877
,527
m
.287
.413
.463
-.106
.463
-22.4
.739
.420
"l
.154
.295
.226
-.259
.355
-44.8
.601
.3)3
Sm
X
X
X
X
X
X
X
"u
-1.35
-.890
-1.81
-.025
-.423
158.
1.48
1.54
b
-1.49
-1.02
-2.07
-.190
-.541
134.
1.33
1.43
bl
-1.63
-1.15
-2.32
-.356
-.658
109.
1.18
1.31
5b
X
X
X
X
X
X
X
X
r2
.089
.206
.075
.010
.279
.021
.377
.246
F
18.2
46.3
15.0
1.90
71.9
3.95
113.
60.8
SF
X
X
X
X
X
X
X
"u
2.04
2.04
2.04
2.04
2.04
2.04
2.04
2.04
h
-.968
-.968
-.968
-.968
-.968
-.968
-.968
-.968
"u
.012
.616
.108
.600
.924
246.8
3.22
2.77
h
-2.01
-1.3Z
-2.01
-1.70
-2.04
26.0
1.28
.979
n
188
188
188
188
188
188
188
188
0%
GO
C " b *
tt,
C b +
Mete: See torments at foot of Table 10 and see text for explanation of colum headings.
Note that In this table does not represent potency factors.
-------
TABLE 18. STATISTICS FOR CORRELATIONS BETWEEN ANTECEDENT DRY DAYS (DD) AND CONSTITUENTS SHOWN
(CONCENTRATIONS OF DISSOLVED AND SUSPENDED SUBSTANCES IN RUNOFF)
FROM VIEWRIDGE 1, SEATTLE, DATA
Constituent,
Units
Ortho P04-P, pp«+
Total P, PR-*
AmnirU N. ppnt
NOj + N03-N, pp«+
Org-N, pp«t
Conductivity,
Turtridlty. JTU*
Lead, ppb^
Zinc, ppbf
Suspended .
solids, ppa
!H
.355
.359
.644
.140
.188
8.59
.278
.458
.237
.434
.288
.298
.523
.053
.118
-4.24
.200
.376
.171
.309
"i
.221
.236
.403
-.034
.047
-17.1
.123
.293
.106
.184
sm
X
X
X
X
X
X
X
X
"u
-1.27
-.E75
-1.73
-.256
-.063
122.
.999
1.99
1.83
1.33
b
-1.32
-.721
-1.82
-.321
-.116
113.
.940
1.92
1.78
1.23
b£
-1.37
-.767
-1.91
-.387
-.169
103.
.832
1.86
1.73
1.14
Sb
X
X
X
X
X
X
X
X
X
X
r2
.279
.332
.236
.008
.056
.002
.125
.302
.127
.114
F
72.0
92.5
74.6
1.48
11.0
.430
26.5
80.6
27.1
23.9
SF
X
X
X
X
X
X
X
X
xu
1.S9
1.59
1.59
1.59
1.59
1.59
1.59
1.59
1.59
1.59
h
-1.02
-1.02
-1.02
-1.02
-1.02
-1.02
-1.02
-1.02
-1.02
-1.02
YU
.012
.616
.103
.63
.798
246.8
2.00
3.22
2.77
3.66
Yt
-2.01
-1.32
-2.0
-1.49
-2.04
26.0
-.968
1.28
9.79
0.0
n
IBB
IBS
IBS
188
IBS
IBS
IBS
IBS
188
188
C - b * m 1«g,0 DD
« b + B log,0 OD
Mote: Sec csmntnts *t foot of Tiblt 10 and see text for explanation of colunn headings.
Hott tint D In this table does not represent potency factors.
-------
There are other implications of antecedent dry days correlations as well.
It might be interpreted that those pollutants correlating well with antecedent
dry days but not with suspended solids do, in fact, accumulate on urban water-
sheds over time, but are not strongly associated with particulate matter. In
this case, suspended solids do not appear to accumulate on the watershed as a
simple function of time, at least on the time scale represented in the View-
ridge 1 data.
In contrast, some pollutants not correlating well with suspended solids
also do not correlate well with number of antecedent dry days. This is true
for NOs + NOg. Apparently, as suggested by this data base, these pollutants
not only do not associate with particulate matter, they also do not appear to
accumulate as a function of dry days. On the other hand, lead and zinc cor-
relate both with suspended solids and with dry days suggesting time-based ac-
cumulation and possible association with particulate. Turbidity, as would be
expected, correlates well with suspended solids, and like suspended solids,
correlates only weakly with number of antecedent dry days.
Seattle; South Seattle
The South Seattle Watershed is an industrialized urban area, geographi-
cally located to the south of Viewridge 1. Table 8 summarizes the data base
for this watershed, and its location is shown in Figure 13. Table 19 shows
results of statistical analysis of runoff data.
For several constituents (orthophosphate, ammonia, and conductivity; see
note a in Table 19), autocorrelation in the residuals suggested that observa-
tions could not be considered independent. The statistics are shown only for
comparison purposes, as taking first differences appears not to be satisfac-
ory under conditions exhibited by this data base (see Appendix A and note a
in Table 19).
For constituents not transformed to first differences (the five at the
top of Table 19) some of the correlations with suspended solids were very
strong. These are lead (r2 = 0.671), total P (r2 = 0.505), and zinc (r2 =
0.390). Comparing with the corresponding entries in Table 16 (Viewridge 1),
the squared correlation coefficients (r2) are 0.430, 0.175, and 0.289, re-
spectivelyall lower, with total P substantially so. It should be pointed
out that these comparisons of rz are subjective and do not represent a statis-
tical test. Statistical comparisons of r2 were not made in this study.
Similarly, for N02 + N03 the correlation appears stronger than for the
Viewridge 1 data. For organic nitrogen, the relationship is reversed, with
the correlation weaker than for Viewridge 1.
Turning now to the 95% confidence limits on the slope (m) and intercept
(b), those for Total P and NO? + NO^ are different comparing South Seattle and
Viewridge 1 (Tables 19 and 16). This is a somewhat subjective comparison made
by examining the ranges between mu and ma and bu and bA for overlap without
making any statements about probabilities that the slopes and intercepts are
in fact different.
65
-------
TABLE 19. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS CONCENTRATION (IN mg/fc) AND
CONSTITUENTS SHOWN (CONCENTRATIONS IN RUNOFF) FROM THE
SOUTH SEATTLE DATA BASE
Constituent,
Units
Total P, ppBf
N02 + NOj-N, ppmt
Org. H. fjaf
Lead, ppbf
Zinc, ppbf
Ortho PO^-P, ppm
Aranonla N, ppm
Conductivity,
l*hos/c«tT
"u
.478
.349
.293
.546
.301
.207
.202
18.7
»*
.419
.252
.211
.497
.256
.132
.106
1.60
Bl
.360
.156
.129
.448
,211
.057
.010
-15.5
Sm
X
X
X
X
X
X
X
bu
-1.30
-.806
-.429
1.50
1.99
.043
.051
3.63
b
-1.40
-.967
-.567
1,42
1.91
.002
-.002
-5.67
»l
-1.49
-1.13
-.704
1.33
1.84
-.038
-.054
-15.0
Sb
X
X
X
X
X
r2
.505
.121
.116
.671
.390
F
201.
27.1
25.8
401.
126.
see note a below
.063
.026
l.BxlO"4
12.3
4.80
.034
SF
X
X
X
X
X
X
X
Xu
3.47
3.47
3.47
3.47
3.47
1.95
1.95
1.95
Xl
0.0
0.0
0.0
0.0
0.0
-1.49
-1.49
-1.49
YU
.409
.654
.987
3.12
3.01
1.47
1.71
372.
\
-1.31
-1.55
-2.04
1.27
1.79
-.918
-1.14
-410.
n
199
199
199
199
199
184
184
184
cr»
'109,0 C b + m log1Q S
nC b + * log,,, S
Not* a: Regressions for constituents listed below this point were done on first differences rather than
on the observations themselves. This was because of a problem of autocorrelated residuals.
Results are presented here primarily for comparison purposes since they are not useful for
developing potency factors. Also, as discussed briefly in Appendix A, taking first differences
(i.e., regressing Y^_i - Y^ on X<_i - X-j rather than Yj on X^) can cause a severe and unrealistic
reduction in r' under some circumstances. Please also see comments at foot of Table 10 and see
text for explanation of column headings.
-------
If the ranges do overlap, then they share values in common. Since both
slopes and/or intercepts could have identical values, we cannot conclude that
the slopes and/or intercepts are different for the fitted lines being com-
pared. The confidence limits for both m and b for lead, comparing South
Seattle and Viewridge 1, are almost exactly coincident. Considering that the
ranges are fairly narrow and r2 values are very high, this suggests strong and
consistent relationships between lead and suspended solids for at least two
Durban watersheds.
Zinc exhibits similar confidence limits for m, comparing the two water-
sheds. The ranges for b appear to be different, however. Despite the rela-
tively low r2 for organic N at South Seattle, the confidence limits for both m
and b overlap those at Viewridge 1, suggesting some spatially consistent wash-
off behavior of organic N in addition to washoff behavior for lead and zinc on
these two watersheds.
Turning, now, to correlations with antecedent dry days, Table 20 shows
correlations between this parameter and runoff quality. Correlations here are
not nearly as strong as at Viewridge 1, possibly suggesting different, or at
least more variable pollutant accumulation phenomena. In this data base, sus-
pended sediment correlates more weakly with antecedent dry days than at View-
ridge 1. Using the model:
Log1Q S = b + m log1Q DD (6)
r2 = 0.098 (not shown in Table 20). For the same relationship with S repre-
senting turbidity instead of suspended solids, r2 = 0.045. Thus, despite the
fact that pollutant correlations with dry days are weaker here than at View-
ridge 1, several of them (Total P. N02 + NOs, and zinc) might still be better
estimated directly from antecedent dry days than from suspended solids, even
given the strong correlation between suspended solids and total P and zinc.
It is interesting to note that at South Seattle, unlike at Viewridge 1, NOa +
N02 was fairly strongly correlated with antecedent dry days. Organic N was
not well correlated at either site.
Seattle; Southcenter
Southcenter is the third and final Seattle watershed to be examined. As
shown in Table 8, it contains a new shopping center and is urbanized; See
Figure 13 for the location of the sampling site. Table 21 provides statistics
for correlations of suspended solids with various other pollutants.
As shown in Table 21, all regressions were statistically significant.
That is, the trends have not occurred by chance alone. Figure 14 shows a com-
parison of the results from Southcenter, South Seattle, and Viewridge 1.
As shown in both Table 20 and Figure 14, the confidence limits for the
slopes for Southcenter all overlap those for South Seattle except Organic N,
while comparing slopes between Southcenter and Viewridge 1, the ranges for
Total P, lead, and zinc overlap. In terms of intercepts, and comparing South-
center and South Seattle, the confidence limits overlap for all but organic N
67
-------
TABLE 20. STATISTICS FOR CORRELATIONS BETWEEN ANTECEDENT DRY DAYS (DD) AND CONSTITUENTS SHOWN
(CONCENTRATIONS OF SUBSTANCES IN RUNOFF) FROM SOUTH SEATTLE DATA BASE
Constituent.
Units
Totil P. ppnf
IK^ + IKyN, ppmf
Org. N. ppnt
Lead, ppbf
21ne. ppbf
Vi
.265
.457
.234
.257
.189
to
.190
.375
.153
.179
.137
"l
.115
.292
.072
.100
.085
Sn,
X
X
X
X
X
"u
-.779
-.690
-.248
.216
2.29
b
-.841
-.760
-.316
.210
2.24
bi
-.904
-.829
-.383
.203
2.20
Sb
X
X
X
X
X
r2
.113
.289
.066
.094
.121
F
25.0
80.3
14.0
20.5
27.2
SF
X
X
X
X
X
*u
1.59
1.59
1.59
1.59
1.59
\
-1.02
-1.02
-1.02
-1.02
-1.02
Yu
.409
.654
.987
3.12
3.03
Yt
-1.56
-1.56
-1.53
1.272
1.79
n
199
199
199
199
199
00
+lo«i0 C b + » log,,, DO
Not*: See comcnts at foot of T«ble 10 ind see text for explanation of colum headings.
Note that 1n this table does not represent potency factors.
-------
TABLE 21. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS (IN mg/£) AND CONSTITUENTS
SHOWN (CONCENTRATIONS IN RUNOFF) FROM SEATTLE SOUTHCENTER DATA BASE
Conititjent.
Units
Total P. pf»t
NOj + I*J3-II. PP"*
Orj. H. ppa+
Lead. ppbf
Zinc. Dpbf
"u
.416
.443
.620
.728
.422
«*
.312
.305
.502
.629
.312
"t
.208
,165
.303
.530
.203
S*
X
X
X
X
X
"«
-1.09
-.098
-.867
1.59
1.60
b
-1.23
-1.17
-1.03
i.ts
1.65
bl
-1.37
-1.36
-1.19
1.32
1.50
Sb
X
X
X
X
X
r2
.152
.089
.265
.446
.140
F
35.4
19.Z
70.9
159.
32.0
SF
X
X
X
X
X
"u
3.06
3.06
3.06
3.06
3. 06
h
0.0
0.0
0.0
0.0
0.0
TU
.644
.987
.987
3.43
3.06
Tt
-2.04
-2.04
-2.04
0.99
0.99
n
199
199
199
199
199
CTi
10
b + ios,0 s
Mate: SM amtnts »t fcot of Table 10 *nd see text for explimtloa of col urn headings.
-------
Total P, ppm
NOg +N03 N, ppm
Organic N, ppm
Lead, ppb
Zinc, ppb
0.75
0.50
r2
O.25
0.0
Total P
ppm
Slope (m)
ITS' f'f^f\
rvw W'V'V'V'VJ
Intercept (b)
*-'v'v^-'v'v>'v'v.J
-.20 0.0 0.20 0.40 0.60
_i_
-2.0 - .0
KEY--
0.0
1.0
2.0
I
ppm
Organic N
ppm
Lead
ppb
Zinc
ppb
South Seattle
3 Southcenter El
0
Figure 14. A comparison of statistical results among Southcenter,
South Seattle, and Viewridge 1.
-------
and zinc, although for zinc, they are close. Comparing confidence limits for
intercepts between Southcenter and Viewridge 1, only those for lead and zinc
overlap.
The ramifications of these comparisons are that if confidence limits are
subjectively compared, and if it is assumed that where confidence limit esti-
mates for two distinct sites share any values in common, they can be consid-
ered to be indistinct, then results for Seattle data are very consistent spa-
tially. This is especially true for slopes, which are of particular interest
here as potency factors. Although the data were not pooled and analyzed, a
good estimate of the potency factor for any pollutant would be one shared in
common by two or even all three sampling sites, bearing in mind, of course,
that values from another urban area or even another watershed in Seattle might
be somewhat different. The fact that the three sites in Seattle give such
consistent results is, however, very encouraging. Later in this report, re-
sults of some multiple regression analyses are presented which further support
(again in a subjective way) the contention that the three watersheds in Seat-
tle behave very similarly in terms of runoff quality and its relationship to
various other factors.
Honey Creek
The Honey Creek basin data base had the greatest number of observations
of any examined in this study. It also provided satisfactory data for the
greatest number of runoff water quality constituents, and because many storms
were monitored with relatively few observations in each (rather than a few
storms intensively sampled), there was also no residuals autocorrelation prob-
lem.
Table 22 shows correlations between suspended solids and other runoff
water quality constituents for Honey Creek.
With the exception of orthophosphate, all regressions were significant.
Many had very high r2 values, some being the highest obtained in the study.
These include iron, total phosphorus, and total phosphorus minus orthophos-
phate phosphorus. The fact that the F values are so high reflects both sig-
nificance of the regressions and the large sample size.
