xvEPA
United States
Environmental Protection
Agency
Environmental Research
Laboratory
Athens GA 30613
EPA/600/9-85/016
May 1985
-------
EPA/600/9-85/016
May 1985
PROCEEDINGS
OF
STORMWATER AND WATER QUALITY MODEL
USERS GROUP MEETING
January 31-February 1, 1985
Edited by
Thomas 0. Barnwell, Jr.
Center for Water Quality Modeling
Environmental Research Laboratory
Athens, GA 30613
ENVIRONMENTAL RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
ATHENS, GA 30613
-------
DISCLAIMER
The work described in these papers was not funded by the U.S. Environ-
mental Protection Agency. The contents do not necessarily reflect the views
of the Agency and no official endorsement should be inferred.
ii
-------
FOREWORD
A major function of research and development programs is to effectively
and expeditiously transfer technology developed by those programs to the
user community. A corollary function is to provide for the continuing ex-
change of information and ideas between researchers and users, and among the
users themselves. The Stormwater and Water Quality Model Users Group,
sponsored jointly by the U.S. Environmental Protection Agency and Environment
Canada/Ontario Ministry of the environment, was established to provide such
a forum. The group has recently widened its interests to include models other
than the Stormwater Management Model and other aspects of modeling water
quality in urban and natural waters. This report, a compendium of papers ^
presented at the users group meeting held on January 31-February 1, 1985, in
Gainesville, FL, is published in the interest of disseminating to a wide
audience the work of group members.
Rosemarie C. Russo, Ph.D.
Director
Environmental Research Laboratory
Athens, Georgia
iii
-------
ABSTRACT
This proceedings includes 17 papers on topics related to the develop-
ment and application of computer-based mathematical models for water quantity
and quality management. The papers were presented at the semi-annual meeting
of the Joint U.S.-Canadian Stormwater and Water Quality Model Users Group
held on January 31-February 1, 1985, in Gainesville, Florida.
Application of stormwater management modeling is examined in North
American and European settings in several of the papers, including Florida
flatlands, North Carolina peatlands, Canadian urban runoff ponds, and Swiss
stormwater runoff tanks. Estuary studies reported include hydrodynamic and
water quality simulations, phytoplankton-nutrient dynamics modeling, and
techniques for assessing reservoir eutrophication. The Hydrological Simula-
tion Program-FORTRAN, is the basis for studies of snow melt simulations, of
deep pumping effects on surface hydrology, and of phosphorus dynamics in
wetlands. Phosphorus concentration data are linked with commonly measured
watershed characteristics in another study.
A model is presented to evaluate the cost-effectiveness of best manage-
ment implementation schemes on two agricultural basins in Florida, and QUAL-II
is applied to a resource allocation project in England. In other studies,
factor analysis is applied to the management of impoundment water quality,
and simulation modeling is used to evaluate aquifer storage recovery in a
regional water supply system. An affordable alternative to a mainframe com-
puter environment for continuous modeling is presented.
iv
-------
CONTENTS
FOREWORD m
ABSTRACT iv
ACKNOWLEDGMENT . vil
STORMWATER MANAGEMENT BY MICROCOMPUTER ..... 1
B.A. Christensen, University of Florida, and A.D. Tilton, Johnson
Engineering, Inc.
AN AFFORDABLE ALTERNATIVE TO A MAINFRAME COMPUTER ENVIRONMENT FOR
CONTINUOUS MODELING 13
W. James and M. Robinson, McMaster University
MULTIOBJECTIVE DESIGN OF STORMWATER IMPOUNDMENTS . . 31
E.A. McBean, University of Waterloo
DETERMINATION OF RUNOFF CHARACTERISTICS OF FLATWOOD WATERSHEDS .... 45
K.L. Campbell, J.C. Capece, and L.B. Baldwin, University of Florida
STORMWATER MANAGEMENT MODEL APPLICATION TO A PEATLANDS REGION IN
COASTAL NORTH CAROLINA 60
R.E. Dickinson, W. Pandorf, and L.J. Danek, Environmental Science
and Engineering, Inc.
AREA-WIDE STRATEGIES FOR STORMWATER MANAGEMENT IN SWITZERLAND: CASE
STUDY GLATTAL 69
V. Krejci and W. Gujer, Federal -Institute of Water Resources
and Water Pollution Control
THE IMPACT OF "SNOW" ADDITION ON WATERSHED ANALYSIS USING HSPF .... 87
S. Udhiri, M-S. Cheng, and R.L. Powell, Maryland-National Capital
Park and Planning Commission
USE OF HSPF TO SIMULATE THE DYNAMICS OF PHOSPHORUS IN FLOODPLAIN
WETLANDS OVER A WIDE RANGE OF HYDROLOGIC REGIMES 116
J.C. Nichols and M.P. Timpe, Water and Air research, Inc.
SIMULATION OF A REGIONAL WATER SUPPLY WITH AQUIFER STORAGE 133
R.L. Wycoff, CH2M HILL
SIMULATION OF POSSIBLE EFFECTS OF DEEP PUMPING ON SURFACE HYDROLOGY
USING HSPF 144
C.N. Hicks, W.C. Huber, and J.P. Heaney, University of Florida
-------
CONTENTS (cont'd)
HYDRODYNAMIC AND WATER QUALITY SIMULATIONS IN AN ESTUARY WITH MULTIPLE
OCEAN BOUNDARIES 157
I.B. Chou, Applied Technology and Management, Inc., and L.J, Danek,
ESE, Inc.
MODELING ESTUARINE PHYTOPLANKTON-NUTRIENT DYNAMICS USING MICROCOMPUTERS 179
W-S. Lung, University of Virginia
APPLICATION OF FACTOR ANALYSIS TO MANAGEMENT OF IMPOUNDMENT WATER
QUALITY 196
C.D. Pollman and R.E. Dickinson, Environmental Science and Engin-
eering, Inc.
THE APPLICATION OF QUAL-II TO AID RESOURCE ALLOCATION ON THE RIVER
BLACKWATER,"ENGLAND 208
B. Crabtree, I. Cluckie, P. Crockett and C. Forster, University
of Birmingham
TECHNIQUES AND SOFTWARE FOR RESERVOIR EUTROPHICATION ASSESSMENT .... 231
W.W. Walker, 'Jr.
MODELING OF PHOSPHORUS CONCENTRATIONS FROM DIFFUSE SOURCES 241
D.J. Andrews, Marshall Macklin Monaghan Limited, K.K.S. Bhatia,
National Institute of Hydrology, and E.A. McBean, University of
Waterloo
A MODEL FOR ASSESSING THE COST-EFFECTIVENESS OF AGRICULTURAL BMP
IMPLEMENTATION PROGRAMS ON TWO FLORIDA BASINS 257
C.D. Heatwole, A.B. Bottcher, and L.B. Baldwin, University of
Florida
LIST OF ATTENDEES 266
VI
-------
STORMWATER MANAGEMENT BY MICROCOMPUTER
B.A. Christensen A.D, Tilton
Professor, Hydraulic Laboratory Project Engineer
Department of Civil Engineering Johnson Engineering, Inc.
University of Florida Fort Myers, FL 33902, USA
Gainesville, FL 32611, USA
ABSTRACT
Stormwater management and floodplain delineation in very flat areas
such as Florida's rapidly developing coastal zones have long depended on in-
put from mainframe computers that may not be directly available to the many
consulting engineering offices that attempt to tackle these problems. Tech-
nological advances in the microcomputer field that make these tasks possible
for owners of the not too expensive latest generation of microcomputers are
therefore welcomed by the engineering profession.
The present paper discusses the development and use of a microcomputer
oriented program that will yield the necessary information needed by civil
engineers and planners dealing with drainage basins of small to intermediate
size in the extremely flat areas along Florida's coastline. The program is
simpler than similar existing programs for larger computers that take into
consideration the many complications that steep or even moderate slopes
usually found in drainage basins at other locations dictate.
The program develops simple synthetic hydrographs for all subbasins and
routes them through overland sheet flow, channels and man-made or natural
reservoirs.
Based on the discharges generated in this way the water surface eleva-
tions are predicted by use of the iterative standard step method throughout
the system. The depths obtained in this way are applied to improve the rout-
ing results in several steps. The initial hydrographs are based on the Man-
ning formula for sheet flow and the assumption that the flat land of the
drainage basin serves two functions, namely water storage and conveyance of
the stormwater towards the ultimate recipient. Man-made as well as natural
watercourses with wide floodplains may be treated.
The method is financially competitive not only during the first flood-
plain mapping of an area but especially during the following updatings called
for by development of the drainage basin.
1
-------
INTRODUCTION
The need for a simple program that will compute the extent of the flood-
plain in a flat area for a given rainstorm has evolved over the last ten to
fifteen years as population pressures and industrial developments have brought
floodplain ordinances and similar legislation into existence. Sophisticated
methods for floodplain prediction have long existed and are reflected by such
major hydrologic simulation models as the U.S. Army Corps of Engineers HEC
programs and the SWMM-program developed for the United States Environmental
Protection Agency. These programs usually require mainframe computers for
their execution.
The program discussed in this paper is designed to meet the needs of the
consulting engineering office without direct access to a mainframe computer.
It is well suited for execution on most of the microcomputers that are avail-
able today (1985) and is limited to the prediction of the floodplain extent
and elevation in a flat watershed of moderate size drained by natural and
man-made watercourses. It is considering the overland sheet flow that often
prevails in such drainage basins in their natural state.
The program consists of four parts:
A. Computation of synthetic hydrographs for the individual subbasins and
composite hydrographs at selected stations. This part includes the
routing of hydrographs through channels and reservoirs shown in Figure 1.
B. Computation of depths, elevations of water surface, piezometric head
line, energy grade line, and widths of the water surface at all selected
stations.
C. Numerical display of results.
D. Graphical display of results.
Since the depths and depth related parameters computed in part B are
functions of the discharges found in part A and these discharges again are
functions of the depths a major iterative scheme must be involved in the solu-
tion. The first set of discharge values is based on an arbitrarily chosen
(but realistic) set of depth values. The resulting discharges are used to
compute an improved set of depth values which are used for improvement of the
discharges in part A and so on in an iterative manner until a satisfactory
accuracy is achieved.
Before the program can be initiated the raw data must be presented in a
form that makes it compatible with the program. From topo maps of the water-
shed water divides must be determined, subbasins defined and basin areas, Ag,
computed. Stations along the watercourses where ther floodplain elevation
and width are to be computed must be defined and the channel cross sections
approximated by well defined curves or straight lines. In the following sec-
tion it is shown how a section with even very wide floodplains may be defined
by six numerical values which together with the section's station number will
describe the section's location and shape completely.
-------
Rainstorm parameters such as rain intensity i, time of the beginning of
the storm t0, duration T and the geographical extend of the storm must of
course also be decided. A runoff coefficient E taking antecedent conditions
into consideration must be evaluated according to the nature and degree of
development of the considered subbasin.
The routing of the runoff handled by part A is demonstrated in Figure 1
showing an example with five subbasins. A hydrograph at a certain location
is indicated by the symbol HG. The subscript (numbers or a combination of
numbers and letters) indicates how that hydrograph is generated.
A single digit (1 through 5) indicates that the hydrograph is a synthe-
tic hydrograph generated for the subbasin having that number. A subscript
consisting of two or more digits, but no letters, refer to a hydrograph
created by simple superposition of the individual synthetic hydrographs for
the subbasins having the numbers referred to by the digits. A letter C in
the subscript shows that the hydrograph represented by the preceding symbols
of the subscript has been channel routed. Channel routing is carried out by
the Muskingum method briefly described by Henderson (1) and in more detail by
Overton (2). Similarly reservoir routing is indicated by the letter R.
For example the symbol HGL 2.C.3.4.R.C.5 for the hydrograph at the re-
ceiving waters in Figure 1 states"that this'composite hydrograph is obtained
by adding the synthetic hydrographs of subbasins Nos. 1 and 2, channel rout-
ing the result, adding the synthetic hydrographs of subbasins Nos. 3 and 4,
reservoir and channel routing the resulting hydrograph and finally adding the
synthetic hydrograph of subbasin No. 5.
Since the program is intended for use in flat coastal areas it is reason-
able to assume that all channel and overland flow is subcritical, i.e. that
the energy of the flowing waters is predominantly potential rather than kin-
etic. Consequently, the computation of the backwater profile mentioned in
part B must begin at the most downstream control section and proceed in the
upstream direction. The standard step method discussed by Chow (3) is chosen.
This iterative method determines the water depth and thereby the floodplain
elevation at the chosen stations along the watershed's channels and creeks.
DRAINAGE BASIN GEOMETRY
It is most important for a successful use of the program that water
divides and cross sections of the watershed's channels are well defined.
While the first are found manually from topographic maps the latter may be
accomplished by approximating the surveyed cross sections by straight lines
and or exponential curves.
An arbitrary cross section is shown in Figure 2. It is divided into
three parts, a central part of width b and constant depth d, and a left (sub-
script 1) and right (subscript 2) bank part. The areas of the two bank parts,
A,, and AB 2» may be determined for a given depth from surveys in the field.
Using the x",y coordinate systems shown in Figure 2 the two bank part may be
approximated by the exponential curves
-------
n,.
alx
(1) and y = a x
X
(2)
where a-^, a^ = constants and n-,, n~ - constant exponents to be determined
from field data,requiring same width of water surface, depth and cross sec-
tional area of section defined by Equations (1). and (2) and the actual section.
Using this concept the area of the section shown in Figure 2 may be
expressed as a function of depth d and known parameters by
1 + n
1
n n n
A 1 1 ' J I
A Dd + 11 " i/'*" ' i _i_
1 + n. 1/n, i +
1 1
a
1
,
2
. , A
n2 l/n2 d
a2
....(3)
HYDROGRAPHS OF SUBBASINS
A synthetic hydrograph for a subbasin may be generated by recalling that
the area of that basin has two functions, storage of water and conveyance of
water.
Separating these two functions as indicated in Figure 3 and assuming
that the flow representing the conveyance of water away from the basin towards
its principal watercourse is uniform with depth do the differential equation
representing conservation of. mass may be written
ei A,, = A,, -rr
£+2Ld Md2/3S//2.
at o o b
(4)
in which e = runoff coefficient taking into consideration infiltration and
other losses such as evapotranspiration, i = intensity of rain, Ag basin
area, h elevation of water surface in horizontal part of basin, t = time
and Q » rate of outflow.
Assuming that h s dQ, the Manning formula is used in Equation (4) to
evaluate the rate of outflow term using the following symbols: L maximum
length of subbasin, M - 8.25 /g/k^'^, g = acceleration due to gravity, k »
equivalent sand roughness of channel bed and submerged land surfaces, and
S. «* sin 3 m slope of land where $ = inclination of land with respect to
horizontal.
Using the boundary conditions that Q = 0 at time t « to, that the rain-
storm start at that time and the rain intensity i remains constant after that
time Equation (4) has the solution
3 A, rQ
I
t » -=
(2LM :/sT)3/5
b
f
dQ
ei
A, Q2/5 - Q7'5
+ t
(5)
-------
This may be approximated by a straight line with good accuracy in a t,Q-
coordinate system.
Letting T denote the duration of the rainstorm and Qmax the maximum rate
of outflow at t » t0 + T the equation of that line may be written
t - t
(6)
where 0 of course is unknown.
Tuax
After the rain has stopped at time to + T the solution of Equation (4)
may be written
t - (t + T)
5/2
(7)
K
where the constant K is given by
Tr _ 3 AB
2 * "\ 1 5
f IT vf /c1 \ J / *
C2LM rS, )
(8)
Equations (6) and (7) represent the rising and falling banks of the
synthetic hydrograph, respectively.
To satisfy continuity it must now be required that
Q dt - ei A, T,
r
(9)
an equation that may be used for determination of
.x.
Introduction of Equations (6) and (7) for Q at times t < t < tQ + T and
t + T < t, respectively, in Equation (9) and integrating yields
(10)
in which the first term on the right hand side may be shown to be predominant.
Consequently Q may be found from
(ID
n as C-H
Tnax
1
^B 1 , 2 K
I+lT *
1
Q275
Tnax
Equation (11) is to be solved by an iterative process for which a sub-
routine is to be included in the program.
-------
To include the effect of the "average" preceding storm condition a base
flow rate of one tenth of Qmax is added to the discharges given by Equations
(6) and (7) resulting in the final equations
Q = Q
t - t +
o
max T
0.1 T
(12)
and
t - (t H
r °
{ K
i
- T)
. 1 i
Q!/S
Tnax
Tnax
2 10
(13)
for the rising and falling limbs of the hydrograph, respectively. K is found
from Equation (8) and 0 by the iterative process outlined by Equation (11).
iH3.X
Figure 4 shows an example of the complete synthetic hydrograph for a
subbasin.
ROUTING OF RUNOFF
The hydrographs are channel and reservoir routed through the system as
indicated by the example of Figure 1.
The programs subroutine for channel routing is based on the classic
Muskingum method.
Reservoir routing is carried out by the storage indication method also
referred to as the modified plus method.
COMPUTATION OF BACKWATER CURVES
The computation of depths and thereby elevations of water surface and
energy grade line at the chosen stations is based on the standard step method,
an iterative method giving the depths at given locations in open water courses
with steady or quasi-steady flow. In the flat regions considered here it is
indeed a reasonable assumption to assume quasi-steady flow. It may further-
more be assumed that the flow is subcritical, i.e., that the predominant form
of energy in the flowing water is potential rather than kinetic. A conse-
quence of the latter assumption is that all step by step integrations of the
differential equation describing the free water surface must be carried out
in the upstream direction.
Using the symbols of Figure 5 and applying the energy equation from
station No. j+1 to station No. j
-------
in which S = slope of the energy grade line at the station indicated by the
subscript, and A£. . , = step length.
J ~~
The energy coefficient a is assumed constant along the entire water-
course and the'areas A. and A , may be found as functions of the correspond-
ing depths from Equation (S).-5
Equation (14) is now solved for the second term, dj+1, on the left hand
side giving the iterative formula for d.,,
S4- 1 1 1+1 '£
; 2 j .j+1
Under normal circumstances this equation provides a rapidly converging
solution. The first estimate of d . is d . .
The slopes, S, of the energy grade line in stations Nos. j and j+1 may
be found from the Manning formula as functions of the corresponding depths,
S -- 5^-770 ...................................... (16)
MA2 R4/3
where M = 8.25 v/g/k ' now must be based on the roughness of the considered
channel and A may be found from Equation (9) for the depths d. and
Instead of the usually applied hydraulic radius a resistance radius R
is introduced in Equation (16) to take the true shear stress distribution
along the wetted perimeter of the combined channel and floodplain into con-
sideration. While the use of the hydraulic radius implies a constant bed
shear stress along the entire wet perimeter of an open channel, the resist-
ance radius is intended to consider the influence of a much more realistic
bed shear stress distribution where the local bed shear stress is propor-
tional to the local vertical depth, i.e. , proportional to d-y in Figure 2.
This is of special importance in the very wide relatively shallow sections
encountered during flooding of the flat lands of Florida's coastal zone.
As indicated by the Danish Hydraulic Institute (4) the resistance radius may
be written
R - (j f (d-y)3/2 dx]2 ...................... (17)
A J0
for wide flat sections.
Applied_to the section shown in Figure 2, Equation 17 yields the resist-
ance radius R for that section
-------
1/n n 1/n
ni ,1 1 no A
1 d - 2 d
-i /
1/n
d
y
Mean Depth d
m
V
Correction Factor for Nonuniform Bed Shear Stress
(18)
Use of this expression for R and Equation (3) for the cross sectional
area in Equation (16) gives the values of the slope of the energy grade line
needed for the iterative determination of the depth d. . from Equation (15).
CONTROL STRUCTURES
The presence of control structures such as culverts, weirs and gates in
the watercourse.Is handled by either computing the upstream water surface
elevation 'fr'oiri the- dbvnstream water surface elevation in cases of submerged
structures or introducing a new control section upstream of the not submerged
structures.
In the not submerged cases the depth in the upstream control section
must be independent of the downstream depth and may be determined from the
discharge and the general geometry of the structure.
Culverts, weirs and gates are considered in the model using the commonly
accepted formulas for the discharge-head relationships for such structures.
CONCLUSION
The program developed in this paper is well suited for floodplain de-
lineations in small to moderately sized coastal watersheds. In spite of the
8
-------
many simplifying assumptions and shortcuts made the method has proven to give
reliable results.
The final display of results include printouts and plots of the energy
grade line, the water surface and the channel inverts along the canals and
natural watercourses in the watershed. Also the width of the floodplain is
plotted along the major watercourses.
Hydrographs may also be plotted at selected stations if it is desired.
V
A drainage basin with 100 to 200 stations may be handled by most advanced
microcomputers equipped with one hard disk drive.
ACKNOWLEDGEMENT S
This research and development effort was sponsored by the Engineering
and Industrial Experiment Station of the University of Florida and by Johnson
Engineering, Inc., of Fort Myers, Florida. This support is gratefully acknow-
ledged.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency and therefore the contents do not necessarily reflect the
views of the Agency and no official endorsement should be inferred.
REFERENCES
1. Henderson, F.M. "Open Channel Flow," The MacMillan Company, New York,
1966.
2. Overton, D.E. "Muskingum Flood Routing of Upland Streamflow," J. Hydrol.
4, No. 3, 1966.
3. Chow, Ven Te. "Flood Routing," in "Open-Channel Hydraulics," Chapter 20,
McGraw-Hill Book Co., Inc., New York, 1966.
4. Danish Hydraulic Institute. "Danish Hydraulics," No. 3, November, 1982.
A more detailed version of this paper was published in Proceedings of the
International Conference on the Use of Micros in Fluid Engineering, BHRA
Fluid Engineering, Bedford, England, 1983.
-------
WATERSHED
HYOROGRAPH
SYMBOL
RECEIVING WATERS
Figure 1. General Lay-Out of Drainage
Basin. Routing Scheme.
LEFT BANK
DAPT
RIGHT BANK
PART
x-B,
Figure 2. Approximation of Natural or Man-made
Channel Cross Section.
10
-------
EVAPOTRANSPIRATION
m n M
PRECIPITATION
INFILTRATION
i- NEARLY HORIZONTAL
STORAGE FUNCTION-
L MILD SLOPE
HORIZONTAL
CONVEYANCE FUNCTION
Figure 3. Generation of Synthetic Hydrograph for a
Subbasin.
RISING LIMB
BASE FLOW ^_
O.IQ
max
TIME t
DURATION OF RAINFALL
to -BEGINNING OF RAINFALL
Figure 4. Synthetic Hydrograph for a Subbasin.
11
-------
(/tj-H -/£=)
PHL-W.S.
CONTROL SECTION
SECTION NO.j
SECTION NO.j + l
BED
Figure 5. Scheme for Backwater Curve Prediction
12
-------
AN AFFORDABLE ALTERNATIVE TO A MAINFRAME
COMPUTER ENVIRONMENT FOR CONTINUOUS MODELLING
by: W. James and Mark Robinson
Computational Hydraulics Group
McMaster University
Hamilton, Ontario Canada LBS 4L7
Telephone: 416/527-6944
ABSTRACT
Continuous modelling is necessary for reliable estimates of
frequencies of flow and pollutional events. When stormwater storage or
diversions are considered, the frequencies ought to be especially carefully
computed. The advantages of continuous modelling in microcomputer control
of combined sewer networks are reviewed. Related activities of the
Computational Hydraulics Group are also briefly covered.
Types and sources of data and special time series management software
necessary for continuous modelling are mentioned. Some general
specifications are suggested for suitable hardware, based on computer
storage capacities, computing speeds and data base requirements.
The local area network of IBM-PC compatibles with 9-track and hard-
disk capabilities used by the group are described, and pleasantly low costs
are cited.
INTRODUCTION - CONTINUOUS MODELLING
The misnamed "Rational Method" is reportedly still widely used in North
America for calculating peak flows due to stormwater runoff. Established
design methodology for urban drainage systems" is based on simplistic,
empirical relations and design storms derived from intensity-duration-
frequency analysis. Many municipalities specify synthetic design storms for
their major and minor drainage systems, e.g. the so-called "Chicago"
synthetic design,storms. The design storm concept, like all single event
modelling, is characterized by the drawback that critical starting values
of field moisture and storage levels are unknown.
13
-------
The proponents of the design storm concept argue on the basis of:
1. Consistency. The technique is traditional practice.
2. Conservative results. Because the method computes excessive peak
flows it is inherently safe.
3. Economy. The method is inexpensive and widely held to he quick and
easy to use.
The opponents' arguments include:
1. Irrelevance. The concept has not been proven correct. Marsalek
(1981, 1977) has shown that design-storm-derived peak flows can be
significantly larger than those obtained for historical storms of
equivalent probability.
2. Unreliability. No reliable probability can be assigned to the
runoff from a single rainfall event (Linsley, 1974).
3. Start-up Probabilities. Current practice specifies a probability
for the design storm event period but does not specify the joint
frequency of the antecedent dry period or the spatial and
kinematic variability of thunderstorm-type rainfall.
4. Inaccurate Input. Design storms do not properly represent long
duration, high volume events, which are significant in
detention/retention basin design.
5. Pollution Control. On the other hand, the most critical periods
for a river or receiving water may occur during dry-weather, low
flow conditions, due to the shock loads produced by intense,
short-duration thunderstorm activity (Medina, 1979; McConnell,
1980; Sullivan, 1977).
6. Benefit/lost Analysis. Facilities designed for low frequency
events will likely prove to be uneconomical when construction and
land acquisition costs are compared with the benefits resulting
from reduced loadings. Facilities designed for higher frequency
events may actually provide a more economical reduction in total
annual loading. The design storm concept does not provide
sufficient probability information to make this assessment.
7. Effect of the Design Itself. The provision of storage (or other
works) changes the response time of the catchment. Thus a
different, longer design storm duration is required for the as-
built storage system, than the original duration specified. The
opposite is true for drainage improvements. This results in an
indeterminate flood frequency for the altered drainage system,
even if it were correct to transpose the design storm frequency.
14
-------
It is essential therefore to account for seasonal hydrologic and
meteorologic variability in designing or evaluating a water quantity
control or water quality management scheme (Sullivan, Heaney et al. 1977;
Shubinski, 1980; Donigan, 1980; Shapiro et al. 1980; Leclerc, 1979; Walesh
and Syder, 1979; Sullivan, Heaney et al. 1978; Litwin et al. 1981; Walesh,
1979; Barlamont and Van Langenhove, 1981). These requirements can be met
by using a properly calibrated continuous model which processes a
continuous precipitation record into continuous hydrologic and pollutant
loading time series'. Other statistical methods also require a long-term
record (Kummler et al. 1981; Shapiro et al. 1980; LeClerc, 1979; Heaney et
al. 1978; Medina and Buzun, 1981; DiToro and Small, 1979).
Continuous modelling requires as input long records of
hydrometeorological data such as rainfall, evaporation, temperature and
streamflow. In Canada, the data are available from government agencies
such as the Atmospheric Environment Service and Water Survey of Canada or
from private groups such as our Computational Hydraulics Group (CHG). The
information is typically distributed on 9-track magnetic tapes since the
amount of data required can be voluminous. For example, an eight year
record of hourly rainfall intensity at the Mount Hope Airport station in
Hamilton consists of approximately 9800 records.
Depending on the situation being analysed, long-term continuous
modelling can generate even larger quantities of output time series. This
volume of data requires special statistical analyses and graphical displays
to be properly understood. As well, archiving of results is often a study
requirement. It should be evident that input and output time series data
management can be difficult.
Continuous urban hydrologic models such as SWMM are constrained to
using only a single station record of rainfall intensity and a constant
computational timestep. This situation has probably resulted from the
difficulties associated with managing multiple long-term
hydrometeorological data sets. Special software is essential for I/O time
series management. This software should be capable of retrieving data from
and archiving data on magnetic storage media, interfacing this data with an
appropriate model package, statistically analysing output time series' and
graphically displaying the model output.
During the course of this study a data base management system, CHGTSM,
was developed by the CHG. The system has been implemented on an IBM-PC
compatible system of microcomputers using a 52-megabyte hard disk and Q-
track tape backup. The single data base is coherent and available to every
member of the modelling group. The data base currently contains all
rainfall, streamflow and pollutant loading data collected by the CHG in the
City of Hamilton for the period 1980 to 1984. CHGTSM is capable of
aggregating and disaggregating rainfall intensity time series', over
variable timesteps for multiple gauging locations. CHGTSM is felt to be
an essential element in continuous modelling using stormwater models such
as SWMM3.
15
-------
INTEGRATED SWMM SOFTWARE
Inexpensive microcomputers permit new approaches to the problems of
pollution and flooding arising when intense storms track across
metropolitan or industrial areas, dumping polluted water into rivers,
harbours, etc.
The necessary techniques include:
(a) Field instrumentation,
(b) Data capture and management,
(c) Atmospheric fallout, pollution build-up and wash-off models,
(d) Storm dynamics modelling,
(e) Large scale continuous modelling of urban runoff,
(f) Algorithms for unsteady flow in complex networks,
(g) Receiving water models,
(h) Statistical post-processing and modelling, and
(i) Real-time control.
There is a critical need for measuring rainfall intensity and
stormwater discharge accurately in urban areas. This requires high spatial
and time resolution, typically at one-minute intervals. To meet this need,
rainfall intensity monitoring equipment was developed by CHG, based on
single-chip microcomputers utilizing reusable cassette tape as a recording
medium. The equipment is currently in use in Hamilton (14 gauges), in
Ontario as part of the acid precipitation network (6 gauges), in the Arctic
(3 gauges), the Ottawa area (6 gauges), Oslo (6 gauges) and Kentucky.
Software (TRANSPLOT) was developed to process the data and to archive it on
tape.
An important interdisciplinary research topic is the relation of
atmospheric pollution to stormwater modelling. Fallout from industry and
vehicular traffic is thought to be a major contributor to surface pollutant
loadings in large metropolitan areas. An attempt was made by our group to
identify the sources of pollutants available for washoff, and the
mechanisms of build-up and washoff. The computer program is known as
ATMDST.
The development of appropriate software to determine the movement of
stormwater pollutants through a city-wide network of combined sewers is a
closely related problem. This software processes information on the dates
of street cleaning activities and incorporates algorithms for processes
such as traffic, interception by the leaf canopy, and so on. The program is
called CH6QUAL.
A computer program package, RAINPAK, has been developed for simulating
storm dynamics. Variations in storm speed and direction were found to
produce storm flows and pollutant concentrations significantly different
from the conventional assumptions of stationary storm distributions. The
model accounts for temporal and spatial variations in storms, such as
ageing, merging, splitting of storm cells, and so on. Analysis of data
16
-------
from rain sensors is performed by the programs STOVEL, THOR3D, THOR40, and
THOR4DPT, in RAINPAK.
A computer model, OVRFL03, was developed for modelling sideweir
diversion structures. This was motivated by the need to accurately model
the first flush loadings from urban areas which will reach the treatment
facility, or be diverted to the outfalls.
A computer model, TOTSED, was developed to predict bed sediment and
suspended sediment load as a function of time and distance along a one-
dimensional quasi-steady-state receiving area near a combined sewer
outfall. The model was calibrated using data obtained through a sampling
program carried out in the Chedoke Creek outfall channel in Coote's
Paradise in Hamilton. TOTSED provides an interface between an urban
drainage network and a receiving water body.
In order to facilitate the use of thse models in a teaching
environment, a series of pre-processing programs has been developed. These
programs, FASTSWSMM3, FASTHEC2 and FASTHYMO, prompt the user for input data
in the appropriate sequence, to which he responds by providing data in
free-format and/or optional commands which direct the job path along
various routes. The pre-processors take care of all system job control
language, design file manipulation, and sequentially execute the programs.
In this way the user can focus on the hydrologic problems without having to
be concerned about the computer system, thereby reducing time expended on
learning and/or carrying out calibration, validation or sensitivity
analyses. The system has been adapted to IBM-PC compatibles and is known
as PCSWMM3. The system is depicted in Figure 1.
The models ATMDST, CHGQUAL, RAINPAK, OVRFLO and TOTSED were designed
to be incorporated as additional "blocks" of SWMM3. Each will be
incorporated in PCSWMM3.