A comment is in order here about the higher r2 values observed in regres-
sion of the non-transformed over the log transformed data. Despite this, the
transformed data regressions are probably more reliable since the resulting
data distributions over both the X and Y axes are much more uniform while be-
ing similarly linear. In the untransformed data, much of the data formed a
cluster with a few remote values strongly biasing r?. Figures 15 and 16 are
examples of plots showing this. Figure 15 is the Honey Creek regression of
runoff iron concentration on suspended solids. Figure 16 shows the regression
of logio iron on lo91Q suspended solids. The data presented at the foot of
the figures correspond to those presented in Table 22. Definitions for these
statistics are provided in Section 3. The plotted numbers represent the num-
ber of observations at each set of coordinates. The line of "+" symbols is
the fitted line, whereas the line of "" symbols represents the confidence
71
-------
TABLE 22. STATISTICS FOR CORRELATIONS BETWEEN SUSPENDED SOLIDS (IN mg/£) AND CONSTITUENTS
SHOWN {CONCENTRATIONS OF SUBSTANCES IN RUNOFF) FROM THE HONEY CREEK WATERSHED
Constituent,
Units
Total P. ppB*
Total P. ppn
Ortho P04-P, ppmt
H02 + NOj-N, ffaf
taraonla N, ppn+
TKN. pp»
TXN. ppnf
Chloride, ppm*
S102. pp.f
Iron, ppb
Iron, ppb
Conductivity.
imtasfaiff
Total P-Ortho
P04. pp»
Total P-Ortho
W4. pp"1"
"u
.432
1.2xlO"3
.062
.256
-.080
2.9xlO'3
.241
-.146
.107
.042
.897
-.134
l.ZxIO"2
.554
n*
.405
l.lxlO'3
.021
.214
-.155
2.lx10-3
.182
-.172
.076
.040
.815
-.152
1.1x10-Z
.474
nt
.378
l.OxIO"3
-.021
.171
-.230
1.4x10'3
.123
-.195
.045
.037
.733
-.170
l.OxlO"2
.394
Sm
X
X
X
X
- x
X
X
X
X
X
X
X
X
bu
-1.26
.194
-1.08
.342
-.609
2.27
.093
1.81
.734
3.95
-.648
2.97
.096
-1.49
b
-1.31
.183
-1.15
.266
-.745
2,01
-.027
1.77
.679
3.17
-.807
2.94
.086
-1.64
bi
-1.36
.172
-1.23
.189
-.880
1.75
-.146
1.73
.624
2.39
-.966
2.91
.077
-1.78
s,
X
X
X
X
X
X
X
X
X
X
X
X
X
r2
.577
.679
.001
.134
.026
.194
.238
.254
.090
.828
.697
.310
.752
.180
F
864.
1340.
.929
98.1
16.4
29.0
37.9
203.
24.0
813.
388.
284.
1897.
135.
IMMRHM
X
X
X
X
X
X
X
X
X
X
X
X
X
3.21
1620.
3.21
3.21
3.21
1620.
3.21
3.21
3.21
1620.
3.21
3.21
1620.
3.21
Xt
-.740
.479
-.740
-.740
-.740
.479
-.740
-.740
-.740
.479
-.740
-.740
.479
-.740
Yu
0.156
1.43
0.154
1.24
.295
7.46
.864
2.11
1.11
56.1
1.78
3.04
1.78
2.91
y
-2.00
0.0
-2.00
-0.37
-2.24
0.56
-.240
.535
.049
.075
-1.14
2.28
0.0
-3.07
n
635
635
628
636
608
123
123
612
244
171
171
633
627
619
no
+loglo C b *
tt.
+ Iog10 S
Note: See cements at foot of Table 10 and see text for explanation of colunm headings.
-------
CO
54.67*999.
40.6T4999«
Iron, ppb
26.67««*t
12*675000
<
** , *» tt
** .. *» tt
* .t *« tt
* tt ** t. **
* ,, »* ,1 *
** t* »* ., **
it ** ,t «
*« tt »« .. **
** tt *»t ,, **
»* t. ** >t **
** t, *+ .. *
* tt ** tt *
tt ** .. **
** . »* -tt *
** tt ** tt *
** tt ++ t. **
** .. ** t. *
,. 11** ,, *
* t. +* t, *
** .. »* .. **
* *. ** .. *
* t." *» tt **
., 1** ,, **
** . .t ** .. *
* , .t *«, .. *
* .1 »* (. **
* tt ** .. *
.. ** rr **
.11 »* .. »*
**.!»..**
! .t 1 t* tt **
1 » 1. 1 ** t. *
** i. i*« §r *
** .. *i .t **
.. in .. *
i i* i,.i u .. **
11 it 11 .. **
** ,n i **i .. **
*t i.' n »i ,. *
*12 .11! 1»* 1 ., *
« 11 .1 1 1*«1 .. »
1,1 11* ,.' «»
.1 11 .. ** 1
. 1 1*23»1 ,. *
1236312 I!.' *
38251 .. *
162411. **
2C .. 1 **
L63 **** ,
16t,«60 5412. «fcO S22.460 T01,«*fl 2.460 1062.460 1242,460 1422,460 1602,460 1782,460
Suspended solids, ppm
.9750 U
tINfc»R RECBE38ION STATISTICS
2.39 Et*)« J.17 ,9750 UL * i,95
,97bO LL B» .366-01 EtB)« ,39b-0l .9750 UL B« ,«23>01
R SQUARED*, 627930 F« 6l3,2 N* {71
Figure 15. Scatter plot and regression of iron concentration on suspended solids concentration
-------
l.'6 Sao oo.
t.218000
.598000
Log1Q Iron, ppb
-.642000
*+ t. »t
** I. *l
** I* +»
** t. +2
* .. +1
' .1 11+ ,.i
,.
1 3*
*
* .. I 11+2 J ..
.. 1+22 1 ,,
** 1 .. 21 11++11 ,,
» 1 .. 1 2211 1 ,, *
I 11 .. Illti 1 ,. **
** 1 t 1 .. 1 I +1 1 .. *
** 1 ii 1 11 .. «
** 1.. t 1+1 ., I *
.. I + + 3I21 ,t ***
* II. 3 2+ 1 t .. *
* 1 i, I3»2 |22 1 1,. *
»* ., I 11 +1 3 ,,, *
** 1 iT. 1211 t+ I I ,, **
» ..11 *+21 I ,. *
* I It ++ I .. **
.. 1 2 1+ ,, ««
, .1 1++ **
r. t+i i .. i *
**
t
+ 2 1
.,
** ., ++ii i. **
** .'« ++ ,» i **
* ... ++ i ii *
* ,. ** ,, 2 ***
* .. ++ ;.* **
«* ,. i ** ,1 *»
* ... ++ i ,, i **
** i. ++ L. i i **
* . ++ i .. a **i
* »+ i .; *
,..' ++ 11 ,. t »*
.. ++ .." i" »*
++ .. l l*
++ ... l *
*+ .. 1 U
++ ..
+ »*
**
»
.1 **
., **
I *» 1
.'. *
** 1
**»
I. 1
1.09J
2,?JS 2.60J 2.071 3,3«S
suspended solids, ppm
RFCRfcSSIUN STATISTICS
."750 Li. Ax..966 t(*3«-.ao7 ,9750 UL As..60«
Lt B* ,733 EtB)« .815 ,9TiO ML B« ,d»7
R SQUAREOs,6966?6 F* 3«H.I N* 171
Figure 16. Scatter plot of 1°g1Q iron concentration on log,Q suspended solids concentration.
-------
interval estimate of the population mean for a given value on the abscissa.
The line of "*" symbols is the corresponding confidence interval estimate for
prediction of individual values on the ordinate axis.
Table 23 presents a summary of potency factors derived from the results
of this study.
MULTIPLE REGRESSION
As shown in the preceding section, suspended sediment concentration can
generally account for only a small proportion of the variability of other run-
off water quality constituents. Multiple regression analysis was used to ex-
amine other variables in concert as predictors of runoff quality. The vari-
ables included those shown in Table 24.
In the regression analyses, it was decided to include water quality con-
stituents as independent variables since the purpose of multiple regression
analysis here was less for prediction than for explication. Each step in mul-
tiple regression amounts to partialing out the independent variable from the
dependent variable (see Appendix A) such that the array of residuals is a vec-
tor orthogonal to (independent of) each independent variable already used in
the regression. That is, the coefficient of linear correlation between the
residuals (difference between observed and predicted values of the dependent
variable) and each of the predictor variables is identically zero. Examining
multiple regression results with respect to variables statistically correla-
ting with the residuals at each step allows consideration of possible physical
phenomena and mechanisms influencing the values of the dependent variable.
Natkinsville; Plot P-04
Table 25 shows correlations (r) between dependent and independent vari-
ables for Watkinsville test plot P-04 (1974) data. Since not all variables
were used as predictors for all regressions, some correlation coefficients are
not available.
Good predictors, overall, were "days since cropping," "cumulative rainfall
since cropping," and "hours since start of storm." The fact that "days since
cropping" was a good predictor variable would tend to make "cumulative rain-
fall since cropping" a good predictor as well (the correlation coefficient be-
tween the two predictors was typically about 0.98, making these variables vir-
tually identical in terms of predictive value). "Hours since start of storm"
was also a fair predictor, especially for atrazine and nitrate. However, it
too was highly correlated jr = 0.74) with "days since cropping," suggesting
that the length of storms (or perhaps sampling times within storms) depended
upon time of the year.
A comment is in order here regarding the high correlations between atra-
zine concentration and the various predictor variables. In this data base,
few satisfactory observations for atrazine concentrations were available
(about 20). Accordingly, these high correlations must be viewed with suspi-
cion, as they may well be spuriousmore a result of special conditions than
75
-------
TABLE 23. SUMMARY LISTING OF POTENCY FACTORS ESTIMATED FROM
AGRICULTURAL, SUBURBAN, AND URBAN WATERSHEDS EXAMINED
IN THIS STUDY. SEE NOTE A
-------
TABLE 23 (continued)
I
£
3
BCD. ppm
Total P, ppm
N02 * N03-N. ppm
Organic K, ppm
Conductivity
umhos/cm
Turbidity. JTU
Lead, ppb
Zinc, ppb
Iron, ppb
Total P-
OP04-P, ppm
S102. ppm
Reference
Page (limber
5? 5? |f °f |f to « |« "c
co. co. co. !-* .SB » T ^ .c *i J: *- ** B
II 11 I! 11 ll IS. II U if
52 '
53
54
1.7xlO"3
(B)
58
.053
(B)
60
-.917
.237
(B.F)
-.108
-.140
(F)
-.447
.280
(B.F)
144.
-24.6
(B.E)
.537
.358
(B.F)
1.42
4.90
(B.F)
1.47
.283
(B.F)
61
-1.40
.419
(B.F)
-.967
.252
(B.F)
-.567
.211
(B.F)
1.42
.497
(B.F)
1.91
.256
(B.F)
66
-.841
.190
(B.F)
-.760
.375
(B.F)
-.316
.153
(B.F)
.210
.179
(B.F)
2.24
.137
(B.F)
69
l.lxlO"3
(B)
.266
.214
(B.F)
2.94
-.152
(B.S)
.040
(B)
i.ixur2
(B)
.679
,076
(B.F)
72
Notes: A.
Potency factors (m from C b t mS) provided 1n this table are for untransformed data unless otherwise
noted Commonly, the transformed data proved to be much more reliable than untransformed 1n terms of
statistical results, as explained 1n the text. Where transformations are noted, regression results for
untransformed data were judged unreliable. Note that C concentration of pollutant, S concentration
of suspended solids. For use with NPS, except at Uatklnsvllle and Michigan State U. Farms, multiply m by
lOOX Where constituent C Is In ppm, result Is In grains C per gram of S. Where C Is In ppb. result Is
in milllarams C per gram of S. For ppt, result Is ralcrograms C per gram of S. For Uatklnsvllle and
Michigan data, Instead of multiplying by 100X, divide by 10 and Interpret as just described. For example,
for aranonla at Watklnsvllle Plot P-04. we have .85 milligrams of ammonia per gram of sediment (because
sediment 1s In g/i) and the patency factor 1s .85/10 .085 ( .00085 gram per gram of sediment). For
dleldrln (Buffalo 8111 shed), the potency factor 1s .04 ppt x lOOt 4 ( .04 mlcrograms per gram of sediment).
I Intercept (b) statistically non-iero at 95t confidence level In a two-tailed test, but Ignored here.
' If CMs not specified. Intercept was not significant.
C. Overall regression not significant at 95% confidence level.
D. Potency factor (m) not significant at 95X confidence level.
I. log.. C b + mS. Upper value Is b, lower Is n.
F. log.. C b « " Iog10 S. UpP««" »«'u« '» D« 'ower f* "
6. C b » " log S- Upper value Is b, lower is n..
77
-------
TABLE 24. VARIABLES EXAMINED IN MULTIPLE REGRESSION AS CANDIDATE
PREDICTORS OF RUNOFF WATER QUALITY
Elapsed time since reference date
Month
0 Elapsed time since storm began
Dry days preceding storm (antecedent dry days)
Suspended solids concentration
Water quality variables
t Flow
Storm rainfall
78
-------
TABLE 25. SIMPLE CORRELATIONS (r) BETWEEN DEPENDENT AND INDEPENDENT
VARIABLES FOR WATKINSVILLE PLOT P-04 DATA
Dissolved »trailne*
Dissolved TKN
Dissolved anmonfi
Dissolved nitrite
Dissolved Phosphate
Dissolved chloride
.971
-.384
'.262
-.592
.347
-.313
-.979
-.492
-.655
-.771
.243
-.595
-.965
-.418
-.601
-.754
.320
-.579
.824
-.424
-.026
-.191
-.225
.282
/
.382
.014
.246*
-.099
-.032
.366*
.078
.609
.430
.444
.264
.236
.868*
-.601
.786*
1.00
.754
.084*
.354*
-.755
1.00
.783
.130*
.391
.405
-.033
-.025
-.415
.026
1
-.267*
-.099
.061*
-.171*
.015*
.681*
.697
1.00
-.269*
.579*
-.170*
-.138
-.077*
-.062*
-.433*
.071
.048
-.269
1.00
.133
-.265*
-.106
-.188*
-.194*
-.267*
.236*
.402
.404
-.031* .
1.00
Variables transferred by 1o9in-
-------
of any real, persistent, physical relationship. For the remainder of depen-
dent variables, there were typically about 66 observations, making correla-
tions somewhat more reliable.
Suspended sediment concentration generally proved a modest predictor with
correlations (r) ranging from 0.444 down to 0.078 except for dissolved TKN
(r = 0.609). In all cases except TKN, the r2 value shows that suspended
solids concentration could account for less than 20% of the variance of the
water quality constituent. Thus about 80% of the variance of dissolved ni-
trate cannot be accounted for by the variance of suspended solids. Only 46%
of dissolved TKN variance may be attributed to variance in suspended solids.
With the exception of dissolved phosphate, all other water quality vari-
ables correlated fairly well with one or more predictor variables. In addi-
tion to suspended solids, TKN correlated fairly well with "days since crop-
ping" (almost 25% of TKN variance), with runoff ammonia, and with runoff ni-
trate (almost 50% of variance in common). Dissolved ammonia correlated well
with "days since cropping" and with nitrate.
Table 26 presents multiple regression results for Watkinsville plot P-04.
The most important non-water quality variables were "hours since start of
storm," and "cumulative storm rainfall." It is of interest to note that in
some regressions (e.g., for dissolved TKN) "hours since start of storm" cor-
related more strongly with the dependent variable than did "cumulative storm
rainfall," suggesting a number of possible phenomena. These would include
volatilization, biodegradation, or photodegradation at the surface, coupled
with initial resistance to transport of sediment and/or pollutant, perhaps as
a result of slow soil wetting. In further support of this, there were cases
(e.g., dissolved TKN and phosphate as dependent variables) where, in later re-
gression steps than that shown in Table 26, both "cumulative storm rainfall"
and "time since start of storm" entered the regression.