A field program has been an essential part of the research. The
activities were generally as follows:
a. Install and maintain flow gauges at urban streams, overflow
structures and outfalls,
b. Install and maintain rainfall intensity recording networks in
metropolitan areas,
c. Collect stormwater samples at flowgauging locations for selected
storm events at approximately five minute intervals. These samples
were subsequently analyzed for suspended solids, BOD5, nitrogen
and phosphorous, as well as, occasionally, other constituents.
Using observed event rainfall intensity, discharge and pollutant
concentration data, obtained* from the field program, a discrete event model
of the Hamilton urban drainage system was constructed, calibrated and
validated. The RUNOFF Block of SWMM3 was used for this purpose. The basic
time-step for the general model is five minutes. A coarse model of the
17
-------
system (time-Step of one hour) was also developed; all diversion structures
were assumed to be operating such that flow was directed to the receiving
waters.
The coarse model of the drainage system was run continuously for a
period of nine years (May to October inclusive) at a time-step of one hour
using rainfall data from the Environment Canada Archives, in order to
develop long-term loadings of pollutants to the Hamilton receiving waters.
Special routines (DATANAL) were written to interface with these data tapes.
The results of the continuous modelling were subjected to statistical
analyses, in order to develop "easy-to-evaluate" equations for predicting
stormwater pollutant loadings for interfacing with a model of the Hamilton
Harbour.
To lessen the impact of pollutant loadings in the receiving waters,
due to sewer overflows, real-time control by microcomputers located in
diversion structures and in storage tanks should be considered. A detailed
study of the design of a microprocessor circuit for installation in a
specific, existing overflow structure was conducted. The software is called
RTCONTROL. This instrumentation integrates with the rainfall and discharge
monitoring equipment mentioned earlier to control overflows and make
maximum use of in-system storage.
EXISTING HYDROMETEOROLOGIC DATABASES
The need for improved access to data and information has been widely
recognized. As a result, electronic data bases have become an essential
service programs
computational diskettes
Drive A: §
Drive B:
-------
tool to fully disseminate or use available information. There are more than
2,000 data bases in North America, which anyone with a microcomputer or
terminal, modem and the appropriate accounts and passwords, can use. The
main objective of an electronic data base is not so much the provision of
information but access to that information.
The distributor of the data bases is the agency that runs the
mainframe computers in which the data bases are stored. Sophisticated
software makes it possible to find the required information within seconds.
One of the larger distributors, for example, is Dialog Information
Services, a subsidiary of Lockheed Corp. of Burbank, California.
In the U.S., climate related information data bases are managed by the
National Oceanic and Atmospheric Administration (NOAA). The NOAA's
"National Environmental Satellite, Data and Information Service" (NESDIS)
established, the National Environmental Data Referral Service (NEHRES) in
1981.
The NEDRES, of which Canada is one of the many international data
contributors, is the nerve centre of a comprehensive climate data archive.
The centre assists in determining the existence, characteristics, content,
location and conditions of data files. It does not supply the data, hut
refers the user to the data holder. Detailed inquiries about NEH.RES can
be directed to the following address:
National Environmental Data Referral Service Office
National Oceanic and Atmospheric Administration
3300 Whitehaven ST. N. W.
Washington, D.C.
(202) 634-7722 (FTS 634-7722)
With reference to climatic data for Canada, most of the information
can be obtained directly from the various divisions of The Atmospheric
Environmental Service (AES) in Toronto. The provision of data involves
mainly the regional climate units and the Canadian Climate Centre (CCC) at
AES. Besides the provision of hard copies and magnetic tapes, CCC offers
direct access to the central digital archive. Qualified users can obtain
the necessary data and undertake a variety of data management functions and
manipulations.
In Canada and Ontario some other data providers are:
Water Survey of Canada (WSC)
Canadian Hydrographic Service (CHS)
Ontario Ministry of the Environment (MOE)
Ontario Ministry of Natural Resources (MNR)
Ontario Hydro
The type of data collected by the above agencies is determined mainly
by their given mandates or specific needs. The more frequent data include:
- hydrometric data (water levels and flows)
- meteorological data
19
-------
- snow surveys
- water temperatures
- ice cover
In addition, information on sediment and water quality is available for a
limited number of stations.
In the Canadian National Climatological Archive, managed by CCC, the
data are retained in the following three forms:
PAPER: The storage of data on paper was for many years the only way of
preserving data and covers a period of operation as far back as the mid-
nineteenth century. The data are published in serial publications which,
for completeness, are summarized below:
. a) Climate Perspectives, contain a weekly summary of national weather
events,
b) Monthly Meteorological Summaries, contain hourly and daily
meteorological data including monthly means and extremes,
c) Canadian Weather Review, permits a preliminary review of monthly
weather based on unverified data,
d) Monthly. Record of Meteorological Observations, a comprehensive
publication of verified meteorological data,
e) Monthly Radiation Sunmary, contains hourly radiation data,
f) Annual Meteorological Summaries, contain a short historical
account of the stations and the meteorological data,
g) Supplementary Precipitation Data, contains listings for long-
duration recording precipitation gauge stations, daily and maximum
amounts from precipitation gauges, and a summary of storage gauge
records, and
h) Snow Cover Data, contains data from over 1800 snow courses
operated across Canada.
In addition to the publications mentioned above, various studies as well as
a comprehensive package of climatic data normals are available.
WSC supplies various publications:
a) Surface Water Data Reference Index, contains descriptive
information for all gauging stations,
b) Surface Water Data, contains daily discharge and daily water level
data for rivers and lakes,
c) Historical Streamflow Summary, contains a summary of monthly and
annual mean discharges, and annual extremes and total discharges,
d) Historical Water Levels Summary, contains a summary of monthly and
annual mean water levels and annual extremes,
e) Sediment Data Reference Index, contains descriptive information
for sediment stations,
f) Sediment Data for Canadian Rivers, contains sediment data,
including daily suspended sediment concentration and suspended
load, and particle size distribution,
20
-------
g) Historical Sediment Data Summary, contains summary of monthly and
annual mean suspended sediment load, annual extremes of suspended
sediment concentration and suspended load, and annual total
suspended load.
MICROGRAPHIC: The storage of meteorological data on microfilm began in the
1940's and today almost all source documents are retained on microfilm.
Besides 12,000 reels of microfilm, the National Climatological Archive
consists of over 10,000 microfiche, containing long-term abstracts of
meteorological data up to 1980, organized chronologically by station.
WSC recently started to provide hydrometric data and sediment data on
microfiche. The hydrometric data include daily streamflow and water level
data prior to 1979. The sediment data cover periods prior to 1978.
DIGITAL: The storage of data in machine-processible form is relatively new
and began in 1950 when 80-column punch cards were introduced for data
storage. In the mid-sixties, the storage of data was transferred to
magnetic tape.
Today, the digital archive contains quality controlled data from all
stations. In addition to the station observations of meteorological data,
digital precipitation observations at 15 minute intervals collected by six
digital precipitation radars (SCEPTRE network) are available (magnetic
tape, 1600 bpi).
The use of the digital archive is facilitated by numerous computer
programs developed by CCC. The standard archival formats permit extraction
of specific parameters from the archival file.
Hydrometric and sediment data are available from WSC in computer-
compatible form. Usually, the data are supplied on magnetic tape, written
in EBCDIC (odd parity) on 9-track at a density of 1600 bpi.
CHG-TIME SERIES STORAGE SYSTEM
When several users require operational data which is being updated,
and which is so large that individual copies would usurp a- significant
share of their computer resources, centralized control of the data bases is
advisable. This avoids redundancy and inconsistency, and encourages
standards, security, integrity and co-operation.
A commercial relational DBMS, designed to organize information into a
collection of different attributes with a common relationship, would not be
efficient for computational hydrology/hydraulics. The basic element of
information in a relational DBMS used to maintain water resources data
consists of a data item such as a water level, flow or sediment
concentration for a particular station at an instant. A better alternative
is to use a block of sequential time series (TS) data as the basic element
of information. The basic concept underlying a TS management system (TSM)
21
-------
is the organization of data into records of continuous, applications-
related elements, as opposed to individually addressable data items.
Our custom made, time series based, pseudo-relational TSM uses a two
layer hierarchial local area network (LAN) of work stations. The time
series store (TSS) resides in the main hard disk of the LAN and has a
tape back-up. The capacity of the hard disk is equal to the capacity of a
"standard" 9-track tape plus sufficient space to store important files.
The top level of the hierarchial LAN is a work station which acts as a
'Boss1. The Boss computer holds the TSS in the hard disk and magnetic tape
drive. The tape drive is main-frame compatible (9-track reels). The Boss
also holds modem connections to external databases. The lower level of
the hierarchy comprises several work-stations which hold the hydrologic
application packages. The overall data flux between the hard disk and lower
level nodes is through the Boss computer, which also holds Distributed
Processing Software (DPS), which control the overall process. A user at a
node can request only high-level operations, such as opening a file, or
writing into a file which will be specified by the Boss computer. Users can
never destroy the integrity of the network file system.
Figure 2 is an overview. The nodes and the boss are all IBM-PC
compatibles, each having the same basic resources, except the 9-track tape
drive, hard disk, modem and printer at the Boss CPU.
CHGTSM is written in ANSI-FORTRAN '77.
A data acquisition system used to load field data into a time series
store is designed to automatically collect large amounts of data from one
or more sources and store or transmit the data for future or immediate
processing. When continuous hydrometeorological data is stored at a fine
time resolution, large quantities accumulate very rapidly and it is
therefore essential that the data collection be computer compatible.
Further
Workstations
-------
Local microcomputer-based data acquisition loggers (DAL), designed to
gather data from sensors, e.g. rain gauges, that partially process the
data, and store the information on inexpensive media such as magnetic
cassette tapes, are used by CHG. The DAL contains one or more acquisition
channels, and it is possible to expand the system to include other
measurements in the future such as temperature, water conductivity, and pH.
Some of the advantages resulting from the use of a local microcomputer
DAL include variable sampling rates, data processing, continuous or
intermittent operation, and programmability to cover other data acquisition
needs. A microcomputer-based data decoder (DD) is used to automatically
retrieve the information stored on tape, verify and transmit it to the base
computer. The advantages of this process over manual methods used
previously include the prevention of random errors and reduced time.
The uses for which data is intended determine the type of data
processing to be performed. For example, the rainfall time series may be
processed to compute storm dynamics, produce intensity-duration-frequency
curves, tables or isohyetal maps. The data processing should facilitate
insertion of the time series into input files to be used by computer
program packages, such as PCSWMM3.
Our data acquisition utilities (CHGOAS) - to receive, interpret,
store, process, and present the information-currently functions on APPLEIIe
and PDP11/23 systems. Two types of processing are carried out: the first
is the derivation of a hyetograph for each of.the storms and for each
monitoring site; the second is input data preparation for computer programs
for storm and streamflow modelling. These programs are used in turn to
determine peak, average, daily, monthly, and annual amounts of runoff and
pollutant. Models are also used to investigate a wide range of design
alternatives and strategies for minimizing floods and pollution.
Individual raingauge records are processed to produce hyetographs,
plotted at various integration time intervals. Long integration time
intervals can be very misleading regarding instantaneous rainfall
intensity, if the timestep covers widely varying instantaneous intensities.
This kind of data tends to underestimate the short period, high intensity
rainfall.
Even a cursory inspection of rainfall distributions illustrates the
wide variety of hyetograph shapes. There is no typical rain sequence even
though the general sequence might consist of a rapidly increasing rate of
rainfall with a maximum intensity reached in the first IB minutes of a
thunderstormon, followed by a period in which intensity decreases to zero
or becomes inappreciable. There are a large number of storm
characteristics important to local hydrology. Accurate information
concerning these characteristics in time and space is indispensable for the
cost-effective design of hydraulic structures. Thus, even though previous
practice indicates a need for rain data at 1 hour time steps, it is
desirable to collect data at (say) 2 minute intervals. Such data are more
useful for urban hydrology, such as deriving regional design storms for
traditionalists.
23
-------
HARDWARE FOR COMPUTATIONAL HYDROLOGY
As indicated above areas of fundamental importance to urban hydrology
include modelling; microprocessor-based instrumentation; field data acqui-
sition, data management and presentation; data archiving; data base
accession; application program systems, incorporating many water discip-
lines; program maintenance and support; documentation and report genera-
tion; text processing; and real-time control of water systems. Application
programs include topics in meteorology; hydrology; hydraulics; municipal
and environmental engineering, for example, water distribution, pump
stations, drainage networks; water and wastewater treatment plants; water
quality control; river systems; and limnology.
As illustrated by the CHG software described earlier, activities
involved in a typical study include:
1) set up microcomputer-based field instrumentation, typically
sampling on a 300 s integration period;
2) acquire a data base of rainfall, runoff, and water quality
parameters for the system to be modelled;
3) collect environmental data as required by the models;
4) develop new models and apply continuous models;
5) verify, perform sensitivity analysis, calibrate and validate the
continuous models;
6) insert the continuous output time series from the models into the
data base system;
7) conduct a statistical analysis of the continuous output time
series;
8) develop transfer function or ARIMA models of the continuous input
and the output time series;
9) develop control programs for diversion and control structures in
the drainage systems being modelled;
10) elaborate the original continuous models to incorporate the new
control programs for the control structures;
11) run the continuous models with the control structures
appropriately modelled for the full time series record;
12) compare the output for various control structure programs and
choose the 'best' control program;
13) summarize the final output;
24
-------
14) prepare documentation;
15) install the 'best' real-time control system in the field; and
16) set up a continuing program of data acquisition, modelling, and
software maintenance to ensure model and prototype performance.
Clearly the data base management system (DBMS) is of central concern.
The ready availability of sophisticated data base management systems,
distributed multiprocessor disk operating systems and integrated graphics
packages offer a variety of tools that should be considered. Although
difficult to estimate, the cost performance of powerful personal
microcomputers, with a central hard disk system, and using the new
generation of integrated software, is evidently better than that of
mainframes, at least for memory requirements up to (say) 4 Megabytes and
hard disks up to 400 megabytes. Our new software is aimed at this
environment, currently functioning in about one-tenth this space.
A simple but powerful user language is being developed for two types
of database: one that handles spatial data, e.g. drainage networks,land
use, soil, and vegetation distribution, etc., and another that handles time
series data, i.e. hydrological data. The dialogue for the two systems
should be designed to be mutually compatible to assist cross reference and
intercomparison between the two resource data archives. Input is menu-
driven. Output includes tabular daily, weekly, monthly or annual
summaries, and graphics.
Commands for the following activities will eventually be included in
the dialogue:
1. summarize contents of file
2. summarize contents of batches of data on the file
3. copy raw data onto file
4. add time series data or a rating curve to a file
5. read magnetic tape
6. convert format of ASCII dataset
7. copy an entire batch of data
8. delete an entire batch of data
9. copy any portion of data from one file to another
10. delete any portion of data from a file
11. process event rainfall data
12. transform series data, multiply, add, apply ratings, etc.
13. run user specified model, e.g. SSARR
14. list series data or rating curves
15. plot series data with time
16. print rating table
17. plot rating curves
18. print-plot values and time
19. print monthly/daily/seasonal/annual summary
20. print distribution of values
21. exit from DBMS
25
-------
At this time our network software has limited capability and is still
being developed and implemented.
CONCLUSIONS
The hardware installed by our group is as follows:
Boss:
Estimated
approx. cost $k
Modem:
Printer:
5 Nodes:
Software:
8088-based portable PC-compatible
512kbytes memory
2 floppy disk drives
8087 co-processor
9-track tape drive
interface card
software
52 Mbyte hard disk (43 formatted)
interface card
software
1200 BAUD, dial-up
High resolution, dot matrix
Total
same as boss CPU
MS-DOS 2.1
MS-FORTRAN 3.2
Network (allow)
Approximate cost per station
5x3
included
3.6
0.6
15
0.3
1.1
29.6
5.0
Regarding large disk storage, the primary constraints are imposed by
MS-DOS, which restricts logical disk volumes to 32 Mbytes. Because of this
the data base may need to be segmented.
The PC-compatibles chosen were Corona portables, reputed to contain a
large power supply and a large fan. The speed of computation with the 80R7
26
-------
math co-processor has been found to be as fast as one-fifth that of a
CYBER 172, as shown in Figure 3. Figure 3 was obtained by programming
measurement techniques.
The 9-track drive unit includes an intelligent interface that utilizes
a Z-80 microprocessor and proprietary firmware to provide data transfer
between a Personal Computer, and an IBM mainframe-compatible 9-track tape
drive with an embedded formatter. Supplied by the manufacturer of the
hardware is a software interface package, TIP (Tape Interchange Program).
Utilizing TIP the user may freely transfer data between Personal Computers
and the 9-track tape drive.
The interface occupies a single slot in the PC Bus, and is interfaced
directly to the 9-track tape drive via two 50-pin data cables.
The tape subsystem is both self-loading and self-threading for ease of
use. The subsystem provides high-speed disk-to-tape transfer at rates of
0.7 megabytes per minute, with up to 42 megabytes of data storage per tape.
350 _
300
250 -
200
150
100
50
EXTRAN
flow duration: 20 mkt
ttmestep: 10 s
20 40 60 80
FIGURE 3
100
27
-------
The tape unit supports 1600 BPI tapes, with user-specified recording
formats from 2 bytes to 16 kilobytes per block. The tape drive will accept
all standard 9-track tape drive reel sizes up to 10-1/2 inches in diameter.
Each tape subsystem includes a single density 5-1/4 inch diskette
containing three comprehensive programs and two utilities that allow the
user to read, write, or inspect a 9-track tape:
1. TREAD - A tape-to-disk-copy/conversion program which will read
ANSI compatible data records from the 9-track tape and convert.
them to DOS compatible ASCII disk records, and store them on disk.
This program allows for conversion of data records from ERCniC to
ASCII format, record segmentation of user's choice, and user-
specified file name and disk selection.
2. TWRITE - The counterpart of TREAD, TWRITE is a disk-to-
copy/conversion program. TWRITE copies DOS compatible disk files
to tape in ANSI compatible format. This program allows for ASCII
to EBCDIC format, and a user-specified tape record structure.
3. TDUMP - This program allows the user to dump the contents of a P-
track tape to either the user's console or to the system printer.
This utility is extremely useful to determine the format in which
a tape has been written.
4. TUTIL - This file contains the primary assembly language
subroutines that are used in TIP. These subroutines may easily be
linked into other 'High Level Language1 programs to create user-
customized tape control programs.
5. TLINK - A tape control utility module designed to be linked into
Basic programs to allow the user to create simple, yet powerful,
customized tape control programs.
REFERENCES
Berlamont, Jean and Van Langenhove, Guido. 1981. Diversion Frequency in
Combined Sewer Systems. Proceedings of the Second International Conference
on Urban Storm Drainage, Urbana, Illinois. 295-303.
DiToro, Dominic M. and Small, Mitchell 0. 1979. Stormwater Interception
and Storage. ASCE Journal of the Environmental Division 105(EE1). 43-54.
Donigan, Anthony S., Jr. 1980. State-of-the-Art Report of the Wasteload
Generation Committee. Workshop on Verification of Water Quality Models.
Environmental Research Laboratory, Office of Research and Development
USEPA. 71-77.
Heaney, James P., Nix, Stephan J. and Murphy, Michael P. 1978. Storage-
Treatment Mixes for Stormwater Control. ASCE Journal of the Environmental
Engineering Division 104(EE4). 581-592.
28
-------
Kummler, Ralph H., Frith, John 6., et al. 1981. Uncertainty Analysis in
Stormwater and Water Quality Management Modelling and SWMM Users Group
Meeting September 28-29 1981. McMaster University, Hamilton, Ontario. 1-
54.
LeClerc, G. 1978. Evaluation of Proposed Urban Runoff Control
Alternatives. Proceedings Stormwater Management Model (SWMM) Users Group
Meeting May 24-25, 1979. Office of Air, Land and Water Use, Office of
Research and Development, USEPA. 28-46.
Linsley, R. and Crawford, N. 1974. Continuous Simulation Models in
Hydrology. Geophysical Research Letters 1(1). 59-62.
Litwin, Yorman J., Lager, John A. and Smith, William G. 1981. Project
Summary Areawide Stormwater Pollution Analysis with the Macroscopic
Planning (ABMAC) Model. Municipal Environmental Research Laboratory,
USEPA. 3pp.
Marsalek, J. 1977. Runoff Control on Urbanizing Catchments. Proceedings
of the Amsterdam Symposium on Effects of Urbanization and Industrialization
on the Hydrological Regime and on Water Quality. IAHS-AISH Publication No.
123. 153-161.
McConnell, James B. 1980. Impact of Urban Storm Runoff on Stream Quality
Near Atlanta Georgia. Municipal Environmental Research Laboratory, Office
of Research and Development, USEPA. 52pp.
Medina, Miguel A., Jr., and Buzun, Jennifer. 1981. Continuous Simulation
of Receiving Water Quality Transients. Water Resources Bulletin 17(4).
549-557.
Medina, Miguel A., Jr. 1979. Level III: Receiving Water Quality
Modelling for Urban Stormwater Management. Municipal Environmental
Research Laboratory, Office of Research and Development, USEPA. 204pp.
Shapiro, Howard M., Blenk, John B. and Allen, Mark P. 1980. CSO Impact
Determination by Long Term Simulation. Proceedings Stormwater Management.
Model (SWMM) Users Group Meeting January 10-11 1980. Office of
Environmental Processes and Effects Research, Office of Research and
Development, USEPA. 142-189.
Shubinski, Robert P. 1980. Use of Models as Projection Tools. Workshop
on Verification of Water Quality Models. Environmental Research
Laboratory, Office of Research and Development, USEPA. 62-67.
Sullivan, Richard H., Heaney, James P. et al. 1978. Evaluation of the
Magnitude and Significance of Pollution from Urban Storm Water Runoff in
Ontario. Research Report No. 81. Research Program for the Abatement of
Municipal Pollution Within the Provisions of the Canada-Ontario Agreement
on Great Lakes Water Quality. Environment Canada and the Ontario Ministry
of the Environment. 183pp.
29
-------
Sullivan, Richard H., Heaney, James P., et al. 1977. Nationwide
Evaluation of Combined Sewer Overflows and Urban Stormwater Discharges
Volume I: Executive Summary. Municipal Environmental Research.Laboratory,
Office of Research and Development, USEPA. 95pp.
Walesh, Stuart G. 1979. Summary - Seminar on the Design Storm Concept.
Proceedings Stormwater Management Model (SWMM) Users Group Meeting May 24-
25, 1979. Office of Air, Land and Water Use, Office of Research and
Development, USEPA. 47-54.
Walesh, Stuart G. and Snyder, Daniel F. 1979. Reducing the Cost of
Continuous Hydrologic-Hydraulic Simulation. Water Resources Bulletin
15(3). 644-659.
Weatherbe, D.G., Marsalek, 0. and Zukovs, G. 1981. Research and Practice
in Urban Runoff Control in Canada. ASCE reprint. 15pp.
30
-------
MULTIOBJECTIVE DESIGN OF STORMWATER IMPOUNDMENTS
by: Edward A. McBean
University of Waterloo
Waterloo, Ontario, N2L 3G1, Canada
ABSTRACT
Considerable variance in design procedures regarding water quality for
Stormwater management ponds for control of urban runoff quantities is reviewed,
as characterized by several surveys. Also, a simple variable-detention time
simulation model is developed to evaluate the water quality behavior of wet
ponds. A design procedure is demonstrated through use of multiple objective
analysis.
INTRODUCTION
Stormwater management ponds for control of urban runoff quantities have
been widely adopted. Nevertheless, there is still considerable variance in
design aspects of these ponds and, to an increasing extent, the multiple use
character of the ponds is being promoted. Potential multiple uses of the
ponds include design for post-development runoff control, for water quality
improvement, water surface recreation, for groundwater recharge and for soil
erosion control. Conflicts between the design aspects of the ponds often
occur, however, since the design requirements for one use are not the same as
the design requirements for another use.
In response to the resulting legal and economic concerns arising in part
due to the multiple use aspects of the ponds, a number of controversial issues
have arisen which involve tradeoffs. Governmental guidelines for Stormwater
runoff are usually quite subjective in nature, and most designers must
therefore interpret and quantify these guidelines, using their own judgment, to
establish the best design tradeoffs.
As a means of better characterizing current practices, several surveys
of practitioners were carried out. The intent of this paper is to discuss
some of the findings of these surveys. In addition, a simple variable-
detention time simulation model is used to consider the water quality behavior
of wet ponds, as a means of demonstrating the use of multiple objective
31
-------
analyses to the problem of design of stormwater ponds. The multiple objective
procedure has considerable merit for assisting the decision-making process in
demonstrating the nature of the tradeoffs being made.
CURRENT DESIGN PRACTICES OF STORMWATER PONDS
Although the use of stormwater ponds has gained increasing popularity for
regulating the quantity and quality of urban stormwater runoff, the number of
fully-documented case studies available in the technical literature is
limited. As a result, surveys of consultants were undertaken to gain some
insight into the current state of the art.
Two questionnaires were distributed, the first to various consulting
engineering companies involved in the design of urban drainage works and the
second survey to all Canadian cities with a population in excess of 40,000.
The responses summarized below represented 32 individual returns associated
with one of the surveys and 84 returns associated with the second survey.
Collectively, they represent a substantive characterization of current design
practices throughout -Canada.
The following indicates summary findings:
PEAK FLOW REDUCTION
While all three factors, the increase in the volume of runoff and peak
flow runoff rate, and the decrease in lag time, have some influence on
downstream flooding it is the magnification of the flood peak that is the
single most influential factor in design practices.
The extent by which the peak flow rates are reduced, varies among
designers. Fo.r the majority of the responses (55%), the capacity of the
downstream channel was known (or determined) and the outlets were designed to
reduce peak flows to this maximum level. Alternately, when the capacity of
the downstream channel was not known, principal outlets were usually designed
to reduce all peak flow rates to their pre-development level (27% of
responses).
WATER QUALITY CONSIDERATIONS
Stormwater is about equivalent in quality to, or better than, effluent
from secondary sewage treatment plants, with the exception of suspended solids
concentrations which far exceed the treatment plant effluent levels. It is
also recognized that many of the pollutants present in the runoff are in the
form of suspended solids; for example, the United States Environmental
Protection Agency assumes that 10% of the suspended solids load is BOD (Heaney
et al, 1975). It is probably for these reasons that most water quality
improvement projects have concentrated solely on the removal of suspended
solids .as a water quality objective. From the surveys it was found that
suspended solids was the only water quality parameter considered in 35% of the
cases. (Note that only 41% of the cases used any form of water quality
design criteria).
32
-------
The criterion used to achieve reduced suspended solids levels varies from
place to place. In Maryland, sedimentation ponds are required to have a 70%
trapping efficiency (removal rate), while a Metro-Toronto Bylaw (#2520)
prevents the discharge of stormwater in.to a natural watercourse when the
concentration of suspended solids is in excess of 30 mg/£. These regulations
are however exceptional, and most cities have no definite limits on the water
quality of urban runoff. Even the Ontario Ministry of the Environment (1978)
has stated that remedial measures for the control of pollution due to urban
non-point sources are required only if "... they are shown to cause or
contribute significantly to violations of the Provincial Water Quality
Objectives". As such, the choice of criteria in regard to water quality used
for the design of retention facilities is frequently left up to the designer.
A pond in Meadowvale, Ontario, was designed to settle out all particles larger
than 0.07 mm, while stormwater management criteria for the Professors' Lake
Project in Brampton, Ontario, called for the removal of particles greater than
40 microns (David, 1978).
The water quality monitoring schedules used by seven municipalities for
in-situ ponds are included in Table 1.
DESIGN AND ANALYSIS OF OUTLET STRUCTUPJES
Maintaining peak flows at their pre-development level is normally the
limiting design requirement, assuming the capacity of the downstream channel
is not a more stringent concern. However, the survey results indicated that
most of the structures are being designed based on a constant (or maximum)
release rate, usually corresponding to the peak pre-development runoff for a
particular storm. Such design methodlogies are used by Theil (1977), Grigg
(1977), Oscanyan (1975) and Rao (1975). As well, the survey found that 39%
of the respondents followed a similar procedure using a peak pre-development
flow rate corresponding to a specific storm, most frequently the 25 year
pre-development storm.
An outlet designed to meet a specific pre-development peak flow rate will
provide little or no peak flow attenuation for events of greater frequency
than the design event and will provide greater attenuation for more infrequent
events (Lafleur and McBean, 1981). In addition, since urbanization also
causes an increase in the volume of runoff, then a specific post-development
flow will have a higher return frequency than the pre-development flow of the
same magnitude. Such an increase in the frequency of flows combined with
higher velocities (resulting from increased peak flows) will cause an increase
in the erosion downstream.
There are numerous types of outlets currently being employed, the most
popular ones being weirs, orifices and closed conduits (or pipes). In the
surveys it was found that 84% of the respondents" used one or more of these
methods. The remaining respondents used some method of manually-controlled
discharge, such as gates, valves or pumps, although current practice is
oriented strongly toward making such ponds self-operating.
33
-------
MULTIPLE USE DESIGN CONSIDERATIONS
The multiple use aspects of the ponds may be divided into primary and
secondary functions. In this respect, flood control (13%), water quality
control (2%), stormwater management (2%) and cost reduction (1%) were each
cited as sole primary reasons for construction of stormwater impoundments.
All other cities (81%) which had stormwater impoundments listed at least one
TABLE 1. WATER QUALITY MONITORING SCHEDULES FOR SEVEN MUNICIPALITIES
Water
Quality
Parameter
B.O.D.5
C.O.D.
Nepean Missi- Winni-
ssauga peg
Weekly
5x/yr
min
City
Regina Saska-
* toon
2x/yr*
Calgary
2x/mo
2x/mo
Edmonton
Ix/mo
3x/yr
Nutrients
Heavy
Metals
Fecal
Co li form
Count
Total
Coliform
Count
Suspended
Solids
Turbidity
Weekly
Weekly
Weekly
Weekly
Weekly
Weekly
4x/yr
min
4x/yr
4x/yr
min
4x/yr
min
4x/yr
min
5x/yr
min
5x/yr
min
lOx/yr
lOx/yr
5x/yr
5x/yr
1 yr. 2x/mo
after
const.
2x/yr* 2x/mo
2x/yr* 1 yr.
after
const .
1 yr.
after
const.
2x/yr*
- 1 yr. 2x/mo
after
const'.
3x/yr
3x/yr
Ix/mo
Ix/mo
Remarks pertinent to water quality monitoring schedules.
The above parameters were listed as question response choices while the
parameters mentioned below were volunteered under the response choice heading
of other parameters tested.
Nepean
Winnipeg -
Regina -
Saskatoon -
Calgary
Monitoring occurs more frequently if storm occurs.
Transparency is monitored 10 times a year (lOx/yr).
* Proposed monitoring schedule to be undertaken in the future.
Turnover or replacement rate will also be noted twice a year.
The information in the above table for Saskatoon is the actual
schedule of Lakeview pond.
D.O. is also tested a year after construction.
Many other parameters are tested.
34
-------
primary and one secondary use. Table 2 illustrates which secondary functions
occur in combination with different primary uses. The figures shown in the
table have been collated for each primary use regardless of whether it occurs
in conjunction with other primary uses. For example, Gloucester, Ontario,
listed flood control and water quality control as the two primary reasons for
creating its stormwater pond and aesthetic and open green space as secondary
uses for it. In Table 2, Gloucester's secondary use responses of aesthetic
and open green space have been recorded twice, once in the flood control row,
and once in the water quality control row.
Either on its own, or in conjunction with another primary use, flood
control (26%) followed by water quality control (10%) was most frequently
cited as a prime reason for impoundment construction. Aesthetic (17%), open
green space (15%) and recreation (12%) were the most prevalent secondary
uses. In Waterloo the pond was not considered aesthetically pleasing.