Table 27 provides regression equations which may be of use for comparing
results of other studies, or as input in designing future investigations.
Watkinsville: Diphenamid and Trifluralin Residues
Data from Watkinsville plots P-01 and P-03 included runoff diphenamid and
trifluralin concentrations. In contrast to the previous discussion (plot P-04
data), analyses here encompassed more than 1 year of data. Data were pooled
so that for P-01, 2 years (1972 and 1973) were represented, and for P-03, 4
years (1972-1975) were represented.
Table 28 shows correlations (r) between diphenamid and trifluralin data
and the nine predictor variables. As shown, four variables were strongly cor-
related with dissolved trifluralin and diphenamid in the runoff. These were
"days since cropping," "cumulative rainfall since cropping," "suspended sedi-
ment," and herbicide (diphenamid or trifluralin) concentration in the sedi-
ment. The fact that time and rainfall since cropping strongly correlated with
the dependent variables suggests that available herbicide decreases (or is
less available for washoff) over time. These two predictor variables ("time"
and "rainfall since cropping") had 83% of their variance in common, and
80
-------
TABLE 26. MULTIPLE REGRESSION STATISTICS FOR WATKINSVILLE
PLOT P-04. ENTRIES ARE t VALUES FOR EACH
REGRESSION COEFFICIENT
Hours since start of storm
Days since cropping
Cumulative rainfall since
Cropping, inches
Cumulative rainfall during
storm. Inches
Rainfall Intensity at
t-l, 1n/hr*«
Rainfall Intensity at
t. in/hr
Suspended solids
concentration, g/t
Sediment atrazine, ug/t
Dissolved amonla N. ng/t
Dissolved TKN, mg/t
Dissolved total phosphorus,
mg/t
Sedinent amwnta N
concentration, ng/1
01ssolved nitrate N
concentration, ng/i
Sed. TKN concentration, mg/t
Dissolved phosphate f
concentration, mg/1
Sediment total phosphorus
concentration, mg/t
Dissolved chloride
concentration, ng/t
R*
a
n"
13.5
*
.816
42
V
-3.00
9.58*
*
*
*
*
*
.666
66
47
-5.12
9.40
.690
103
12
7.93
4.45
.629
66
66
3.92
*
*
3.32*
*
*
*
*
.2S2
66
13
-6.45
2.55
*
*
*
*
-3.79*
*
.465
108
25
Variables '0910 transformed
"t-l Implies runoff sampling t1m minus one hour.
^Substantial autocorrelation of residuals was noted. Estimated number
of Independent observations very much less thtn n. Estimated p. based
on an approximation due to Hold (see Appendix A).
Table entries are values of t for the coefficients. Selection of regressions
was made such that the greatest number of predictors was used subject to at
least 30 observations per predictor variable and all t significant at
a .OS. two-tailed test.
81
-------
TABLE 27. REGRESSION EQUATIONS WITH CONFIDENCE INTERVALS
FOR SLOPE AND INTERCEPT (SIMPLE REGRESSIONS ONLY,
a = .05, TWO-TAILED TEST) FROM
WATKINSVILLE PLOT P-04
Equations R2
1og,Q [A] » .405 t .12 * (.00478 ± .0006) KSS .943
1°9,0 [A] - 9.13 t .79 - (.047 i .0046) D .960
l«g,0 [A] - 3.49 t .32 - (.311 t .04) CR .931
[A] - (2.14 X ID'5 * .025) [A$ed] (2'*4 * '67) .754
[TKNp] - 4.12 i .92 - (.0082 i .0037) D .242
[TKNp] - 2.64 ± .23 + (1.19 ± .317) log,Q [NOj] .464
[TKND] 2.80 ± .21 + (1.62 ± .32) log]0 [NH, ] .619
[TKNp] = 3.13 - .213 CR + 1.50 log,Q [NH4 ] .666
[NH. ] - 2.96 ± .56 - (.0094 ± .002) 0 .429
4D
[NH, ] ' 1.50 i .29 - (.035 t .014) CR .361
[NH4 ] « 1.80 ± .126 + (.480 4 .095) [N03] .486
[NH4 ] - -.745 ± .212 t (.661 ± .101) [TKNp] .613
[NH4 ] * .779 - .0047 0 + .512 [TKNp] .690
[NO,] " .109 + .898 [NH. ] + .083 [Cl] .630
3 4D
[P04] - .053 + .0001 HSS t .088 [SS] .252
log,0 [Cl] 1.53 t .32 - (.0045 t .0012) D .354
1°910 [Cl] - .852 t .155 - (.018 ± .0049) CR .335
[Cl] (3.71 J 1.24) {[N03] + l.)('585 * -158)* '335
'«9,0 [Cl] - 1.79 - .0037 D + .084 CRS - .190 log,Q ([TKNS] + 1.)* .465
Explanation of symbols on following page.
82
-------
TABLE 27 (continued)
Explanation of symbols:
A = dissolved atrazine, ppb
HSS = hours since start of storm
D = days since cropping
CR = cumulative rainfall since cropping, in.
A d = sediment atrazine, ppb
TKNr, = dissolved total kjeldahl nitrogen, ppm
NH* dissolved ammonia nitrogen, ppm
NO, = dissolved nitrate nitrogen, ppm
Cl = dissolved chloride, ppm
P04 = dissolved phosphate phosphorus, ppm
SS = suspended solids, g/8,
TKN = sediment total kjeldahl nitrogen, ppm
*A constant (small relative to the values of
the variables themselves) was added to
independent variable to prevent log1Q 0.
83
-------
TABLE 28. SIMPLE CORRELATIONS (r) BETWEEN DEPENDENT VARIABLES
(TRIFLURALIN AND DIPHENAMID) AND INDEPENDENT VARIABLES
DATA ARE FROM WATKINSVILLE TEST PLOTS
THfluralln (P-01)
Dlphenamld (P-01)
Trlfluralln (P-03)
Dlphenamld (P-03)
-.180
.126
-.561**
-.377**
-.812
-.758
-.760**
-.825**
-.771
-.810
-.884**
-.870
-.371
-.254
-.187
-.582
-.366
-.211
-.100**
-.214**
.002
-.227
.025**
-.324**
.665
.666
.538
.615**
.713
.513
.141
.633**
*(t-l) Implies rainfall data at one hour before runoff was sampled.
"All variables marked "**" are log,, transformed. Data for P-01 are for 1972 and 1973.
Data for P-03 are for 1972-1975.
'10
^Elapsed time Is days sine* the first data collection. Thus, this variable should contain
Information on long-term changes 1n runoff such as caused by monotonic Increases
or decreases In amount of herbicide applied each year or monotonic cllmatologic trends.
84
-------
partial ing "cumulative rainfall since cropping" out of "time since cropping"
and then correlating the latter with logjQ trifluralin (plot P-03) reduced the
shared variance between predictor and dependent variables by 77% (from 78% to
0.9%). Accordingly, as much as 98% of the variance of plot P-03 trifluralin
shared with "time since cropping" might be caused by storm washoff. It is
possible, of course, that less is directly caused by washoff, since such ef-
fects as wind erosion losses and natural degradation would also be functions
of time and would correlate with both time and cumulative rainfall.
The correlations with suspended solids and sediment herbicide concentra-
tions have been discussed earlier.
Table 29 presents t statistics for multiple regression results with tri-
fluralin and diphenamid as the dependent variables. Table 30 presents regres-
sion equations for this data set.
Buffalo Bill Watershed
Table 31 presents correlations (r) between dependent and independent var-
iables for the Buffalo Bill Watershed. As shown, there are several variables
which correlate very poorly with virtually all others (i.e., phosphate, DDT,
DDE and nitrate) while some (i.e., fecal coliforms. BOD, dieldrin) correlate
moderately well with a number of other variables. Dieldrin, for example,
shares 47, 49, and 32% of its variance with suspended solids, turbidity, and
fecal coliforms, respectively (based on r2). For BOD correlations with sus-
pended solids, turbidity, and fecal coliforms, the values are 32, 30, and 40%,
respectively. In contrast, nitrate correlates most strongly with nitrite (22%
shared variance) and ammonia (11% shared variance) and phosphate most strongly
correlates with TKN, sharing only 10% common variance. In the case of BOD,
dieldrin, and fecal coliforms, some commonality is suggested in the mechanisms
and/or rates of deposition, degradation, and runoff; while for those not cor-
relating well, other factors not accounted for by the available data must be
involved.
Table 32 presents multiple regression statistics for the Buffalo Bill
Watershed data base. As shown in the table, even in the best regressions
(highest R? values), only slightly over 50% of the variance of the dependent
variables could be explained, and in many cases R2 was substantially lower.
Unfortunately, only water quality variables were available in this data base.
It is likely that land use, farming practice, and storm information would im-
prove expli cati on.substanti ally.
Table 33 presents regression equations for the Buffalo Bill Watershed
data base.
Michigan State University Test Plots
Because of uncertainties and errors found in the data, multiple regres-
sion results (like results from this data base presented earlier) are suspect.
Accordingly, multiple regression results will not be presented.
85
-------
TABLE 29. MULTIPLE REGRESSION STATISTICS FOR WATKINSVILLE
PLOTS P-01 AND P-03 DATA. ENTRIES ARE VALUES OF
t FOR EACH REGRESSION COEFFICIENT
Elapsed time
Cays since cropping
Cumulative rain since cropping, In.
Cumulative storm rainfall, In.
Rain Intensity at (t-1)**, in./hr.
Rain Intensity at (t-0). In./hr.
Suspended sediment, g/l
Sed. dlphenamtd concentration, ppb
Sed. trlfluralln concentration, ppb
R*
A
n
3.99 *
-13.1 *
*
*
*
*
6.08 *
*
.762
134
*
*
*
*
-2.12 *
12.2 *
*
.382
260
-6.47
-24.3
8.29 *
*
5.12
.892
170
2.05
-11.9
-10.5 *
-7.27 *
-2.16
5.87 *
.898
223
Variables not logjg transformed.
**(t-l) Implies runoff sampling time minus one hour.
"^Autocorrelation statistics not available for these regressions. See footnote of Table 28.
Table entries are values of t for the coefficients. Selection of regressions
was made such that the greatest number of predictors Mas used subject to at
least 30 observations per predictor variable and all t significant at
a .05, two-tilled test.
86
-------
TABLE 30. REGRESSION EQUATIONS WITH CONFIDENCE INTERVALS FOR SLOPE
AND INTERCEPT (SIMPLE REGRESSIONS ONLY, a = .05, TWO-TAILED TEST)
FROM WATKINSVILLE PLOTS P-01 AND P-03 HERBICIDE DATA
Equations* f?
Plot P-01
Iog10 ([THf] + 1.) - .937 ± .035 - (.0082 * .001) D .65g
log,0 ([Tr1f] + 1.) .954 * .046 - (.099 * .015) RSC .594
log]0 ([Tr1f] + 1.) .735 + 4.6 x 10"4 ET - 7.33 x 10"3 D + 9.1 X 10"3 SS .762
log,0 ([D1ph] + 1.) 2.21 ± .07 - (.024 t .0025) 0 .574
loglfl ([01ph] + 1.) 2.12 ± .07 - (.181 t .016) RSC .655
log1Q ([Dlph] + 1.) .595 t .075 + (1.45 x 10'3 * 1.8 x 10'4) [D1phse(J] .509
Plot P-03
([Ofph] + 1.) (1175 J 1.38) (D + I.)1'1'12 * -n) .681
log,0 ([Otph] + 1.) 2.69 ± 0.1 - (.246 ± .018) RSC .756
10910 ([Dlph] + 1.) 2.42 ± 0.18 - (.759 ± ,141) RSS .338
log1Q <[01ph] + 1.) 1.03 ± .164 + (7.88 x 10'4 ± 1.75 x 10'4) [D1phsed] .263
loglo ([Dlph] + 1.) 3.20 - .515 log,,, (D * 1.) - .139 RSC - .305 RSS .874
([Trlf] t 1.) (32.4 J 1.51) (ET + I.)1"443 * 100) .315
(CTr1f] + l.) (14.8 J 1.17) (D + l.)(-357± 046) .577
([Trlf] + 1.) " (13.5 J 1.10) (RSC + 0.1)(--853 * "m} .782
([Trif] + 1.) 0.56 5 1.29) ([THfsed] + l.)('315 * -059) .400
1<>9,0 ([Trlf] + 1.) -824 - .823 1og,0 (RSC + 0.1) + .102 RSS + .114 1og,0 ([Trif$e<|] + 1.) .864
*Nott: ContUnti (small relative to values of the variables themselves)
«er« added to prevent 1og1Q 0.
87
-------
TABLE 30 (continued)
Explanation of symbols:
Dlph = dissolved diphenamid, ppb
D = days since cropping
RSC = cumulative rain since cropping, inches
RSS = cumulative storm rainfall, inches
Diphsec| = sediment diphenamid, ppb
Trif = dissolved trifluralin, ppb
ET = elapsed time since 7/02/72, days
Trifsecj = sediment triflural in, ppb
SS = suspended sol ids, g/4
88
-------
TABLE 31. SIMPLE CORRELATIONS (r) BETWEEN DEPENDENT AND
INDEPENDENT VARIABLES FOR THE BUFFALO BILL WATERSHED
Oleldrln*
DDT*
DDE
Phosphate*
Nitrate*
Nitrite*
Ammonia
TKN
Fecal conforms*
BOD
.685
.195
.143
-.289
.016
.325
.387
.339
.463
.566
.697
.181
.074
-.249
.115
.473
.508
.392
.502
.545
.482
.185
.132
.059
.124
.375
.339
.438
.632
1.00
.569*
.230*
.226*
.159*
.115*
.285
.197*
.393*
1.00*
.632*
.317
-.055
-.004
.324
.113
.316
.207
1.00
.393
.438
.300
.207
-.074
-.049
.330
.632
1.00
.207
.335
.339
.282*
.362*
-.024
.178*
.425*
1.00*
.632*
.316*
.373
.375*
.088*
-.235*
-.124
.195*
1.00*
.469
.332
.193
.144
.094
.020*
.084*
.015
1.00*
.195*
.178*
-.049*
.324*
.230
.059*
'Variables Iog10 transformed.
89
-------
TABLE 32. MULTIPLE REGRESSION STATISTICS FOR THE BUFFALO BILL WATERSHED.
ENTRIES ARE VALUES OF t FOR EACH REGRESSION COEFFICIENT
vo
o
Suspended solIds concentration, rag/1
Turbidity, JTU
BOD concentration. mg/i
Fecal conforms. MPH/100 ml
TNI concentration. ng/Jt
Amonla-Ji concentration, ng/1
Nltrlte-M concentration, mg/l
NUrate-H concentration, mg/l
Phosphate concentration, ng/t
6.61
3.68 *
*
*
*
.559
as
14
*
2.45 *
*
.131
AZ
42
2.09 *
.051
83
61
-5.29
2.38 *
4.12
*
*
*
.32a
89
89
*
4.38 *
*
.181
39
S3
2.61
4.46
3.82
*
.514
89
35
2.99
5.60 *
*
.456
89
42
4.60
*
*
3.27 *
.231
89
62
6.89
2.67
.446
39
89
3.65
4.74 *
2.02
*
*
.518
89
89
Variables not log]0 transformed.
^Substantial autocorrelation of residuals was noted. Estimated number of Independent observations
very much less than n.
Estimated n, based upon an estimation due to Hold (see Appendix A) shown as n""'.
Table entries are values of t for the coefficients. Selection of regressions was made such that
the greatest number of predictors was used subject to at least 30 observations per predictor
variable and all t significant at a - .05, two-tailed test.