TABLE 2. MULTIPLE USE OF STORMWATER PONDS
Secondary Use
Primary
Use
Real
Aesthetic Estate Wildlife
Open
Green
Space
Recreation
Flood
Control
Water
Quality
Control
10
4
4
2
13
6
11
2
Recreation 2
Soil Erosion
Control
Stormwater 2
Management
Discharge 2
Regulation
Groundwater
Recharge
Cost 2
Savings
1
2 2
1 2
1 1
1 1
2 2
3
1
3
3
Remarks
Markham, Mississauga and Winnipeg listed recreation as both a primary and
secondary use of stormwater ponds.
35
-------
EROSION CONTROL
Erosion is most severe immediately downstream of urbanized areas and the
degree of erosion diminishes with distance from the urban area. A design
parameter involves requirement of wide buffer strips to assist in preservation
of natural channels. The wide buffer limits the detrimental impact of man's
activities on the banks, such as fill encroachment, garbage dumping, and tree
removal. Buffer strips with landscaped lawns and gardens are not equivalent
to the natural wooded areas since they do not slow runoff flow as much as
natural leaf mulch and do not inhibit access to the channels as would a
natural forest bush area.
WATER QUALITY SIMULATION
The design of retention pond facilities for water quality control is not
straightforward. The quality of pond effluent is a function of both the
influent water quality and the treatment administered to the flow, which is a
direct function of the detention time of the water. Both of these parameters
vary within the duration of a particular storm event, and it is therefore
necessary to identify the naturally occurring combination of these parameters
which govern the effluent water quality.
To allow comparisons of the pollutant removal efficiences of alternative
pond designs, Lafleur et al (1981) developed a model to simulate retention
pond operation. The model is premised on Camp's (1945) clarification theory.
The model employs the concept of "plug flow" which assumes delivery of
the flow on a first-in-first-out basis and allows no mixing between plugs.
The detention time (t*) of each plug is required for solids removal
calculations and so it is imperative that the identity of each plug remains
separate and identifiable. The procedure used in the present program to
obtain this can be explained using Figure 1. Consider the nth time increment
of the inflow hydrograph corresponding to t.^. The cumulative inflow at this
time can be found by summing the area under the inflow graph. This can be
determined using a finite difference approach:
n-1
Zcum n = .^ XiAt + 2 V* at t - t± (1)
From the assumption of plug flow, it follows that the volume of water
contained within any time increment must remain in the pond until the water
corresponding to the previous inflows has been discharged. This implies that
as the plug (of time increment n) begins to discharge at some time (t = O
then : 2
°cum = ^cum n at t = fc2
The detention time can then be easily determined since,
36
-------
t* =
t
C
(3)
If the influent suspended solids concentrations, C. ., is known for a
particular time increment, then the effluent concentration can be found by,
out
Jin
where X _ is the average removal rate for time increment i and is a function
c 1 1
of:
Xm , = F(H
T i
t*. v a)
avg i it
(5)
where v represents the terminal velocity of the particles. H is the
avg
average depth over the detention time, and a is a correction factor to
account for non-idealized conditions, such as short-circuiting, resuspension
of sediments or the non-spherical nature of the particles. It is noteworthy
that this simulation model reflects the unsteady nature of the pond's
operation.
Since it was found that the variation of particle size distribution with
time was small, this parameter was assumed constant throughout the storm. It
should also be noted that due to a lack of data, no attempt was made to
evaluate a and for current purposes it has been assumed to equal unity.
30-
-20
>
o
10
FLO* HfDMWAM
o
20
8 K,
3
§
OUTFLOW HVMMMPH
^CUMULATIVE INFLOW
- '
CUMULATIVE OUTFLOW
DETOmON TIME «*
I I |
0 24 10 12 14 M IB
TIME (krt)
Figure 1. Method of establishing detention time.
37
-------
PREDICTION OF PARTICLE REMOVALS
The settling behavior of suspended discrete non-flocculating in a
laminar flow condition can be expressed by (after Camp, 1945) :
Vt =
where v is the terminal velocity, d = particle diameter, g = acceleration due
to gravity, p = particle density, p. = fluid density and y = fluid viscosity.
s **
To utilize equation (6) in predicting the quantities of discrete
particles removed in a pond it is necessary to assume that the particles are
distributed randomly throughout the cross-section of the pond and that plug
flow conditions exist.
For a particular time interval the average depth of water in the pond
(H ) and the average outflow from the pond (0 ) is known. It is then
avg ° avg
possible to calculate the critical settling velocity (v ) for the interval.
The critical settling velocity may be described as the velocity of the
smallest particle which will be completely removed during the time increment.
This velocity may also be defined as:
H
avg f-,^.
v = .° (7)
c t*
where t* is the detention time of the time interval and is equal to:
t* - J- <8>
avg
where S is the storage volume in the time increment.
Those particles possessing a terminal velocity greater than the critical
settling velocity will be completely removed from the pond. If, however, the
terminal velocity of a specific group of particles is less than the critical
settling velocity, then only a fraction of the particles in the group will be
removed. Mathematically this may be expressed as:
v = _. x (9)
"R vc j
where X^ is the percentage of particles of the group removed and X. is the
th
total number of particles in the j -group (expressed as a percentage of the
total). If the first n groups have a terminal velocity greater than the
critical velocity, the fraction of particles removed during the time increment
can be given by:
38
-------
X
T i
11
= i
V
oo
E
j=n+l
(10)
Using this equation in conjunction with the calculated concentration
levels of the influent, as described in the previous section, the effluent
concentration levels may be easily obtained.
To design the pond, it is necessary to decide which areas of the storm
govern water quality. This decision is not obvious since maximum pollutant
concentrations do not always coincide with the shortest detention time. For
the case study, it was initially assumed that the concentration corresponding
to peak flow conditions would be the controlling element of effluent water
quality. Even if such an assumption was found to be invalid through
simulations of the designs operation, such simulations would enable the
quality controlling portions of the design storm to be identified. The
detention time of this incremental volume of water corresponding to the peak
of the inflow hydrograph is then estimated in the procedure outlined above.
Using SWMM (Huber et al, 1975), inflow hydrographs and pollutant loads
were synthesized for storms corresponding to 5, 10 and 25 year return periods.
Figure 2 summarizes the results.
Figure 2. Relationship between flow, concentration and mass flux
39
-------
MULTIOBJECTIVE ANALYSIS
Most water resource projects try to achieve maximum net benefits. This
involves estimating all the costs and benefits for all feasible alternatives.
Project selection is then determined by identifying the project which
provides the best return on the capital invested. The purpose is to identify
optimal projects as a function of numerous relevant objectives. However, in
many problems, many components of the costs are relatively easy to quantify
in monetary terms. These include items such as construction costs, land
costs and maintenance costs.
On the other hand, the benefits of increased water quality control are
not only more difficult to identify, but are almost impossible to evaluate in
monetary terms. These benefits include such aspects as increased property
values downstream, higher values for aesthetic appreciation and an increase
in biological life forms. In severe cases, pollutant removal may also
decrease the cost ox water treatment facilities downstream.
Although it is difficult to put a dollar value on these benefits, they
do contribute significantly to the overall economic impact of the project and
should not be ignored. The procedure used for decision-making should
incorporate these values. One methodology useful for this purpose is
multiple objective planning.
Multiple objective planning techniques were developed using economic
production theory concepts (e.g. Major et al, 1970). Its purpose is to
identify optimal projects as a function of numerous relevant objectives.
Consider, for example, the usage of stormwater management ponds where water
quality improvement, as measured by suspended solids removal, is at issue.
There are two main objectives:
(i) to minimize the cost of the structure while meeting the post-development
flow requirements,
(ii) to maximize the removal of suspended solids.
A particular design of the pond will result in responses with respect to both
these objectives. A number of different design configurations exist
consisting of different sizes and shapes. One particular design may best meet
the objective 1 but it will not be best in meeting objective 2. To determine
the nature ofthe tradeoff between the objectives, the first step of the
multiobjective analysis involves the construction of a transformation curve.
This involves evaluating a number of possible alternatives and plotting the
points as a function of the objectives. The boundary of this technologically
feasible set is the net benefit transformation curve.
The transformation curve developed for a typical case can be seen in
Figure 3(a). The cost (on the vertical axis) consists of the construction
and land costs in excess of those required to meet peak flow constraints. The
suspended solids concentration (on the horizontal axis) is the highest
concentration (in the weir outflow) of the events considered. Figure 3(b)
shows the range of the transformation curve contrplled by each storm.
40
-------
Or
25
50
er
u
a.
8 75
u
100
~ 2
m
O
§
TRANSFORMATION
SURFACE
Rl I I I I I I I I
110 100 90 80 70 60 50 40 30
SUSPENDED SOLIDS CONCENTRATION (mg/l)
Figure 3(a). Transformation curve between additional costs and
suspended solids concentration.
O
M
6
o
S 7
8
I
25 YEAR STORM
COMBINED CURVE
I
110 100 90 80 70 60
SUSPENDED SOLIDS CONCENTRATION (mg/l)
50
Figure 3(b). Elements creating combined transformation curve.
41
-------
The next step in the use of multiple objective planning involves the
development of preference functions. These functions represent the attitudes
of the various decison-making bodies such as the developer, governmental
authority, and the public. For a particular decision-maker, the decision-
maker is equally content with any two points on the curve (see Figure 4),
The point of tangency between the transformation curve and the preference
function indicates the trade-offs willing to be made by the decision-maker.
Thus, together the transformation curve and the preference function indicates
which design is desired by that group.
The construction of the transformation curves generally requires the
interviewing of the various groups to obtain an indication of their perception
of appropriate trade-offs between the objectives. The various curves
illustrate the different attitudes of the various groups. The developer, for
25
50
oc
III
Q.
75
100
~ 2
ID
O
K 3
W-
«
fc
o
O 7
O '
8
DEVELOPER PREFERENCE FUNCTION
GOVERNMENTAL AGENCY PREFERENCE
FUNCTION
TRANSFORMATION
SURFACE
PUBLIC PREFERENCE
FUNCTION
I I I
I
I I
1
110 100 90 80 70 60 50 40 30
SUSPENDED SOLIDS CONCENTRATION (mg/l)
Figure 4. Illustration of multiobjective tradeoffs
42
-------
example, is not willing to sacrifice much money for additional suspended
solids removal and at some point he is unwilling to sacrifice any more money
regardless of the level of suspended solids removal. This attitude is
characterized by the horizontal nature of the preference function for low
suspended solids concentrations and may be attributed to a minimum profit
level. At some point, the additional money spent on water quality improvement
will reduce the return on the developer's investment to such an extent that
the overall development will represent a bad investment..
The public preference function usually places a much higher value on
improved water quality. Usually such a curve has a maximum limiting value
above which residents are unwilling to allow further degradation of the water
body. The value could correspond to the concentration level at which severe
ecological damage (e.g. fish kills) occurs. The other limiting case (that
occurs when the public is unwilling to pay additional costs for water quality
improvement) may correspond to a level when the pollutant or its effects are
no longer visibly noticeable.
The preference function of the governmental authority may usually be
found somewhere between, that of the public and that of the developer. If
governmental guidelines concerning the quality of urban runoff exist, then
this curve will also possess an upper bound for the quality of urban runoff.
The presence of an upper bound on the money spent to achieve improved water
quality may or may not exist depending on current governmental policies.
CONCLUSIONS
Stormwater management ponds have received widespread implementation in
Canada. With increasing frequency, however, the criteria used in the ponds
are going beyond flood-related concerns to encompass many other uses.
Multiple objective planning techniques are useful to identify optimal
(and near optimal) projects as a function of numerous relevant objectives.
These procedures have utility to improve the understanding of how different
the various decision-makers are in their perception of the best design
configuration.
REFERENCES
1. Camp, T.R. Sedimentation and the design of settling tanks. Proceedings:
American Society of Civil Engineers, Vol. 71, No. 4, April, 1945,
pp. 445-486.
2. David, I. Professors' Lakemethods of stormwater management.
Proceedings of Stormwater Management Seminar '78. Kleinfeldt Engineering
Ltd., Edmonton, Alberta, June, 1978.
3. Grigg, N.S. Evaluation and implementation of urban drainage and flood
control. Proceedings International Symposium on Urban Hydrology,
43
-------
Hydraulics and Sediment Control. University of Kentucky, UKY BU 114,
July 1977, pp. 1-6.
A. Heaney, J.P., Huber, W.C., Sheikh, H., Medina, M.A., Doyle, J.R., Peltz,
W.A. and Darling, J.E. Urban stormwater management modeling and decision-
making. EPA-670/2-75-022, U.S. Environmental Protection Agency,
Cincinnati, Ohio, 1975.
5. Huber, W.C., Heaney, J.P., Medina, M.A., Peltz, W.A., Sheikh, H. and
Smith, C. Storm Water Management Model Users Manual Version II. EPA-
670/2-75-017, U.S. E.P.A., Cincinnati, Ohio, March, 1975.
6. Lafleur, D.W., McBean, E.A., and Al-Nassri, S.A. Design and analysis of
stormwater retention ponds based on water quality objectives. Presented
at International Conference on Numerical Modelling of River, Channel and
Overland Flow for Water Resources and Environmental Applications,
Bratislava, Czechoslovakia, May 4-8, 1981.
7. Lafleur, D., and McBean, E.A. Multi-stage outlet design of stormwater
retention facilities. Canadian Water Resources Journal, Vol. 6, No. 1,
1981.
8. Major, D., Bravo, C., Cohon, J,, Grayman, W., Harley, B., Lai, D.,
McBean, E., and O'Brien, T. Multiple objective redesign of the Big
Walnut Project. WRC-69-3, Water Resources Council M.I.T. Cooperative
Agreement, Cambridge, Massachusetts, April, 1970.
9. Ontario Ministry of the Environment. Water managementgoals, policies,
objectives and implementation procedures of the Ministry of the
Environment, Water Resources Branch, Toronto, Ontario, 1978.
10. Oscanyan, P.C. Design of sediment basins for construction sites.
Proceedings National Symposium on Urban Hydrology and Sediment Control.
University of Kentucky, UKY BU 109, June, 1975, pp. 101-117.
11. Rao, V.V. Methods for sizing storm water detention basinsa designers
evaluation. Proceedings National Symposium on Urban Hydrology and
Sediment Control. University of Kentucky, UKY BY 109, June, 1975,
PP. 91-100.
12. Theil, P.E. Urban drainage design for new development. Proceedings
Modern Concepts in Urban Drainage. Canada-Ontario Agreement on Great
Lakes Water Quality, No. 5, March 1977, pp. 261-299.
44
-------
DETERMINATION OF RUNOFF CHARACTERISTICS
OF FLATWOODS WATERSHEDS1
o
K. L. Campbell, J. C. Capece and L. B. Baldwin
Agricultural Engineering Department
Institute of Food and Agricultural Sciences
University of Florida
Gainesville, Florida 32611
ABSTRACT
Several methods of estimating stormwater runoff total volume and peak
discharge are evaluated as to their performance on watersheds of Florida's
Flatwoods Resource Area. Characteristics of these watersheds include extreme-
ly flat relief, sandy soils, dynamic water tables, and scattered wetlands.
Data collected by the U. S. Geological Survey and South Florida Water Manage-
ment District from five small (20-3600 acres), agricultural watersheds (im-
proved and unimproved pasture) served as the basis of evaluation. All total
volume estimation techniques examined rely upon the SCS runoff equation. Best
results were achieved with methods which included antecedent depth to the
water table as a measure of watershed storage potential. Runoff peak rate
estimation techniques ranged in approach from empirical formulas to an over-
land flow simulation model. For the original methods examined, standard
errors of estimate were inversely proportional to model sophistication. Two
peak rate estimation methods, the CREAMS hydrologic model equation and the SCS
unit hydrograph method, were modified to better reflect observed data.
Contribution from the Institute of Food and Agricultural Sciences,
University of Florida, as a part of Southern Region Project S-164
of the USDA-CSRS with support from South Florida Water Management
District.
o
The authors are, respectively: Associate Professor, Instructor and
Associate Professor, Agricultural Engineering Department, University
of Florida, Gainesville, Florida 32611.
45
-------
INTRODUCTION
Many techniques have been developed to estimate stonnwater total runoff
volume and peak discharge rates from small watersheds. However problems arise
when these methods are applied to the unusual hydrologic conditions found in
Florida's Flatwoods Resource Area. Watersheds of this area typically have
very flat slopes, extremely permeable sandy soils, high water tables, and
wetlands scattered throughout their basins. Such characteristics are unlike
those of the watersheds which served as the models for the development of most
runoff prediction methods. The problems introduced by these atypical water-
shed conditions are often compounded when the methods are called upon to
predict runoff resulting from rainfall events for which they were not Intended
i.e., frequent, instead of extreme (design), events.
Studies which document the accuracy of standard runoff prediction tech-
niques as applied to Florida's flatwoods watersheds under a range of rainfall
events are not currently available. Hydrologists, engineers, and water re-
source managers are therefore forced to make decisions based upon runoff
estimates resulting from methods which, although generally accepted, are not
necessarily accurate under these particular watershed conditions. The users
often appreciate the errors and limitations associated with their runoff
estimates, but do not have sufficient Information with which to offer Improve-
ments. The research described in this paper represents an effort to help fill
the existing information gap.
Data collection sites for this study are within the Lower Kissimmee River
and Taylor Creek-Nubbin Slough Basins north of Lake Okeechobee. The pre-
dominant soil associations for both basins are Myakka-Immokalee-Waveland and
Wabasso-Felda-Pompano (1). Natural vegetation consists primarily of wet and
dry prairie grasslands and pine-palmetto forests. In the depressional areas,
wetlands species predominate. Land use in the two basins Is dominated by
improved and unimproved pasture, claiming about 75% of the total area by 1980
(2, 3).
The means of transformation from a natural marsh and slough system to
agricultural use has been drainage Improvement achieved through ditching.
Extensive channel networks combined with extremely low watershed slopes
(<0.5%) make delineation of watershed boundaries a difficult task in some
cases. Drainage patterns can, in fact, shift depending upon rainfall patterns
and runoff magnitude.
Hydrologic data from five watersheds ranging In size from 20 to 3600
acres and located within the Lower Kissimmee River and Taylor Creek-Nubbin
Slough Basins were collected between 1979 and 1983 in conjunction with the
Upland Detention/Retention Demonstration Project. More detailed site and data
descriptions can be found in the project report (4).
4fr
-------
METHODS
TOTAL RUNOFF VOLUME
The SCS runoff equation serves as the basis for the total volume estima-
tion methods examined in this report and analyzed by Konyha et al. (5) and can
be written as:
(P - 0.2S)2
Q - [1]
(P + 0.8S)
where P 24-hour rainfall depth and
S - watershed storage parameter.
Each of the following methods uses a distinct technique for arriving at
the storage parameter, S. These methods are referred to in this paper as NEH-
4 (6), SCS-Florida (7), .DRM (8), ARS (4, 9), CR-1 (5, 10), CR-2 (5,11), and
CR-WT (12). NEH-4 is the standard SCS curve number method as described in the
SCS National Engineering Handbook-Section 4, Hydrology (6). SCS-Florida is a
revision of this method from the SCS Florida Interim Procedure report (7)
which uses antecedent moisture condition II for all events. DRM is a method
described in the South Florida Water Management District (SFWMD) Regulatory
Manual (8) which makes the storage parameter, S, in equation 1 a direct func-
tion of the depth to water table. ARS is a method developed by modifying the
available storage relationship in DRM based on research data from Speir et al.
(9). CR-1 uses an algorithm from the CREAMS model (10) to adjust the water-
shed storage parameter as determined by the ARS method to allow for influences
of management practices on runoff volume. CR-2 is similar to CR-1 except that
it uses a simple soil moisture accounting model to estimate the storage para-
meter instead of relying on water table data. CR-WT uses the full hydrology
component of the CREAMS model in a modified form to account for a fluctuating
shallow water table by preventing deep percolation out of the soil profile
(12). Each of these methods is described in more detail in the project report
(4).
STORM RUNOFF PEAK RATE
A variety of approaches are available for the routing of stormwater to
arrive at peak discharge rates. Several techniques representing a range of
complexity levels have been applied to the Florida flatwoods. The following
methods were evaluated, beginning with the very empirical and progressing
through to the more theoretical approaches:
1. Cypress Creek Formula (11),
2. CREAMS peak rate equation (10, 13),
3. SCS Chart Method (7, 14),
4. SCS Unit Hydrograph Method (6, 15), and
47
-------
5. SFWMD Model (16, 17) used to generate graphs In District
Regulatory Manual IV (8).
Each of these methods is described by Capece et al. in the project report
(4). Two methods, the CREAMS equation and the SCS Unit Hydrograph method,
were modified in an effort to improve their performance. The results are
summarized in the following section of this paper and the procedures used in
modifying the methods can be found in the project report (4).
RESULTS
STORM RUNOFF TOTAL VOLUME
Seven techniques for estimating stormwater runoff volume were applied and
evaluated on an event basis. The selected storms measured 0.70 or more inches
of rainfall in 24 hours and may or may not have produced measurable runoff.
The seven runoff volume estimation methods applied to the data set were: NEH-
4, SCS-Florida, DRM, ARS, CR-1 , CR-2, and CR-WT as described earlier.
Tables 1 and 2 present the same results as standard error of estimates
determined with the following equation:
(Q- Q)
<>»
I (Q;-
[2,
n - 1
where e - standard error of estimate in inches,
Q* predicted runoff volume for event i in inches,
Q. - measured runoff volume for event i in inches, and
n » total number of storm events.
Each table represents technique performance as applied to selected classes of
events: all daily rainfall events equal to or exceeding 0.70 inches, and a
subset which produced runoff equal to or exceeding 0.50 inches.
Standard errors corresponding to "all" sites do not weight each watershed
equally. Instead, the overall method standard error of estimate is most
heavily weighted toward the sites which had more usable events. The ranking
48
-------
TABLE 1. STANDARD ERRORS OF RUNOFF VOLUME ESTIMATES, IN INCHES, FOR
ALL EVENTS
Site
Armstrong
Peavine
SEZ Dairy
Bass West
Bass East
All
Site Ranking
NEH-4
0.57
0.66
0.31
1.12
0.64
0.73
(6)
SCS-FL
0.52
0.57
0.57
0.86
0.54
0.63
(4)
DRM
0.40
0.45
0.52
0.50
0.57
0.47
(2)
Method
ARS
0.38
0.45
0.40
0.46
0.55
0.44
(1)
CR-1
0.44
0.45
0.46
0.53
0.57
0.48
(3)
CR-2
0.70
0.70
0.61
0.64
0.69
0.66
(7)
CR-WT
0.52
0.61
0.30
1.11
0.84
0.74
(5)
TABLE 2. STANDARD ERRORS OF RUNOFF VOLUME ESTIMATES, IN INCHES, FOR
EVENTS WITH MEASURED RUNOFF EQUAL TO OR EXCEEDING 0.50 INCHES
Site
Armstrong
Peavine
SEZ Dairy
Bass West
Bass East
All
NEH-4
1.38
1.09
0.71
1.85
1.31
1.36
SCS-FL
0.79
0.58
0.64
1.37
0.91
0.93
DRM
0.81
0.65
0.71
0.78
1.01
0.76
Method
ARS
0.54
0.59
0.47
0.72
1.04
0.67
CR-1
0.93
0.65
0.76
0.83
1.08
0.79
CR-2
0.75
1.07
0.60
0.88
1.06
0.90
CR-^WT
1.12
1,05
0.72
1.62
1. 53
1.32
Site Ranking (7)
(2)
(5)
(1)
(4)
(3)
(6)
49
-------
corresponding to "all" sites was determined by comparing the sum of the
methods' performance ranking for each site and, therefore, weights performance
on each watershed equally.
Generalizations can be drawn from these rankings regarding technique
overall performance and trends through changing runoff volume. The ARS method
consistently performed better than all other methods. The SCS-Florida method
demonstrated improved accuracy as runoff volume increased, as would be expec-
ted of a method intended for design applications. The CR-1 method performed
very well on the smaller events, but not as well on the larger events. The
DRM method gave results very similar to ARS and CR-1 for small events and also
demonstrated decreased accuracy when applied to the larger runoff events. The
CR-2 method performed poorly on the small events, but improved somewhat on
larger events. Both the NEH-4 and CR-WT methods produced consistently in-
accurate estimations of runoff volume.
STORM RUNOFF PEAK RATE
The following sections present results of the peak rate estimation tech-
niques as applied to each watershed and all runoff events. Performance is
quantified as standard error of estimate, in percent, and average error of
estimate, in percent:
[31
n - 1
M * !
i (-V-^
5 = [-^ i ]
n
where e standard error of estimate in percent,
§ « average error of estimate in percent,
q' » predicted peak rate for event i in cfs,
q. » measured peak rate for event i in cfs,
n - total number of runoff events.
The average error for each site describes a method's tendency to over-
predict or underpredict while the standard error quantifies error absolute
magnitude. Rainfall events greater than 0.70 inches and having measurable
runoff were included in the data base for this analysis. Results described as
applying to "all" sites are biased toward the sites having more usable runoff
events. Standard errors of estimate for each peak rate method are summarized
in Table 3.
50
-------
256.*
946.
656.
363.
1050.
1002.
2770.
1764.
2069.
7166.
46.
142.
117.
201.
1075.
106.
430.
192.
188.
44.
67.
55.
64.
29.
479.
21.
33.
34.
21.
86.
21.
77.
51.
18.
26.
TABLE 3. RESULTS SUMMARY FOR PEAK RATE ESTIMATION TECHNIQUES AS APPLIED
TO EVENTS WITH MEASURED RUNOFF EQUAL TO OR EXCEEDING 0.50 INCHES
SitePeak Rate Estimation Technique
Cypress CREAMS SCS-Chart SCS-UH SFWMD CR-Mod UH-Mod
Armstrong
Peavine
SEZ Dairy
Bass West
Bass East
All 715. 3511. 461. 254. 181. 42.' 45.
* Results are reported as standard error of estimate, in percent.
Runoff events measuring less than 0.50 inches tended to produce erratic
results. Because estimation errors are expressed as a percent of measured
peak, very small events are prone to produce large errors of estimate.
Another problem was that with small quantities of measured runoff, ground
water discharge becomes more significant and produces atypical hydrographs.
For these reasons, emphasis is placed on peak rates predicted for runoff
events equal to or exceeding 0.50 inches.
Cypress Creek Formula^
Predictions from the Cypress Creek Formula are compared against measured
peaks in Table 3. As previously described, prediction errors associated with
the smaller events were larger than the error reflected in Table 3.
The standard and average errors were comparable in magnitude and average
errors are all positive. Thus, the method consistently resulted in large
overpredictions of measured peak discharge. Standard errors ranged from 200%
for Armstrong (the largest watershed) to 1000% for Bass East (the smallest
watershed). Even when the effect of transforming a 24-hour maximum rate into
an instantaneous rate was removed, the method still overpredicted.
CREAMS Equation
The standard CREAMS equation performed worse than any other method ex-
amined in this study. It consistently overpredicted by an order of magnitude
or more (see Table 3). The results when examined graphically indicated that
the estimation error was fairly consistent for all sites.
51
-------
A regression of the CREAMS model formulation against measured data
yielded this modified version of the equation:
qp - 4.52(DA-)(CS-)(LW-)(Q') [5]
where q_ = peak runoff rate in cfs,
DA = drainage area in mi ,
CS = main channel slope in ft/mi,
LW = watershed length to width ratio, and
Q = daily runoff volume in inches.
When reapplied to the data base, performance of this equation was good.
CR-Mod in Table 3 does not represent an independent evaluation of the modified
CREAMS equation, but simply reflects the regression fit to the data. This
equation resulted in an R2 of 0.96. The standard error of estimate associated
with equation 5 ranged from 20% for Bass West to 85% for Bass East.
SCS Chart Method
Performance comparisons for the SCS Chart Method are presented in Table
3. Unlike the Cypress Creek Formula, large events produced results similar to
those for all events. This method also tended to overpredict peak
discharge. Maximum and minimum standard errors of estimate were 1000% and
50%, corresponding to Bass East and Armstrong Slough, respectively.
SCS Unit Hydrograph Method
Evaluation of the unit hydrograph method was conducted in three steps:
(1), evaluation of standard SCS methodology; (2), evaluation and modification
of certain aspects of the method; and (3), re-evaluation of the unit hydro-
graph method implementing various modifications.
Based on the triangular unit hydrograph, the standard estimate for Kf
(484) describes a hydrograph whose recession is 1.67 times as long as its time
to peak. Mockus (6) notes that this K1 value has been known to vary from 600
in steep terrain to 300 in flat swampy country. For the Delmarva peninsula,
which includes Delaware and parts of Maryland, Welle et al. (18) concluded
that a value of 256 is more appropriate. The watersheds they examined were
small with sandy soils and slopes in the range of 2%. The U.S. Army Corps of
Engineers (19) studied records from several large watersheds in Central and
South Florida (the entire Kissimmee River Basin being one) and determined an
appropriate time factor for use in a similar peak discharge equation. Miller
and Einhouse (20) translated this factor into the SCS form, arriving at a
value of 284 for K1.
52
-------
Table 4 presents results from application of the standard K1 factors (300
and 484). SFWMD assumed rainfall distributions were used for all events.
Best peak estimates were obtained using a K' of 300. However this method
still tended to overpredict discharge peaks by about 200%.
TABLE 4. RESULTS SUMMARY FOR SCS UNIT HYDROGRAPH METHOD
Site K*- 300 F - 484
Std. Error Avg. Error Std. Error Avg. Error
106.*
430 1
192.
188.
44.
88.
380.
142.
175.
28.
230.
724.
352.
337.
112.
200.
646.
272.
316.
94.
Armstrong
Peavine
SEZ Dairy
Bass West
Bass East
All 254. 195. 439. 354.
* Errors are reported in percent.
The second step of the evaluation was to determine best-fit K1 factors
using various estimates of time to peak, both measured and assumed rainfall
time-distributions, and different classes of runoff events. The SFWMD assumed
rainfall distribution yielded almost identical results to those derived from
measured distributions. Differences in best-fit K' values were apparent when
comparing the two runoff event classes. Factors for events less than 0.50
inches were 20-30% higher than factors for the larger class of events. The
scatter among Kf values for each site is also greater for smaller events.
Focusing in on the K1 solved using assumed rainfall distributions and events
greater than 0.50 inches, almost all were computed to be less than 100.
Table 5 shows the site variability of best-fit factors. Here the average
K1 is calculated over n events for each site and s is the standard deviation
associated with that average. The trend among sites was for an Increasing K1
value with decreasing watershed percent wetlands. Results reported for Pea-
vine Pasture differentiate between the large and small drainage area condi-
tions. The smaller factor associated with the 1800 acre watershed may be due
to channel block effects.
Also included in Table 5 are the best-fit factors for the large runoff
events associated with Hurricane David. Runoff from this event ranged between
2.5 and 5.2 inches, depending upon the specific site. Armstrong data for this
period are documented as being "estimated." Questions also exist regarding
the actual contributing area for this event as well as other events at other
sites. Examination of the SCS unit hydrograph equation shows that when mea-
sured runoff data are used, the influence of errors in drainage area estimates
53
-------
upon peak rate calculations is confined to the T term. The multiplication of
the depth and area terms in the numerator of the equation yields volume. This
is the inverse of the calculation used to estimate runoff depth (measured
volume divided by estimated contributing area) and thus negates the influence
of drainage area estimates.
TABLE 5. RESULTS SUMMARY FOR SCS UNIT HYDROGRAPH K' FACTOR OPTIMIZATION
USING SFWMD ASSUMED RAINFALL DISTRIBUTION, RUNOFF EVENTS EQUAL
TO OR EXCEEDING 0.50 INCHES, AND MODIFIED-FIXED LAG ESTIMATES
Site
n Events Maximum Minimum'
Average
72.
13.
Davids
Armstrong
Peavine (1800)
Peavine (775)
SEZ Dairy
Bass West
Bass East
5
10
3
4
16
7
119.
74.
76.
101.
107.
84.
62.
27.
53.
39.
59.
66.
83.
50.
66.
70.
88.
72.
22.
15.
12.
26.
12.
8.
88.
66.
-
-
118.
70.
Maximum observed optimized event K1 factor.
* Minimum observed optimized event K1 factor.
t K values represent an average of all events for a site.
§ Based on available data for rainfall associated with Hurricane David
(9-3-79).