-------
TABLE 33. REGRESSION EQUATIONS WITH CONFIDENCE INTERVALS FOR SLOPE
AND INTERCEPT (SIMPLE REGRESSION ONLY, a = .05, TWO-TAILED TEST)
FROM THE BUFFALO BILL WATERSHED DATA BASE
Equations R2
log]0 [D1el] .913 ± .091 + (6.11 x 10"4 ± 1.37 x 10'4) T .486
log]0 [D1«1] .889 ± ,149 + (.030 ± .012) [BOD] .232
[D1el] (.658 5 2.79) FC(<309 * -098) .323
loglo [Dlel] 4.8 x 10"4 T * .167 log,0 FC .486
[DDE] 1.91 + 1.01 log]0 FC .051
log,Q [DOT] 1.02 + .276 Iog10 [N02] - .361 log,0 [NOj] .250
Iog10 [P04] -.910 - 8.0 x 10'5 SS + .074 [TKN] .283
log,0 [N03] - .433 + .316 log)0 [NOj] .181
Iog10 [N02] » -1.38 t .09 + (3.43 x 10'4 ± 1.36 x 10"4) T .223
1og,0 [N02] -1.49 i .08 + (.585 * .153) [NH4] .400
'09,0 CN02] -1.48 ± .12 + (.198 ± .079) [N03] .220
1og1fl [N02] - -1.61 + .495 [NH4] + .123 [N03] .476
[HH+] .261 ± .093 + (3.99 x 10"4 * 1.44 x 10'4) T .259
[NH4] 1.28 t .22 + (.684 t .179) log,0 [H02] .400
[NH4] 1.02 + 2.0 x 10'4 T + .546 Iog10 [NOj] .456
[TKH] 3.39 + .154 [BOO] + 1.92 log,,, [P04] .281
log]0 FC " 4.15 t .20 + (1.66 x 10"4 i 6.77 x 10"5) SS .214
FC 4.05 t .20 + (8,62 x 10"4 ± 3.17 x 10"4) T .252
1o9,0 FC 3.67 t .25 + (7.65 x 10'2 ± .02) [BOD] .400
91
-------
TABLE 33 (continued)
o
Equations R
1og10 FC = 3.58 + .069 [BOD] + 1.94 [N02] .446
[BOD] = 6.71 ± 1.65 + (7.75 x 10'3 ± 2.54 x 10"3) T .297
[BOD] = -12.9 ± 6.2 + (5.22 ± 1.36) log]0 FC .400
[BOD] = -8.96 + .001 SS + 3.89 log]0 FC .495
Explanation of symbols:
Die! = dissolved dieldrin, ppt
T = turbidity, JTU
BOD = biochemical oxygen demand (not specified whether
5-day, ultimate or other), ppm
FC = fecal coliforms, MPN/100 ml
DDE = dissolved DDE, ppt
DDT = dissolved DDT, ppt
N02 = dissolved nitnte-N, ppm
NO, = dissolved nitrate-N, ppm
TKN = dissolved total kjeldahl nitrogen, ppm
NH. = dissolved ammonia-N, ppm
P0d = dissolved phosphate, ppm
(unknown whether as P or as PO^)
SS = suspended solids, ppm
92
-------
Seattle: Southcenter
Tables 34 and 35 show multiple regression results for Southcenter. Table
34 provides simple linear correlations between dependent and independent vari-
ables. Table 35 shows multiple regression statistics.
In Table 34, it can be seen that in all cases, there are better predic-
tors of runoff water quality than suspended solids, although in general, such
data may not be as readily available or as practical from a standpoint of
modeling applications. In the case of l.ead, organic nitrogen has a higher r
value. The value of r2 for lead versus suspended solids is 0.446 (r = 0.668)
while r2 for organic nitrogen is 0.503, which is probably significantly bet-
ter. Zinc concentration is much more strongly correlated with lead concentra-
tion than with suspended solids, suggesting common phenomena other than asso-
ciation with suspended matter. The common phenomena may include those related
to zinc and lead deposition, and these may not be the same as those involved
in dust and dirt deposition. The upper accumulation limits or rate of ap-
proach to those limits for particulate and for lead and zinc may be very dif-
ferent. Further, washoff characteristics for lead and zinc may also differ
from those of suspended solids.
Both lead and zinc correlate strongly with nitrite plus nitrate and with
organic nitrogen. They also correlate well with total phosphorus.
Urban runoff water quality models, such as SWMM, model water quality as a
simple linear function of runoff suspended sediment, the accumulation and
availability of which are, in turn, simulated as a function of the number of
dry days preceding each storm. Lead and zinc concentrations are only weakly
correlated with the number of dry days, suggesting that "antecedent dry days"
is not a very good predictor variable for lead and zinc. In this data base,
suspended solids shares only 2% (r2 = 0.02) of its variance with the number of
antecedent dry days, making dry days a poor predictor of suspended solids as
wel 1.
Why the correlation is so poor is not clear since it would be expected
that runoff lead and zinc concentration would be proportional to the amount
available for transport. Since washoff should, in turn, affect amount avail-
able to wash off, it is reasonable to expect runoff lead and zinc concentra-
tions to be a function of dry days preceding storms.
Total phosphorus and organic nitrogen similarly correlate weakly with
"antecedent dry days." In contrast, the relatively soluble species, nitrite
plus nitrate nitrogen, orthophosphate phosphorus, and ammonia nitrogen all are
much better predicted by "antecedent dry days" than by suspended solids con-
centration. Since the latter is itself poorly correlated with "antecedent dry
days," it seems that simulating the species nitrate, nitrite, orthophosphate,
and ammonia as a function of dry days preceding storms through the intermedi-
ary suspended solids is very tenuous. It should be noted here, however, that
the poor correlation between "antecedent dry days" and suspended solids may
well be due-to dilution effects related to variations in flow. "Antecedent
dry days" may, in fact, correlate well with dust and dirt accumulation. None-
theless, the poor correlation between sediment and some water quality
93
-------
TABLE 34. SIMPLE CORRELATIONS (r) BETWEEN DEPENDENT AND INDEPENDENT VARIABLES
FOR SOUTHCENTER, SEATTLE
<£>
Lead
Z1nc
Nitrite
plus Nitrate N
tamonla H
Total Phosphorus
Orthophosphate
Phosphorus
Organic Nitrogen
.084
.203
-.014
-.251
-.130
.113
.225
.186
-.040
.229
.374
-.161
.401
.267
-.282
-.202
-.131
.106
-.292
,070
-.287
.231
.140
.483
.523
.138
.402
.265
.668
.374
.293
.135
.390
.112
.514
1.00
.666
.602
.351
.486
.381
.709
.666
1.00
.700
.331
.520
.427
.654
.602
.700
1.00
.583
.472
.614
.662
.351
.331
.583
1.00
.207
.546
.286
.486
.520
.472
.207
1.00
.540
.532
.381
.427
.614
.546
.540
1.00
.481
.709
.654
.662
.286
.532
.481
1.00
.151
.495
.420
.062
.534
.391
.316
.548
.489
.271
.075
.568
.346
.471
-.073
-.266
-.273
-.128
-.273
-.277
-.153
-.273
-.311
-.285
-.211
-.230
-.308
-.239
**.!! variables but those narked vlth "*" are log1Q transformed.
-------
TABLE 35. MULTIPLE REGRESSION STATISTICS FOR SOUTHCENTER,
SEATTLE. ENTRIES ARE VALUES OF t FOR EACH
REGRESSION COEFFICIENT
.0*'
ff
Days since 1/1/73
Month of the year
Hours since start of storm
Cumulative storm rainfall. Inches
Dry days before storm
Suspended solids concentration.
mg/1
Turbidity. JTU
Lead concentration, ug/i
Zinc concentration, yg/1
Nitrate and nitrite N
concentration, mg/i
Ammonia N concentration, mg/1
Organic nitrogen
concentration, ng/i
Total phosphorus concentration,
ng/i
Orthophosphate P
concentration, mg/i
Conductivity, umhos/cm
Flow, eft
R2
n
n"
*
*
-4.85*
8.52
4.83
2.61
3.80
-3.61
*
.738
99
99
8.40
-8.67*
*
4.23
4.64
9.32
3.9S
.751
199
98
*
-3.06
*
5.50
7.96
3.90
2.83
3.41
*
.744
199
32
6.29*
-2.51*
3.79
2.49
6.66
3.77
.684
199
199
-3.94
-3.42*
*
*
7 63*
5 37
6.82*
*
.634
199
199
Long-term trends
Storm-related
variables
Suspended matter
Heavy metals
Nitrogen variables
Phosphorus variables
Variables not log,Q transformed.
Substantial autocorrelation of residuals was noted. Estimated number of Independent observations
very much less than n (199), probably equivalent to about 32 based upon an estimation due to
Hold (see Appendix A). Estimated n shown as n+t.
Table entries are values of t for the coefficients. Selection of regressions was made such that
the greatest number of predictors was used subject to at least 30 observations per predictor
variable and all t significant at a .05, two-tailed test.
95
-------
constituents and the suggested low correlation between "antecedent dry days"
and dust and dirt accumulation seem to warrant a dry days - water quality
estimation procedure.
Regarding other predictors, "elapsed time," "hours since start of storm,"
"runoff flow," and "cumulative storm rainfall" were not well correlated with
any of the dependent variables. "Month of the year" was modestly correlated
with soluble species, perhaps suggesting the effect of seasonality of "ante-
cedent dry days" or of deposition phenomena. As might be expected, turbidity
was correlated with the dependent variables to an extent similar to suspended
solids. Conductivity correlated well with some dependent variables.
Multiple regression results shown in Table 35 suggest that phenomena in-
fluencing runoff water quality with respect to the dependent variables shown,
are quite complex. This is suggested by the fact that different predictor
variables were selected for the several dependent variables. Further, and
perhaps more importantly, no single (or small number of) independent variables
was adequate to explain water quality variability. If simple phenomena were
involved, it would be expected that perhaps two or three predictors would ac-
count for much of the variability of runoff water quality. It is possible, of
course, that the set of predictors used here simply missed one or two critical
ones. It is of interest to note that if the predictor variables are grouped
as shown in Table 35 (e.g., into storm, suspended matter, heavy metal, nitro-
gen, and phosphorus variables), then some consistency appears in the set of
predictor variables chosen for each dependent variable. In all five regres-
sions, at least one storm variable and one nitrogen variable were selected.
In four of the five regressions, at least one heavy metal, and one suspended
solids variable was selected. This suggests that although different nitrogen
variables were chosen in different regressions, it may well have been possible
to obtain nearly as high values of R2 using only one candidate variable in
each variable category. This suggests, further, that the concentration of any
of the water quality variables in runoff depends upon phenomena reflected in
information contained in perhaps four general variables (i.e., storm charac-
teristic, heavy metal, suspended solids, nitrogen).
The suggested complexity of phenomena influencing runoff water quality
apparently does not preclude some spatial consistency. When regression coef-
ficients from Viewridge 1 (Seattle) were used in conjunction with Southcenter
data, the equations from Viewridge 1 were, in some cases, reasonably good pre-
dictors of runoff quality at Southcenter. Table 36 shows "validation-general-
ization" results.
Table 37 presents predictive equations which may be of use in estimating
runoff water quality from urban watersheds, or which may be of interest in
comparing with results of other studies.
Seattle: South Seattle
Table 38 shows simple correlations (r) between all regression variable
pairs, with ratios of these to the corresponding r values for Southcenter. As
shown, the ratios of r values for the two sites are quite variable and often
of opposite sign for those variables which are modest to poor predictors (e.g.,
96
-------
TABLE 36. RESULTS OF VALIDATION-GENERALIZATION (V-G) USING DATA
FROM SOUTHCENTER AND COEFFICIENTS FROM VIEWRIDGE 1
(BOTH SEATTLE)*
Variable Viewridge 1 R'
Lead concentration
Zinc concentration
Nitrite plus nitrate N
concentration
Total phosphorus P
Organic nitrogen
.792
.733
.754
.678
.564
> ?
" V-G IT -
}
.447
.545
.239
.010
.461
V-G R2 .. 1QQfl/
/iewridge 1 R^
56.
74.
32.
1.5
81.7
*"Validation-generalization" is a term used for the application of a
previously developed regression equation to a new set of data, as
from a different site. The value of R2 so obtained, when compared
with R2 from the new regression equation suggests the degree of
consistency from data set to data set.
97
-------
TABLE 37. REGRESSION EQUATIONS WITH CONFIDENCE INTERVALS FOR SLOPE
AND INTERCEPT (SIMPLE REGRESSIONS ONLY, a = .05, TWO-TAILED TEST)
FROM SOUTHCENTER, SEATTLE, DATA
Equations R2
[Norg] = (.00891 ? 1.70) [Pb]1-735 * J03) .503
[Norg] - (.0109 ? 1.82) [Zn]('764 * -125) .428
[Norg] - (1.26 f 1.24) [N02 + N03]('631 * -10" .438
log]0 [N ] 1.78 + .00068D + .477 1og]0 [Pb] + .369 1og)0 [N02 * NOj] .626
[N02 + N03] - (.00549 ? 1.91) [Pb]('654 * '122) .363
[N02 + M03] (.00288 J 1.82) [Zn](>858 * >124) .491
[N02 + N03] - (.305 * 1.17) [N ]('694 * -111' .438
[N02 * N03] (2.40 * 1.66) [0-PO,]'"787 * '143) .377
log,0 [N02 + H03] -1.48 + .287 Iog1() DO - .078 HSS + .610 log,0 [2n] + .338 1o9]0 [0-PO,] .716
[Pb] (5.13 * 1.77) [In]1'750 * -118> .444
[Pb] - (479. ? 1.26) [N02 + N03]('554 * '104) .362
[Pb] (324. i 1.15) [Norg](-685 * -096) .503
[Pb] - (525. ? 1.38) [Pjot]''572 * >146) .236
log,0 [Pb] - 1.28 + .422 Iog10 [SS] + .314 log,,, [Zn] + .247 Iog10 [Norg] - .210 CR .713
[Zn] - (5.20 J 1.63) [Pb]C<59' * -093) .444
[Zn] - (309. J 1.17) [N02 + N03]('572 * 082) .491
[Zn] > (182 * 1.12) [Norg](>561 * -092) .428
[Zn] - (309. J 1.29) [Pt,,,.]'-543 * '"' -27'
1091Q [Zn] - 1.59 + 1.26D + .507 log1Q [N02 + NOj] + .351 loglg T .631
loa,0 [Ptot] 2.79 - 1.300 + .201 1081Q [Nflrg] + .001 C + .012 T .612
98
-------
TABLE 37 (continued)
Explanation of symbols:
N = organic nitrogen, ppm
Pb = lead, ppb
Zn = zinc, ppb
N0« + N03 = nitrite plus nitrate N, ppm
D = days since 1/1/73
0-P04 = orthophosphate phosphorus, ppm
DD = antecedent dry days (dry days preceding each storm)
HSS = hours since start of storm
p.. = total phosphorus, ppm
SS = suspended solids, ppm
CR = cumulative rainfall per storm, inches
T = turbidity, JTU
C = conductivity, ymhos/cm
n = 199 for all equations
99
-------
TABLE 38. SIMPLE CORRELATIONS (r) BETWEEN DEPENDENT AND INDEPENDENT VARIABLES
(SOUTH SEATTLE)
o
o
Entries in table are
Seattle
/
South Seattle
Southcenter
All variables except those narked with "*" ire logla transformed.
-------
days since 1/1/73, month of the year, hours since start of storm, and flow).
In contrast, good predictors (e.g., nitrite plus nitrate nitrogen, orthophos-
phate phosphorus, and organic nitrogen) give ratios fairly close to 1.0, sug-
gesting similar degrees of correlation from site to site. Moderately good
predictors (e.g., suspended solids and antecedent dry days) give positive r
ratios (connoting relationships of consistent sign of slope) but ratios which
are less nearly equal to 1.0. Thus, as might be expected, the degree of spa-
tial consistency of correlations improves as r approaches ±1. That is, the
better the predictor, the more spatially consistent the correlation between
predictor and predicted variable.