Incorporating results from the K1 analysis, the incremental unit hydro-
graph method was re-applied to the data base. Like the initial evaluation,
only SFWMD assumed rainfall distributions and runoff events greater than 0.50
inches were examined. Best results shown in Table 3 under UH-Mod were
achieved with a K1 of 75.
SFWMD Model
The SFWMD overland flow computer model was applied to runoff events
exceeding 0.50 inches. Results are summarized in Table 6. The model under-
predicted on most sites. However, where it did overpredict, the percent error
was high (Bass East). Best results were associated with the Bass West and
Peavine watersheds. The large Peavine Pasture watershed is believed to re-
spond in a sheetflow manner and should be described well by an overland flow
model. For the large watersheds with significant channel effects (Armstrong
54
-------
Slough and SEZ Dairy), the model underpredicted as would be expected. However
for the smallest watershed (Bass East), where the overland flow approximation
would appear most applicable, results were not good. The observed overpredic-
,tion is probably due to the length-to-width ratio of the pasture. It is wide,
about 1700 feet, and only 500 feet long. This shape would simulate as a very
high peak-producing watershed.
TABLE 6. RESULTS SUMMARY FROM SFWMD OVERLAND FLOW COMPUTER MODEL
AS APPLIED TO RUNOFF EVENTS WITH MEASURED RUNOFF EQUAL
TO OR EXCEEDING 0.50 INCHES
SiteStandard ErrorAverage Error
Armstrong
Peavine (1800)
Peavine (775)
SEZ Dairy
Bass West
Bass East
67.*
59.
50.
64.
29.
479.
-60.
10.
-35.
-46.
-8.
410.
All 181. 48.
* Errors are reported in percent.
Summary
When compared with one another (Table 3), the methods demonstrate magni-
tudes of error inversely proportional to their degree of complexity. With
decreasing overall standardjerror of estimate, the original methods line up
as: CREAMS, 3500%; Cypress Creek Formula, 700%; SCS Chart, 400%, SCS unit
hydrograph, 250%; SFWMD, 180%. The CREAMS and Cypress Creek Formula should be
reversed based upon complexity level, however the CREAMS equation was not
developed using Florida data, while the Cypress Creek Formula is described as
being applicable to the Florida flatwoods.
Fitting the CREAMS equation and SCS unit hydrograph approach significant-
ly improved the performance of both methods. Each achieved between 40% and
45% standard error of estimate with little bias'toward over- or under-predic-
tions.
55
-------
SUMMARY AND CONLUSIONS
TOTAL VOLUME EVALUATION
The SCS runoff equation was developed for application to large design
events occurring on small watersheds. However, in many instances it has been
applied to smaller rainfall events with little or no consideration given to
accuracy implications. Specific techniques employed to determine .the water-
shed storage parameter, an input to the SCS runoff equation, have not been
sufficiently evaluated as to their suitability for atypical watershed condi-
t ions.
Evaluations of the SCS equation and specific methods for determining its
inputs demonstrate that large errors can be associated with runoff estimates
for smaller events. For the seven methods examined, overall standard error
of estimates ranged from several hundred to fifty percent. For both the
larger and smaller events, best estimates of runoff volume resulted from
techniques which incorporated antecedent water table conditions.
Three of the methods (DRM, ARS, and CR-1) relied upon measured water
table elevations and performed similarly on small events. The ARS method
consistently performed best on all event classes. The CR-1 method incor-
porates the ARS storage relationship, but has the added advantage of account-
ing for factors other than water table depth via the SCS curve number. This
offers latitude useful in evaluating changes in runoff volume resulting from
alternative land use patterns and agricultural practices. The CR-2 method has
the same advantages, but rather than water table history or assumptions, a
rainfall history is required. This method did not perform as well as CR-1.
The SCS-Florida method considers strictly land use and soils, ignoring varia-
tions in watershed wetness. Therefore its use could lead to significant
runoff estimation errors when applied to large events falling upon saturated
watersheds. Neither the NEH-4 nor CR-WT methods should be used for runoff
estimation on an event basis.
Estimates of runoff volume and the evaluation of prediction methods are
more sensitive to errors in data collection and drainage area determination
than are peak rates* However results demonstrate that techniques which incor-
porate water table levels (total available soil storage) can be expected to
yield more accurate estimates of runoff volume for flatwoods watersheds.
PEAK RATE EVALUATION
Results of this study demonstrate that more accurate estimates of runoff
peak rates can be expected as models progress from the empirical to the more
physically based. However when empirical models are tailored to specific
watershed conditions, results may be comparable to those from more complex
models. As watershed conditions change or changes are anticipated, physically
based models again become more reliable than empirical techniques.
56
-------
The two extremes of empirical and physical models are represented by the
CREAMS equation and the SFWMD overland flow computer program. The overland
flow model performed best of all the original methods examined, however, it
still demonstrated considerable overall error.
With modifications, estimation error was significantly reduced in the
CREAMS equation and SCS unit hydrograph method. For the CREAMS equation,
overall standard error of estimate was reduced from 3500% to 42%. For the
unit hydrograph method, modifications reduced the overall standard error
estimate from 250% to 45%. Between the two modified methods, the SCS method
is more versatile and should be more transportable to other flatwoods water-
sheds. The SCS technique is capable of handling multiple-day (complex)
events, whereas the CREAMS equation does not allow superposition.
Significant unit hydrograph results indicate that the SCS recommended
triangular hydrograph factor, 300, is too high. Analyses indicate that a
value less than 100 is more appropriate for Florida's flatwoods watersheds.
Also noteworthy were the almost identical peak rate estimates derived from
measured and assumed rainfall time-distributions. Discharge hydrographs from
flatwoods watersheds are much more attenuated and produce much lower peaks
than most other small watersheds of the United States.
The work described in this paper was not funded by the U. S. Environmen-
tal Protection Agency and therefore the contents do not necessarily reflect
the views of the Agency and no official endorsement should be inferred.
REFERENCES
1. Caldwell, R.E., and R.W. Johnson. General Soils Map: Florida. USDA-SCS
and Institute of Food and Agricultural Sciences, Gainesville, 1982.
2. Huber, W.C., J.P. Heaney, P.B. Bedient, and J.P. Bowden. Environmental
Resource Management Studies in the Kissimmee River Basin. Central and
Southern Florida Flood Control District, West Palm Beach, 1976.
3. Allen, L.H., J.M. Ruddell, G.H. Ritter, F.E. Davis, and P. Yates. Land
Use Effects on Water Quality. In; Proceedings on the Speciality Confer-
ence on Environmental Sound Water and Soil Management. American Society
of Civil Engineers, New York, 1982.
4. Capece, J.C., K.L. Campbell, and L.B. Baldwin. Estimation of Runoff Peak
Rates and Volumes from Flatwoods Watersheds. Final Report. South
Florida Water Management District, West Palm Beach, Florida, 1984.
5. Konyha, K.D., K.L. Campbell, and L.B. Baldwin. Runoff Estimation from
Flat High-Water-Table Watersheds. Coordinating Council on the Restora-
tion of the Kissimmee River Valley and Nubbin Slough Basin, Tallahassee,
Florida, 1982.
57
-------
6. U.S. Department of Agriculture-Soil Conservation Service. National
Engineering Handbook, Section 4, Hydrology. USDA-SCS, Washington, D.C.,
1972.
7. U.S. Department of Agriculture-Soil Conservation Service. Interim Runoff
Procedure for Florida. Florida Bulletin 210-1-2. USDA-SCS, Gainesville,
1980.
8. South Florida Water Management District. Permitting Information Manual -
Volume IV, Management and Storage of Surface Waters. SFWMD, West Palm
Beach, 1983.
9. Speir, W.H,, W.C. Mills, and J.C. Stephens. Hydrology of Three
Experimental Watersheds in Southern Florida. ARS Publication No. 41-
152. USDA-Agricultural Research Service, Washington, D.C., 1969.
i
10. Knisel, W.G. CREAMS: A Field-Scale Model for Chemicals, Runoff, and
Erosion from Agricultural Management Systems. Conservation Research
Report No. 26. USDA-Agricultural Research Service, Washington, D.C.,
1980.
11. Stephens, J.C., and W.C. Mills. Using the Cypress Creek Formula to
Estimate Runoff Rates in the Southern Coastal Plains and Adjacent Flat-
woods Land Resource Areas. ARS Publication No. 41-95. USDA-Agricultural
Research Service, Washington, D.C., 1965.
12. Heatwole, C.D., J.C. Capece, A.B. Bottcher, and K.L. Campbell. Modeling
the Hydrology of Flat, High-Water-Table Watersheds. In; Proceedings of
the Fourth Annual AGU Front Range Branch Hydrology Days, ed. by H.J.
Morel-Seytoux. Colorado Water Resources Research Institute, Colorado
State University, Fort Collins, 1984.
13. Smith, R.E., and J.R. Williams. Simulation of Surface Water Hydrology.
In: CREAMS: A Field-Scale Model for Chemicals, Runoff, and Erosion from
Agricultural Management Systems, ed. by W.G. Knisel. Conservation
Research Report No. 26. USDA-Agricultural Research Service, Washington,
D.C., 1980.
14. U.S. Department of Agriculture-Soil Conservation Service. Urban
Hydrology for Small Watersheds. Technical Release No. 55. USDA-SCS,
Washington, D. C., 1975.
15. Kent, K.M. A Method for Estimating Volume and Rate of Runoff in Small
Watersheds. SCS-TP 149. USDA-Soil Conservation Service, Washington,
D.C., 1973.
16. Higglns, R.W. The Development of a General Watershed-Highway Mathe-
matical Model. Unpublished Master's Thesis, University of South Florida,
Tampa, 1976.
58
-------
17. South Florida Water Management District. Memorandum Report on Peak
Runoff Estimation for Undeveloped Lands. SFWMD, West Palm Beach, 1979.
18. Welle, P.I., D.E. Woodward, and H.F. Fox. A Dimensionless Unit Hydro-
graph for the Delmarva Peninsula. ASAE Paper No. 80-2013. American
Society of Agricultural Engineers, St. Joseph, Michigan, 1980.
19. U.S. Army Corps of Engineers. Central and Southern Florida Project for
Flood Control and Other Purposes: Part VI, Section 9. Office of the
District Engineer, Jacksonville, 1955.
20. Miller, H.D., and J.D. Einhouse. Little Econlockhatchee River Restora-
tion Study. Miller and Miller Inc., Orlando, Florida, 1984.
59
-------
STORMWATER MANAGEMENT MODEL APPLICATION TO A
PEATLANDS REGION IN COASTAL NORTH CAROLINA
by: Robert E. Dickinson
Warren Pandorf
Larry J. Danek
Environmental Science and Engineering, Inc.
Gainesville, Florida 32602
ABSTRACT
The runoff- from a North Carolina peatlands region, an area with a
high drainage density, was simulated using a modified version of the
Storm Water Management Model (SWMM) RUNOFF block. The RUNOFF block was
changed to simulate two types of pervious flow: (1) surface and overland
flow from the peat areas adjacent to the drainage ditches, and (2) slower
subsurface flow from the inter-ditch peat areas. In general, the peat
areas near the canals and ditches contribute to the peak portion of the
runoff hydrograph while the inter-ditch areas contribute to the recession
curve of the runoff hydrograph. The model was calibrated to 28 rainfall-
runoff events measured at two small peat sites: (1) a 10-acre shrubland
(gallberry) site and (2) a 10-acre harvest (bare soil) site.
INTRODUCTION
The Pamlico-Albemarle peninsula of North Carolina has extensive
deposits of peat soil along with freshwater marshes, swamps, and upland
hardwood, pine, and mixed forests,. The majority of the peninsula was
once covered with dense verdant vegetation that impeded stormwater runoff
to the North Carolina estuaries (1). The peninsula was first developed
for farming in 1787, with the completion of a canal that drained
10,000 acres of peatland (2). This development process accelerated
during the period from 1930 to 1950 with the construction of numerous
main canals, cross-connector canals, and drainage ditches. This canal
system has a high drainage density (nearly 20 miles of canals and ditches
per square mile of land area) which increases the peak storm flows and
the sediment load to the estuaries of North Carolina.
60
-------
Ironically, much of the ditched pocosin area was found to be
unsuitable for farming. The ditched peat areas were either converted to
pasture, left fallow, or never developed. During the 1980"s, a plan was
developed to harvest 15,000 acres of the ditched peat area for conversion
to methanol, with the methanol being used as a fuel additive (3). As
part of the permitting process for the peat-to-methanol plant, an
environmental monitoring network for water quality and quantity was set
up at the project site (see Figure 1 for a location map of the project
area).
SWMM Version III (4) was chosen to simulate the runoff from the
project site for the following reasons:
1. Its output was compatible with the input requirements of RECEIV
(5), which was, being used to simulate water quality in the
rece'iving water estuary;
2. It was capable of both single-event and continuous rainfall
simulation; .and
3. It could simulate the complex canal and lake system to the
extent of detail required by permitting agencies.
The problem with using the model, however, was that the overland
flow portion of the SWMM RUNOFF block was physically unrepresentative of
the runoff process from the peat area. The runoff from a peat area is
primarily subsurface or interflow. The runoff-generating mechanism of
RUNOFF was modified to successfully simulate stormwater runoff from
ditched peat areas. The remainder of this paper describes the modified
SWMM application to two small (10-acre) sites in the project area. The
two sites, one covered by gallberry shrub (Site C on Figure 2) and the
other a bare soil peat area (Site H on Figure 2), are 165 feet wide and
2,640 feet long. The sites are bisected by a 4-foot-wide drainage ditch
which flows to a cross-connector canal and then to a main drainage canal.
The land area is mildly sloped (1-percent slope) to the drainage ditches
and the ditches have a slope of 0.1 percent. Rainfall and runoff data
for 28 storm events were gathered between February 1983 and July 1983 and
were used to calibrate the runoff model.
KINEMATIC WAVE STORAGE MODEL
A peat area can be described as a perched water table in which there
is little hydrologic interaction with the deep aquifer because of the
impermeable soil underlying the peat (1). The only mechanisms for the
removal of rainfall are evapotranspiration and runoff.
The peat runoff mechanism is influenced by the presence of the
bisecting drainage ditches. When the water table is below the bottom of
the ditch, there will be no runoff except for a small amount due to
direct rainfall in the near ditch area. During the seasons of the year
in which the water table is above the bottom of the drainage ditch, the
runoff into the ditch will come from the saturated peat areas adjacent to
the drainage ditch. The peat is saturated due to a rising water table
fed by infiltration from above and by laterally inflowing baseflow (6).
61
-------
WINSTON-SALEM» GREENSBORO
CHAPEL
HILL* RALEIQH
MAYSVILLE
CAROLINA
PROJECT
SITE
\
40 MILES
20 0 40 KILOMETERS
WILMINGTON
APE
HATTERAS
Figure 1. Location map.
SCALE
01 2345 MILE
STRIAM aAOINO, NONHICOROINQ RAIN
OAQi. AND WATIR QUALITY UHN.INO
A CONTINUOUS IteCOROINO MAIN QAQE
DRAINAOE IA1IN IOUNOARIU
OUAINAOI PLOW TO lAWUNa
TAIIOHI
Figure 2. Storm event sampling locations February-July 1983.
52
-------
The existing SWMM runoff-generating mechanism, which uses a version
of the kinematic wave equation with Manning's velocity equation, is best
suited for turbulent overland flow (4, 7). However, subsurface flow
models using alternative kinematic wave equations for a runoff mechanism
were found to be appropriate for soils with large macropores (8, 9, 10).
The peat area, which has large macropores due to undecayed tree trunks,
branches, and twigs, is modeled using a kinematic wave storage model.
The kinematic wave equation relates depth of water in storage to
outflow. Kinematic wave equations are applicable in situations in which
all the terms of the mass continuity equation are negligible compared to
bottom slope and friction. The equation has the general form of:
(1) q » adm
where: q * flow,
a = coefficient,
d = depth of water in peat, and
m = exponent.
The coefficient a and the exponent m vary according to flow regime, i.e.
laminar or turbulent flow. The exponent m has a value of 1.5 using the
Chezy turbulent flow equation, a value of 1.66 using Manning's turbulent
flow equation, and a value of 3.0 for laminar flow (7). Horton in 1983
found an average value of m « 2.0 for a mixture of laminar and turbulent
flow (11). The calibrated exponent for runoff from the two peat areas
was 2.66. Two a values were calibrated per drainage area: (1) the first
a is the coefficient for the subsurface and overland flow from the peat
areas adjacent to the drainage ditch (aj), and (2) the second a is the
coefficient for the slower subsurface flow from the inter-ditch area
(02). The ai and 02 coefficients incorporate the effect of slope,
roughness, and undefined factors of influence in the flow model.
The drainage sites were modeled with two contributing runoff source
areas. The peat area nearest the drainage ditch contributes to the peak
portion of the storm hydrograph (first source area), while the inter-
ditch area contributes to the recession curve portion of the storm
hydrograph (second source area). There is no interconnection between the
two areas; however, there is a direct model connection between each
source area and the outlet of the drainage ditch.
The first source area was assumed to always have an initial depth of
zero. The second source area has an initial water depth corresponding to
the base flow immediately preceding the calibrated storm event. The
initial depth is calculated as:
(2) d0 - (Q0/a2)3/8
where: d0 - initial depth of water in peat (feet),
a? » kinematic wave coefficient, and
Q0 » base flow (cfs).
63
-------
An initial depth of zero corresponds to a water table elevation
equal to the elevation of the bottom of the drainage ditch.
The vertical boundaries of the model extend from the surface to the
bottom of the drainage ditch. The model is constrained to never have a
negative depth. A water balance is maintained, i.e. runoff plus
evaporation and storage equals rainfall.
RESULTS AND DISCUSSION
A visual comparison of the measured runoff at the shrubland site
(Figure 3) versus the measured runoff at the harvest site (Figure 4)
shows that the harvest site had larger peak flows. Because of an
extensive root mat, the shrubland site attenuates the peak discharge and
lengthens the runoff time compared to the bare soil harvest site.
The predicted versus measured calibration results for 13 storm
events from the shrubland site are presented in Table 1, and the results
for 15 storm events from the harvest site in Table 2. Presentations of
the comparison of measured versus predicted runoff for selected events at
the control site and harvest site are shown in Figures 5 and 6,
respectively.
Each drainage area had a rain gage located next to the V-notch weir
but due to the shape of each drainage area (2,640 feet long and 165 feet
wide), a significant portion of the runoff error may be the result of
rainfall variability during localized storms. For example, the rainfall
for Storms 7, 8, and 9 was observed to be quite variable among the onsite
rain gages. If this rainfall was underestimated from the shrubland site,
this may explain the high runoff rate (90.6 percent) measured for these
three storms.
The model closely predicted runoff from both sites for three seasons
of storm event data (winter, spring, and summer). The shapes of the
hydrographs, peak flows, and total volumes were matched to the field data
by adjusting the coefficients of the kinematic wave storage model.
The result was a slightly more complicated runoff model that was
more representative of the actual physical conditions than the original
SWMM runoff mechanism.
The work described in this paper was not funded by the
U.S. Environmental Protection Agency and therefore the contents do not
necessarily reflect the views of the Agency and no official endorsement
should be inferred.
ACKNOWLEDGMENTS
The work described in this paper was funded by Peat Methanol
Associates of Pittsburgh, PA. We wish to thank Robert Honstein, Peter
64
-------
10'
FEBRUARY
MARCH
APRIL MAY
MONTH
JUNE
JULY
Figure 3. Flow hydrograph at Station C (Control Site) February-July 1983.
2
o
s1
o
10'z_
10'?.
a
£
o
o
I
lid
':
APRIL MAY
MONTH
FEBRUARY
MARCH
JUNE
JULY
Figure 4. Flow hydrograph at Station H (Harvest Site) February-July 1983.
65
-------
Storm*
1, 2, 3
7, 8, 9
13,14,15
16
17
21
24
TOTAL
Beginning
date Rainfall
(1983) (in)
2/07
3/17
4/08
4/23
5/04
5/30
6/07
3.71
3.95
4.09
0.90
0.50
0.50
1.70
15.35
Simulation
time
(day 9)
11.3
10.0
10.0
3.0
2.0
4.0
3.0
43.3
Initial
storage
(in)
1.05
1.02
1. 13
1.12
0.98
0.00
0.00
5.30
Ending
storage
(in)
1.28
0.98
1.39
1.27
1.03
0.00
1.06
7.01
Measured runoff
(in)
2.62
3.58
2.57
0.36
0.097
0.00
0.056
9.28
tt)
70.6
90.6
62.8
40.0
19.4
0.0
3.3
60.5
Predicted runoff
(in)
2.61
2.77
2.14
0.388
0.132
0.010
0.235
8.30
«)
70.3
70.1
52.3
43.0
26.4
2.0
13.8
54.0
Runoff
errort
«)
0.3
-20.5
-10.5
3.0
7.0
2.0
10.5
-6.4
First Source Area:
37.1;
30 percent. Second Source Area: a2 - 7.4; P,, - 70 percent.
*ESE storm number.
tNormalized runoff error - 100.0 * (predicted runoff - measured runoff ) /rainfall.
Source: ESE, 1985.
Table 1. Predicted versus measured runoff for Control Site (shrubland).
Storm*
1
2
5
9
10
11
16
17
18
19
20
21
22
26
30
TOTAL
Beginning
date
(1983)
2/06
2/10
2/24
3/24
3/27
3/31
4/24
5/04
5/16
5/23
5/26
5/30
6/01
6/20
7/19
Rainfall
(in)
1.04
0.90
0.65
0.85
0.30
1.09
1.20
0.51
0.40
0.25
0.30
1.18
1.10
1.35
1.05
12.17
Simulation
time
(days)
2.0
2.5
2.0
2.0
2.5
2.0
2.5
3.0
5.8
2.6
5.8
4.0
4.0
3.0
2.5
46.2
Initial
storage
(in)
0.83
0.91
0.86
0.99
0.86
0.79
0.45
0.38
0.00
0.44
0.46
0.53
0.71
0.46
0.07
8.74
Ending
storage
(in)
1.24
1.23
1.15
1.23
0.92
1.23
0,96
0.46
0.00
0.38
0.07
0.58
0.77
0.88
0.35
11.45
Measured
(in)
0.50
0.65
0.31
0.49
0.08
0.34
0.35
0.01
0.01
0.01
0.02
0.03
0.10
0.29
0.35
3.54
runoff
(X)
48.0
72.2
47.8
57.6
26.7
31.2
29.2
2.0
2.5
4.0
6.7
2.5
9.1
21.5
33.3
29.0
Predicted
(in)
0.50
0.44
0.23
0.40
0.11
0.50
0.30
0.08
0.05
0.02
0.05
0.14
0.40
0.33
0.26
3.81
runoff
«)
48.0
48.9
35.3
47.0
36.7
45.9
25.0
15.7
12.5
8.0
16.7
11.9
36.3
24.4
24.8
31.3
Runoff
errort
^
0.0
-23.7
-12.5
-10.6
10.0
14.7
-4.2
13.7
10.0
4.0
10.0
9.4
27.2
2.9
-8.5
2.2
First Source Area: c^ - 2,972; Pj - 45 percent. Second Source Area: a 2.97; P - 55 percent.
*ESE storn number.
tNormalized runoff error - 100.0 * (predicted runoff - measured runoff)/rainfall.
Source: ESE, 1985.
Table 2. Predicted versus measured runoff for Harvest Site.
66
-------
r o.ooe -
I*
0.353
0.282
- 0.141 £
- 0.071
0.883
0 30 60
120 150 180 210 240 270 300
TIME OF DAY (hours)
0.030
<> 0.018
0.0
i 1 r 1
STORM EVENT: 4/8 - 4/18/83
i ""T 1 1 1
1.059
0.847
0.635 |
1
0.424 II
0.212
30 60 90 120 150 180 210 240 270 300
TIME OF DAY (hours)
0 30 60 90 120 150 1SO 210 240 270 300
TIME OF DAY (hours)
0.0040
0.0032
0.0024
0.0016
o.oooe
0.0
I 1 i i r
STORM EVENT: 4/23 - 4/25/83
"-.
'i ~'~'"^
\ " -:... _.
4*
i , 1111 i i
0.1130
0.0847 £
*
0.0565 i
0.0282
U..I
21 28 35 42 49 56 63 70 77 64
TIME OF DAY (houn)
Figure 5. Predicted versus measured runoff for Control Site.
t^- 0.015 -
Q
- 0.706
0.530 ~ m 0.0024
1 i
- 0.353 [^ -J 0.0016 -
- 0.177
5 10 15 20 25 30 3S 40 45 50 55
TIME OF DAY (hours)
0.035
<> 0.021
I
0.007
M
-I 1 1 1 I
STORM EVENT: 4/24 - 4/25/83
1.236
0.030
0.988 0.024
.?
0.741 -5
0.494
0.247
JO)
0.018
0.012
0.006
JO 24 28 32 38 40 44 48 52 58
TIME OF DAY (hours)
- 0.085
- 0.057
10 16 20 25 30 35 40 45 50 55
TIME OF DAY (hours)
1.0S9
STORM EVENT: 6/20/83
- O.S47
O
- 0.212
Figure 6. Predicted versus measured
67
14 21 21 35 42 49 68 63 70 77
TIME OF DAY (hours)
runoff for Harvest Site.
-------
Sawchuck, Arch Merritt, and Andrew Middleton of Peat Methanol Associates
for their support and assistance. All of the field work was performed by
Michael Tomlinson, Kathleen Ingram, Michael Geden, Bill Vogelsong, and
others at Environmental Science and Engineering, and we are thankful for
their excellent support.
REFERENCES
1. Heath, R.C. Hydrology of the Albemarle-Pamlico Region, North
Carolina, a preliminary report on the impact of agricultural
developments. USGS Water Resources Investigations, 9-75, 1975.
2. Sharpe, B, A new geography of North Carolina Vol. 2. Raleigh,
Sharpe Publishing Company, 1966. pp. 535-1114.
3. Environmental Science and Engineering, Inc. Runoff Assessment
Report. ESE No. 83-600-601, 1983.
4. Huber, W.C., Heaney, J.P., Nix, S.J., Dickinson, R.E., and Polmann,
D.J. Storm Water Management Model User's Manual Version III.
Project No, CR-805664, EPA-Cincinnati, Ohio. 1981.
5. Huber, W.C., Heaney, J.P., Medina, M.A., Peltz, W.A., Sheikh, H.,
and Smith, G.F. Storm Water Management Model User's Manual
Version II. Project No. R-802411, EPA-Edison, New Jersey. 1975.
6. Freeze, R.A. Role of subsurface flow in generating surface runoff,
Part 2: Upstream source areas. Water Resources Research,
8(5):1272-1273, 1972.
7. Eagleson, P.S. Dynamic Hydrology. McGraw-Hill, New York, 1970.
8. Bevin, K. Kinematic subsurface stormflow, Water Resources Research.
17(5), 1419-1424, 1981. ~
9. Bevin, K. On subsurface streamflow: Predictions with simple
kinematic theory for saturated and unsaturated flows. Water
Resources Research. 18(6), 1627-1633, 1982.
10. Sloan, P.G., and Moore, I.D. Modeling subsurface stormflow on
steeply sloping forested watersheds. Water Resources Research.
20(12), 1815-1822, 1984.
11. Horton, R.E. The interpretation and application of runoff plot
experiments with reference to soil erosion problems. Proceedings of
the Soil Science Society of America, 3:340-349.
68
-------
AREA-WIDE STRATEGIES FOR STORMWATER MANAGEMENT IN SWITZERLAND
CASE STUDY GLATTAL
by: Vladimir1 Krejci and WillI Gujer
Federal Institute of Water Resources and
Water Pollution Control (EAWAG)
Associated with The Swiss Federal Institutes of Technology
CH-8600 Dubendorf-Zurich, Switzerland
ABSTRACT
An area of 260 km with 230,000 inhabitants was analyzed for the feasibi-
lity of storm-water tanks to reduce the pollution during rain events. Due to
the number of parameters (sources of pollution, constructed and projected
technical measures for water pollution control), the pollution load and the
construction and exploitation costs were computed with the help of simulation
models.
Simulated results for the "today situation" agreed well with experimental
results, thus verifying the simulation models. Then, operating and investment
costs as well as residual pollutant load were simulated for different design
and operating strategies. Case study results demonstrate the inadequacy of
storm-water tanks for cost efficient water pollution control:
the physical stress on river ecology during rainy weather (flow velo-
city, bed-load and sediment-transport) can hardly be influenced by
the small size tanks utilized in Switzerland.
the chemical parameters (concentration and load of pollutants in
riverwater, DOC, NHA, PO,,, ...) are only slightly reduced with the
exception of substances In street surface runoff (e.g. Pb).
Storm-water tanks may however help to solve local esthetic problems.
The study demonstrated, that programs which reduce pollutant sources
(e.g. lead in gasoline) or slow down and reduce surface runoff (dispersed
retention and improved infiltration) are more efficient than a strategy of
uniform and area wide application of stormwater tanks.
1 Visiting Research Engineer at Department of Civil Engineering, University of
California in Davis (1984-1985)
69
-------
INTRODUCTION
Stormwater Management and Control is one of the most complicated aspects
of water pollution control. The large number of parameters makes the study of
this problem very complex and expensive. Furthermore, conditions within any
given watershed vary greatly during wet weather, consequently, local con-
ditions are very important and the results and conclusions are not easily
transferred to other areas. This is another reason why research in this area
is not very attractive and why most of the studies performed deal with only
one part of the much more complex problem.
In Switzerland combined sewer systems (municipal sewage plus stormwater
runoff) predominate over separate sewer systems. The hydraulic capacity of
wastewater treatment plants is generally twice the maximum dry weather flow.
Due to limited hydraulic capacity, excess stormwater is discharged directly to
the receiving water. As a result, several times a year untreated, combined
sewage is discharged to the receiving water.
In Switzerland two types of stormwater tanks are used as an element of
combined sewer systems which have a high frequency of overflow. The first
type of tank stores the "first flush'* and, after the rain event returns the
water to the treatment plant. The second type provides clarification (and
some storage) before discharging to the receiving water. The sediments are
returned to the treatment plant after the rain ceases.
In Switzerland stormwater tanks are generally concrete; ranging in speci-
fic capacity from 15-25 cubic meters per hectare of impermeable surface. They
can store approximately 2 mm of rainfall.
The main purpose of this study was to explore the benefits gained from
the use of storage tanks in combined sewer systems in relation to the high
costs associated with construction of the tanks.
STORMWATER RUNOFF AS WATER POLLUTION PROBLEM FOR WATER COURSES
Studies have shown that during storm events, stormwater runoff results
in an increase in the flow and the pollution load in receiving water. The
runoff comes from several different sources (Figure 1).
As a consequence of higher flow-velocity, shear stress, transport of
sediments, turbidity, change of the concentration of chemical substances and
disposal of coarse substances, paper, etc. a stormwater pollution problem ari-
ses. The effects of the stormwater runoff can be divided into three areas
ihysical (mechanical) effects (flow-velocity, shear stress,
ransportof sediments,erosion, turbidity) influence the water
course as an ecological environment and also affects groundwater
infiltration;
70
-------
POINT SOURCES ^NON POINT SOURCES*!
Domestic and industrial
waste water
WASTEWATER
COMBINED
! SYSTEM
I
Deposit in
wastewater
network
Pollution of
rain
Runoff of
urban
area
r
Runoff of
agriculture,
forest, etc.
Receiving water
Wastewater treatment possible
Wastewater treatment non possible
Figure 1. Sources and transport of pollution load during wet weather.
chemical effects (change of concentration of different chemical
substances as DOC, Nfy, Oxygen, Phosphorus, heavy metals, etc.) is
significant e.g. for interaction between chemical and biological
processes in receiving water, for raw water quality for
groundwater-infiltration, drinking water treatment, etc.
esthetic and health effects (sludge, coarse substances, paper,
germs, etc.) can cause local problems (swimming, recreation).
The relative significance of these three aspects of the problem and the
results of various methods of stormwater pollution control in Switzerland are
demonstrated using the Glatt River Watershed Case Study.
GLATTAL-AREA
The Glatt is a small river in northeastern Switzerland. The river has an
average discharge ranging from 3.3 m*/s at the point of outflow from Lake
Greifensee to 8.7 rn^/s at the point of confluence with the Rhine. The
watershed is approximately 260 km2, and is heavily urbanized (about 900
71
-------
persons/km2). About 40 km2 (16% of total area) are impervious surfaces.
There are 13 wastewater treatment plants in the area and nearly 100% of the
wastewater receives secondary treatment (Figure 2).
10km
boundary of Glattal-Area
boundaries of subareas
Check Point in river
A Sample Point in river
O KA (Name) : Wastewater Treatment
Plants and its relative capacity
Figure 2. Glatt River Watershed (Switzerland). Survey of catchment of River
Glatt with the important tributaries and survey of the existing
wastewater treatment plants in the areas.
In spite of nearly complete wastewater treatment the quality of the Glatt
River is poor. In 1976, the emphasis of Swiss regulation changed from one of
setting effluent standards to one of regulating receiving water quality.
With regard to esthetic, biological and chemical criteria, the Glatt River
does not satisfy the standards set by Swiss Law.
In response to this situation, a two phase study was initiated to deter-
mine possible pollution control options. The first phase of the study cen-
tered on upgrading dry weather treatment to the tertiary level (nitrification,
chemical precipitation and coagulation, partly also filtration). The second
72
-------
phase centered on possible stormwater runoff pollution control measures.
During the second phase, four major questions were:
1. What is the volume and duration of stormwater runoff in the Glatt
River and in its important tributaries?
2. How large are the pollution loads from the point and non-point sour-
ces during stormwater runoff?
3. What are the effects and the costs of stormwater pollution control
measures, especially those of overflow storage tanks in a combined
wastewater network?
4. What is the priority of different measures of water pollution
control (dry weather and stormweather) in this area?
THE EFFECT OF WATER RUNOFF ON RECEIVING WATER IN THE GLATT RIVER
The effect of stormwater runoff on the Glatt River are based on detailed
analyses of:
basic hydrologic conditions;
wastewater collection networks;
wastewater treatment plants in the area; and
results of chemical and biological analysis in sewers, treatment
plants, and receiving waters.
The costs of management alternatives and the resulting pollution loads
were calculated with an EAWAG simulation model and the results verified
against existing data. The verification and the sensitivity analysis proved
the feasibility of the simulation model to answer the questions posed above.
PHYSICAL EFFECTS
As a consequence of agricultural drainage, channelization of the Glatt
and its tributaries, and development of urban and traffic areas, the discharge
and the flow-velocity in the Glatt is significantly increased during wet
weather periods.
Figure 3 shows the distribution of discharge and flow-velocity in the
river Glatt over one year. Several times a year the river bed is loosened,
resulting in drastic changes in the environment for the river biota as well as
infiltration into the groundwater, etc.
73
-------
f
o>
ref
O
Q
40
35
30
25
20
15
10
5
V)
Si^r
Q>2.
re >
-
- 2.0
1.8
-
1.6
-
1.4
-
1.2
' 1.0
0.8
0.6
0.4
a>
i',
100
90
80
. 70
60
50
40
30
20
10
Gravel
Size of
particle
coarse
Sand
fine
Sand
15mm
1mm
Jan. Feb. March April May June July Aug. Sep. Oct. Nov. Dec.
Figure 3. Depth, flow velocity, discharge and the approximate sediment
transport in the river Glatt (1974).
The relative significance of the different sources of the flow and tur-
bidity in the river Glatt during wet weather is shown in Table 1. As can be
seen, the greater part is derived from non-point sources.
TABLE 1. SIGNIFICANCE OF THE DIFFERENT SOURCES OF THE FLOW AND TSS LOAD
(TURBIDITY) IN THE RIVER GLATT DURING WET WEATHER
Source of Load
Wastewater treatment plants
Stormwater overflows (combined
syst.)
Stormwater runoff (separative
syst.)
Non point sources
average load during wet
weather in %
Flow TSS
22
10
6
62
15-25
>50
74
-------
^ The results of this study showed that during stormwater runoff the
ecological systems" in the river suffer from considerable physical stress.
The high velocity of the flow causes an interruption of natural development
and an unnatural selection of river-biota.
The high erosion of agricultural surfaces and the runoff from the
impermeable urban and traffic areas causes a high turbidity in the receiving
water and influences sunlight penetration in the river (Figure 4).
Duration curve of Turbidity
in 1972-1974
w
[Duration of I
p wet weather 4-
I period |
90 180 270
Duration of dry weather period
365
Figure 4. Duration curve of concentration of suspended solids (TSS) in
Chriesbach (tributary of Glatt River).
CHEMICAL EFFECTS
The following chemical parameters were used to represent groups of che-
mical substances:
Dissolved Organic Carbon (DOC) - represents dissolved che-
mical substances with their origins from domestic and
industrial wastewater, runoff from impermeable urban
areas, and non-point sources;
Ammonium - represents dissolved substances with their ori-
gins mainly from domestic wastewater;
Total-Phosphorus - represents partly dissolved substances
from domestic wastewater;
Lead - represents small particulate substances from the
runoff from impermeable urban and traffic areas.
Figure 5 shows the relative load of different sources during 13
hypothetical rains which statistically represent the various storm events
experienced in one year. The high relativer load from point sources is
striking.
75
-------
Q (flow)
DOC
- P
Pb
Rainfall
w Uk w
Mu»oiovS««?S --K.MM5«o5»»^i: annual frequency [-]
-K.^-. ^ JM _ £ _-
"Vu """«-£ "-"»S duration |hrs ]
po.Sja.BS^.uaaB .s,5i..S.ca»«us»5 p.s»a_ss^»o,s<..s average intensity [l/s-h]
Load origin:
B lake outflow
Q non point sources
Q] treatment plants
surface water sewer
(separate system)
storm overflow
(combined system)
Runoff
^ -"SI «-*!"£ .*-*** -*'-ol-t'-01-*' .*,*-".*'* PPppPppppoooo ooooooooooooo ooooooooooooo
*
I**!** ******««
§«(> *. «IX « <» to> » W W k)
u<*^ua*uo0aoQ
pppoppoppppp
OOOOOOOOOOOOO »>j<>ai Cj«>-'>>'QBnoO
0000 000000$ 00 W^1UI*-«0«»0>F=>M»^
ooooooooooooo oooo ooooooooo
Figure 5. Calculated wet weather pollutant load and concentration in the river
Glatt during 13 different rainfalls (percentage from important sour-
ces).
Figure 6 shows the annual pollution load in the Glatt River versus the
Chriesbach River. The Chriesbach River was chosen for comparison due to its
small watershed area (approximately 16 km2) as compared to the Glatt River.
The Chriesbach area is also heavily urbanized. In both Rivers, dry weather
sources represent the major sources of the selected parameters, except lead.
Figure 7 shows typical flow concentration changes with time during a sum-
mer storm event in the river Glatt. The concentration of particulate substan-
ces (represented by TSS and heavy metals tied to TSS) increase greatly. The
concentration of substances from domestic and industrial wastewater
(represented by DOC. Nfy-N, and P) stayed nearly constant, some other substan-
ces (Cl~, Ca2+, Mg2*) were diluted.
We can assume that these relatively small, short-term changes in con-
centration of DOC, ammonium, nitrate and phosphorus have negligible effects on
water quality. However, the concentrations of heavy metals reach signifi-
cantly higher values than during the dry weather period. Although the greater
part of the heavy metals load is in particulate form there may still be toxic
effects.
76
-------
Q (Flow)
100
I
90-
80-
70-
60-
50
<^
)0
20-
10-
0-^J
Stormoverflow
(combined system)
DOC
I
NH4-N
sj w O
u> p w
o g o
ges.-P
Chrinbidi GUll
2 I
Pb
TOTAL LOAD IN KG ( Flow
* ^
S 5
Xui 6
w ^i
O O O
Surface water sewer
(separate system)
Wastewater treatment
plants
QNon- point sources
I Outflow lake
Figure 6. Calculated annual load in the river Glatt and in Chriesbach
[tributary of the Glatt), percentage from important sources.
24 08 16 24 08 Time
16 24 08 16 24 08
16 24 08 16 24 08
16 24 08 16 24 08 Time
Figure 7. Typical flow and concentration charges with time during a summer
storm event in the river Glatt (July, 19-20, 1976).
77
-------
During wet weather periods other substances also get into receiving water
(e.g. many different hydrocarbons) from different sources. We could not esti-
mate the amount of these substances with the simulation model. However, we
have some hints about the significance of these substances in groundwater, for
example, as shown in Figure 8.
90
373.00
90-
80
A)
18 20 22 0 2 4 6 8
Time
1000
ng/l
100
10
o
o
1
10000
ng/l
1000
£ 100
3
10
ng/l
1000
L L 1 I I I I
I I 1
2-HEPTANON
1,3-XYLOL
CH3
1,4-DICHLORBENZOL
Groundwitttr
I I I
18 20 22 0 2 4
8 10 12
Time
Figure 8. Time varying flow and concentration in the river Glatt and in
groundwater (distance of groundwater sampling to river bank ~2.5m)
during stormwater runoff.
78
-------
Another shortcoming of the model is it's inability to accurately model
suspended solids (TSS) and particulate organic carbon (POC), due to the lack
of necessary basic information (the load from non point sources versus the
load from deposits in sewer lines).
The short-term change in dissolved oxygen concentration in the river
Glatt from 9-10 mg 02/1 to 4-6 mg 02/1 several times a year during a storm-
water event may result from the DOC and POC load during wet weather. However,
the evidence is not conclusive. On one hand, many easily decomposable organic
substances reach the receiving water from the sewage overflows; on the other
hand the anaerobic deposits in the river are stirred up. It was not possible
to simulate this problem.
ESTHETIC EFFECTS
The physical and chemical loads during stormwater runoff influence large
areas of the water course. Esthetic problems occur only in the local area (in
the vicinity of outfalls). In the Glatt-Area we are aware of only a few local
problems.
POSSIBLE MEASURES FOR STORMWATER POLLUTION CONTROL
The effects of four different stormwater pollution control strategies and
their costs were investigated:
reduction of pollution load from individual sources;
changes in the character of runoff;
treatment of storm overflow water (in the combined wastewater
network); and
enlargement of the hydraulic capacity of wastewater treatment plants
to accommodate wet weather flows.
REDUCTION OF POLLUTION LOAD FROM INDIVIDUAL SOURCES
Table 2 demonstrates, that the substitution of phosphorus in detergents
and the removal of lead from gasoline have about the same effect on water
quality in the River Glatt as a very large increase in storage of stormwater
overflow.
79
-------
TABLE 2. POLLUTANT LOAD DURING WET WEATHER: OVERFLOW STORAGE VS. REDUCTION
OF POLLUTION FROM INDIVIDUAL SOURCES
RAINFALL:
average intensity: 12.5 Jl/s-ha
duration: 3 hrs
PHOSPHORUS
Phosphorus in detergents , +
volume of overflow-storage m /ha
Total Load
Load of overflow
Outflow of treatment plants
LEAD
Lead in Gasoline _ ^
volume of overflow-storage m /ha
Total Load
Load of overflow
Load of separate sewers
Phosphorus
0
306
111
79
0
46
34
6.5
and Lead Load in River
Glatt in kg
yes
35
281
66
100
yes
35
34
18
6.5
no
0
205
59
31
no
0
5.2
1.8
0.0
35
191
32
43
35
4.3
0.7
0.0
of impermeable surface
CHANGES IN THE CHARACTER OF RUNOFF
There are several possibilities for changing runoff:
1. The impermeable surface can be reduced (for example if stormwater from
roofs and parking areas is infiltrated). Table 3 shows the effect of a
20% reduction in the runoff-coefficient. Despite the reduction in pollu-
tion load, a relatively larger part of the residual stormwater will be
treated in the wastewater treatment plants.
2. A periodical flooding of flat roofs, not intensively used traffic sur-
faces, and parking areas, and the use of storage capacity in the sewage
system could retard the runoff-discharge. The effect of these changes
can not be calculated yet, but it is certain that all retardation
measures have a positive influence on the receiving water.
80
-------
TABLE 3. POLLUTANT LOAD DURING WET WEATHER: OVERFLOW STORAGE VS. REDUCTION
IN THE RUNOFF-COEFFICIENT
RAINFALL:
average intensity: 12.5 £/s-ha
duration: 3 hrs
Runoff-Coefficients
Volume of overflow-
storage m^/ha*
Total Load
Q m3
DOC kg
NHA-N kg
Tot.P kg
Lead kg
Load of overflow
Q m?
DOC kg
NH.-N kg
Tot.P kg
Lead kg
Existing
0
594 000
3 168
395
306
46
92 500
1 420
160
111
34
runoff
35
594 000
2 760
365
281
34
59 400
802
102
66
18
Diminuation of
surface
0
560 000
2 820
363
285
36
70 700
1 125
133
93
26
impermeable
of 20%
35
560 000
2 420
332
260
25
37 000
513
73
46
11
of impermeable surface
3. The drainage of urban areas through separate systems instead of combined
systems would significantly reduce pollution in receiving waters (Table
4).
No one of these measures could be realized in a short time. However,
long-term, systematical inclusion of these measures during the process of
renewing old systems will result in a significant contribution to the protec-
tion of receiving water.
TREATMENT OF OVERFLOWS
There are also several possibilities for the treatment of stormwater
overflows. All these alternatives tend to eliminate coarse particulate
substances. Dissolved substances (DOC, Nfy-N and a large portion of the
phosphorus) can be reduced only through the storage and subsequent treatment.
For the treatment of stormwater overflow in Switzerland different kinds
of storage tanks are used. These tanks can significantly affect the pollution
load from stormwater overflows, (depending on rainfall, storage tank volume,
etc.). However, the effect on the total pollution load in the receiving
water is minor (Figure 9 and 10).
01
-------
TABLE 4. POLLUTANT LOAD DURING WET WEATHER: OVERFLOW STORAGE VS.
SEPARATE SEWAGE SYSTEM
RAINFALL:
Average intensity: 12.5 i/s
duration: 3 hrs
separate system
in the area
volume of overflow-
storage m /ha*
Parameter
Total Load kg
Load of overflows
of separate system
outflow of treat-
ment plants
of non-point source
(agriculture, etc)
ha
existing situation
(~ 25*)
35
DOC
2764
802
193
949
s 819
NH4-N Tot.P Pb
365 281 36
102 66 19
23 66
172 100 6
69 109 2
100%
None
DOC NH,-N Tot.P Pb
2061 226 193 26
693 82 23 23
549 75 61 1
819 69 109 2
of impermeable surface
POLLUTANT LOAD FROM STORM OVERFLOWS TOTAL POLLUTANT LOAD
DOC
During 3 different
rainfalls
100
50-
(£ of point and nonpomt sources in study area
During 3 different Annual
rainfalls load
50-
0
STRATEGIES (VOLUME OF STORAGE TANKS) 0 0 m3 / ha (of impermeable surface}
015m3/ha
H 35m3/ha
RAINFALLS: AVERAGE INTENSITY A= 35 l/s.ha BM2.5l/s.ha C = 9l/s.ha
DURATION A= Ihr B=3hrs C = 2hrs
Figure 9. Effect of stormwater overflow storage tanks reduction of DOC-Load in
the river Glatt (similar effects for NH^-N, Tot. P and Lead).
82
-------
POLLUTANT LOAD FROM STORM OVERFLOWS
During 3 different Annual
TOC
rainfalls
load
POLLUTANT LOAD FROM ALL POINT SOURCES
(TREATMENT PLANTS AND STORM OVERFLOWS)
During 3 different Annual
rainfalls
100i
50-
n.
-
vl
7
!
r
-
<
/
:'
;
|
ABC
load
% l
TOC-DOC
so-:
Participate
org. Carbon
o-
STRATEGIES (VOLUME OF STORAGE TANKS) 0 m3/ha(of impermeable surface)
0 35 m3/ ha
RAINFALLS: AVERAGE INTENSITY A--35 l/s.ha B=12.5l/s.ha C = 9l/s.ha
DURATION A= Ihr B=3hrs C = 2hrs
Figure 10. Effect of stormwater overflow storage tanks on reduction of TOC and
POC load in point sources (wastewater network and plants) in the
Glatt River Watershed (particular org. carbon (POC) = TOC-DOC).
ENLARGEMENT OF THE HYDRAULIC CAPACITY OF WASTEWATER TREATMENT PLANTS
The hydraulic capacity of wastewater treatment plants in Switzerland is
generally twice the maximum dry weather flow. An enlargement of the hydraulic
capacity of treatment plants requires an enlargement of their primary and
secondary settling tanks.
Calculation of reduction in pollution load versus the costs of enlarge-
ment shows, that in watersheds with a small specific dry wastewater flow per
unit of impermeable surface (for example in rural areas), storage tanks are
more cost-effective than enlargement of treatment plants. For the large city
areas the opposite is true.
In treatment plants with a capacity in the primary settling tanks of 3-5
times the dry weather peak, and with overflow after mechanical treatment, a
good removal of particulate substances can be achieved. However, at the same
time a higher load of dissolved substances will be displaced from the primary
settling tanks to the receiving water (Figure 11).
83
-------
mg/l
mg/l
3LJ
40
30
20
10
0
1
g/s
160
120
80
40
0
r
i
DOC- CONCENTRATION
1 ,_ OVERFLOW AFTER
1 f PRIMARY SETTLING
1
1
1
INFLOW
OUTFLOW -
7 18 19 20 2
Time
1 i
_.. DOC -LOAD
j
U OVERFLOW AFTER
"I PRIMARY SETTLING
I
|
Is-' '
INFLOW
OUTFLOW
1 18 19 20 21
Time
ouu
400
300
200
100
0
1 1
kg/s
3.0
2.5
2.0
1.5
1.0
0.5
0
1
-
1 i 1
f~| TSS -CONCENTRATION
L"
r.fl
LJ
p"! INFLOW
^__ ^^^-.^^
..) ""' "VOVERFLOW AFTER
i | PRIMARY SETTLING
i i
|.1 _ J f OUTFLOW
7
18 19 20 21
Time
-
7
JL,
r>
y^
\\
1 1
TSS -LOAD
-OVERFLOW AFTER
/PRIMARY SETTLING
" 1
[ I . INFLOW
"i=t _ OUTFLOW
18 19 20 2
Time
Figure 11. In treatment plants with a primary settling tank capacity 3-5 times
dry weather peak and with overflow after mechanical treatment a
good removal of participate substances can be achieved. At the
same time a higher load of dissolved substances will be from the
primary settling tanks.
COST EFFECTIVENESS OF DRY WEATHER VS WET WEATHER POLLUTION CONTROL
Table 5 shows the costs and the resulting load reduction of wastewater
treatment plants versus storage tanks in the Glatt River area.
The existing treatment plants are designed for removal of the suspended
solids and organic substances (secondary treatment). The proposal for
advanced wastewater treatment in the area will bring a relatively efficient
and favorable removal of ammonium and phosphorus.
04
-------
TABLE 5. THE COSTS AND EFFECTS OF WWTP VS. STORMWATER OVERFLOW STORAGE
MEASURES
Existing
treatment
Extended
TOTAL
Stormwater
storage ta
15 m3/ha**
35 m3/ha**
Cost of Cost
investment annual
Mio Fr. Mio Fr.
122 9.9
47 3.6
169 13.5
nks
12 0.6
26 1.3
Removal in
t/a
TSS DOC NH4-N Tot.P
9000 1570 180 110
350 150 365 240
9350 1720 545 350
220 10 2.0 1.4
500 22 4.4 3.7
Specific costs
FrAg
TSS DOC NH4-N Tot.P
1.1 6 55 90
10 24 11 15
1.4 7.8 25 39
2.7 60 300 400
2.6 60 300 400
^ specific costs: 1 Fr = 0.5 U.S. $
specific volume of an impermeable area
SUMMARY AND CONCLUSIONS
Table 6 shows the different measures and their effects on stormwater
pollution control.
The table shows, that we can not decrease the physical/mechanical effects
with conventional measures. However, we have the possibility of reducing both
the chemical as well as the esthetic effects on receiving waters.
In many water courses the physical effects during stormwater runoff cause
the most important stress on biota in the receiving water. Conventional
measures in the sewage network and in treatment plants are not suitable to
reduce this stress.
In practical terms, control of pollution from stormwater runoff has two
aspects:
1) chemical loads and resulting effects which impact the entire reach
of the river; and
2) local problems (well defined esthetic and biological effects of
sludge, sediments and coarse substances).
The chemical load during stormwater runoff in Switzerland could be signi-
ficantly reduced with long-term measures (elimination of causes, retention,
infiltration, etc). All these long-term measures also have a significant
impact on the esthetic effects and partly reduce the physical stress.
85
-------
TABLE 6. EFFECT OF DIFFERENT MEASURES ON THE REDUCTION OF POLLUTANT LOAD
IN WATER COURSES DURING STORMWATER RUNOFF
Measures
Physical
Effect of Pollution
Chemical Esthetic
Reduction of Pollution Load
from individual sources (P,Pb)
Changes in the character of Runoff
- Retardation (flooding)
- Storage tanks
- Reduction of impermeable
surfaces
- Installing separate sewer
system
Treatment of stormwater runoff
- Enlargement of treatment plants
- Treatment of overflow
0
0
0
+ positive effect
0 no effect
- negative effect
The local problems will have to be identified and defined, then solved
with specific and suitable measures.
ACKNOWLEDGMENTS
We thank Professor G.T. Orlob and Mr. Mark Tumeo for helpful discussions
and their support in the final lay-out of this publication. Greatfully
acknowledged is the effort of Dinah Pfoutz for typing the manuscript and Heidi
Bolliger for drawing the figures efficiently.
1. Krejci, V. und Gujer, W.: Probleme und Massnahmen des Gewasserschutzes
bei Regenwetter, Wiener Mitteilungen, Band 53, 1984.
2. Dauber, L. und Novak, B.: Quellen und Mengen der Schmutzstoffe in
Regenabfluss einer stadtischen Mischkanalisation, EAWAG Dubendorf-Zurich,
1983.
3. Gujer, W. et al.i Von der Kanalisation ins Grundwasser, Gas-Wasser-
Abwasser, Mr. 7/1982.
4. Roberts, P.V.: Pollutant loadings in urban Stormwater IAWPR - Workshop,
Vienna, September 8-12, 1975.
86
-------
THE IMPACT OF "SNOW" ADDITION ON WATERSHED ANALYSIS USING HSPF
by: Stan Udhiri1
Mow-Soung Chengl
Richard L. Powell1
ABSTRACT
The hydrological Simulation Program - FORTRAN (HSPF) is a
powerful tool for in-depth analyses of the hydro!ogic and associated
water quality processes in a river basin. The model use requires a
thorough understanding of the various algorithms that make-up the
model. One of the most complicated algorithms is the section referred
to as "SNOW"; not only because of the massive amount of calculations
involved, but also due to the several additional meteorologic time
series which are usually not easy to obtain. The "SNOW" section in
the HSPF model simulates the hydro!ogic processes involved in the
accumulation and melting of snow and ice on a land segment. It is
important, as a significant amount of runoff is derived from snow,
especially in the northern part of the United States.
The model was used to assess the hydrologic and water quality
impact of planned development within Piscataway Creek in southern
Prince George's County, Maryland. Prince George's County is located
in the Washington Metropolitan area which has a humid, continental
type of climate with the temperatures falling below the freezing
point, approximately 90 days each year. During the study, one of the
questions dealt with was wh(at effect the inclusion of the "SNOW"
option would have had on the study result. To address this question,
the following four cases were studied: (1) Develop and calibrate the
mode! without SNOW; (2) Add SNOW to case (1) and calibrate SNOW
parameters only; (3) Develop and calibrate model Including SNOW
option; and (4) After the model in case (3) is well calibrated, remove
SNOW calculation.
1 Water Resources Engineers, Maryland-National Capital Park and
Planning Commission, 14741 Governor Oden Bowie Drive, County
Administration Building, Upper Marlboro, MD 20772.
87
-------
The results of the analyses, including both aspects of hydrology
and water quality, for all the above four cases are presented here.
INTRODUCTION
This study is intended to evaluate the impact of snow/ice
accumulation and melting processes on Piscataway Creek in the Washing-
ton Metropolitan Area using the HSPF model. Piscataway Creek is
located in Prince George's County, Maryland. The Piscataway Creek
Study was originally undertaken by the Metropolitan Washington Council
of Governments (COG) in 1983 (Reference 1) to investigate the stream's
water quality and to identify stormwater and nonpoint source
management needs within the watershed. The COG watershed model was
developed as a water resource planning tool for the'evaluation of land
use planning alternatives and applied management practices, and for
the assessment of their impacts on stormwater runoff and water
quality. However, the effects of accumulation and melting of snow/ice
were ignored in the COG study.
From long-term statistics, the average number of days that the
temperature is 32°F or less in the watershed is approximately 90 days
every year (Reference 2). For an area where 25 percent of the year
the temperature is 32° F or below, the snow/ice accumulation and melt-
ing might be of some significance. It is known that from a hydrologic
point of view, the major difference between snow and rain is that snow
is not transformed to surface runoff immediately resulting in (1) a
less peak flow rate and a later peak time; (2) a higher surface
detention capacity; and (3) a higher soil moisture content when snow
melts. Furthermore, the phenomena of surface pollutant accumulation
and removal/washoff are vastly different between snow and rain. For
example, most of the nonpoint pollutants are transported from the land
surface in the early portion of rainfall events; while most of the
pollutants remains on the land surface during a snow storm (covered by.
snow) until the accumulated snow starts to melt. Also, a snow cover
reduces soil detachment and washoff and therefore reduces soil loss.
On the basis of a possible significant effect on "SNOW" addition,
it was decided that a sensitivity analysis on the effects of snow/ice
acumulation and melting is important. Four cases were involved in
this study. The first case is to recreate COG's Piscataway Creek
Watershed model using the same parameters, meteorologic time series,
and input sequence as those of COG. In the second case, "SNOW" relat-
ed parameters and time series were added to the input sequence of case
1 and Time Series Store (TSS) files. At this point in time, only
"SNOW" related parameters were adjusted for calibration purposes.
Other parameters were kept the same. The next case (case 3) was the
adjustment of other parameters as well. For example, monthly upper
zone storages were modified to a more realistic profile; INFIL was
08
-------
reduced; Manning coefficients were changed to reflect the seasonal
effects on vegetation, etc. The final case (case 4) was the removal
of "SNOW" calculation from case 3. The primary difference between
case 1 and case 4 is that the "monthly upper zone storages" for the
winter period were manually increased in case 1 to reflect the effect
of snow accumulation on the land segment, although the snow accumula-
tion might not always be on the ground for the entire winter period.
As a result, the simulated storm runoff for case 1 would be smaller
than that for case 4 in the winter. The results from the four cases
were then analyzed.
The Piscataway Creek watershed encompasses a complex mixture of
land uses from heavily urbanized areas at the upstream portion of the
watershed to rural, forested and agricultural areas downstream. Cur-
rent land uses in the watershed include: agricultural, which occupies
15 percent, residential and commercial developments at 15 percent,
with the rest of the watershed primarily forest and open unforested
land.
Piscataway Creek has a relatively well-defined valley, with
channel elevations ranging from sea level to 250 feet above mean sea
level. The watershed lies within the Atlantic Coastal Plain Province.
Soils are principally sandy loams and silt loams derived from the
underlying sedimentary materials (Reference 3). The watershed has a
humid, continental type of climate by reason of over 40 inches of
average annual precipitation and its location in the middle latitudes
where the general atmospheric flow is from west to east across the
North American continent. The long-term mean annual temperature in
the watershed is approximately 54° F (Reference 4).
GENERAL MODELING PROCEDURES
The Hydrological Simulation Program views the processes within
the hydrologic cycle as a series of mathematical representations with
the physical characteristics of the watershed entered as parameters.
The programming is divided into four load modules: TSSMGR, PERLND,
IMPLND, and RCHRES. TSSMGR performs data management for any time
series stored in the system. PERLND and IMPLND simulate snowpack and
soil profile processes and calculate continuous soil moisture/evapo-
transpiration, groundwater accretion, sediment detachment and washoff,
soil and water temperatures, and water quality constituents genera-
tion. RCHRES assembles and routes the information (inflow) determined
by PERLND and IMPLND, as well as certain lateral inflows and outflows,
through the drainage network, including reservoirs.
Values of several required parameters for PERLND, IMPLND, and
RCHRES can be determined directly from physical characteristics of the
watershed. The parameter values for land slope and land cover, for
example, can be determined directly from topographic and land use
89
-------
maps. FTABLES for channel or reservoir routines can be calculated
though hydraulic analyses from channel geometry. However, values for
some parameters, such as nominal soil moisture storage and infiltra-
tion rates are usually obtained through the calibration process by
first assuming an initial value for each of these parameters. The
simulated streamflows and water quality constituents are compared with
observed values at sampling sites. If there is a substantial differ-
ence between the simulated and recorded data, the parameters have to
be changed and the procedure repeated.
The explicit purpose of model calibration is to adapt the HSPF
model to the hydrologic, meteorologic and physiographic characteris-
tics of the watershed. The process essentially entails the adjustment
and fitting of varius parameters and constants used in the mathemati-
cal expressions. This fitting procedure begins with the "LAND" phase
{PERLND and IMPLND) of the hydrologic cycle and then proceeds through
the channel/reservoir routines.
MODEL INPUT DATA
The application of HSPF watershed model requires a great deal of
data. This data, which will be discussed in detail in subsequent
sections, were scrutinized before input to the model and then re-eval-
uated as unusual results developed during the calibration process.
This data includes meteorologic time series and steady state data.
Meteorological time series data are critical inputs for both
hydrologic and water quality simulation. All hydrologic simulations
of runoff require precipitation and potential evapotranspiration data.
Hydrologic studies which simulate snowmelt and water quality studies
require additional time series data for air temperature, wind speed,
solar radiation, cloud cover, and dewpoint temperature. Plankton sim-
ulation requires solar radiation data. Wind speed may be required for
simulation of dissolved oxygen (Reference 5). The detailed descrip-
tion of the meteorological time series is provided in Reference 6.
The steady state data include land usage (land cover), soil type,
surface slope, channel slope, and channel geometry. These data are
used to establish the values of the model parameters. If the physical
characteristics of the watershed remain the same, the parameters are
constant over the simulation period. However, if substantial changes
in physical characteristics, such as land uses, occurred during the
simulation period the corresponding changes would be made in the model
to reflect the changes. The detailed information of the steady state
data is also described in Reference 6.
The entire 39.5 square miles of drainage above the USGS stream
gauge at Pi scat away was divided into 7 channel reaches. Each reach
represented an Idealized drainage path through a tributary area and
was assumed to have relatively constant hydraulic properties through
its ^length. Each reach could accept: runoff from Its tributary area;
90
-------
inflow from upstream reaches; and municipal and/or industrial
discharges, if any. At the same time, both consumptive and noncon-
sumptive withdrawals might occur from each reach to simulate
irrigation or water supply. The resulting net runoff or water pollu-
tants from each reach was then discharged to the next downstream
reach. A function table (FTABLE) was created using the geometric data
of each channel reach with the Manning Equation. The FTABLE is a HSPF
input used to represent the hydraulic and geometric characteristics of
each reach.
CALIBRATION RESULTS
The general calibration procedures of the HSPF model are
relatively standardized. The detailed procedures and guidelines are
presented in Reference 5 and are not repeated here. The following
sections present the calibration results from each of the four cases. ,
SOURCES OF ERROR
The ability of the hydrologic model to produce results that
correlate well with the recorded data involves three considerations:
model error, data error, and calibration error. Model error results
when one or more of the mathematical relationships do not adequately
describe the prototype process. The results of numerous applications
and tests indicate that this error for the HSPF model is insignifi-
cant.
Data errors are of two types: measurement and random.