Table 39 presents t statistics for multiple regression coefficients for
South Seattle data. In four cases (except organic nitrogen), R2 is higher in
this data base than for Southcenter. The values are generally similar, how-
ever, with lead and zinc being among the best explained using the independent
variables available. Also similar to Southcenter is the fact that every re-
gression selected one or more nitrogen variables, and four selected one or
more representing suspended matter. In this case, however, all five regres-
sions used at least one heavy metal. In contrast, storm-related variables in
this data base appeared to contribute relatively little, overall, to predic-
ting runoff quality, whereas at Southcenter, at least one storm-related vari-
able was selected for each regression.
Table 40 shows results of validation-generalization of South Seattle data
using Viewridge 1 regression equations. Like Southcenter, the best valida-
tions are for heavy metals and organic nitrogen. Nitrate plus nitrite vali-
dates modestly, while total phosphorus validates poorly. This suggests that
similar physical/chemical phenomena determine runoff lead, zinc, and organic
nitrogen concentrations, and concentrations of nitrate plus nitrite, to an ex-
tent as well. In contrast, total phosphorus concentrations in runoff appear
to be highly variable spatially both comparing Southcenter to Viewridge 1 and
South Seattle to Viewridge 1 (based on validation R2) and comparing South-
center and South Seattle (based on variables used in the regressions).
Table 41 presents predictive equations developed for South Seattle data.
Seattle; Viewridge 1
Table 42 shows simple correlations (r) between dependent and independent
variables for Viewridge 1 data, and ratios of r for Viewridge 1 to South Seat-
tle and to Southcenter r values. The r values for correlations between water
quality variables and the first three predictor variables ("days since 1/1/73,"
"month," and "hours since start of storm") are relatively low, as they are for
Southcenter and South Seattle. Also, the ratios are quite variable suggesting
not only weak, but also spatially inconsistent and possibly spurious relation-
ships. Conductivity, "cumulative storm rainfall" and especially, flow, are
similarly weak and spatially inconsistent predictors. In contrast, and again
like South Seattle and Southcenter, "antecedent dry days," suspended solids,
and turbidity are better runoff water quality predictors, with r being com-
monly in the 0.5-0.6 range. Here, as has been observed earlier, high r values
are associated with spatially consistent values as shown by ratios fairly
close to 1.0. Correlations among water quality variables, themselves, are
101
-------
TABLE 39. MULTIPLE REGRESSION (t) STATISTICS AND R2
FOR SOUTH SEATTLE DATA
lioyi ilnce 1/1/73
Mjnth of the year
Hours since start of storm
Cumulative storm rainfall, 1nche
Dry days before storm
Suspended sol Ids concentration,
mg/4
Turbidity, JTU
Lead concentration, p.g/l
21nc concentration, ug/i
Nitrate and nitrite N
concentration, mg/l
Ammonia H concentration, mg/l
Organic nitrogen
concentration, mg/l
Total phosphorus concentration,
mg/t
Orthophosphate P
concentration, mg/i
Conductivity, vim ho s/ cm
Flow, cfs
R2
n
ntt
*
*
*
5.19
6.02
9.33
2.48
2. 59
-4.70
*
,850-
99
63
*
-3.02*
2.79
-4.81
9.63
4.07
7.65
.814
199
47
-9.29
*
*
-5.16
3.69
5.99
4 fin
6.17
#
.762
199
41
2.76*
*
*
3.85
5.22
2.88
*
4.67
199
155
4.95
*
4 85*
3.06
3.36
2.85
*
*
.£79
199
104
Long-term trends
Storm- related
variables
! Suspended matter
Heavy metals
Nitrogen variables
/ Phosphorus variables
'Variables not log,, transformed.
Substantial autocorrelation of residuals was noted. Estimated number of Independent observations
very much less than n (199).
t*t ++
Estimated n, based on a method due to Wold (See Appendix A) shown as n .
Table entries are values of t for the coefficients. Selection of regressions was made such that
the greatest number of predictors was used subject to at least 30 observations per predictor
variable and all t significant at a * .05, two-tailed test.
102
-------
TABLE 40. RESULTS OF VALIDATION-GENERALIZATION (V-G) USING
DATA FROM SOUTH SEATTLE AND COEFFICIENTS FROM
VIEWRIDGE 1*
V-G R2
Viewridge 1 Rz
Lead concentration
Zinc concentration
Nitrate plus nitrite N
concentration
Total phosphorus P
Organic nitrogen
.792
.733
.754
.678
.564
.688
.605
.322
.021
.333
86.9
82.5
42.7
3.1
59.0
*"Validation-generalization" is a term used for the application of a
previously developed regression equation to a new set of data, as
from a different site. The value of R2 so obtained, when compared
with R2 from the new regression equation suggests the degree of
consistency from data set to data set.
103
-------
TABLE 41. REGRESSION EQUATIONS WITH CONFIDENCE INTERVALS FOR SLOPE
AND INTERCEPT (SIMPLE REGRESSIONS ONLY, a = .05,
TWO-TAILED TEST) FROM SOUTH SEATTLE DATA
Equations R2
[Norg] (.041 * 1.86) [Pb]('526 * J24) .266
[Norg] (.00490 ? 2.5) [Zn]('896 * >172) .352
[Norg] - (1.14 ? 1.18) [N02 + N03]('515 * -096> .365
[Norg] (1.74 { 1.30) [NH4](>453 * -097' .304
[N ] (3.45 ? 1.68) [OP04-P]'-562 * >158) .202
[Norg] - (.02652.00) C('764 * J65) .301
Iog1fl [Norg] -1.52 + .0004 D + .490 log,0 tZn] + .403 log1Q [N02 + N03] .452
[N02 + N03] - (3.3 x 109 ? 2.24 X 102) o'"3>4° * l79' .269
[N02 + N03] (.174 ? .069) DD('375 * '083) .289
[N02 + N03] = (.010 * 2.09) [Pb]''649 * J42) .294
[N02 + N03] - (1.56 5 1.27) [NH4]('723 ± 090) .564
[N02 + N03] - (6.61 x 10'4 ? 2.81) [Zn](1'13 * '196) .405
[N02 + N03] - (.392 ^1.15) [Norg]('708 * '132> .365
[N02 + N03] - (1.25 J 1.33) [P^]1'641 * '149) .270
[NOZ + N03] (4.26 * 1.73) [OP04-P]('87° * '167) .353
[N02 + N03] 6.12 - 1.95 1og]0 D + .374 log1Q [NH4] + .280 log]0 [Norg] t ,328 log1Q [OP04-P] .723
[Pb] (.299 ; 1.98) [Zn](1'17 * '13) .627
[Pb] (468. J 1.29) [NH4](<462 * -093) .329
[Pb] ' (275. J 1.17) [N02 t N03]('454 * 100> .294
[Pb] (199. * 1.12) [Norg]<-505 * -119) .266
104
-------
TABLE 41 (continued)
Equations
[Pb]-(575. Jl.23)[Ptot]<-7661 '097) .553
[Pb] - (1412. J 1.58) [OP04-P](-701 * -142) .327
[Pb] (12.9 J 2.05) C(>61Z * >172) .201
[Pb] (18.6 * 1.29) T(<752 * 088) .592
log,0 [Pb] -.00*7 + .163 log]0 [SS] + .823 log,0 [Zn] - .187 log,,, C + .301 log,0 T .835
[Zn] (13.8 * 1.35) [Pb](>535 * -058) .627
[Zn] (501. * 1.15) [NH4]<-370 * -056) .463
[Zn] (331. J 1.12) [N02 + H03]('359 * -061) .M5
[Z,,] (251. « 1.07) [Norg]<-393 * -076> .352
[Zn] - (479. J 1.15) [P^]'"492 * '^ .501
[Zn] (832. J 1.38) [OP04-P]('444 * '098) .289
[Zn]- (15.5 J 1.48) C('635* -095) .474
milM/r.tJi.w)^349*-079* .279
log10 [Zn] .967 + .475 1og]0 [Pb] + .091 1og,0 [NH^] + .327 log,0 C - .143 log,,, T .800
[Ptot] - (.00468 J 1.58) [Pb]<-722 * 09J> .553
[Ptot] (7.94 x 10'* J 2.14) [Zn]'1'02 * -146> .501
[Ptot] - (.313 J 1.18) [N02 + H03]{'422 * -098) .270
[Ptot]. (.547 I 1.27) [HH4](>4S8± 089) .343
log)0 [Pj,,,.] (-1.03 t .07) + (.0118 ± .002) T .37S
Iog10 [Ptot] -2.65 + .00033 HSS + .268 log,0 [Pb] + .498 1og]0 [Zn] + .005 T - .666
Explanation of $/nbo1$ on following pagt
105
-------
TABLE 41 (continued)
Explanation of Symbols:
N = organic nitrogen, ppm
Pb = lead, ppb
Zn = zinc, ppb
ML + N03 = nitrite + nitrate N, ppm
NH. = ammonia N, ppm
OPCh-P = orthophosphate P, ppm
C = conductivity, ymhos/cm
D = days since 1/1/73
DD = antecedent dry days
P.. = total phosphorus, ppm
T = turbidity, JTU
HSS = hours since start of storm
106
-------
TABLE 4E. SWPLE CORRELATIONS (r) BETWEEN DEPENDENT AND INDEPENDENT VARIABLES
(VIEWRIDGE 1)
Lead
concentration
Zinc
concentration
Mnonta
concentration
Nitrite + nitrate
concentration
Organic
nitrogen
concentration
Total phosphorus
concentration
Or thopho spin te
phosphorus
concentration
-.333
.073
-.3&S
/'**
-.284
1.13
-.002
.005,
.175
-.10
-.337
.291
-1.6
12.58
-.16$
-.257
-.190
-.185
-.321
-.344
-.198
-.355
-,059
.113
-.238
-3.3J
.982
.154
.550
.357
.530
.089
.237
.576
-.048
.655
.537
.319
-.2J4
.513
.419
1.00
.746
.356
-.216
.Ml
.563
.757,
1.16
.434
.745
\.\i
1.00
.315
-.016
.579
.480
.368
.356
.315
.463
.953
1.00
.438
.304
,640
.685
-.216
-.016
.438
.533
.753
1.00
.176
.261
.345
.541
1.05>
.579
.97J,
.889
.304
.552
f.07
.176
1.00
.437
.326
.B63
.757^.
.480
.921
.640
1.09
3.09
.261
,487
l.OQ
.715
1.09.
1.33
.434
.759,
.358
.665,
.863
.685
1.26
1.25
.345
.580
.563
.326
.715
n.33
1.00
-.253
.013
.01B
.026
.179
.757
1.81
,427
.T3&
.139
.105
.614
.799
1.12
.496
.273
-.101
.528
1.77.
Ull
.454
.299
.137
,400.
-.034
-.no
s.re
-.525
1.89
-.127
-.129
-.131
-.051
.190
-.209
.762
.671
-.219
l.ll/
Xl.04
-.169
.814
.5*
-.075
-.035
*A11 vsriables except those marked with "*" are log transformed.
1Q
-------
again high (although not as high overall as at South Seattle), with many val-
ues in the 0.5-0.7 range, and with r ratios comparing Viewridge 1 correlations
with those of Southcenter and South Seattle often being near 1.0. Notably,
whereas nitrite plus nitrate was a good predictor for other water quality con-
stituents at Southcenter and South Seattle, and r ratios were close to 1.0,
this variable is a poor predictor of other water quality variables at View-
ridge 1.
Table 43 presents multiple regression statistics for Viewridge 1. Like
Southcenter and South Seattle, it appears that for a given regression, vari-
ables are selected from among the general groups (e.g., heavy metals, nitrogen
variables), with only one case of more than one from any group (i.e., sus-
pended solids and turbidity as predictors of organic nitrogen).
In all five regressions, a heavy metal variable was selected (lead, zinc,
or cadmium) and four out of five selected a suspended solids variable. In
contrast to South Seattle and Southcenter, nitrogen variables seem relatively
less important (as evidenced by only three regressions using a nitrogen vari-
able), and the occurrence of "days since 1/1/73" in all five regressions sug-
gests that conditions at Viewridge 1 are changing over time.
Regression R2 values for Viewridge 1 were high (all greater than 0.7 ex-
cept for organic nitrogen) as they were at South Seattle and Southcenter. At
all three sites, regression R2 values for lead, zinc, and nitrite plus nitrate
were in excess of 0.7. Generally, the value of R2 was greater for total phos-
phorus than for organic nitrogen which-was lowest (except at Southcenter).
Table 44 presents regression equations for Viewridge 1 data.
108
-------
TABLE 43. MULTIPLE REGRESSION (t) STATISTICS AND
(SEATTLE) DATA
FOR VIEWRIDGE 1
Days since 1/1/73
Month of the year
Hours since start of storm
Cumulative storm rainfall, Inche
Dry days before storm
Suspended solids concentration,
9/t
Turbidity. JTU
Lead concentration. V0/1
Zinc concentration, 119/1
Nitrate and nitrite N
concentration, no/I
Ammonia N concentration, ng/l
Organic nitrogen
concentration, mg/l
Total phosphorus concentration,
g/i
Orthophosphate P
concentration, mg/t
Conductivity, umhos/cm
Flow, cfs
«'
n
.«
-5.30
*
*
*
2.70
3.01
12.1
-5.95
2.S4
»
,792
188
188
9.22
*
*
*
14.3
3.91
2.56
*
-2.44
.733
188
30
-2.46
*
*
*
-3.63
-4.65*
7.83
12.3 *
-3.01
.753
188
188
-3. 24
*
*
*
-2.50
4.06
3.20
5.65
7.24*
.564
188
46
-4.18
*
*
-7.73*
2.27
4.89
10.0
5.33*
.741
188
150
Long- tern trends
Storm-related
variables
Suspended matter
Heavy metal s
Nitrogen variables
Phosphorus variables
Variables not log1Q transformed.
"Cadmium data which were not used In other regressions, correlated with this
variable, and this 1s the value of t for the cadmium coefficient (ug/i).
^Substantial autocorrelation of residuals was noted. Estimated number of Independent observations
very nuch lets than n (188).
"Estimated n shown as nn~ based upon as estimation due to Wold (see Appendix A).
Table entries are values of t for the coefficients. Selection of regressions ms mile such that
the greatest number of predictors was used subject to at least 30 observations per predictor
variable and all t significant at a .05. two-tailed test.
109
-------
TABLE 44. REGRESSION EQUATIONS, WITH CONFIDENCE INTERVALS
FOR SLOPE AND INTERCEPT (SIMPLE REGRESSIONS ONLY, a = .05,
TWO-TAILED TEST) FROM VIEWRIDGE 1 (SEATTLE) DATA
Equations R
[Norg]' (.130? 1.56) [Pb]<'394± -089' ,292
[Norg] (.066 ? 1.70) [Zn]('MO * <123) .335
[NorgJ 0.63 { 1.21) [P^]*'470 * -123> .238
D»OPB]-{.M8;i.31)T<-463± -108> .279
log,0 [Norg] -1.29 + .161 Iog10 SS + .294 log,,, [Zn] + .002 C + .234 log,0 T .537
log)0 [N02 + N03] (-.869 ± .079) + (.0051 i .0006) C .574
[N02 + MOj] (.888 J 1.19) Q(-'538 * J27) .276
[Pb] (83.2 I 1.15) DD('376 * -0831 .302
[Pb] - (1.30 ? 1.07) [Zn]"'06 * l137> .557
[Pb] - (135. J 1.12) [Morg](-741 * -167) .292
[Pb] - (339. ? 1.29) [Ptot](-744 * -159) .317
[Pb] (21.4 J 1.41) T('739 * '138) .377
'O9,o [Pb] 6.05 - 1.90 0 * .947 log,0 [Zn] -.296 log,0 [N02 + M>3] + .272 Io9,0 [Ptot] .774
[Zn] (5.78 J 1.40) [Pb]<>525 * -068) .557
[Zn] (77.6 J 1.07) [Norg](>S59 * '115) .335
[Zn] (132. « 1.20) [P^]'-446 * 'm) -230
[Zn]- (26.9?!.32)T<-4Mt '107) .246
log,0 [Zn] 3.93 * 1.65 log)0 0 + .494 log)0 [Pb] + .203 lcg,0 [Norg] * .120 log,Q [OP04-P] .725
110
-------
TABLE 44 (continued)
Explanation of Symbols:
N = organic nitrogen, ppm
Pb = lead, ppb
Zn = zinc, ppb
Ptot = total PnosPhorus
T = turbidity, JTU
SS = suspended solids, ppm
C = conductivity, umhos/cm
N02 + N03 = nitrite plus nitrate N, ppm
Q = flow, cfs
DD = antecedent dry days
D = days since 1/1/73
OPO^-P = orthophosphate phosphorus
111
-------
REFERENCES
1. Metcalf and Eddy, Inc., University of Florida, and Water Resources Engi-
neers, Inc. Storm Water Management Model (four volumes) 11024 DOC 07/71,
11024 DOC 08/71, 11024 DOC 09/71, and 11024 DOC 10/71, U.S. Environmental
Protection Agency, Water Quality Office, 1971.