Measurement error occurs when the recording instrument malfunctions or
the observer misreads the instrument and no value (or an incorrect
value) is recorded. Every attempt has been made to correct these
errors but some residual errors always remain. Random errors occur in
may ways. The most common random errors occur in the measurement of
climate conditions and particularly in the measurement of precipita-
tion. While these discrepancies are termed "errors", they are actual-
ly variations which do in fact occur 1n the real world. A single rain
gauge rarely represents the true precipitation over a watershed. Even
the use of multiple rain gauges does not insure precise representation
of the spatial and temporal variations of rainfall and snow. The
hourly precipitation time series used in this study was developed
through the application of the National Weather Service Mean Areal
Precipitation program (Reference 7) using the recorded data at thir-
teen nearby gauging stations. This composited hourly precipitation
time series might not truly represent the actual precipitation for the
entire watershed. Also, the stage-discharge relationship for the
stream flow gauging station might not accurate, especially during high
flood stages. For example, the maximum discharge of 5,000 cfs record-
ed at the Piscataway Creek gauge on September 26, 1975 was estimated
91
-------
from rating curve extended above 1700 cfs. Furthermore, required
solar radiation information is not available for all areas of the
State of Maryland. Several assumptions were therefore necessary to
generate this information from "cloud cover" data recorded at Balti-
more-Washington International Airport. Some errors might have been
introduced into the model, from this data.
Calibration error occurs when incorrect parameter values are
chosen. Such an error can lead to persistent bias where simulated
flows are too high or too low, or it can lead to sporadic bias when
the conditions of an infrequent phenomenon are misrepresented. Cali-
bration error can be minimized by comparing simulated and observed
information which has short time intervals but long-term simulation
periods.
Very often in river basin simulation, a great deal of time is
expended in an effort to match recorded data which are in themselves
questionable. Therefore, at the onset of this calibration process,
the integrity of all the data was reviewed. The two data series of
prime importance in any hydrologic model are obviously precipitation
and stream flow. Just as obvious is the need to scrutinize these
records thoroughly so that weak or misleading data can be identified.
Figure 1 depicts the spatial variation in the monthly
precipitation for September 1975, over Prince George's County,
Maryland. The isohyetal lines in this figure were drawn from the
information recorded by more than 50 voluntary observers. If fewer or
more gauges had been used, the isohyetal would have been presented
differently. This is a result of the variation in precipitation which
exists over the watershed. Thus, it can be seen that an infinite
number of rain gauges would be required for the definition of the
spatial and temporal variation. This, of course, is practically
infeasible.
The accuracy of streamflow data depends primarily on (1) the
stability and accuracy of the stage-discharge relationship, and
(2) the accuracy of observations of stage and the interpretation of
records. The USGS rates their streamflow data according to the
following criteria:
Excellent = about 95 percent of the daily discharges
are within 5 percent of the true values
tiood = within 10 percent
Fair = within 15 percent
Poor = less than "Fair" accuracy
The accuracy ratings of the Piscataway Creek USGS gauge used for the
calibration-verification period are reported as "Fair", in general.
With this stream flow rating and the spatial variation in precipita-
i
92
-------
12
Figure 1. Monthly Precipitation (in.) over Prince George's
County, Maryland in September 1975
93
-------
tion, attaining a perfect match with recorded stream flow is nearly
impossible. It should therefore be recognized that the discrepancies
were not necessarily due to modeling errors. They could, in fact,
have been due to errors in the recorded input data.
On this basis, the parameter adjustments for the hydrologic
calibration were based on the results of the following analyses:
0 Annual Water Balance
0 Monthly Water Balance
0 Mass Curves
0 Comparison of Monthly Runoff
o - - - - - *
o
Comparison of Daily Runoff Duration Curves
Comparison of Hydrographs
The above methods were used to compare the simulated data with the
recorded data.
ANNUAL WATER BALANCE
Following the efforts to characterize the accuracy of the
precipitation and streamflow data, the next step in the calibration
process was to adequately approximate the annual water balance. In
this phase, the loss terms such as evapotranspiration and deep ground
water, were adjusted so that the volume of discharge was approximately
correct with no concern yet for timing.
Figure 2 shows the comparison of annual water yields during the
cal\brftion-verification pen'od for the four cases at the Piscataway
Creek USqs gauge. Generally speaking, the first few years had simu-
lated annual runoff volumes higher than the recorded ones; while the
simulated runoff volumes were lower than the recorded runoff volumes
for the last few years. The next step of the annual water balance
review required the preparation of mass curves. Mass curves provide a
relatively straight-forward method for analyzing stream yield. While
generally used for reservoir analyses, their use in this study was
intended to graphically compare the annual variation between the simu-
lated and historical data. Figure 3 shows the .mass balance between
the accumulated, simulated and recorded water yields. From the above
two figures, it can be seen that including the effect of snow/ice
accumulation and melting in a simulation study will certainly cause a
better fit with the recorded data although this is not significant.
The major reason for discrepancies between recorded and simulated
flows is that the actual storm patterns are often quite different from
those used in the model. While the variations from storm to storm
have a tendency to even out over time, Figure 1 shows that a spatial
variation does exist. This variation could have a substantial effect
on the annual water balance. Also, the land use conditions were
assumed to be constant over the entire study period while in reality
the conditions were constantly changing. To incorporate solutions to
94
-------
PISCATAWAY CREEK, 1967 to 1975
Simuloted Q
COG no SNOW
Simulated Q
COG w/ SNOW
V777A
Recorded
Discharge
Simulated Q
MNCPPC SNOW
100
AVERAGE ANNUAL DISCHARGE, (cfs)
10
en
Simulated Q
MNCPPC (no)
1967
1968
1969
1970
1971
1972
1973
1974
1975
YEARS
Figure 2. Comparison of Average Annual Discharges
-------
PISCATAWAY CREEK. 1967 to 1975
Simulated Q
COG no SNOW
Simulated Q
COG w/ SNOW
Recorded
Discharge
Simulated Q
MNCPPC SNOW
Simulated Q
MNCPPC (no)
10
CT>
600
500
400
300
200
100
CUMULATED ANNUAL WATER YIELD. In cf»
1967
1968
1969
1970
1971
1972
1973
1974
1975
TIME, (YEAR)
Figure 3. Comparison of Annual Mass Water Yields
-------
these problems into the existing data base would have required an
effort far in excess of the intended purpose. The current modeling
effort wis not intended to perfectly match historical records, but
rather, to produce a tool for use in evaluating environmental planning
alternatives.
MONTHLY WATER YIELDS
The best method of evaluating the closeness of fit between
simulated and recorded monthly discharge volumes is to present the
data on a simulated vs. recorded plot. If the simulation is perfect,
all of the plotted points would fall on the 45-degree line (1:1) indi-
cating that each simulated monthly value was exactly equal to the
recorded monthly value. The plot of monthly values of simulated and
recorded runoff volumes for the four cases (conditions) is shown in
Figure 4. The points tend to be randomly scattered and evenly distri-
buted around the 45-degree line. The best fitting curve (straight
line) for each of the four cases has been determined and also present-
ed in Figure 4. This figure indicates that the simulated water
balance on a monthly basis is quite representative. Also, this figure
indicates that it is more representative to include the effects of
snow/ice accumulation and melting.
CALIBRATION OF DAILY FLOWS
The rainfall event displays two results: (1) surface runoff
and (2) groundwater replenishment. Surface runoff is a transient
phenomenon and passes through the river system within several days of
the rainfall event. Once this water passes through the system, the
flow in the river is that water which has infiltrated and is now leav-
ing the system through ground water depletion. It is obvious that if
the annual and monthly water balance is reasonably good but the low
flow periods are either oversimulated or undersimulated, the differ-
ence in the water balance must be accounted for somewhere. Thus, a
few cfs errors over a dry period must be counteracted in a few wet
weather events. To avoid this type of misrepresentation, the daily
average discharge as simulated was printed out and compared -to the
recorded data. Particular attention was paid to dry periods during
the summer months.
When this analysis was first performed, it was found that in the
simulation, stream flows during the dry periods were oversimulated.
Upon further investigation, it was determined that the original INFIL
parameter was too high. This original INFIL parameter was estimated
based on the available soil information. When this parameter was
97
-------
PISCATAWAY CREEK: 1967 TO 1975
250
COG w/o
"SNOW"
SIMULATED DISCHARGE, (cfa)
COG w/
"SNOW"
- - Q.- __
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
200 -
150 -
100 -
200
250
300
OBSERVED DISCHARGE, (cfs)
Figure 4. Simulated & ObservPd Monthly Water Yields
-------
properly adjusted, the simulated daily flow duration curves matched
the recorded data extremely well (see Figure 5). From this figure, it
can be seen that the simulated daily flow duration curves for all four
cases converge very well.
At this point in the calibration procedure, the simulated daily
average discharge value matched the historical data well. Further
refinement of the model calibration requires matching hydrographs.
This process is critical for flood analyses.
CALIBRATION OF SIMULATED HYDROGRAPHS
When calibrating a continuous simulation model, it is rarely
expected to match every hydrograph exactly. The real variation in the
prototype varies far more than can be explained by any modeling
effort. This variation is the major reason continuous simulation is
far superior to single event simulation. When matching a hydrograph
in a single event model, it is easy to justify increasing the antece-
dent soil moisture. This results in "turning up" hydrograph peak and
hydrograph volume. Timing of the hydrograph can be adjusted easily by
varying the time of concentration or any related parameters.
Continuous simulation, on the other hand, does not easily lend
itself to such superficial calibration methods. Any assumption made
for a given event must hold true for the entire period. Oftentimes,
adjusting one parameter to fit a particular hydrograph would jeopar-
dize the hydrograph shape of other storm events. Thus, a careful
selection of the parameter values is extremely important in a continu-
ous simulation model.
In the HSPF model, hydrograph shapes for selected storm events
can be effectively altered with the UZSN (nominal upper zone soil
moisture storage) and INTFW (interflow inflow parameter) parameters to
better agree with observed values. Minor adjustments to the INFILT
parameter can also be used to improve simulated hydrographs. Since
adjustments to INFILT will change the established annual and monthly
water balance, the changes to INFILT should, we felt, be somewhat
minimal. Parameter adjustment was concluded when changes, did not
produce an overall improvement in the simulation.
The observed and four simulated hydrographs (one for each case)
of the February. 2, 1973 storm event are presented in Figure 6. This
storm event was selected for not only is it a winter storm, it is also
the maximum storm event for that year. From this figure, it is very
evident that the consideration of snow/ice accumulation and melting
processes is necessary. The peak flow is closer to the observed data
99
-------
o
o
PISCATAWAY CREEK, 1967 to 1975
Recorded
Discharge
Simulated Q
COG no SNOW
Simulated Q
COG w/ SNOW
Simulated Q
MNCPPC SNOW
Simulated Q
MNCPPC (no)
100
EXCEEDED PERCENTAGE, (X)
100
200
300
400
500
600
MEAN DAILY FLOW, (cfs)
Figure 5. Comparison of Daily Flow Duration Curves
-------
PISCATAWAY CREEK AT PISCATAWAY, MARYLAND
1400
1200
1000
800
600
400
200
Recorded
Discharge
STREAM FLOW, (cfa)
Simulated Q
COG no SNOW
Simulated Q
COG w/ SNOW
Simulated Q
MNCPPC SNOW
Simulated Q
MNCPPC (no)
I I
I I I I | I
J I I
I I I
00
06
12
18
00
06
12
18
00
06
12
Feb. 02 Feb. 03
Figure 6. Comparison of 02/02/73 Storm Hydrographs
-------
so also is the timing. It should be noted that the peak flow "time"
for case 1 is almost identical to that for case 4. However, due to
the use of a higher "upper zone storage" value in the winter period,
the peak flow rate for case 1 is much smaller than the flow rate for
case 4.
WATER QUALITY
The generation and transport of sediment and water quality
constituents are very closely related to watershed hydrologic proces-
ses. Since all of the simulated annual water balance, monthly water
yields, daily flows, and individual storm hydrograph appear to be very
close to the recorded data, it is considered that the watershed model
is very capable of simulating the water quality related responses of
Piscataway Creek to all meteorologic conditions including winter
periods. As the primary purpose of this study was to assess the
fufecutos^of accumulation and melting of snow/ice on a watershed using
the HSPF model, water quality and sediment calibrations were not
performed.
^ .In.thl's study, the set of water quality parameters developed by
COG in its Piscataway Creek Watershed study (Reference 1) was used for
each of the input sequences of the four study cases to analyze the
impact of "SNOW" addition on the generation and transport of sediment
and nonpoint source constituents. The water quality constituents
included in ttns study are total suspended solids, nitrate, ammonia,
total organic nitrogen, total phosphorus, and BOD-5. Comparison of
the results of the four cases includes monthly total loads and water
quality hydrograph for two selected storms: one in the winter and the
other in the summer.
Owing to the complexity of the model, it was felt that an
extremely long time period would be required to simulate water quality
analysis. It was estimated that, with a HP-3000 version of the HSPF
model, it would take approximately 4 hours of computer time to com-
plete one whole year of water quality simulation. Therefore, it was
decided that this water quality analysis would cover only a two-year
period; instead of the 10-year period in which the hydrologic proces-
ses were simulated. A two-year period from February 1971 to February
1973 was selected for the following reasons:
(1) The simulated annual water balance, monthly water yields, daily
flows, and individual hydrographs are very close to recorded
data;
(2) The annual event (with the highest peak flow) of 1972, which
occurred from June 22 to June 23, is a representative summer
storm; and
(3) The annual event of 1973, which occurred from February 2 to
February 3, is a good winter storm.
102
-------
MONTHLY TOTAL LOADS
Since the water quality processess are very closely related to
stream flows, a comparison of simulated monthly average discharge
among the four cases is presented for the same period in which water
quality simulation was performed (see Figure 7). From this figure,
the monthly average stream flows do not significantly change when
snow/ice accumulation and melting are considered. Because of this
insignificant difference in monthly stream flows when a completely
identifical set of water quality related parameters is added to the
HSPF input sequences, the monthly total loads of selected water
quality constituents do not significantly change with the addition of
the "SNOW" option. This finding is graphically presented in Figures 8
through 10. Monthly loads of BOD-5, total phosphorus, ammonia, and
total organic nitrogen all showed very similar results.
INDIVIDUAL HYDROGRAPHS
As in the hydrologic process, one of the most important
comparisons in water quality simulation is a comparison of individual
hydrographs. Efforts are emphasized on the hydrographs of the largest
storm event occurred in each year. In this case, the largest storm
event for 1972 is the June 21 event, and the largest storm event for
1973 is the February 2 event. The former storm event is a typical
summer storm event while the latter one is a winter snow storm event.
The individual hydrographs for these two completely different types of
storms were studied separately.
Figure 11 and Figure 12 present a comparison of bi-hourly
hydrograph of total suspended solids and a comparison of bi-hourly
hydrograph of nitrate, respectively, for the June 21, 1972 storm
event. As expected, from these two figures, the time distributions of
sediment loads and water quality were not significantly affected for a
summer storm event when snow/ice accumulation and melting effects were
included in the watershed model. The bi-hourly hydrographs of nitrate
particularly were almost identical with and without the "SNOW"-option.
A similar situation has been observed for other water quality
constituents as well.
On the other hand, however, the time distributions of sediment
loads and water quality constituents would be quite different during a
winter storm if snow/ice accumulation and melting effects are consid-
ered in the watershed model. The majority of water pollutants are not
transported from the land surface in the early portion of a winter
event. The pollutants are gradually conveyed from the land surface
when the accumulated snow/ice starts to melt. In addition, with
snow-covered land surface, soil detachment, washoff, and total sedi-
ment, are not as prodigous.
103
-------
PISCATAWAY CREEK. 1971 TO 1972
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
250
MONTHLY STREAM FLOWS, In CFS
200 -
150 -
100
Jan Feb Mar Apr May Jun Jul Au» Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sap Oct Nov Dec
1971
1972
Figure 7. Monthly Average of Total Stream Flows
-------
o
en
PISCATAWAY CREEK, 1971 TO 1972
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
25000
MONTHLY SEDIMENT LOAD, In tons/month
20000 -
15OOO -
10000 -
5000 -
Jon Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Fob Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1971 1972
Figure 8. Monthly Mass of Total Suspended Solids
-------
o
CTi
PISCATAWAY CREEK. 1971 TO 1972
MWCOG
w/o "SNOW"
MWCOG
*/ "SNOW"
M-NCPPC
w/ "SNOW"
20000
MONTHLY NITRATE LOAD. In Iba
15000 -
10000 -
5000
M-NCPPC
w/o "SNOW"
Jan Feb Mar Apr May Jun Jul Aug Sop Oct Nov Dec Jan Fob Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1971 1972
Figure 9. Monthly Mass of Nitrate (Load)
-------
PISCATAWAY CREEK. 1971 TO 1972
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
250
MONTHLY 5-DAD BOD LOAD, In 1000 Iba/month
200 -
150 -
i I i II II
100
Jan Fob Mar Apr May Jun Jul Aug Sap Oct Nov Dec Jon Fob Mar Apr May Jim Jul Aug Sep Oct Nov Dec
1971 1972
Figure 10. Monthly Mass of 5-Day BOD (Load)
-------
PISCATAWAY CREEK, 06/21/1972 STORM EVENT
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
O
C»
2000
BI-HOURLY TOTAL SUS. SOUPS, In tona/2-hr
1500 -
1000 -
500 -
Jun. 22 Jun. 23
Figure 11. Bi-Hourly Hydrograph of Total Suspended Solids
-------
PISCATAWAY CREEK, 06/21/1972 STORM EVENT
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW-
400
BI-HOURLY NITRATE LOAD. In lbe/2-hour
300
200
100
IB 00 06 12 18 00 06 12
18 00
Jun. 22
Jun. 23
Figure 12. Bi-Hourly Hydrograph of Nitrate
-------
Figure 13 graphically demonstrates the bi-hourly hydrographs of
total suspended solids of the four different cases for the February 2,
1973 storm event. When snow/ice accumulation and melting effects were
considered, the total suspended solids load was much less and the peak
loading time was much later. For this particular event, "SNOW"
effects reduced the total sediment yields by almost 50 percent,
although the reduction of sediment load did not significantly decrease
the total monthly load.
Figures 14, 15, and 16 present the bi-hourly hydrographs of total
nitrate, phosphorus, and BOD-5, respectively, of the four cases for
the February 2, 1973 storm event. From these three figures, shapes
and trends of the hydrographs among the three water quality constitu-
ents appear very similar. Case 1 (MWCOG without "SNOW") generated the
highest peak load, case 2 (MWCOG with "SNOW) was the next highest, and
case 3 (M-NCPPC with "SNOW") generated the lowest peak load. When the
"SNOW" option was considered, the time of the peak load was approxi-
mately 12 hours later than it was when "SNOW" was ignored in the
watershed analysis.
On this particular storm event, it is of interest to note that
because some relatively higher values of "monthly upper zone storage
(MON-UZSN)1' in winter periods were used in the COG study (case 1) to
reflect the snow accumulation, the simulated peak flow for case 1 is
much less than the peak flow simulated from case 4 (See Figure 6).
Theoretically, higher flow rate should produce higher pollutant loads
at a given time. In this particular event, however, it did not occur.
Figure 14 through Figure 16 indicate that the generated pollutant
rates for case 1 were larger than those generated for case 4 although
the flow rate for case 1 is smaller. After a careful analysis, it was
discovered that another storm had occurred 5 days previously. The
simulated total flow volumes during that 5-day period (prior to this
storm) were 1736 acre-ft for case 1 and 2001 acre-feet for case 4.
And the simulated total nitrate loads during that same period were
2226 Ibs. for case 1 and 3012 Ibs. for case 4. In other words,
between case 1 and case 4, when the February 2, 1973 storm event
occurred, the available pollutant on the ground surface was approxi-
mately 800 Ibs. less for case 4. Thus, even though the simulated
flow volume during this February storm event was much higher for case
4, the generated pollutant loads were much smaller simply because the
major portion of the available pollutants had been washed off before
the storm arrived.
CONCLUSION AND SUMMARY
The following conclusions have been reached:
0 The inclusion of "SNOW" option can only slightly improve the
simulated annual water balance, monthly water yields and daily
flows. However, it can significantly improve both shapes and
110
-------
PISCATAWAY CREEK, 02/02/1973 STORM EVENT
MWCOG
w/e "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
300
250
200
150
100
B1-HOURLY TOTAL SUS. SOUPS, In tons/2-hr.
00
06
12
18 00
06
12
18
00
06
12
Feb. 2
Feb. 3
Figure 13. Bi-Hourly Hydrograph of Suspended Solids
-------
PISCATAWAY CREEK. 02/02/1973 STORM EVENT
MWCOG
w/o "SNOW"
MWCOG
W/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
350
300
250
200
150
100
50
BI-HOURLY NITRATE LOAD, In lb«/2-hour
I I i I
j i
i i i I
j I
I I I I
I I
00
06
12
18
00
06
12
18
00
06
12
Fob. 2
Feb. 3
Figure 14. Bi-Hourly Hydrograph of Nitrate
-------
PISCATAWAY CREEK, 02/02/1973 STORM EVENT
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
1200
BI-HOURLY PHOSPHROUS LOAD, In lba/2-hr.
1000
BOO
600
^ \ \
\
400 -
200 -
Feb. 2
Feb. 3
Figure 15. Bi-Hourly Hydrograph of Phosphorus
-------
PISCATAWAY CREEK, 02/02/1973 STORM EVENT
MWCOG
w/o "SNOW"
MWCOG
w/ "SNOW"
M-NCPPC
w/ "SNOW"
M-NCPPC
w/o "SNOW"
10000
BI-HOURLY 5-DAY BOD UDAD. In lb«/2-hr.
8000 -
6000 -
4000 -
2000 -
» \ x
N. \
Feb. 2
F«b. 3
Figure 16. Bi-Hourly Hydrograph of 5-Day BOD Load
-------
volumes of each individual flow hydrographs for those winter
storms.
0 Since the generation and transport of water quality constituents
are highly dependent on the hydrologic process, the total annual
and monthly pollutant loads will not be significantly altered when
snow/ice accumulation and melting processes are ignored in the con-
tinuous watershed simulation study. But, if each individual
pollutant hydrograph is the primary study goal, the "SNOW" option
is certainly necessary.
0 The HSPF model has been used successfully to simulate the
hydrological response of Piscataway Creek. Furthermore, a good
simulated hydrologic response tends to enhance a reasonable water
quality simulation because of the close relationship between water
quality and hydrologic processes.
REFERENCES
1. Metropolitan Washington Council of Governments, The Piscataway
Creek Watershed Study, Final Report, prepared for the
Metropolitan Washington Water Resources Planning Board and the
Maryland-National Capital Park and Planning Commission, Sept.
1983.
2. Baldwin, J.L., Climates of the United States. National Oceanic
and Atmospheric Administration, U.S. Department of Commerce,
December 1974.
3. Soil Conservation Service, Soil Survey: Prince George's County,
Maryland. U.S. Department of Agriculture, April 1967.
4. National Oceanic and Atmospheric Administration, Climatplpgic
Summary, U.S. Department of Commerce in cooperation with Maryland
Experiment Station, Department of Agronomy, Climatology of U.S.
No. 20-18, March 1973.
5. Donigian, A.S., et al.( Application Guide for Hydrological-
Simulation Program - FORTRAN (HSPF). EPA-600/3-84-065. U.S.
Environmental Protection Agency, Atnens, Ga., June 1984.
6. Udhiri, S., Cheng, M.S., and Powell, R.L., Piscataway Creek
Watershed Study Using HSPF, Maryland-National Capital Park and
Planning Commission, Upper Marlboro, MD, 1985.
7. National Weather Service. National Weather Service River Forecast
System, User's Manual Part'2.6: Precipitation Model, Office of
Hydrology, Silver spring, Maryland, 1975.
115
-------
USE OF HSPF TO SIMULATE THE DYNAMICS OF PHOSPHORUS
IN FLOODPLAIN WETLANDS OVER A WIDE RANGE OF HYDROLOGIC REGIMES
By
James C. Nichols, P.E., Staff Engineer
Michael P. Timpe, Senior Technician
Water and Air Research, Inc,
Gainesville, Florida
ABSTRACT
HSPF was utilized for the purpose of simulating the phosphorus dynamics
within the floodplain wetlands of a major South Florida watershed in order to
assess the water quality impacts of various proposed restoration alterna-
tives. In order for the model to produce a realistic simulation, a number of
modifications and enhancements were required within most of the program
modules utilized. This paper will discuss the general approach taken to
perform the water quality simulations as well as the modifications required
to the HSPF source code in order to produce reasonable hydrologic, hydraulic
and water quality simulations of the alternatives on a Harris 500 mainframe
computer.
INTRODUCTION
A modified and extensively enhanced version of HSPF, release 7.0, was
utilized to simulate the phosphorus dynamics within the floodplain wetlands
of a major, channelized river in South Florida. The purpose of the study was
to assess the potential water quality impacts of various restoration alterna-
tives. The restoration alternatives under consideration incorporated a wide
range of structural and nonstructural control techniques, including abandon-
ment of the existing canal and attendent control structures, impoundment of
major tributaries, recreation of floodplain wetlands by means of levees and
control structures. An examination of baseline conditions and future no
action conditions was also performed. Agricultural land use predominates
within the watershed under study.
PROGRAM OVERVIEW
The HSPF program is a system of separate, task-oriented program modules
designed to perform a continuous networked simulation of the hydrologic cycle
and water quality behavior within a watershed and its receiving waters. The
previous land segment module (PERLND) was used to simulate the upland and
wetland subbasins and the RCHRES module was used to simulate the channel
116
-------
reaches. HSPF is a very large FORTRAN program written in a structured
programming style. The program uses a large random access file to store
intermediate results for subsequent input to the program when the entire
simulation network could not be run in one job.
PROGRAM SELECTION CRITERIA
The HSPF program was selected for use after careful screening and
comparison of the capabilities of existing water quality models in the public
domain. Other models given serious consideration included:
o Agricultural Runoff Management Model II (ARM-II),
o Dynamic Estuary Model (DEM),
o EXPLORE-I,
o Nonpoint Source Model (NPS),
o QUAL-II,
o Stormwater Management Model (SWMM-III),
o Storage, Treatment, Overflow and Runoff Model (STORM), and
o Water Quality for River-Reservoir Systems (WQRRS).
Each model was subjected to a rigorous technical and applicability
evaluation procedure. The technical evaluation considered the type of model,
time variability involved, transport processes, water quality processes,
constituents simulated, forcing processes, underlying assumptions and limita-
tions of the model, and the data requirements. The model applicability eval-
uation considered the ease with which a given model could be applied, given
the study objectives. An assessment was made of the availability of model
source code, quality of model documentation and technical support, and the
type and form of output generated by each program. Further consideration was
given to computer hardware and software required for program implementation
and the estimated run time and resulting computer costs.
HSPF was found to incorporate the following major advantages:
o Use of a single model program package to perform the simulations for
each of the major hydrologic components identified within the study
area;
o The program was capable of simulating, with modifications, the major
features of each of the proposed restoration alternatives;
o The water quality components of the model simulate clearly ..defined
(although simplified) physical, chemical and/or biological
processes;
o Modern, topdown, modular program structure;
o Output format flexibility and analysis capabilities;
o Program developed and supported by the Environmental Protection
Agency (EPA) Environmental Research Laboratory;
o Flexible internal time steps;
o Flexible time series data file manipulation capabilities; and
o The utilization of a common data base structure.
In any modeling work there are inherent limitations and assumptions
which may impact the usefulness of the results. These limitations include
both structural limitations which result from initial model simplifications
117
-------
and/or assumptions which affect the ability of the model to accurately simu-
late reality and assumptions made relative to model input data that cause
deviations from reality in the model results. HSPF has a number of signifi-
cant structural limitations, not all of which were fully appreciated when the
program was originally selected, including:
o Momentum is not considered in the receiving water portion of the
model, thus flows must be unidirectional and backwater effects are
not taken into consideration;
o Individual model elements are essentially one dimensional with only
vertical variability taken into consideration;
o Empirical hydrologic and nutrient transfer/transport process simula-
tion algorithms greatly dependent upon the simulation time step;
o Considerable program modification and enhancement was required to
simulate hydrologic conditions found in wetland systems;
o Considerable program modification and enhancement was required to
simulate phosphorus processes occurring in wetland systems;
o Cumbersome method to input external sources of water quality constit-
uents into the model;
o Lack of nutrient flow routing capabilities between certain program
modules;
o Dispersion is not taken into consideration,
o Inability to simulate gradually varied reservoir stage fluctuation
schedules;
o Rigid program structure centered around a central common block cross
linked to external data files;
o High initial setup costs due to program complexity and use of non-
standard FORTRAN extensions;
o High computer run costs due to the large program overhead, highly
segmented program structure coupled with mainframe virtual operating
system characteristics, and the large number of data input/output
operations.
HSPF PROGRAM MODULES UTILIZED
The PERLND Module of HSPF was utilized to simulate hydrologic and water
quality behavior of the upland and wetland subbasins in the watershed. The .
following program sections were invoked: PWATER, to simulate major compo-
nents of the water budget; PHOS, to simulate the phosphorus cycle and trans-
port of phosphorus from the land surface; MSTLAY, to simulate soil moisture
and fractional solute fluxes; and SEDMNT, to simulate erosion and transport
of sediment and soil.
Section PWATER forms the key component of the PERLND module. The output
time series generated by this subroutine are used as input by other PERLND
module sections. The model performed a continuous simulation of the state of
each of six moisture storage compartments within each subbasin: intercep-
tion, surface detention, interflow, and upper, lower and active groundwater
zones. A number of data sources were exploited to establish the analytical
framework, including a computerized grid cell data base with a spatial
resolution of 12.36 acres for fundamental framework information such as sub-
basin delineations, land use, topographic, vegetation, and soils association
118
-------
information; U.S.D.A. Soil Conservation Service (SCS) Soils Characterization
Data for soils physical and hydrologic characteristics; and an extensive
review of the literature.
Section PROS was used to perform a continuous simulation of the state of
four phosphorus storage compartments: plant, phosphate adsorbed to soil
particles, detrital organic, and soil phosphate in solution. These phospho-
rus storages are simulated separately for each of four soil horizons, as well
as soil solution phosphate phosphorus in the interflow compartment of each
subbasin. Data sources exploited to establish the framework for this section
included the Project GCDB; SCS Soils Characterization Data; and an extensive
review of the literature to estimate soil adsorption isotherm coefficients as
functions of soil properties, the seasonal magnitude of phosphorus standing
crop for each major land use/vegetation group in the watershed, the magnitude
and seasonal variation of the detrital phosphorus storage compartment for
each land use/vegetation group, detrital mineralization and solution phos-
phate immobilization rates under oxidizing and reducing conditions, seasonal
uptake rates of soil solution phosphate by viable plants, and seasonal plant
dieback rates.
The RCHRES Module was utilized to simulate the hydrodynamics and water
quality behavior of open water systems in the watershed. The following
program sections were invoked: HYDR, to simulate the hydraulic behavior of
receiving waters; ADCALC, to simulate longitudinal advection of dissolved and
entrained constituents; and GQUAL, to simulate instream attenuation of
phosphorus assuming first-order reaction kinetics.
Section HYDR forms the key component of the RCHRES module with the
output time series generated by this subroutine used as input by other RCHRES
module sections. The model utilizes the kinematic wave method to route
streamflows and ignores momentum. In order to circumvent this limitation, a
series of water surface profiles were developed using the HEC-2 program for
each alternative. An analysis of historical streamflow records was performed
to develop the expected range of streamflow conditions in the watershed. The
results of these analyses were used to synthesize depth-discharge relation-
ships for each model element within the floodplain, including open channel
segments and wetland subbasins. In addition, these results were used to
route flows between open channel RCHRES elements and floodplain wetland
PERLND subbasins. Data sources exploited to establish the framework for this
section included the cross sectional data to set up the HEC-2 simulations and
a review of the literature.
SEGMENTATION OF THE WATERSHED
The study area is characterized by extremely flat slopes, seasonal rain-
fall, and long-term flooding. It lacks a well-developed dendritic drainage
pattern and natural drainage pathways are typically poorly-defined. Soil
moisture and surface depression storages and evapotranspiration processes
dominate the hydrology of the watershed. Considerable man-made alterations
include both modifications to the natural drainage system as well as a
119
-------
considerable development of entirely man-made drainage networks. The study
area was conceptualized as consisting of three major hydrologic components:
o An Upland, or Runoff Loading component;
o A Wetland System component; and
o A Receiving Water component.
HSPF has an internally dimensioned limit of 75 "operations" which can be
performed during a simulation run. This limitation determines the total
number of PERLND subbasins, RCHRES channel segments, DISPLY time series,
etc., which can be incorporated into a given run. Therefore, the simulation
runs for each of the alternatives were performed sequentially in a series of
steps, with output from upstream simulations used as time series input to
downstream model elements. The watersheds formed by existing canal control
structures provided a logical breakpoint for the model. This allowed a
reasonable spatial resolution to be utilized, a greater amount of output
detail to be saved for-'subsequent analysis, and a shorter run time for a
given simulation run.
UPLAND, OR RUNOFF LOADING COMPONENTS
The Runoff Loading components were simulated utilizing the PERLND Module
of HSPF. These watersheds were simulated independently of the Wetland and
Receiving Water Model components with the output time series generated by the
simulated watersheds used as input to the appropriate downstream elements.
This approach was used since differences between alternatives only occur in
the wetlands and receiving waters of the watershed. Since HSPF is hydrauli-
cally unidirectional, any backwater affects cannot be simulated. The purpose
of the Runoff Loading Model component was to simulate the hydrology and
processes affecting the transport of phosphorus from the land surface in the
"upland" portions of the watershed. In this study, the upland portions of
the watershed were considered to be those areas outside of the floodplain of
the river and major tributaries, and/or upstream of areas directly impacted
by existing or proposed structural controls.
A total of 22 upland runoff loading subbasins were employed to perform
the water quality simulations for all of the alternatives. Since none of the
alternatives simulated directly impact the hydrology or land use projected
for the upland subbasins, the runoff loading model simulations were performed
one time for each land use scenario for the projected impact scenarios. The
resulting output time series from each of these simulations were then saved
and utilized as input time series to the wetland/receiving water model for
the appropriate impact scenario. Upland subbasin delineation, for the most
part, was determined by the delineation of subbasins within the floodplain.
Where feasible, a separate upland subbasin was delineated for each wetland
subbasin. Tandem upland subbasins, i.e., subbasins hydrologically in series,
were not utilized.
WETLAND SYSTEM COMPONENTS
The purpose of the Wetland System Model component was to simulate the
hydrology and processes affecting the transport, retention, and release of
120
-------
phosphorus through the wetland systems impacted or created by the various
alternatives. These systems included such features as tributary impound-
ments, impounded wetlands, flow-thru marshes, and the floodplain wetlands
under various hydrologic regimes. Since most of the alternatives simulated
in this study were designed to impact the hydrology of the floodplain wet-
lands, the establishment of the framework for the Wetland System Model ele-
ments received a disproportionate share of attention. In this study, the
wetland component of the watershed was considered to consist of the areas
within the floodplain and/or major tributaries. Between 23 and 26 wetland
subbasins were employed to perform the water quality simulations for the
various alternatives.
The extensively modified and upgraded version of the PERLND Module of
the HSPF Program was utilized to simulate hydrologic and water quality
behavior of the wetland system subbasins. The simulation of the hydrologic
behavior and transport of water from the land surface was performed by
Section PWATER. The primary difference between the upland and wetland sub-
basins was the algorithm used to simulate surface hydrology. The original
algorithm treating overland flow as a turbulent process was used to simulate
direct surface runoff in the upland subbasins. A new algorithm employing a
functional depth-discharge relationship was utilized to simulate direct
surface runoff from the wetland subbasins.
RECEIVING WATER COMPONENTS
The major open surface water components of the watershed were simulated
utilizing the RCHRES Module of HSPF. The Wetland System and Receiving Water
components of each alternative were simulated simultaneously. In this study
the receiving water portions of the watershed were confined to open channels
as well as any operational control structures located on these waterways.
The Receiving Water component incorporated between 15 and 18 separate reach
segments to simulate the hydrodynamic and phosphorus transport processes.
No mechanism was available in HSPF to simulate a continuously varying
upstream pool elevation, in either the floodplain wetlands or in the receiv-
ing waters of C-38, as proposed in a number of alternatives. Therefore, a
feature was added to the HSPF to approximate this operation in a step-wise
fashion on a monthly basis. The operating rule was assumed to be applied in
discrete, 0.5-foot monthly increments.
MODEL NETWORKING AND SIMULATION LOGISTICS
The pathways by which the above model components were connected were
made by means of the NETWORK Block, between upstream and downstream model
elements within a given model run, and/or by means of the EXT SOURCES and EXT
TARGETS Blocks, between upstream and downstream elements simulated during
separate runs. Figure 1 illustrates the physical arrangement of the model
elements within a major segment of the system modeled and a schematic of the
networked model elements utilized to perform the simulations.
121
-------
PI*
m*
T1
IEOENO
POOL WATERSHED IOUNOARV
MODEL IUMAIIN IOUNOARY
UPLAND TYPE MODEL (UUAtlM NUMICR
WETLAND TVH MODEL IUMAIIN MUUtt*
FLOW-THRU MA HIM IDENTIFICATION NUMIER
TRIIUTARV IMPOUNOMCNO IDENTIFICATION NUMIER
RESTORED FLOOOfLAIN. FLOW-THRU UARIH AND TRIIUTARV
IMPOUNDMENT TYPE WCTLANDt
POOL IMPOUNDED WETLAND
NOMINAL CONTROL ELEVATION
CHANNELIZED KIUIMMCE RIVER. CM
OLD IPRE-CHAMNELIUTIONI KlttlMMEE RIVER CHANNEL
WOIL TO REMAIN IN PIACC
MOOIL ITNEAMtEOMINT AND REACM. NUMIER
MEC J CROU (ECTION LOCATION AND NUMIER DELIMITING
MOBIL MACK
LEVEE
IX CONTROL ITRUCTURE AND IDENTIFICATION NUMIER
GATED CULVERT TYPE STRUCTURE AND IDENTIFICATION MUUtER
FIXED CNEtT WEIR TYPE STRUCTURE AND IDENTIFICATION NUMNER
ffl
FIGURE 1.
122
-------
First, the simulation runs for the Upland/Runoff Loading Model subbasins
were performed. The times series representing the predicted outflows of
water and phosphorus fluxes were saved in the proper Time Series Store for
subsequent input into the appropriated downstream model segments. The model
simulations for each alternative were then performed, proceeding from
upstream to downstream. The predicted model outputs of discharge and
phosphorus fluxes from upstream model elements were used as inputs, and the
predicted outflows of water and phosphorus fluxes were saved in the proper
Time Series Store for subsequent input into the appropriate downstream model
segments or impact analysis.
MODEL TEMPORAL AND SPATIAL RESOLUTION
MODELING TEMPORAL RESOLUTION
A total period of 13 months was employed for all simulations. The
calendar time employed for each simulation run was from September 1 through
September 30 of the following year. The first month of the simulation was
used to stabilize the hydrology and water quality of the model. This served
to reduce the impact of errors in estimating initial conditions. The model
predictions for the 1-month "warm-up" period were not taken into considera-
tion in the subsequent evaluation of alternatives. The simulation period
from October 1 through September 30 of the following calendar year, used in
the evaluation of alternative impacts, was selected to correspond to the
standard water year, which begins roughly at the end of the wet season. The
externally input time series, employed as model forcing functions, were
identical during the first and last month. Simulated output time series,
used as input into downstream model elements, were used directly for the
entire 13-month simulation period.
MODEL SPATIAL RESOLUTION
The Project GCDB provided the data base used to establish the framework
for the PERLND subbasins of the water quality model. The spatial resolution
of the model was determined by the 12.36-acre size of the individual grid
cells. Similarly, subbasin; location; existing and future land use; vegeta-
tion groups; topographic data; and soil association related data affecting
the development of the various hydrologic model parameters as well as the
model parameters used to simulate the application of phosphorus to the land,
the cycling of phosphorus within a watershed, and the transport of phosphorus
from a watershed were developed around the information contained within the
project GCDB.
HSPF is capable of simulating each of four soil horizons, the surface
layer upper zone, lower zone, and active groundwater storage. The soils
association characteristics were compiled into four horizons. Based on an
analysis of the available soils data, previous modeling conventions, and
site-specific conditions, the surface layer was taken as the 0-10 centimeter
horizon, the uper zone as the 10-50 centimeter horizon, the lower zone as the
50-150 centimeter horizon, and the active groundwater storage to be the
150-180 centimeter horizon. Average soils characteristics for each soils
123
-------
association were calculated using a weighted average of the corresponding
characteristics for component soil series. Weighting factors were based on
the estimated percent coverage of each soil series within the association and
a depth weighting factor based upon the proportion contained within a given
horizon. Composite soil parameters for each ^ubbasin were generally deter-
mined by calculating a weighted average of soil association parameters, based
on the number of grid cells in the PERLND assigned to each association.
MODEL FORCING FUNCTIONS
Externally input data used to drive the dynamics of the simulations
included precipitation, evaporation, air temperature, outflow and phosphorus
loading from the upper watershed, as well as diffuse inputs of phosphorus to
the watershed.
A total of 13 daily precipitation input datasets were synthesized using
statistical modeling techniques and a 21-year historical rainfall record for
nine National Oceanic and Atmospheric Administration stations surrounding the
watershed. Average daily pan evaporation and air temperature data for the
watershed were also estimated using data from surrounding NOAA stations.
Inflows from the upper watershed were estimated based on a statistical analy-
sis of streamflow records at the outflow from the upper watershed, prior and
subsequent to the installation of controls. Phosphorus loadings from the
upper watershed were estimated based on an analysis of historical water qual-
ity at the same location. Diffuse, external inputs of phosphorus to the
watershed were estimated for present and future conditions based on an analy-
sis of bulk precipitation data and supplemental cattle feeding and fertilizer
application practices in the watershed.
PROGRAM MODIFICATIONS
A number of modifications were made to the HSPF program to improve its
simulation capabilities for utilization in wetland systems. In addition, a
number of programming errors; model deficiencies encountered as the modeling
work proceeded; and hardware, operating system, and compiler related incon-
sistencies were identified and corrected. Model enhancements, corrections,
and/or modifications made to the hydrologic algorithms included:
o A revised flow routing algorithm employing a depth-discharge, func-
tional relationship from up to two possible exits;
o An enhancement to the revised flow routing algorithm to allow inde-
pendent control of outflow on a monthly basis;
o The incorporation of a surface outflow time series corresponding to
the second possible outflow exit to allow complex networking schemes
between model components to be implemented;
o Modifications to the infiltration/percolation algorithms to improve
simulation capabilities for saturated soil conditions.
Enhancements, corrections, and/or modifications made to the water
quality algorithms included:
o A phosphorus "dieback" loop coupling the plant phosphorus compartment
to the detrital phosphorus compartment with a time variate process
function;
124
-------
o The algorithm used to simulate uptake of plant phosphorous from the
available solution phosphorous storage was modified to depend on the
plant phosphorous storage;
o The capability to simulate the upward transport of nutrients between
soil horizons corresponding to the nutrient pumping action of plants
and to vary the ratio of aboveground to belowground plant phosphorus
storage on a monthly basis;
o Revision of the algorithm simulating detrital organic phosphorus
mineralization in order to take into account reducing conditions
typically found in flooded wetlands soils.
o An improved mechanism for the input of external nutrient loadings and
the incorporation of solution phosphorous inflow time series corre-
sponding to the hydrologic flows to provide nutrient flow routing
capabilities lacking in the original program.
The surface layer input was made to serve as the input point for all
external nutrient loadings, such as from adjacent subbasin inflows, adjacent
channel overbank inflows, atmospheric fallout, fertilizer inputs, etc. As a
result of the sequence in which calculations are performed by HSPF, the phos-
phorous loadings to the surface layer were added directly to the storage
compartment of phosphate adsorbed to the soil in the surface layer.
Fortunately, many localized algorithm changes in the program were fairly
easy to implement on the Harris system and were thus readily attempted.
PROGRAMMING STRATEGY AND CONSTRAINTS
The initial intention was to make modifications to the program in con-
formance with the original structure of the system and with full retention of
existing capabilities. However, this approach could not be followed because
of the number and extent of program revisions which would have been required.
The main COMMON block used by HSPF is employed in literally hundreds of loca-
tions throughout the program and is intricately cross-linked to the external
library file INFOFL. This file is also internally cross-linked. The cross-
linking is so extensive that major program modifications would have been
required to incorporate even minor changes affecting this underlying
structure.
As a result, modifications to the program made use of unused portions of
the source code, such as existing input tables and time series, connections,
wherever possible. A number of new arrays were created in the limited free
space available in the main COMMON block or in the space formerly allocated
to inactive sections. Available time series and COMMON block space from
section SNOW of the PERLND module were put to new uses, effectively deacti-
vating that section. The land use block capability of the PWATER section was
found to be of little practical use in its current form. Several time series
originally used by this option were utilized for other purposes in the
modified program. A number of PWATER and PHOS input data tables were used to
provide data needed by the modified algorithms, effectively eliminating the
unused original options. Non-functional changes pertaining to inactive
sections and options were made as required to satisfy compiler syntax
125
-------
constraints and to maintain the required program structure framework.
Fortunately, a number of the required algorithm enhancements had only local
effect in the program and could be designed to work around the limiting
structural elements.
HYDROLOGIC SIMULATION ENHANCEMENTS
The original PWATER section did not have the capability to adequately
simulate the controlled surface hydrology which will occur in the various
impounded wetland systems incorporated into a number of the alternatives. As
a result, the use of function tables supplying stage-discharge relationships,
"PTABLES", was implemented. Difficulties encountered in optimizing a
networking scheme indicated that a second outflow, also under PTABLE control,
was needed in the model. This allowed wetland systems with multiple outlet
control structures to be simulated, and the capability of routing floodplain
wetland outflows independently to both downstream reach and/or wetland
elements.
The PTABLE routing algorithm was developed for the PERLND module similar
in concpet to the FTABLE system used in the RCHRES module. A functional
table of surface water depth values and hourly outflow rates are supplied to
control the outflow from each of two outflow points. Multiple columns of
outflow rate data may be included, to be selected for use on a monthly basis.
This provides for variations in the control regime of impounded wetlands as
well as for the seasonal fluctuations of naturally flooded wetlands and
allows more complex networking schemes to be implemented. The upland
subbasins used the original routing algorithm, A time borrowed series,
SUROB, was required for the newly created second outflow gate. This provided
the networking capabilities required to simulate the complex floodplain
systems being simulated.
The modifications outlined here required program changes in the input
section PPWATR, addition of the input subroutine PTABLE, changes in the
calculation routines of Sections PWATER, SURFAC, DISPOS and PROUTE, and
updates to the independently installed INFOFL used by HSPF. The PTABLE data
was input to the program in the same format as an FTABLE. The first column
corresponding to depth, in inches, the remaining columns corresponding to
outflow rates, in inches per hour. Input table type MON-INTERCEP provided
the monthly outflow column selector for use with both possible outflow
PTABLES. The FTABLE/PTABLE to be used by the first and second exits was
specified in the input table PTABLE-PARMS. A value of zero was used for this
parameter to deactivate the pertinent calculations if a second outflow exit
was not utilized. The modified time series SUROB was used to represent the
outflow time series corresponding to the second outflow exit.
In the course of model calibration, further hydrologic modifications
were found to be necessary. It was discovered that it was not possible to
maintain the wetland elements in a flooded condition. The original program
hydrologic algorithms placed no limit on the magnitude of the Active
Grouhdwater storage compartment. With the hydrologic conditions simulated
and the time step used in this study, this storage tended to increase
126
-------
steadily, draining the upper soil layers. This resulted in negligible
surface layer storages and virtually eliminated overland flows, so important
in the floodplain wetland systems being simulated. Even the upland subbasins
did not function properly until an upper limit was placed on the total inflow
to the lower and active groundwater layers in order to keep the storages
within expected limits. The algorithm implemented to limit this downward
flow, now passes all inflow to the lower soil horizons through the Upper Zone
storage. The original model included this a bypass as a means of allowing a
volume of flow greater than the available upper zone storage to be routed
downward directly to the Active Groundwater compartment.
With this modification, land surfaces may saturate completely under
flooded conditions and will still function properly under drier conditions.
All downflow was routed directly through the upper zone storage compartment,
eliminating the bypass to the Active Groundwater horizon. The model origi-
nally provided this bypass to satisfy Lower Zone demand that exceeded the
size of the Upper Zone storage capacity, particularly when the Upper Zone is
modeled as a relatively thin layer. In order to correct an algorithm defic-
iency related to the large surface layer storages simulated, subroutine UZINF
was modified to compare and limit the calculated Upper Zone inflow to the
maximum inflow allowed by the algorithm. This problem was caused by opera-
tion with larger surface layer storages than the original algorithm was
developed to simulate, and is otherwise unrelated to the modification
controlling the total percolation to the lower layers.
These modifications required changes in PWATER, DIVISN, and UZINF. The
lower zone storage limitation is now implemented by specifying the size of
the active groundwater storage as the third parameter in table PTABLE-PARMS.
The maximum lower zone storage is now the sum of:
o Three times the Upper Zone nominal storage (UZSN),
o Three times the Lower Zone nominal storage (LZSN), and
o A new input parameter, the Active Groundwater storage.
The net effect of these modifications was that the upland-type systems
function as they did previously with the original surface routing algorithm
until the soil became completely saturated, in which case somewhat greater
surface runoff would be stimulated. No problems were experienced in satisfy-
ing the limited lower zone water demands by the modified method. A similar
result could have been accomplished by arbitrarily reducing the infiltration
capacity of the soils. However, this approach would not have produced the
same overall effect because the soils found in the study area are generally
very porous, have a limited available storage capacity, and quickly become
saturated when sufficient moisture is available.
These modifications were considered to be essential to produce reason-
able hydrologic functioning of the wetland systems. The hydrologic algorithm
modifications had no effect on the subsequent quality calculations, since the
nutrient fluxes were already simulated in this manner in Section MSTLAY.
Figure 2 illustrates graphically the program modifications and enhancements
made to Section PWATER.
127
-------
00
-------
NUTRIENT SIMULATION ENHANCEMENTS
Initial review of the applicable program algorithms indicated several
deficiencies in the agrichemical section PHOS of the PERLND module. Although
a plant phosphorous uptake mechanism to remove phosphorous from the solution
storage was available, no corresponding dieback mechanism existed. As a
result, the plant phosphorous storage compartment (PLTP) functioned as an
ever increasing dead-end storage. In the systems being simulated, plant
phosphorus cycling was considered important enough to warrant the inclusion
of a plant phosphorous dieback mechanism in order to recycle phosphorous to
the organic detrital storage.
The original program was only capable of simulating the downward and
outward flows of nutrients', the upward transport of nutrients between soil
horizons was not possible. However, plants function as nutrient "pumps",
transporting nutrients to the surface layer from the lower soil horizons via
their root systems. This was simulated by allowing phosphorus uptake to
occur within the various soil horizons, adjusting the plant phosphorous layer
storages based on ratios of the aboveground to the soil horizon storages.
The overall ratio of aboveground to belowground biomass may be varied
monthly.
The original program plant uptake algorithm calculated plant phosphorous
uptake as a first-order reaction on the basis of the available solution phos-
phorous storage, independent of the magnitude of the plant storage compart-
ment. In the wetland systems simulated, the magnitude of the storage
compartments of surface layer "soil" phosphate phosphorus in solution often
fluctuate widely. This resulted an unstable simulation of the plant phospho-
rus storage compartment. To remedy this situation, the demand basis for the
plant uptake from the solution compartment was modified to depend on the
existing plant phosphorous storage.
Mineralization rates in the surface layer*corresponding to oxidizing
(aerobic) or reducing (anaerobic) conditions can now be selected for use by
the program based on the degree of surface flooding.
Examination of the program capabilities revealed that nutrient loading
input time series capabilities for the PERLND module agrichemical section was
inadequate. The existing mechanism for external nutrient loadings, SPEC-
ACTIONS, was found to be unsuitable because the data format used by SPEC-
ACTIONS was foreign to all other program systems, the SPEC-ACTIONS data
format cannot be generated by the program in time series format, SPEC-ACTIONS
is not operational in the networking portion of the system, and performs a
direct memory address variable modification, which may change with program
modification. As a result, a set of phosphorous input time series were
developed to facilitate the simulation. These time series also provide the
nutrient flow routing capabilities lacking in the original program. The
surface layer input now serves as the input point for all external nutrient
loadings, such as from adjacent subbasin inflows, adjacent channel overbank
inflows, atmospheric fallout, fertilizer inputs, etc. Program modifications
affecting the nutrient simulation are illustrated in Figure 2 for Section
129
-------
MSTLAY and solute transport and in Figure 3 for Section PROS intra- and
inter-compartmental fluxes.
The original networking scheme used to input solution phosphorous inflow
time series corresponding to the various hydrologic flows placed these
inflows directly into the solution phosphorous storage compartment. As a
result of the sequence in which calculations are performed by HSPF, the phos-
phorous loadings were found to be quickly transported out of the system or
incorporated into the organic phosphorous or plant phosphorous storages. The
surface layer phosphorus inputs are now added directly to the phosphate
adsorbed to the soil in the surface layer. This implies the soils reactions
attain equilibrium immediately, avoiding the flushing and rapid uptake pheno-
mena. The assumption is that phosphorus inputs rapidly become bound to the
soil. The resulting model is also operationally more robust than the
original method.
These difficulties were alleviated by the redirection of three existing
input time series from the unused SNOW section to the PHOS section of the
PERLND module. The DTMPG, SOLRAD, and WINMOW time series now serve as
surface, interflow and active groundwater layer solution phosphorous inflow
times series, respectively. These correspond directly to the hydrologic flow
time series in these layers and provide greatly expanded networking capabili-
ties. An additional nutrient outflow time series (IFWOB) was converted from
its original function and- added to the PHOS section to correspond to the
second outflow added to the surface flow routing capability of the PWATER
section.
The modifications described above required program changes in subrou-
tines PHOS, PHORXN, MSTLAY, TOPLAY, TOPMOV, PWATPB, and PWAACC and effective-
ly disabled the SNOW section of the PERLND module. Modifications were also
required in the external file INFOFL. No additional input data was required
to use the modified time series DTMPG, WINMOW, SOLRAD, and IFWOB. They were
handled exactly like the existing time series in the TSS and in the EXT
SOURCES, NETWORK and EXT TARGETS blocks to perform their new functions.
SUMMARY
The HSPF program is somewhat unsophisticated hydrologically, having been
developed around a number of empirically derived algorithms. The program was
considerably modified in an attempt to reduce any potential impacts which
might have resulted from some of the hydrologic algorithm deficiencies. More
serious structural limitations of HSPF were first, the model is one-
dimensional and second, flows are assumed to be unidirectional. The
one-dimensionality of the model necessitated simulating floodplains as flat,
amorphous regions with constant surface areas. Thus the program can only
determine whether a particular part of the system is inundated and to what
depth.
Flow through the model is constrained to be unidirectional, since momen-
tum is not taken into consideration. It is assumed that an element cannot
hydrologically or hydraulically influence an element upstream from it and
130
-------
_lntirrul SurfMt Liytr md Abovt Ground_
Phoiphorut Stortf* Compcrtmtnit
in l*Krtte« tfeMfMlMi
Intfrnil Upp«r Livtr Trmwiory (of Initrllowl
Phoiphoreui Stor«*t C
LEGEND
INTERNAL STORAGE
COMPARTMENT
.INTERNAL FLUX
PHOSPHORUS INPUTS
_i.*^ PHOSPHORUS OUTFLOWS
-"»> PROGRAM
*' MODIFICATION
rnal Upper ZoiM/teww Zone/
Active QfoundwaMf
orvH Comp«rtin*nti
"<*:::.-
7T-
n>«Uf»M
I.UPMVMI.
SS.L-
I-
-*, lMMMlTnM»V1
f tMMt»»tM
PM^MrMtK.
I HMHM«
.J I»-*^C^
FIGURES.
131
-------
that downstream conditions, external to the model, cannot affect the hydro-
logic or hydraulic behavior of the system. In this study, outflows from
reach and wetland subbasin elements were assumed to be functions of volume or
depth. The methodology developed to define these relationships utilized the
HEC-2 program, and an analysis of historical streamflow records. Using this
methodology, it was possible to approximate the complex routing of surface
water flows and attendant phosphorus loads between floodplain wetlands and
the restored or channelized Kissimmee River over a wide range of hydrologic
conditions. Thus many of the limitations of the hydrologic and hydraulic
structural simplifications and assumptions of HSPF were circumvented.
The resulting model framework was such that those alternatives which
route more water with its attendant phosphorus load to the "wetland"
subbasins, predict lower downstream phosphorus loadings. In effect, the
simulation models predict that downstream phosphorus loads are to be closely
related to the hydrology of the impacted wetland areas. This supports the
popular notion of the value of wetlands in regards to their nutrient removal
capability. Thus, if the hydrologic/hydraulic assumptions which have been
incorporated into the framework of the simulation models result in an
improper allocation of either the absolute amount of water outside the
channel or the time that such water spends out of the channel, it can be
anticipated that a corresponding reduction in the capability of the model to
accurately simulate phosphorus dynamics within the system will result.
Another major constraint was that the simulation of the system was
limited to one year. While this appears to be adequate for comparison of
alternatives, the simulations say nothing about the long-term viability of
any of the alternatives regarding phosphorus removal and retention. For any
alternative to continue to effectively remove phosphorus in the long run,
there must be long-term "sinks" or storage compartments for phosphorus within
the system. However, the simulation models for the various restoration
alternatives, were not specifically designed to address this issue and no
conclusions can be drawn from the results regarding the long-term phosphorus
removal effectiveness of any of the alternatives.
132
-------
SIMULATION OF A REGIONAL WATER SUPPLY WITH AQUIFER STORAGE
by: Ronald L. Wycoff, P.E.
CH2M HILL
Gainesville, Florida 32601
ABSTRACT
The storage of treated water in a suitable aquifer for later recovery
and use may be an economically feasible component of a regional water supply
system. This concept, known as aquifer storage recovery (ASR), can be used
to reduce the required treatment and/or raw water storage capacities of a
water supply system such that a net production cost savings is achieved.
This paper describes a computer simulation model constructed for the General
Development Utilities (GDU) Peace River Regional Water Supply System,
located near For£ Ogden, Florida. Principal hydrologic components of the
simulation include streamflow records extension and statistical analysis,
generation of synthetic streamflow traces, and simulation of raw water
diversions subject to regulatory, water quality, and pumping constraints.
Major components of the water supply system considered include the water
treatment plant, off-line surface storage of raw river water, and
underground storage of treated water. The purpose of the simulation is to
develop relationships between the size of major system components and
overall system reliability. These relationships, along with component cost
data, will be used in future analysis to identify the least-cos.t combination
of major system components that will meet design demands at an acceptable
reliability.
INTRODUCTION
The GDU Peace River Water Supply Facility is located on the west bank
of the Peace River, near Fort Ogden, Florida. Total tributary area at the
raw water intake is approximately 1,787 square miles, and average daily
streamflow is approximately 1,540 cfs (990 mgd). All of the tributary area
except for about 70 square miles is gaged by three U.S. Geological Survey
(USGS) stream gages. The recording stations are the Peace River at Arcadia,
Joshua Creek at Nocatee, and Horse Creek near Arcadia. Records are
available beginning in 1932 for the Peace River gage and in 1951 for Joshua
133
-------
Creek and Horse Creek. The locations of the three USGS stream gages, as
well as the water treatment plant diversion point, are shown in Figure 1.
Existing water supply facilities include a diversion pumping station
with a maximum capacity of 22 mgd, a 6-mgd water treatment plant, and an
off-line raw water storage reservoir with an effective storage volume of
1,920 acre-feet (625 mg). These facilities serve a rapidly developing area
of southwest Florida. Ultimate maximum day water demand for the Peace River
facility is projected to reach 60 mgd by the year 2011 (1).
The planning and design of water supply facilities to meet these future
demands in an economically optimum manner represents a significant
challenge. The objective of the current project is to investigate the
hydrologic and economic feasibility of ASR at the Peace River plant site.
The project consists of two major parts. The first part involves
construction and operation of test wells to quantify the storage
characteristics and hydraulic response of the potential storage zones, as
well as the background quality of the native groundwater. The second part
of the project consists of development and application of the water supply
systems simulation model presented in this paper. This work is ongoing, and
all simulation techniques and results reported herein should be considered
preliminary and subject to change.
PEACE RIVER WATER SUPPLY SIMULATION MODEL
The Peace River water supply simulation model includes two major
modules, which operate independently and sequentially to ensure flexibility
in the analysis. The main objective of the first module, known as PEACE, is
the calculation of potential diverted streamflows (quantity and quality)
from the historical and/or generated total flow time series. The second
module, known as PLANT, utilizes these flow and quality arrays as inputs for
the simulation of the Peace River Water Supply System, including the
evaluation of the reliability of the overall system.
For this project, total dissolved solids (TDS) was chosen as the water
quality parameter of primary interest. Preliminary analysis of the raw
Peace River water as well as the native groundwater indicates that the
drinking water standard for TDS (i.e., 500 mg/L, maximum) would likely be
exceeded before any other drinking water limits were violated. Therefore,
TDS is considered to be the controlling water quality parameter, and the
term "quality", in this paper, refers to TDS concentration.
THE PEACE MODULE
The basic computational sequence for the PEACE module is summarized in
Figure 2. The main input data are the historical monthly flows and the
regression equation parameters for flow record extension and river quality
(TD.S) calculation. Monthly flows at the diversion structure are calculated
first; then, optionally, the corresponding potential diverted flows and
their quality are calculated using subroutines DIVERS and QUALTY, respect-
ively. Both of these time series are then saved on an off-line computer
134
-------
Total
Gaged Tributary
Area 1717 sq ml
H Peace River Gage
D.A. 1,367 sq mi
Water Treatment
Plant Diversion
Total Tributary
Area -1787 sq ml
. '
~*fm
, ..
*. -\
;
.
- .
.
< '". . j
;
FIGURE 1. Location of Streamflow Recording Stations and
Water Treatment Plant Diversion.
135
-------
Start
I Inputs /
A)
B)
1
sssion,
ua Creek
>e Creek
I
Historical Flows
Time Series (Monthly)
A) Peace
River @
Arcadia 51 yr
B) Joshua Ck. 32 yr
C) Horse Ck. 32 yr
Regre
Parame
River Fk
Concen
Complete Series of
Total Stream Flows
at Plant Diversion
(51 yr)
Statistical Analysis
of Monthly Stream
Flows (STAT)
Generate
Monthly
Flows?
Generate Synthetic
Flow Series
(STOCH)
Calculate TDS
Concentration of
River Flows
(QUALTY)
Calculate Diverted River Flow
Regulatory Constraints
River Algae Load
Pumping Capacity
Flows Available for Storage/Treatment
(DIVERS)
Write
DFLOW (I, J)
DQUAL (I, J)
(OF LINE)
I Hard
Copy
Output
End
FIGURE 2. General Flow Chart for Streamflow Portion of Water
Supply Simulation (PEACEModule).
136
-------
file for later use as input to the PLANT module. If generated flows are
needed for further simulation, the subroutine STAT is used for statistical
analysis of the historical total flows and the estimation of the parameters
for the flow generation module (STOCH), The optional stochastic generation
of monthly streamflows is based on a first-order Markov Model. TDS and
diverted monthly flows are then calculated by subroutines QUALTY and DIVERS,
respectively. All time series defined within the PEACE module may be
printed, along with their statistical summaries, by calling the STAT
subroutine.
Observed Flow Records Extension and Adjustment
Two simple power functions are used for the flow record extension at
the Joshua Creek and Horse Creek stations, where observed monthly flows from
1932 to 1950 are unavailable. Regression equation parameters were estimated
using the 384 monthly observations of concurrent streamflow data extending
from 1951 to 1982. These parameters are estimated by simple linear
regression on the logarithms of the observed concurrent monthly flows. The
resulting equations are then used to estimate monthly streamflow at Joshua
Creek and Horse Creek, based on the observed monthly flow at the Peace River
Arcadia gage for water years 1932 through 1950. Monthly flows at all three
stations are then summed, and this total is increased 4.1 percent to account
for the ungaged tributary area (see Figure 1). The resulting flow series is
considered an observed 51-year streamflow trace at the Peace River treatment
plant diversion structure.