2. Donigian, A.S., Jr., and N.H. Crawford. Modeling Nonpoint Pollution from
the Land Surface. EPA-600/3-76-083, U.S. Environmental Protection Agency,
Athens, Georgia. July 1976.
3. U.S. Army Corps of Engineers. Storage. Treatment. Overflow Runoff Model
"STORM." 723-SS-L7520, Hydro!ogic Engineering Center, Davis, California.
1976.
4. Donigian, A.S., Jr., and N.H. Crawford. Modeling Pesticides and Nu-
trients on Agricultural Lands. EPA-600/2-76-043, U.S. Environmental
Protection Agency, Athens, Georgia. February 1976.
5. Hydrocomp. Model currently under development under the support of U.S.
Environmental Protection Agency, Athens, Georgia. Work performed at
Hydrocomp, Inc., Palo Alto, California.
6. Donigian, A.S., Jr., and N.H. Crawford. Simulation of Nutrient Loadings
in Surface Runoff with the NPS Model. EPA-600/3-77-065, U.S. Environ-
mental Protection Agency, Athens, Georgia. June 1977.
7. Cowen, W.F., K. Sirisinha, and G.F. Lee. "Nitrogen Availability in Urban
Runoff." Journal WPCF, Vol. 48, no. 2, February 1976. Pp. 339-45.
8. Cowen, W.F. and G.F. Lee. "Phosphorus Availability in Particulate
Materials Transported by Urban Runoff." Journal WPCF, Vol. 48, no. 3,
March 1976. Pp. 580-91.
9. Sartor, J.D., G.B. Boyd, and F.J. Agardy. "Water Pollution Aspects of
Street Surface Contaminants." Journal WPCF, Vol. 46, no. 3, March 1974.
Pp. 458-67.
10. Colston, N.V., Jr. Characterization and Treatment of Urban Land Runoff.
EPA-670/2-74-096, U.S. Environmental Protection Agency, Cincinnati, Ohio.
December 1974.
112
-------
11. Smith, C.N., R.A. Leonard, G.W. Langdale, and G.W. Bailey. Transport of
Agricultural Chemicals from Small Upland Piedmont Watersheds"EPA-600/
3-78-056, U.S. Environmental Protection Agency, Athens, Georgia, and U.S.
Department of Agriculture, Watkinsville, Georgia. May 1978.
12. Ellis, B.G., A.E. Erickson, A.R. Wolcott, M. Zabik, and R. Leavitt.
Pesticide Runoff Losses from Small Watersheds in Great Lakes Basin. EPA-
600/3-77-112, U.S. Environmental Protection Agency, Athens, Georgia.
October 1977.
13. Rb'mkens, M.J.M., D.W. Nelson, and J.V. Mamnering. "Nitrogen and Phos-
phorus Composition of Surface Runoff as Affected by Tillage Method."
J. Environ. Quality, Vol. 2, no. 2, 1973. Pp. 292-5>
14. Burwell, R.E., G.E. Schuman, K.E. Saxton, and H.G. Heinemann. "Nitrogen
in Subsurface Discharge from Agricultural Watersheds." J. Environ.
Quality, Vol. 5, no. 3, 1976. Pp. 325-9.
15. Burwell, R.E., D.R. Timmons, and R.F. Holt. "Nutrient Transportation in
Surface Runoff as Influenced by Soil Cover and Seasonal Periods." Soil
Sci. Soc. Amer. Proc., Vol. 39, 1975. Pp. 523-8.
16. Morris, R.L. and L. Johnson. Buffalo Bill Watershed Agricultural Run-Off
Study. Preliminary Report, Iowa State Hygienic Laboratory (University of
Iowa), Iowa City, Iowa. February 1974.
17. Ellis, B.G., A.E. Erickson, and A.R. Wolcott. Nitrate and Phosphorus
Runoff Losses from Small Watersheds in Great Lakes Basin. EPA-600/3-78-
028, U.S. Environmental Protection Agency, Athens, Georgia. March 1978.
18. Iwatsubo, R.T., K.M. Nolan, D.R. Harden, G.D. Glysson, and R.J. Janda.
Redwood National Park Studies, Data Release Number 1. Redwood Creek,
Humboldt County. California, September 1, 1973 - April 10, 1974JU.S.
Geological Survey, Menlo Park, California.December1975.
19. Iwatsubo, R.T., K.M. Nolan, D.R. Harden, and G.D. Glysson, Redwood
National Park Studies, Data Release Number 2, Redwood Creek, Humboldt
County, and Mill Creek, Del Norte County, California, ApriVll, 1974 -
September 30, 1975.U.S. Geological Survey, Menlo Park, California.
December 1976.
20. Huber, W.C. and J.P. Heaney. Urban Rainfall-Runoff-Quality Data Base.
EPA-600/8-77-009, U.S. Environmental Protection Agency, Cincinnati, Ohio.
July 1977.
21. Heidel burg College River Studies Laboratory. Water Quality Data for
Sandusky River Material Transport Stations. U.S. Arniy Corps of Engi-
neers, Buffalo, New York.Undated.
113
-------
APPENDIX A
REGRESSION ANALYSIS THEORY
In order to understand the results of analysis presented in the body of
this report, it is necessary to understand the statistical methods used, what
the individual specific results suggest about water quality relationships,
what the statistical assumptions mean, and what the techniques employed can
and cannot do.
A considerable risk is associated with the interpretation of statistical
results if the implications of assumptions and details of techniques are not
understood. For example, it is always possible to fit a straight line to data
consisting of three or more coplanar points. The slope and intercept will
seldom be exactly zero, and as a result, it is all too easy to assume that the
fitted line has some meaning when it, in fact, does not. It is through proper
interpretation of statistical results that significant relationships are dis-
tinguished from the non-significant relationships.
In general, regression analysis may be considered as a two part process.
The first part is nonstatistical in the sense that probabilities are not in-
volved and assumptions about data distributions are not made. This is the
fitting of a model (a predictive equation) to the data. The second part in-
volves making assumptions about the underlying distributions from which the
data are drawn. Where the assumptions are valid, statements can be made about
the probability that relationships are real or that apparent relationships
have probably occurred by chance alone.
MODEL FITTING
The first step in regression analysis consists of deciding upon the model
to be used based upon assumed or known relationships in the data. Consider,
first, a simple linear model which is assumed to represent some data set.
where Y. = ith observed value of the dependent variable
Xi = ith observed value of the independent variable
3 ,3 = coefficients representing the true population
0 l relationship
e-j = ith error term
114
-------
The equation to be fitted to the data is:
Y1 = bQ + bjXi (A-2)
where Y. = ith predicted value for the dependent variable
b ,b = fitted coefficients and estimates of $ and 3 , respectively
01 0 1 r j
Now the "least-squares prescription" specifies that the total sum of squares
of deviations of the data values from the fitted line be minimized. To do
this, first express the sum of squared deviations (or errors):
Z e2 = Z (Y. - Y.)2 = Z (bQ + bjX1 - Y.)2 = f(x,y) (A-3)
where n = the number of observations
Set the partials of the function with respect to b and b equal to zero:
o i
affx v) n n n
AU"' =2Zb +2b ZX. -2ZY.=0 (A-4)
3Do 1-1 ° H-l n 1-1 n
or
n n
nbo + bi.Z Xi = .Z Yi ^A"5)
and
^fi.. ,.\ n n « n
= 2b Z X. + 2b Z XT - 2 Z X.Y. - 0 (A-6)
or
n n « n
b Z X. + b Z Xf = Z X.Y, (A-7)
°M ! H-l 1 i=l 1 1
Equations (A-5) and (A-7) are called "normal equations," and solved simultan-
eously to give the coefficients b and b .
The extension to multiple regression is straightforward. For any number
of independent variables with a linear model, the equation to be fitted is:
Y = b + b X . + b X . + + bj( , (A-8)
o II»T 2 z»> mm,l \ i
Solving for the matrix of b coefficient values is commonly done in a somewhat
different way. In one approach, the matrix of correlation coefficients (r)
is found as follows.
115
-------
M
l-2
' r
l-m
2-l
2-2
' r
2-m
(A-9)
m-l
m-2
'
m
where if r. is an element of jr then r. . is given by:
I «J III ''I 9 I J
£v*
2 2
/ n Y
( E Xk h)
\h=l k>n/
L (A-10)
and for i=j, r. . = 1.
' J
Then the system of normal equations for the determination of the b's is:
or
r 6 = r
nrm m ym
R = r r
m -ym -nrm
where r m = the symmetric matrix of correlation
m m . .. . .
-^ -- -
coefficients among the independent variables
as defined by equations (A-9) and (A-io)
fi = the column matrix of b to be found for equation (A-8)
vm
co^umn matri'x °f correlation
coefficients between the dependent
variable, y and each of the independent'
variables
Computation of the constant term, b is performed as follows:
o
b =7-bx" -bX" - . . . - b I
o 1122 mm
(A-ll)
(A-12)
(A-13)
116
-------
It is clear that conceptually, the procedure for fitting any model by
least squares is equally straightforward and consists of the following steps.
1. Define the model relating the dependent variable to the
independent variable(s).
2. Define the function to be minimized (sum of squared devia-
tions of observed values from the fitted line, surface, etc.).
3. Take the partials with respect to each model parameter
and set equal to zero.
4. Solve the resulting equations simultaneously to give
values for model coefficients.
However, depending upon the form of the equation, it may be difficult to ex-
plicitly solve for the desired coefficients.
It is customary as a next step, in any regression other than simple lin-
ear regression, to compute R2, the squared multiple correlation coefficient.
In simple linear regression (fitting an equation of the form Y = b0 + biX),
R2 = r2. R2 can be interpreted simply as the proportion of the total variance
of the dependent variable which may be accounted for by the model. This is
shown schematically in Figure A-l. R2 may be computed as follows:
(A-14)
where SSR = sum of squares due to regression
SST = total sum of squares
SS = 2 Y^ - *'"An (A-15)
where Y, = predicted (for "Reg") and observed (for "T") values
1 of the dependent variable
It should be noted that to this point nothing has been defined in terms
of probabilities or "statistics." The fitting of a model by, least squares is
a purely mathematical process, and embodies no assumptions about underlying
distributions. Similarly, the computation of R2 is entirely devoid of assump-
tions about distributions and probabilities.
117
-------
INDEPENDENT VARIABLE
Figure A-l. Schematic showing an interpretation of
118
-------
REGRESSION STATISTICS
A number of important statistics are ordinarily generated in regression
analyses. These include:
F-statistic for the overall model.
F-statistic for improvement in R2 for each step in
stepwise multiple regression.
t-statistic for the coefficients or b-weights.
Confidence intervals about the simple linear regression
line for predicting individual values of the dependent
variable or the mean, for a given value of the independent
variable.
Confidence intervals for the slope and intercept in
simple linear regression.
Partial correlation coefficients.
Autocorrelation coefficients as a measure of independence
of residuals.
0 Confidence intervals about predicted values of the
dependent variable in multiple regression.
Tests of hypotheses (the fundamental utility of computed statistics) in-
volve assumptions about underlying distributions and characteristics of the
data. The assumptions are:
0 The data, for a given value of the independent variable,
are normally distributed.
0 The error component is independently distributed for a
given value of the independent variable.
0 The error variances are constant (homoscedastic) for
different values of the independent variable.
0 The independent variable has no error associated with it.
The last assumption is specific for certain regression applications.
The first three assumptions may be summarized as follows: Given that:
Y< = b + b X, + e., 1-1, n (A-16)
i o i ' i
where n = the number of observations.
119
-------
Y. = the value of the dependent variable for each value
1 of X.
e. = The error component of Y^ after accounting for the
deterministic component, b + b Xj.
then the assumptions specify that the EJ are normally and independently dis-
tributed with mean equal to zero and a standard deviation of a . a 2 is, in
turn, uniform over X. e e
The F-Statistic
Provided the requisite assumptions are met, a number of important statis-
tics may be generated and interpreted. Probably the most important of these
is the F-statistic. The F-statistic for the overall regression is computed as
2/dfN SSr/
=
F = Tl^P/dTp - syTn-k-1)
where dfN = degrees of freedom for the numerator = the number
of independent variables (k) in the regression
dfn = degrees of freedom for the denominator = the
number of observations (n) minus the number of
independent variables (k) minus one
P
(1-R ) = the coefficient of non-determination
SS = regression sum of squares
See equation (A-15)
SS = total sum of squares
k = number of independent (predictor) variables
n = number of observations
The value of F is compared with FCritical from a table of the F distribution.
If F>Fcr1*ica-,, then the regression relationship is a significantly better
estimate than is the mean at the specific confidence level.
F for comparing any step in the regression with the prior step is given
by: / \ /
VRy-1.2, tk " Ryl,2, ,k-l//DfN
F = 77-2 Y7 {A~18)
(l-Ryl,2, k)/ DfD
p 2
where R , 2 ... k = R for the current regression step
p 2
R i « L - = R for the preceding step
y*ijt» ,K~i
120
-------
DfH-l
Dfn = the number of observations (n) minus the
number of independent variables (k)
minus one
The t-Statistic
With respect to regression, the t-statistic is commonly used for deter-
mining whether each coefficient in multiple regression is significant. If so,
then at the selected confidence level, there is reason to believe the value is
truly non-zero. If it is non-zero, then the best estimate of the coefficient
is that determined in the regression. If the coefficient is not significant,
then there is no reason to believe it is non-zero, and the variable should not
be entered into the regression.
The t-statistic for each coefficient is given by:
(A-19)
\l \JI ">J
where b. = the coefficient
SE. = the standard error of the coefficient
bj
SE2
SEM " / .. "V (A-20)
where SS . = the sum of squares of variable j
R? = the squared multiple correlation coefficient regressing
J the jth variable on the remaining independent variables
(A-21)
i r\ A,
where SE 4. = standard error of estimate
est
SS = the residual sum of squares (sum of squares of
res differences between predicted Y and observed Y)
To estimate the confidence limits for predicting individual Y values and T for
a given value of X, where u and d designate upper and lower limits,
Vd = Y * *sy
where S = standard error of sample mean, 7 for a given value of X:
121
-------
1 t (X-X)2'
" 17"
or for S = standard error of individual Y values for a given X value:
(A-23)
's2
bE
1
, 1
n
. (X-X)21
XX2 J
(A-24)
2
In equations A-23 and A-24, Ex is the sum of squares of deviations of X:
;z_x
n
(A-25)
Using these equations (A-22 through A-25) it is possible to establish confi-
dence limits about the fitted line.
Partial Correlations
The partial correlation coefficients are useful in that they permit cor-
relations of variables to be examined while removing the effect of any number
of other variables on that correlation. For example, if it is desired to ex-
amine the correlation between variable 1 and variable 2, but it is suspected
that some of the correlation of variables 1 and 2 is contaminated with vari-
able 3, the effect of variable 3 may be removed from variables 1 and 2.
An example will clarify the utility of the partial correlations. Suppose
we are interested in the correlation between the amount of a pesticide remain-
ing on soil as a function of time with losses due to effects other than wash-
off. Correlating pesticide-remaining with time also includes the effect of
stormwater runoff transport. Partialling out cumulative rainfall (or other
rainfall variable) from time gives the correlation between time (and not rain-
fall) and the amount of pesticide remaining on the soil. If the correlation
is substantial, this suggests a real loss of pesticide on soil over time due
to factors other than washoff, s>ueK as biodegradation, photolysis, erosion,
and volatilization.
The computation of the partial correlation coefficient is performed as
follows:
rij*k
" rikrjk
(A-26)
where r
.. k
J
the partial correlation coefficient for the
correlation between variables 1 and j, both
independent of variable k
122
-------
Autocorrelation Coefficients
Autocorrelation of residuals and its evaluation are very important in re-
gression analysis, particularly where time-series are involved. Autocorrela-
tion statistics measure the degree of serial correlation in a data series.
Thus it can be considered as rx.-xl+j where if x represents individual obser-
vations, n is the number of such observations, and j is the lag, then i varies
from 1 to n-j.
The autocorrelation coefficient, which can be designated as ra is given
by
ra = ~- (A-27)
2dt
where d = residual (difference between observed and predicted
value of Y)
j = selected lag
Clearly j can be set to 1 for the smallest degree of lag in the observations
or it may be increased. If there is a periodicity to the autocorrelation,
then ra is likely to be similarly periodic.
Another measure of autocorrelation is the von Neumann Ratio:
V.N.R. = - ~ - (A-28)
In general, assessments of autocorrelation using the two statistics tend to be
consistent. In both cases, the statistic is computed and the value compared
with tabulated critical values. Within a defined region, significant autocor-
relation is indicated.
In regression, observations are assumed to be independent as discussed
earlier, and evaluation of statistical results is predicated upon some numbers
of degrees of freedom. Degrees of freedom are based, in turn, upon numbers of
observations and variables. If observations are not independent, then the
number of degrees of freedom, or, in essence, the number of sets of truly in-
dependent measures, can be severely overestimated, and data relationships may
be Judged meaningful when they are not.
123
-------
Several methods have been cited in the literature for dealing with the
problem of autocorrelation of residuals in time-series. Ezekiel and Fox*, for
example have cited the use of "first differences" between observations rather
than the observations themselves. This technique, which has been used suc-
cessfully with econometric data, usually eliminates the autocorrelation of
time-series regression residuals. However, in the author's experience, the
technique can also seriously affect results of correlation analysis, and the
value of the method is sensitive to features of the data. An example will
clarify this. Suppose the following is a data set of observations and first
differences with results of regression (Y = a + bX) as shown.
1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
X1
.00
.02
.03
.06
.08
3.21
3.23
3.26
3.31
3.40
7.20
7.80
7.90
8.00
8.10
a
b
rz
F
»1
37.3
39.2
37.6
39.6
40.1
40.8
41.$
47.0
46.2
47.8
50.0
55.0
55.2
54.9
53.2
- 36.84
2.192
.9070
- 126.8
»1
39.03
39.08
39.10
39.16
39.21
43.88
43.92
43.99
44.10
44.29
52.62
53.94
54.16
54.38
54.60
df d*
1.73 2.99
-0.12 0.014
1.50 2.25
-0.44 .194
-0.89 .792
3.08 9.49
2.42 5.86
-3.01 9.06
-2.10 4.41
-3.51 12.32
2.62 6.86
-1.06 1.12
-1.04 1.082
-0.52 .270
1.40 1.96
r, -0.0102
(rf not significant
at 95* level )
d1 di+l
-.208
-.18
-.66
.392
-2.74
7.45
-7.28
6.32
7.37
-9.20
-2.78
1.10
.541
-.728
-
AX,
0.02
0.01
0.03
0.02
2.13
0.02
0.03
0.05
0.09
3.80
0.60
0.10
0.10
0.10
a
b
r2-
F
AY,
1.9
-1.6
2.0
0.5
0.7
0.7
5.5
-0.8
1.6
2.2
5.0
0.2
-0.3
-1.7
--
.9756
.3157
.0263
.325
«,
0.98
0.98
0.99
0.98
1.65
0.98
0.99
0.99
1.00
2.18
1.17
1.01
1.01
1.01
~
V
<%
at
d1
-.92
2.58
-1.01
0.48
0.95
0.28
-4.51
1.79
-0.60
-0.02
-3.83
0.81
1.31
2.71
-
A
.846
6.66
1.02
.23
.90
.08
20.3
3.20
.36
0.0
14.7
.66
1.72
7.34
«
"l Vl
-2.37
-2.<1
-.48
.46
.27
-1.26
-S.07
-1.07
.01
.08
-3.10
1.06
3.55
--
-.233
not significant
95* level)
Regardless of whether the residuals are autocorrelated (they are not, in this
case), it is clear that taking differences can cause a severe reduction in the
correlation (r) in the data. This is particularly true when observations tend
to be clustered in a number of groups, as in data taken frequently over time
but only during storm activity. Because of this, the technique of using first
differences, as is sometimes suggested, may not be useful in many analyses.
*Ezekiel, M. and K.A. Fox. 1959. Methods of Correlation and Regression
Analysis. John Wiley and Sons, New York.pp. 340 et seq.
124
-------
Another alternative is that suggested by Wold* and reported by Ezekiel
and Fox (cited earlier in this Appendix). In this approach, the degrees of
freedom are corrected using the formula
N" =
" = NM1 + 2ra + 2ra + 2r& (A-29)
where IT = number of independent observations
N = number of observations
r = the autocorrelation coefficient for lag j.
aj
This is the technique used in this study.
Confidence Intervals in Multiple Regression
In simple regression, confidence limits are constructed on Y for each
value of X, permitting confidence envelopes to be described about the fitted
line. In a similar fashion, it is possible to generate confidence interval
estimates of predictions for each set of values for the independent variables
in multiple regression. Although the process is conceptually the same, vis-
ualizing it becomes difficult because of the multidimensional nature of multi-
ple regression. In three dimensions, the confidence limits consist of two
curved surfaces located above and below the fitted surface. The formulas for
the computation are presented in Ezekiel and Fox (cited earlier in this Appen-
dix, pages 320-1).
*Wold, Herman, in association with Lars Jureen. 1953. Demand Analysis. A_
Study In Econometrics. John Wiley and Sons, New York. pp. 43-5; pp. 209-13,
125
-------
APPENDIX B
RUNOFF WATER QUALITY ANALYSIS METHODS
WATKINSVILLE*
Analytical Methodology
Physiochemical Characterization of Soil, Sediment and Runoff--
Runoff samples for sediment analysis were acidified in polyethylene buck-
ets with a few drops ^04 to about pH 3 to 4 to promote flocculation of the
suspended sediment. Trie clear supernatant was removed and discarded and the
sediment was air-dried. The dried sediment was removed, weighed, and stored
for later use. The sediment concentration in the original runoff sample was
computed knowing the sediment weight and the volume (mass) of the runoff sam-
ple.
Particle size distribution, surface area, and organic carbon content were
determined on sediment samples from selected runoff events. Similar analyses
were also conducted on composite soil samples from each of the sampling areas
of the watershed as shown in Figures A2, A4, and A5. .Particle size distribu-
tion was determined by the hydrometer method,21 except that dispersion was ac-
complished using ultrasonic vibration.22 Organic matter was determined by wet
oxidation and potentiometric titration.23'21* Specific surface area was deter-
mined by N2 gas desorption,25'26 which measures external surfaces only. This
method was chosen as an indicator of total adsorptive capacity because of its
rapidity and reproducibility and small sample requirement. Non-expanding clay
minerals were predominant in the watershed soils. In preliminary comparative
studies of methods, total surface area determined by an ethylene glycol mono-
ethyl ether procedure27 gave values averaging about three times those of the
N2 desorption procedure.
Pesticide Residue Analysis in Soil, Sediment, and Runoff
During the project planning stage, it was anticipated that large numbers
of runoff and soil core samples would be collected for chemical analysis.
After planting, runoff samples were analyzed from each event until the parent
pesticide decreased in concentration to a level (depending upon the compound)
below the detectable range of the measuring instrument. Each runoff sample
received was recorded, a laboratory number assigned, and the samples placed
under refrigeration at 4°C pending analysis.
*This is a direct excerpt from the following report: Smith, C.N., R.A.
Leonard, G.W. Langdale, and 6.W. Bailey, 1978. Transport of Agricultural
Chemicals from Small Upland Piedmont Watersheds., U.S. EPA (Athens, GA) and
U;S.D.A. (Watkinsville, GA) EPA-600/3-78-056 pp. 57-60.
126
-------
Soil core samples for persistence and mass balance computations were ob-
tained after runoff events. Each core sample was recorded, a laboratory num-
ber assigned, and placed in a freezer at -18°C pending analysis. All core
data were reported on a moisture free basis.
An analytical method was needed to analyze the parent pesticides in run-
off (water and sediment) and soils at a minimum sensitivity in the low parts
per million (ppm) for paraquat and the low parts per billion (ppb) for triflu-
ralin and diphenamid. "Production line" analysis was necessary to provide a
large sample throughput in a minimum amount of time. In addition, a rapid
analytical procedure would reduce the risk of trifluralin loss by volatiliza-
tion and degradation.
An integrated method fulfilling these requirements was developed.28'29
This method was later used for the herbicides atrazine, propazine, and cyana-
zine. These compounds, however, required adjustment of the soil moisture to
at least 20 percent to ensure efficient extraction.
2,4-D was analyzed by a modification of the method of Woodham ejt al_.30 as
follows: Residues of 2,4-D were determined in soil, sediment, and water by
solvent extraction, acidification, and esterification to the methyl ester us-
ing diazomethane. The amount of the acid herbicide present was determined by
electron capture gas chromatography. A series of 2,4-D-fortified soil and
water samples as the free acid were analyzed using the final method. Recov-
eries run in replicate ranged from 96.7 to 98.4% in soil and water.
Fortified soil and water samples using the 2,4-D formulation (dacamine)
consistently ranged from 87 to 91% recovery on duplicates ranging from 2 ppb
to 400 ppm. This broad range of levels was run to assure that the length of
reaction time of the herbicide with the esterifying reagent and the amount of
reagent used would not affect the increase or decrease of the ester recovery.
Interferences from soil extractions were eliminated by a ^O/CI^Clo shakeout
of the acetone/soil extract at the time of acidification. The CH2&12 extract
was evaporated to 1 to 2 ml and transferred to 15-ml conical centrifuge tubes.
The remaining CH2C12 was evaporated just to dryness, and 2 to 3 ml of ether
was added at the time of esterification. Fortified samples and 2,4-D stan-
dards were run as controls with each set of 20 samples extracted and esteri-
fied.
The analyses were performed by using a Tracer MT-220 gas chromatograph
equipped with a Coulson electrolytic conductivity detector operating in the
nitrogen mode. Colorimetric determinations of paraquat were made by using a
Perkin-Elmer Model 202 recording spectrophotometer equipped with an auxiliary
recorder and scale expansion accessory.
Chloride and Plant Nutrient in Soil and Runoff
Soil and runoff samples were stored at -10°C until ready for extraction.
Subsamples of unfiltered runoff were stored at -10°C and the remaining subsam-
ples were filtered through a 0.60-um Nucleopore membrane and the filtrate
stored at -10°C. .Sediment was not analyzed separately because sediment con-
centrations in runoff were occasionally so low that collection of an adequate
sample by filtration was impractical.
127
-------
Nutrients were extracted from the frozen soil and runoff samples by plac-
ing a 5-gram sample of the frozen material into a 125-ml Erlenmeyer flask with
50 ml of distilled water and shaking the suspension for 1 hour on a wrist-
action shaker. Sample weights were corrected for water content from values
determined on separate samples taken during the initial sampling (see Table
C22 through C35). The suspension was filtered through Whatman Number 41 fil-
ter paper, and the filtrate was returned to storage at -10°C until analysis.
Chemical analysis was later accomplished by allowing the frozen test sol-
ution to equilibrate to room temperature before proceeding with the selected
automated procedures. Technicon auto-analyzer procedures were used exclu-
sively, varying analytical manifold configurations, reaction solutions, and
absorption cell lengths as required to give the required sensitivity in the
particular colorimetric method.31"35
Nitrate-N and ChlorideNitrates and chloride were determined on a dual
channel system using the ferric-mercuric thiocyanate color complex for chlo-
rides and the cadmium reduction procedure for nitrates.
Ortho-phosphorusFi1tered runoff samples were analyzed for ortho-P using
the phosphomolybdenum-ascorbic acid blue, color complex. Values reported as
ortho-P are often referred to as molybdate reaction phosphorus (MRP).
NH.yNAmmonia was determined in the filtered and unfiltered runoff sam-
ples using the Berthelot color reaction. Differences between the filtered and
unfil tered samples are assumed to represent exchangeable N^-N and reactive
amines displaced from the particulate phase in the alkaline medium.
Total Kjeldahl Nitrogen (TKN)Filtered and unfiltered runoff samples
were predigested in a Technicon BD-40 block digester with subsequent measure-
ment of the ammonia produced. The quantisation of ammonia was achieved by
the Berthelot reaction.
Total PhosphorusPhosphorus in the filtered runoff samples was hydro!-
ized with ammonium persulfate and sulfuric acid in a pressure cooker at one
bar for 30 minutes prior to colorimetric determination of P. The unfiltered
samples were digested in a mixture of 1:4 (HC10^:HN03) acid until fumes of
HC104 appeared. The residue was then taken up in distilled water and analy-
zed for total P.
Acid Extractable Phosphorus (Available P)Available soil P was extracted
with a double acid (0.05N HC1in 0.025N ^$04) solution and determined color-
imetrically.36
Watkinsville Data Base References
21. Day, P.R. Particle Fractionation and Particle Size Analysis. In: Method
of Soil Analysis, Part I, Black, C.A. (ed.). Agron. Monograph No. 9.
22. Bouget, S.J. Ultrasonic Vibration for Particle-Size Analysis. Can. J.
Soil Sci. 48:372-373. 1968.
128
-------
23. Jackson, M.L. Soil Chemical Analysis. Englewood Cliffs, Prentiss-Hall
Inc. 1958.
24. Ravek, A. and Y. Aurimelach. Potentiometer Determination of Soil Organic
Matter. Soil Sci. Soc. of Amer. Proc. JJ6:967. 1972.
25. Cremer, E. and H. Huck. Determination of Very Low Surface Area. Insti-
tute of Physical Chemistry, University of Innsbruck, Innsbruck, Austria.
1965.
26. Kremen, J., J.S. Lararias, and U.R. Dirtz. Surface Area Determination by
Equilibrium Gas Adsorption in Nitrogen-Helium Systems. Rev. of Scien-
tific Instruments. 37:1265-1266. 1966.
27. Neilman, M.D., D.L. Carter, and C.L. Gonzalez. The Ethylene Glycol
Monoethyl (EGME) Technique for Determining Soil Surface Area. Soil Sci.
100.: 409-413. 1965.
28. Payne, W.R., Jr., J.D. Pope, Jr., and J.E. Benner. An Integrated Method
for Paraquat, Diphenamid, and Trifluralin in Soil and Runoff from Agri-
cultural Land. J. Agr. Food Chem. 22_(l):79-82. 1974.
29. Pope, J.D., Jr., and J.E. Benner. Colorimetric Determination of Paraquat
Residues in Soil and Water. J. Assoc. Offic. Anal. Chem. 57(1):202-240.
1974. ~~
30. Woodham, D.W., W.G. Mitchell, C.D. Loftis, and C.W. Collier. An Improved
Gas Cbromatographic Method for the Analysis of 2,4-D Free Acid in Soil.
J. Agr. Food Chem. 19.(1): 186-188. 1971.
31. Technicon Autoanalyzer Methodology. Individual Simultaneous Determina-
tion of Nitrogen and/or Phosphorus in BD Acid Digest. Industrial Method
No. 329-74W, Tarrytown, NY. 1975.
32. Technicon Industrial Systems. Chlorides in Water and Wastewater. Indus-
trial Method No. 99-70W, Tarrytown, NY. 1971.
33. Technicon Industrial Systems. Ammonia in Water and Seawater. Industrial
Method No. 154-71W, Tarrytown, NY. 1973.
34. Technicon Industrial Systems. Orthophosphate in Water and Seawater. In-
dustrial Method No. 155-71W, Tarrytown, NY. 1973.
35. Jackson, W.A., C.E. Frost, and D.M. Hildreth. Versatile Multirange
Analytical Manifold for Automatic Analysis of Nitrate-Nitrogen. Soil
Sci. Soc. Amer. Proc. 39:592-593. 1975.
36. Procedures Used by State Soil Testing Laboratories in the Southern Region
of the United States. North Carolina State University, Raleigh, NC.
Southern Cooperative Series Bulletin No. 190. 23 p.
129
-------
BUFFALO BILL WATERSHED*
Experimental Method
Since it has been-our hypothesis that farm chemicals are attached to silt
particles, our sampling program was predicated to reach the stream in the
early stages of increased flow. . We set up telephone contacts with farmers
living on the Buffalo Bill Watershed asking them to advise us when they antic-
ipated sufficient rain to cause significant rise in stream levels. Upon re-
ceiving their call, our sampling teams were immediately activated and were
sampling the various chosen sampling points in approximately one hour. We set
up nine (9) sampling points over the watershed as indicated on the attached
map and attempted to sample these nine points three times during a given run-
off incident. Approximately one hour was.consumed in sampling the nine con-
secutive sampling points and several of the rainfall run-off incidents started
and finished within the space of 4-6 hours. The samples were collected for
various parameters including pesticides, nitrogens, solids, fecal coliforms,
phosphorus, dissolved oxygen, BOD and pH. These samples were taken back to
the laboratory in Iowa City and analyzed by chemists and microbiologists on
our regular, full-time staff.
Rainfall data was collected on three (3) sections of the 3500 acre water-
shed and flow measurements were made on the final downstream aggregate sam-
pling station in Jones Creek (Station #9).
Information regarding the application of farm chemicals was collected
during the fall operation and this information appears on the map. Our sam-
pling regimens were designed timewise to show water quality at low-flow-non-
runoff periods as well as the degraded water quality during runoff and high
turbidity stages.
The various sample stations (1 through 9) were selected to fractionate
out the various types of land as well as its usage. Sample sites 1, 2, 3 and
4 were sited to establish runoff water quality characteristics in upland high
slope corn and bean fields, while sample site #5 drains essentially pasture
land. Sample sites 6, 7 and 8 were chosen because they drain the flatter, low
slope corn and bean lands on the downstream part of the watershed. Sample
site #9 is the composite of all this drainage and served as the single flow
measurement point. We had originally hoped to have flow measurements on more
points than just number 9 but the suddenness with which the proposal was
funded made the procurement of flow sampling devices impossible.
*This is a direct excerpt from the following report: Morris, R.L. and L.
Johnson, 1974. Buffalo Bill Watershed Agricultural Runoff Study, Iowa State
Hygienic Laboratory, University of Iowa, Iowa City, Iowa. pp. 4-9. Report
and data used with kind permission of Dr. Roger Splinter of the Iowa State
Hygienic Laboratory. Note that materials presented here provide only lim-
ited information about methods but were included for the insights they pro-
vide into overall study methods. Figures and tables were not included here.
130
-------
Rainfall and Flow
Rainfall and flow data are shown in Figure 1 and Table 1 with the turbid-
ity and flow being graphed for station #9. The rainfall data are the average
of rainfall collected at the three (3) stations over the watershed. Rainfall
data were collected in standard, commercial rainfall gauges and the flow mea-
surements were made according to cross sectional areas and velocity calcula-
tions recommended to us by personnel of the U.S. Geological Survey. Because
these are not instrumental finite measurements, we have labeled the graph and
table as estimated flows. The flow measurements do correlate well with the
logic of actually measured rainfall and turbidity so that they appear to have
considerable validity.
Perusal of Figure 1 indicates that significant rainfall peaks correlate
quite well with increases in estimated flow and that the turbidity values also
follow this same flow increase pattern. Absolute conformity of flow and tur-
bidity with rainfall should not be expected because the manner in which the
watershed received the rainfall is quite variable. Sometimes an inch of rain
falls in a matter of a few minutes and at other periods the one inch rain in-
come occurred over several hours. This variability of rainfall produces dif-
ferent lag periods and peak heights with respect to both flow and turbidity.
Figure 1 indicates that there were four (4) significant storm episodes during
the five month period. These episodes occurred on August 13th, September
26th, October 10th and December 4th. It will be noted by scrutiny of follow-
ing data that various levels of rainfall under different conditions produce
different levels of silt movement and attendant farm chemicals.
Dissolved and Suspended Solids
Figure 2 and Table 2 contain information correlating turbidity levels and
both dissolved and suspended solids. A direct correlation between the sus-
pended or non-filterable solids is shown to a high degree. The filterable or
dissolved solids show the same high degree of correlation except in an inverse
manner as would be expected. This figure certainly indicates that in a water-
shed where poor conservation procedures exist, we demonstrate the expected
high siltation rates which are so noticeable in midwestern streams. Since the
August 13th rain was approximately three inches and the September 26th and
October 10th rains were about 0.9 inches, it shows that there is not a direct
relationship between the amount of rainfall and the movement of dissolved and
suspended solids. Again, this is due in part at least to the duration of
rainfall income. Other factors such as plant cover and soil moisture are un-
doubtedly highly influential in the siltation effect produced by a given
amount of rain.
Pesticides
Figure 3 shows the relationship between turbidity and the movement of
dieldrin during runoff periods. It should be noted here that aldrin was ap-
plied above station 8 or 9 yet the upstream sampling points indicated consid-
erable involvement of dieldrin attached to the moving silt particles. This
obviously can be attributed in part to the durability of aldrin applied on
the watershed the previous year. Suffice it to say, rises in turbidity
131
-------
produced by the rainfall incidents in every case resulted in significant in-
creases of the dieldrin content. This is in direct agreement with other stud-
ies performed by our laboratory in different areas of the state over the past
several years. There were no commercial applicators or manufacturing opera-
tions associated with this watershed area. The possibility of dieldrin move-
ment by volatility or wind transport from areas outside the Buffalo Bill
Watershed was not explored during this five month study, but we certainly in-
tend to review this possibility in the next seven month extended study period.
Air samples and rainfall will be analyzed for aldrin-dieldrin content before
and after application seasons. Non-durable pesticides were not detected due
to the fact that our first samples were taken in mid-August, three months or
more subsequent to actual application.
Fecal Coliform
Figure 3 shows the high correlation between turbidity, flow and fecal
coliforrn density. It is interesting to note that the highest levels of fecal
coliform at station #9 were not achieved during the greatest rainfall period
on August 13th. We have always felt that the earlier part of rainfall pro-
duced runoffs probably contained the largest amount of surface leached fecal
material and the highest levels were achieved during the two 0.9 inch rains on
September 4th and October 10th. Also, it can be noted that the earlier sam-
ples taken during the August 13th runoff contained the highest fecal coliform
levels which degraded as the runoff incident preceded.
Nitrogen Series
Figure 4 and Table 2 indicate a distinct, direct correlation between to-
tal organic nitrogen and turbidity as well as a consistent but lower level ef-
fect for ammonia nitrogen. Nitrate content does not appear to vary in this
instance. The ammonia nitrogen levels recorded are within the Lower Water
Quality Standard limits and are not nearly as high as we have frequently re-
corded on other streams within our state. Without having adequate information
on the application of commercial and natural fertilizers, we are unable to
really determine the source of these nitrogenous materials. It is hoped that
the time afforded for this type of application data in the next surveillance
period will permit us to render valid judgments on nitrogenous material
source. Many of the farmers involved were not exactly sure what they had used
on their fields and in what amounts.
Phosphates
Figure 5 shows the relationship of phosphate with respect to turbidity
variation and this parameter does not exhibit the correlation found with pre-
viously discussed constituents. It does indicate however that phosphorus
materials do move off a watershed such as the Buffalo Bill but the variability
in phosphate levels is less. As one would obviously expect, pH exhibits an
inverse function with respect to runoff due primarily to dilution effects.
Elevated pHs are undoubtedly a result of groundwater recharge of the streams
and the moving siltation is not sufficiently soluble to offset the effect of
volume dilution.
132
-------
SEATTLE, WASHINGTON
All water quality analyses represent standard procedures indentifiable
through EPA STORET codes. The codes for Seattle data are as follows:
STORET
code
665
10501
605
610
630
530
70
94
1051
1052
PARAMETER
Total P
OP04-P
Organic N
NH3-N
(N02 + N03) - N
Suspended solids (SS)
Turbidity
Conductivity
Pb
Zn
HONEY CREEK*
The analytical method used on each parameter sampled at the Sandusky
River material transport stations are given below.
Total Phosphorus
Total phosphorus analyses was performed using the automated colorimetric
ascorbic acid reduction method for Technicon Autoanalyzer II systems as de-
scribed in the Methods for Chemical Analysis of Water and Wastes, U.S. Envir-
onmental Protection Agency, 1974, beginning on page 256.The persulfate di-
gestion was performed by heating in an autoclave for 30 minutes at 121°C. For
samples with high suspended solids, the samples were filtered through a pre-
washed glass fiber filter (millipore AP2504700) following removal from the
autoclave and before cooling. Standard solutions and blanks were included
with each batch of samples and underwent the same digestion procedures as the
samples. Reported as P.
*This is a direct excerpt from: U.S. Army Corps of Engineers (Buffalo, NY),
undated. Water Quality Data for Sandusky River Material Transport Stations.
Lake Erie Wastewater Management Study. Data collected and analyzed by
Heidelburg College, Tiffin, OH. pp. VIII-X.
133
-------
Dissolved Orthophosphate
The analytical procedures for dissolved orthophosphate were the same as
those used for total phosphorus as described above except that sample pre-
treatment consisted of filtration of the raw sample through a prewashed Milli-
pore HAWP filter. The filtrate was then directly analyzed by the colorimetric
procedure cited above. Reported as P.
Residue, Total Non-Refilterable (Suspended Solids)
Suspended solids were analyzed according to the procedures outlined in
the Methods for Chemical Analysis of Water and Wastes (U.S.E.P.A., 1974, page
268 and following). A well mixed sample was filtered through a preweighed
glass fiber filter, (Reeve Angel 934AH) and the residue retained on the filter
was dried to constant weight at 103-105°C. Weighings were done on a Mettler
H20T balance with digital readout to the nearest 0.01 milligrams.
Nitrogen, Nitrate-Nitrite
The automated cadmium reduction method was employed as described in the
Methods for Chemical Analysis of Water and Wastes (U.S.E.P.A., 1974 page 207
and following).The analysis was run on the same filtrate as used for dis-
solved orthophosphate. The value reported included both nitrate and nitrite
nitrogen.
In this method a filtered sample is passed through a column containing
granulated copper-cadmium to reduce nitrate to nitrite. The nitrite (that
originally presented plus reduced nitrate) is determined by diazotizing with
sulfanilamide and coupling with N-(l-napthyl)-ethylenediamine colorimetrical-
ly. Reported as N.
Nitrogen, Ammonia
The automated colorimetric phenate method was employed as described in
the above manual (E.P.A., 1974, page 168 and following). The analysis was run
on the same filtrate as used for dissolved orthophosphate and nitrate-nitrite.
In this method Alkaline phenol and hypochlorite react with ammonia to form
indophenol blue that is proportional to the ammonia concentration. The blue
color formed is intensified with sodium nitroprusside. Reported as N.
Specific Conductance
Specific conductance was measured using a Barnstead Model PPM 70CB Con-
ductivity Meter and a YSI conductivity cell. Samples were brought to 25°C
prior to measurements. Details of the procedure are outlined in Standard
Methods for the Examination of Water and Hastewater, 13th Edition, (1971),
page 323.
Silica, Dissolved
Silica analysis was performed using the automated method (Technicon In-
dustrial Method #182-72W) on the same filtrates as used for dissolved orthos-
phate. Rep'orted as Si02-
134
-------
Chloride
Chloride analysis was performed using the procedures outlined in the
Methods for Chemical Analysis of Water and Wastes (U.S.E.P.A., 1974, page 31
and following),Thiocyanate ion (SCN) is liberated from mercuric thiocynate,
through sequestration of mercury by chloride ion to form unionized mercuric
chloride. In the presence of ferric ion, the liberated SCN forms highly col-
ored ferric thiocynate, in concentration proportional to the original chloride
concentration. Reported as Cl.
Iron
The analysis of iron was done according to the procedure given in Stan-
dard Methods for the Examination of Water and Wastewater, 13th Edition, (1971),
with some modification.A Technicon Autoanalysis was utilized according to
the automated method (Technicon Industrial Method #109-71W).
Total Kjeldahl Nitrogen
An ultramicro technique was used in the analysis of Total Kjeldahl Nitro-
gen (TKN). This procedure is outlined in the Ultramicro Semi-Automated Method
for the Simultaneous Determination of Total Phosphorus and Total Kjeldahl Ni-
trogen in Wastewaters (Environmental Science and Technology. October, 1976).
The ammonia is analyzed in the range of .05 to 10.0 mg N/l using the indo-
phenol blue method with automated spectrophotometry at the rate of 30 samples
per hour. Reported as N.
135
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA-600/3-80-022
2.
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
Sediment-Pollutant Relationships in Runoff from Selectejd
Agricultural, Suburban, and Urban Watersheds:
A Statistical Correlation Study
5. REPORT DATE
January 1980 issuing date
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
Stanley W. Zison
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Tetra Tech, Incorporated
3700 Mt. Diablo Blvd.
Lafayette, California 94549
10. PROGRAM ELEMENT NO.
A34B1B
11. CONTRACT/GRANT NO.
68-03-2611
12. SPONSORING AGENCY NAME AND ADDRESS
Environmental Research LaboratoryAthens GA
Office of Research and Development
U.S. Environmental Protection Agency
Athens, Georgia 30605
13. TYPE OF REPORT AND PERIOD COVERED
Final, 9/77-9/78
14. SPONSORING AGENCY CODE
EPA/600/01
15. SUPPLEMENTARY NOTES
16. ABSTRACT
Data from agricultural, suburban, and urban watersneds were subjected
to statistical correlation analysis to estimate potency factors. These
factors are coefficients that, when multiplied by sediment mass emission
rates (transported in runoff), provide estimates of mass emission rates for
other pollutants. The potency factors are required input for such lumped-
parameter runoff models as the Nonpoint Source (NPS) Model and the Storm-
water Management Model (SWMM).'
The data were also subjected to multiple regression analysis to examine
the effect of storm parameters on runoff water quality and the inter*relation-
ship among runoff water quality constituent concentrations themselves (other
than sediment load). The multiple regression analysis was primarily explora-
tory with the objectives of explaining variance of water quality and identi-
fying important independent or predictor variables rather than developing
predictive expressions.
7.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
Simulation
Water Quality
Runoff
Model Studies
Nonpoint Pollution
Agricultural Runoff
Agricultural Watersheds
Urban Runoff
02A
08H
12A
68D
18. DISTRIBUTION STATEMENT
RELEASE TO PUBLIC
19. SECURITY CLASS (ThisReport)
UNCLASSIFIED
21. NO. OF PAGES
148
20. SECURITY CLASS (Thispage)
UNCLASSIFIED
22. PRICE
EPA Perm 2220-1 (9-73)
136
ft U.S. GOVERNMENT HUNTING OfFICt: 1980 -657-146/5566
------- |