Computation of River TDS Concentration
The monthly river water TDS concentrations are related to the total
monthly river flows by a quality rating curve which is also represented by a
simple power function. The parameters of the TDS rating curve were
estimated using 4 years of flow and TDS data from the Peace River at Arcadia
gaging station. These parameters are estimated by simple linear regression
of the logarithms of the concentrations and the flows. The resulting rating
curve is applied to the observed or generated monthly streamflow series in
order to establish the estimated TDS concentration of each monthly flow.
Computation of Available Flows
Only a small fraction of the total monthly river flow is available for
water supply purposes. Divertable river flows are a function of three major
constraints, as well as the total river flow. These are a regulatory
low-flow constraint, the pumping capacity of the diversion structure, and
the algae content of the river water.
Minimum monthly low flows have been defined for the Peace River
diversion by the Southwest Florida Water Management District, which is the
responsible state regulatory agency. These monthly minimums are included in
the diversion simulation, and diversions are not allowed when streamflows
are less than these values.
137
-------
Maximum diversion rates are controlled by the pumping capacity of the
diversion structure, which is currently 22 mgd. However, other maximum
rates may be simulated. In addition, the Peace River is subject to periodic
algae blooms in the spring and summer months. During these algae blooms,
river water is not diverted in order to prevent operational problems in the
raw water storage reservoir and plant. This situation is simulated with the
aid of a simple monthly probability distribution. The monthly probabilities
of not diverting river water due to high algae content were determined by
analysis of plant operating records and concurrent river algae data.
Synthetic Streamflow Generation
For water supply system analysis, more than one trace of flow series
may be required for the assessment of reliability of the different
components of the system under investigation. Given the limited length of
historical records, the generation of synthetic traces of monthly flows
series "similar" to the historical one has been included in the PEACE module
as a user option. For this study, a first-order Markov Model was chosen (2,
3). The parameters of the stochastic model are computed by the PEACE module
using the extended and adjusted flow record at the diversion point discussed
previously.
THE PLANT MODULE
The PLANT module is essentially a flow and quality mass balance
accounting procedure operating on a monthly time step. Computations are
governed by a predefined flow distribution logic, which is given in
Figure 3. Also shown in Figure 3 is a schematic diagram of the proposed
Peace River Water Supply System. This system consists of five major
components: 1) the river diversion structure; 2) the water treatment plant;
3) the aquifer storage/recovery system, which stores treated water in the
aquifer; 4) the surface storage reservoir, which stores raw water in an
off-line open storage basin; and 5) product water distribution.
The diversion of flows from the Peace River is addressed in the PEACE
module, and these flows, therefore, are a known input to the PLANT module.
They are referred to as "potential" diverted flows because they may or may
not actually be diverted, depending upon the status of the surface storage
reservoir at a given time step.
Flow distribution in any time step is a function of the potential
diverted river flow available and the system demand. If the available
diverted river flow is greater than the demand, then the river flow is
treated and distributed, and the remaining available flow is treated and
stored underground or stored as raw water in the surface reservoir,
depending on the status of system components.
If the available diverted flow is less than the demand, then the
deficit is obtained from the aquifer storage/recovery system if the water
quality goal can be met. If the water quality goal cannot be met, then a
blend of aquifer storage/recovery water and surface storage water is used
that will meet the water quality goal. If this is not possible (i.e.,
138
-------
CO
V£>
FLOWS DISTRIBUTION LOGIC
A) DFLOW (I.J) > DEMAND (J)
1) Treat & Distribute - [DEMAND (J)]
2) Treat & Inject - [OPLANT-DEMAND(J)]
3) Remaining Flow [if any] to SSR.
If SSR is full then water is lost
from system.
B) DFLOW (I.J)< DEMAND (J)
1) Treat DFLOW (I.J) A distribute
2) Treat additional flow from SSR to
maintain minimum plant flow if necessary.
QSSR = PLANTCM - DFLOW (I,J)
3) Obtain remaining required flow from ASRS.
4) Check quality of product water.
If quality criteria is not met, then
solve for blend of SSR Waters and ASRS
waters which will meet quality goat.
5) If quality goal cannot be met then
satisfy DEMAND (J) from ASRS.
(This is a system failure.)
Aquifer Storage
Recovery System
(ASRS)
v /*
Product
Water
Distribution
Defined By
WSS1
DFLOW (I.J)
DQUAL (I,J)
0>
>
0>
o
I
Demand (J)
FIGURE 3. Schematic Diagram of Peace River Storage Treatment System.
-------
surface reservoir empty), then the required water is obtained from the
aquifer storage/recovery component, even if the water quality goal cannot be
met. In this case, a failure occurs when the water supply system does not
produce the desired quality water. Since native groundwater is available
for pumping during extreme droughts, the system is fail-safe hydraulically,
assuming that installed well field capacity matches demand.
This flow distribution logic is based on two basic assumptions. First,
the aquifer storage/recovery system will be the primary means of water
storage. The raw water surface reservoir will serve as a secondary or
backup storage system. Second, the system will be operated in a manner that
maximizes its reliability from a water quality standpoint. That is, if it
is physically possible to deliver water within the quality goal, such water
will be produced and delivered.
Aquifer Storage/Recovery System Mixing
A unique element of this simulation is the aquifer storage/recovery
system. When treated water is injected into the aquifer, it will displace
and mix with the native groundwater. Therefore, when waters are withdrawn
from the aquifer, they will be a blend of the treated injected waters and
the native groundwaters. The quality of the recovered water, therefore,
will be a function of this mixing process. The approach utilized in PLANT
consists of development of an empirical or conceptual model rather than a
theoretical model. Data used to develop this mixing model were taken from
onsite aquifer storage/recovery testing.
The components of the aquifer storage/recovery system mixing model are
shown schematically in Figure'4. Three major components are identified:
native water storage, injected water storage, and a mixing process. Inputs
to the aquifer storage/recovery subsystem include the monthly time series of
injection volumes (VI(I,J)) and corresponding quality (QI(I,J)). Outputs
include the monthly time series of recovered volumes (VR(I,J)) and
corresponding quality (QR(I,J)).
Native water is assumed to occupy an infinite reservoir and have a
constant quality (QN). Therefore, the volume of native water will never be
depleted, nor will its quality vary with time.
When treated water is injected into the aquifer, it will displace the
native waters near the injection wells. The maximum volume available for
storage of injected waters is unlimited. However, the amount in storage in
any given time step (RRV(I,J)) will be the summation of all waters
previously injected, less the summation of all injected waters previously
recovered. The quality of the injected waters (RRQ(I,J)) is assumed to be
the composite quality of all injected waters in storage at any given time.
The blend of injected and native waters withdrawn from the aquifer
during "a recovery period is computed by application of an exponential
relationship as illustrated in Figure 4. This empirical relationship does
not simulate the mixing process itself, but only defines the results (i.e.,
140
-------
Injection
Cycle
Limits of
Aquifer Storage Recovery Subsystem
of Peace River Water Supply Model
VI (I. J)
QI(I.J)
Native
Water
Storage
Vol. = »
QN = Const.
Injected
Water
Storage
RRV(I.J)
RRQ(U)
Mixing
Process
1.0 -
Native
Water
Injected
Water
I
1.0
VRT/VIB
2.0
Recovery
Cycle
VR
QR
FIGURE 4. Conceptual Sketch of Aquifer Storage Recovery Mixing Model.
141
-------
response) of the aquifer mixing. The relationship and its parameters are
defined as follows:
PI = a e 6(VRT/VIB) {1)
Where:
FI = instantaneous fraction of injected water contained in
recovered water mixture
VRT = total volume recovered since beginning of recovery cycle
VIB = volume of injected water in aquifer storage at beginning
of recovery cycle
a and $ = aquifer mixing parameters defined from analysis of
observed onsite aquifer storage/recovery test data
Raw Water Storage Reservoir
The surface reservoir simulation is a simple mass balance of both the
quantity and quality of raw river water. Inflows accounted for include
diverted river flow and direct rainfall. Outflows include water supply
withdrawals and evaporation. Both rainfall and evaporation are defined on a
monthly basis. If the reservoir is full and potential diverted flows are
available, then these flows are lost from the system and will not be
available at a later time step.
RESULTS
To date, the simulation has been applied to define relationships
between treatment plant/ASR capacity, off-line raw water storage capacity,
and overall system reliability for the year 2011 conditions (i.e., 60 mgd
maximum day and 37.5 mgd average daily flow). Twenty different combinations
of treatment capacity and off-line storage capacity have been simulated and
a family of system reliability curves developed.
Preliminary economic analysis, based on the results of the simulation
as well as an estimated annual cost of alternative systems, indicates
substantial overall costs savings using ASR when compared to the traditional
surface-water supply approach, which utilizes raw water storage only.
Savings on the order of 25 percent using ASR appear feasible. These savings
result from a reduction in required treatment capacity as well as in raw
water storage volume. However, at this time, the overall best solution
using ASR is not obvious from the economic analysis because differences
between alternatives are small. The final selection will involve the
evaluation of many factors in addition to system economics and reliability.
Future plans for this work include various refinements in the
simulation, as well as application to design demand levels between the
current system capacity and the ultimate (i.e., year 2011) system demand.
142
-------
ACKNOWLEDGMENT
This work is being funded by General Development Utilities, Inc. The
Peace River Water Supply Simulation Model was developed by Dr. Khlifa Maalel
of the Environmental Engineering Sciences Department, University of Florida,
under the direction of R. David G. Pyne, P.E., and the author, both of CH2M
HILL, Gainesville, Florida. The work described in this paper was not funded
by the U.S. Environmental Protection Agency, and, therefore, the contents do
not necessarily reflect the views of the Agency, and no official endorsement
should be inferred.
REFERENCES
1. General Development Utilities, Inc. Long range water supply planPort
Charlotte Region. Miami, Florida, April, 1982.
2. Kottegoda, N.T. Stochastic Water Resources Technology. John Wiley &
Sons, Inc., New York, New York. 1980.
3. Fiering, M.B., and B.B. Jackson. Synthetic Streamflows. Water
Resources Monograph 1. American Geophysical Union, Washington, D.C.
1971.
143
-------
SIMULATION OF POSSIBLE EFFECTS OF DEEP PUMPING
ON SURFACE HYDROLOGY USING HSPF
by: C. Nancy Hicks, Wayne C. Huber
and James P. Heaney
Department of Environmental Engineering Sciences
University of Florida
Gainesville, Florida 32611
Submitted to the
EPA Storm and Water Quality Modeling Users Group Meeting
University of Florida
Gainesville, Florida
January 31 - February 1, 1985
ABSTRACT
The continuous simulation program HSPF is being used to model
the hydrology of the Cypress Creek watershed in Pasco County, Florida
north of Tampa. Various strategies have been considered for simulating
possible effects on surface water of pumping 30 mgd from deep wells
within the Floridan Aquifer beneath the watershed boundaries. Modeling
complications include the fact that HSPF was not specifically designed
to simulate withdrawals from groundwater by pumping; hence, indirect
methods have to be applied. Other factors that affect the surface
hydrology are also being considered, especially surface drainage
activities and recurrent droughts.
CYPRESS CREEK WATERSHED
INTRODUCTION
Cypress Creek Watershed, located north of Tampa in Pasco County
(Figure 1), has undergone extensive development over the last decade.
Landowners in the basin have made major modifications of surface drainage,
and the West Coast Regional Water Supply Authority has established a
30 mgd capacity wellfield in the center of the watershed, pumping from
the deep groundwater aquifer. The Department of Environmental Engineering
Sciences at the University of Florida has undertaken a two-year project
to study the possible effects on the basin of these types of development
144
-------
under the varying meteorological conditions of the region. Recurrent
droughts are a particular problem.
Previous studies of the basin include seven years of hydrobiological
monitoring of the Cypress Creek Wellfield by the Southwest Florida Water
Management District, Biological Research Associates and Conservation
Consultants, Inc. (Rochow, 1983). Two models for simulation of steady
state groundwater flow for a 932 square mile area containing Cypress
Creek Wellfield and nine other municipal wellfields were developed by the
USGS: a two dimensional model (Hutchinson et al., 1981) and a
quasi-three-dimensional model (Hutchinson, 1984). In addition, a
model patterned after the Prickett-Lonnquist aquifer simulation model
is currently maintained by SWFWMD. Previous studies have been inconclusive
as to the effects of either drainage development or deep aquifer pumping
on the surface hydrology of the basin.
Cypress Creek Watershed consists of 117 square miles of sandy ridges,
flatwoods, hammocks and swamps drained by Cypress Creek. The creek runs
through an area of low-lying swamps and wetlands known as Big Cypress
Swamp. The five soil associations generally found in the area contain
soils which tend to be sandy throughout and are characterized by
wetness, poor filtration, and ponding. These soils constitute the
surficial zone and range from 20 to 40 feet in thickness (SWFWMD, 1982).
Under these unconsolidated deposits is a clay layer between 2 to 25 feet
in thickness which acts as a semipermeable confining layer (Ryder, 1978).
Under the clay lies the consolidated rock of the Floridan aquifer.
BASIN RECORDS
Two rain gages located in the basin provide daily records for a
period of seven years, 1977 to 1983. One gage located just east of
the basin at St. Leo provides an hourly rainfall record from 1944 to
the present. This gage gives a long term average annual rainfall for
the area of 54 inches. Outside the western boundary of the watershed
at Lake Padgett is a pan evaporation station with eleven incomplete
years of record, 1972 to 1983; the mean annual pan evaporation for the
seven complete years is 56 inches.
There are three stream gages on Cypress Creek which offer some
discharge and stage data. The Worthington Gardens gage is at the
mouth of the watershed with a period of record from June 1974 to the
present. The Drexel gage lies just south of the Cypress Creek Wellfield
and has been discontinued but provides records from 1977 to 1981.
The San Antonio gage just north of the wellfield provides the longest
record for the stream, from 1963 to the present. The various gaging
stations are shown in Figure 1.
SIMULATION
OBJECTIVES
As stated, the objectives of the study are to look at the effects
on the surface hydrology of both on-going drainage development and deep
aquifer pumping. A long term continuous simulation model was needed to
span the period prior to wellfield operation (pre-1976) to the present.
A groundwater model could best simulate deep aquifer pumpage, but
could not as accurately capture the effects of continuous changes in
145
-------
. A
0
Q
Watershed Boundary
Scream Gage
Rain Cage
,Pan Evap. Gage
PERLND Boundary
Y^- ,&^ 1
,/' ' ^7 PEHLND 3
tr f^"^ l
J Cypr/T /Cyp. Crk.
Cre/k / Wellflelc
Figure 1. Gaging stations in the vicinity of Cypress
Creek watershed.
146
-------
surface drainage; the fact that the Prickett groundwater model of the
basin was already in operation at SWFWMD was an added consideration.
For these reasons, it was decided that a surface model could best allow
for the examination and comparison of hydrographs and stages before and
after various hydrologic modifications and conditions, which might
include surface drainage, deep aquifer pumping, general development
in the watershed, and droughts.
CHOICE OF MODEL
Given the aforementioned characteristics desired in the model,
HSPF (Hydrological Simulation Program - Fortran) was chosen. The
initial release of HSPF was prepared by Hydrocomp Incorporated. The
revised version used for this project, Release 7.0, was prepared by
Anderson-Nichols and Company and obtained through EPA's Environmental
Research Laboratory in Athens, Georgia (Johanson et al., 1981). HSPF is a
long-term continuous simulation model which is both well documented
and well supported. It was expected that HSPF could simulate the surface
hydrology well, although that has proved challenging for the swamp
conditions existing in Cypress Creek Watershed.
CALIBRATION
Parameters
The HSPF component which controls the hydrologic simulation of the
basin is called PWATER. In section PWATER there are over 25 parameters
which are used in the description of each subcatchment. Cypress Creek
Watershed was divided into 6 permeable land segments or sub-catchments,
designated as PERLND 1 through PERLND 6 (Figure 1). A great difficulty
arises in connecting over 150 parameters with numbers that are physically
meaningful. Suggestions for parameter values according to watershed
location and land use can be found in the user's manual for the ARM
(Agricultural Runoff Management) Model. (Donigian et al., 1978).
Some of the more important parameters affecting runoff volumes are the
water storage capacities for the unsaturated soil zones of the surficial
aquifer.
Storages
Water in the unsaturated zone can be divided into six storage
blocks in HSPF (Figure 2). Interception storage (CEPS) is water retained
on plant surfaces. Surface storage (SURS) is depression storage on the
surface. Interflow storage (IFWS) contains water that flows below the
surface to streams and other surface water bodies. The upper zone
storage (UZS) is the upper few inches of the unsaturated zone from
which water infiltrates to lower zone storage (LZS) or leaves as
evapotranspiration. From the lower zone water may also leave by
evapotranspiration or percolate into active groundwater storage (AGWS).
Active groundwater may become runoff, evapotranspiration, or be lost
to the simulation as deep percolation.
Each storage has a state variable which is given a value to
reflect the amount of water in the soil at the beginning of the simulation
period. The upper and lower zones, however, also have nominal storage
capacity parameters whose values indicate the maximum amount of water
which can be stored in these zones. These nominal capacities greatly
influence the amount of runoff generated by the simulation and thus
the entire water balance.
147
-------
To determine the nominal storage capacities, the soil groups
occurring in the basin were determined from the Soil Survey of Pasco County
(Stankey, 1982), and the percent of each soil group occuring in each
of the six land segments was estimated. SCS Soil Interpretation Sheets
provided soil storage capacities for soils in each soil group. The
SCS data divided the capacities into two horizons; the upper horizon
consisted of the first 5 inches of soil and the lower horizon consisted
of the soil from the 5 inch depth to the water table. It was decided
that these two horizons would correspond well to the HSPF upper and
lower zone storage blocks (Figure 2).
The horizon storage capacity information for each soil group
was combined. Each group was then weighted as a percentage of its
occurrence in each land segment and the storage capacities for each
horizon were multiplied by the weights to determine total storage
capacity horizons for each land segment. The final HSPF nominal storage
capacities were obtained by multiplying by the soil zone depths,
5 inches for the upper zone horizon and the remaining inches between the
5 inch depth and the average water table depth for the lower zone horizon.
Average depth to the water table was obtained for each soil group from
the SCS soils data. The upper zone storage capacities ranged from
0.25 to 0.70 inches. Lower zone storage capacities ranged from
0.08 to 6.05 inches. For each land segment the minimum capacity for the
upper or lower horizon was used because HSPF allows overfill of soil
storages; the nominal capacities input into the model are indicators,
not maximum capacities.
COMPARISON
Parameters reflecting the best available data about the watershed
are shown for two land segments in Table 1. Some method of comparison of
simulated values against measured or predicted values was then needed.
Starting at San Antonio, Cypress Creek gage discharges data have been used
to calibrate measured annual and mean monthly discharge volumes in acre-ft
against simulated annual and monthly volumes.
The gage data were available in the form of daily flows in cfs.
To obtain the mean monthly volume of runoff in acre-ft, the mean
monthly discharge rate was multiplied by days per month and by the factor
1.983 acre-ft per cfs-day. The annual volume was the total of the 12 mean
monthly volumes over the course of a water year, October 1 through
September 30. .
Four methods were used to compare the measured and simulated volumes.
Table 2 shows the simulated and measured values indicating which water
years were being over- or under-predicted and to what extent. Then, three
plots were generated using SAS (Statistical Analysis System) on the
University of Florida computers. The first (Figure 3) was a scattergram of
monthly measured volumes versus monthly simulated volumes; best fit would
occur if the scatter of points approximated a 45 degree line. Next, a
hydrograph was prepared of the monthly volumes versus time, with measured
and simulated values overlaid (Figure 4); here, the emphasis would be to
match timing and volume of flow, and peaks if possible. The third plot
(Figure 5) was a double mass curve of measured versus simulated values,
again looking for a 45 degree line. The results of the closest simulation
to date for the upper third of the basin are shown in Figures 3, 4 and 5.
148
-------
/ V
Intercept/ion \
Storage/ CEPS \
SUROif-
IFWO x
AGWO x
AGWLI
Surface Storage ,
Interflow Sto'rage,
Upper Zone Storage,
Lower Zone Storage,
AGWI
1 t
M*
->
J
s
URR.
IFWS
UZS
LZS
7
T
\
Nominal
y Water
A. Storage,
inches
.3-. 7
\
.08-6,05
s
J- =
Active Groundwater
Storage, AGWS
Actual
Depth,
inches
V
5
t
\
1-55
f
Confining Layer
.
V
>)
:r
i
I SURO + IFWO + AGWO - PERO
PERO » total runoff -
IGWI DEEPFR (IGWI + AGWI)
Figure 2. Storage blocks for permeable land segments in HSPF
149
-------
Table 1. PWATER PARAMETERS DESCRIBING LAND
SEGMENTS 5 AND 6
Parameter
Interception:
CEPSC
CEPS
Surface Detention:
SURS
Interflow:
IFWS
Upper Zone:
UZSN
UZS
Lower Zone:
LZSN
LZS
Active Groundwater:
ACWS
Infiltration and
Perolation:
INFILT
INFILD
IRC
AGWRC
DEEPFR
Evapo transpiration:
FOREST
BASETP
AGWETP
LZETP
Lateral Transport of
Water:
LSUR
SLSUR
NSUR
Units
inches
inches
inches
inches
inches
inches
inches
inches
inches
in/hr
/day
/day
-
feet
Value 5
0.15
0.0
0.0
0.0
0.25
0.017
3.30
2.99
1.51
6.0
2.0
0.9
0.985
0.25
0.3
0.0
0.6
0.4
6880
0.0007
0.25
Value 6
0.15
0.0
0.0
0.0
0.30
0.001
4.40
4. 43
2.04
10.0
2.0
0.9
0.985
0.30
0.3
0.0
0.6
0.4
28960
0.0140
0.25
Description.
Interception Storage Capacity
Initial Interception Storage
Initial Surface Detention
Storage
Initial Interflov Storage
Upper Zone Nominal Storage
Initial Upper Zone Storage
Lower Zone Nominal Storage
Initial Lower Zone Storage
Initial Active Groundwater
Storage
Index to Mean Infiltration Race
Ratio Max/Min Infiltration Race
Interflow Recession Rate
Active Groundwater Recession Rate
Fraction of Groundwater to Deep
Aquifer
Fraction Winter Forest Transpiration
Fraction ET from Active GW Outflow
Fraction ET from Active GW Storage
Lower Zone ET Parameter
Length of Overland flow Plane
Average Overland Flow Plane Slope
Manning's n For Overland Flow
150
-------
TABLE 2. MEASURED AND SIMULATED VOLUMES
OF STREAMFLOW, REACH 1
Water Year Month
1978 Oct.
Nov .
Dec.
Jan .
Feb.
Mar.
Apr .
May
June
July
Aug.
Sept.
Total
1979 Oct.
Nov .
Dec.
Jan .
Feb.
Mar.
Apr .
May
June
July
Aug.
Sept.
Total
1980 Oct.
Nov .
Dec.
Jan .
Feb.
Mar.
Apr .
May
June
July
Aug.
Sept.
Total
Measured Flow
Acre-f t
1,107.0
84.5
312. 3
1,063.0
2,338.0
2,528.0
139.8
304.9
223.1
651.6
2,533.0
239.1
11,524.3
11.1
0.0
0.0
84.2
149.4
670.1
11.3
2,742.0
89.2
18.4
3,418.0
7,793.0
14,986.7
4,217.0
562.8
377.4
369.5
300.2
435.2
339.4
99.0
101.1
799.1
779.6
94.6
8,474.9
Simulated Flow
Acre-f t
805.8
6.7
44.3
531.1
1,526.6
2,196.2
77.0
310.5
348.2
1,447.8
3,271.9
975.7
11,541.8
384.8
5.1
28.8
734.4
439.3
618.0
0.9
2,439.4
586.0
113-.8
2,943.3
5,330.4
13,624.2
4,693.3
1,598.8
593.8
92.9
26.4
42.2
21.2
55.4
8.0
436.7
669.3
187.1
8,425.1
151
-------
8000-j
7000-
60QD-
5000-
H
E
fl
S
U
R UOOD
E
D
F
I
D
u aooo
20QD-
tOOD-
Q
<°
P _° D O
a
a
Q 500 LOOD L5DO 2DOO 2500 3000 350D UODO U50Q 5000
SIMULflTED FLOW
Figure 3. Scattergram for 3 years of monthly measured and
simulated flows, volumes in acre-ft.
152
-------
aoocH
7DOQ-
QDOQ-
5DOQ-
L UDOQ-
3
H
3DOQ-
2000-
1DOQ-
0-
o
-h
ID
\ A.
\
nn
15
" " " ' * ir
20
MONTH
-r-rjr
25
30 35
Figure 4. Hydrograph of measured and simulated flows,
volumes in acre-ft.
153
-------
SUMMERS
35000 -I
32500 -
3DGGO -
275QO -
25000 -
22500 -
2:0000 -
17500 -
150QO -
12500 -
1DOOO -
7500 -
5000 -
2500 -
0 -
Q 5QDO IQOOD 15000 200DO 25000 30DOO 3500D
SUMS IH
Figure 5. Double mass curve of 3 years of measured and
simulated monthly flows, volumes In acre-ft.
154
-------
DEEP AQUIFER PUMPING
Any water leaving active groundwater storage by percolation into
deep groundwater is lost to the HSPF simulation. HSPF cannot add or
subtract specific water volumes directly from a deep aquifer. The
parameter which controls the rate at which water leaks from the surficial
to the deep aquifer is called DEEPFR; DEEPFR is entered as a fraction
between 0 and 1.0 and governs the percentage of water entering the active
groundwater storage and leaving this block via deep percolation. One
possible effect of deep aquifer pumping might be to increase this leakance
rate through the semi-confining layer.
This study has broached three methods to capture this possible
effect. The first method is to calibrate the model during years when
the wellfield was in operation, estimating the appropriate leakance
rate, DEEPFR, for this period. Verification of the simulation during
years prior to wellfield operation would either confirm that no change
in the leakance rate had taken place or suggest that the rate DEEPFR
should be lowered for the pre-pumping period.
The second method would involve the use of a continuous lateral
outflow from active groundwater storage using a time series. HSPF
has available a time series, AGWLI, which can be used to input a
continuous inflow into the surficial aquifer; the substitution of
negative numbers can change this time series to an outflow. A percentage
of the daily pumpage, transformed into units of inches per day, would be
used in this time series to mimic the possible increased leakance of
water into the deep aquifer, over and above the normal recharge rate.
A third possibility involves the coupling of the HSPF surface
hydrologic simulation with a groundwater model. The Prickett model
of the Cypress Creek basin used by SWFWMD would be used to generate
numbers for HSPF, either a leakance rate to input as DEEPFR or a daily
groundwater storage loss which could be input using the AGWLI time series.
A quasi-three-dimensional groundwater model of ten municipal wellfields
including Cypress Creek estimated that pumping 133 mgd increased
leakage to about 9 inches per year (Hutchinson, 1984). A digital
model of predevelopment flow by Ryder (1982) estimated a downward
leakage of 5 inches per year with no pumping.
CONCLUSION
The number of parameters involved in hydrological simulation using
HSPF do not readily lend themselves to physical interpretations. All
available data should be used to estimate values for parameters, such
as nominal soil storage capacities, but these values should be viewed
for the most part as indicators, subject to change during calibration.
With this in mind, HSPF can be used with some success to simulate the
peculiar conditions of Florida watersheds.
Simultaneous simulation of surface and groundwater flows must be
accomplished using.indirect methods. The parameter DEEPFR or the time
series AGWLI are available in HSPF; consideration should be given to
linkage of these values with output from a groundwater model of the
watershed.
155
-------
REFERENCES
1. Donigian, A.S. Jr., Davis H.H. Jr., 1978, User's Manual for
Agricultural Runoff Management (ARM) Model, EPA-600/3-78-080,
EPA, Athens, Georgia.
2. Hutchinson, C.B., Johnson, D.M. , Gerhart, J.M., 1981,
Hydrogeology of well-field areas near Tampa, Florida,
Phase 1. Open-File Report 81-630, USGS, Tallahassee, Florida.
3. Hutchinson, C.B., 1984, Hydrogeology of well-field areas
near Tampa, Florida, Phase 2. Water Resource Investigations
Report 84-4002, USGS, Tallahassee, Florida.
4. Johanson, R.C., Imhoff, J.C., Davis, H.H. Jr., 1981, User's
Manual for Hydrological Simulation Program-Fortran (HSPF)
(Release 7.0), EPA, Athens, Georgia.
5. Rochow, T.F., 1983, Vegetational monitoring at the Cypress
Creek well field, Pasco County, Florida. Environmental Section
Technical Report 1983-2, SWFWMD, Brooksville, Florida.
6. Ryder, P.B.,.1978, Model Evaluation of the hydrogeology of the
Cypress Creek wellfield in west central Florida, Water Resource
Investigations 78-79, USGS, Tallahassee, Florida.
7. Ryder, P.B., 1982, Digital model of predevelopment flow in the
Tertiary limestone (Floridan) aquifer system in west-central
Florida, Water Resource Investigations 81-54, USGS, Tallahassee,
Florida.
8. Stankey, D.L., 1982, Soil Survey of Pasco County, Florida,
Soil Conservation Service, USDA, SCS, Tallahassee, Florida.
9. SWFWMD, 1982, Evidentiary Evaluation, Cypress Creek Wellfield,
CUP No. 203650, West Coast Regional Water Supply Authority,
Brooksville, Florida.
156
-------
HYDRODYNAMIC AND WATER QUALITY SIMULATIONS
IN AN ESTUARY WITH MULTIPLE OCEAN BOUNDARIES
by: Ivan B. Chou, P.E.
Applied Technology and Management, Inc.
Gainesville, Florida 32607
and Larry J. Danek, Ph.D.
ESE, Inc.
Gainesville, Florida 32602
ABSTRACT
RECEIV-II model was used to assess the hydraulic and water quality
impacts on Pungo River, North Carolina due to a proposed facility in the
Albermarle-Pamlico Peninsula. The study area is hydraulically connected with
the Atlantic Ocean by both the Albermarle Sound at the north and the Pamlico
Sound at the south. The quantity module of the model was calibrated and
verified by the results of a dye study data, 5 tide gages, 2 in situ current
meters, and the wind, precipitation and flow data. The quality module of the
RECEIV-II was calibrated and verified by two years of salinity, DO, and BOD
measurements at 12 water quality stations. The temperature, solar radiation,
chlorophyll a, and secci depth data were used to establish the photosynthesis
and respiration rate constants. The benthic oxygen demand was also measured
on site.
The effects of wind, tide, and plant discharge were assessed by the
calibrated model. It was found that under low flow conditions, the tide
range was the most sensitive hydraulic parameter in water quality predictions.
Also, the wind direction, under certain circumstances, could have significant
effects on the water quality in a multiple ocean boundary estuarine system.
The wind induced setup will cause net circulation in the estuary and
consequently increase the ocean exchange.
157
-------
INTRODUCTION
North Carolina has an estimated 1,000 square miles of peatland contain-
ing about 600 million tons of moisture-free peat. The largest deposits are
located on the Albemarle Pamlico Peninsula, which contains an estimated
360 square miles of peatland and 210 million tons of moisture-free peat.
These large, concentrated deposits have made the Peninsula an area of primary
interest for peat production.
Peat Methanol Associates proposed a synthetic fuels project on the
eastern peninsula of North Carolina near Creswell. The development would
include the construction of a peat-to-methanol plant. To evaluate the
potential environmental impacts of the plant discharge, a comprehensive
environmental study was conducted between October 1981 and December 1983 to
establish the baseline information for hydrology, water quality, terrestrial
and aquatic ecology in the study area, and to predict the hydrological and
water quality impacts caused by the proposed facilities.
The hydrological and water quality impacts were assessed by model
simulation using RECEIV-II model.
STUDY AREA
The Albemarle-Pamlico Peninsula (Figure 1), located in the tidewater
region of North Carolina, is bounded on the east by Pamlico and Croatan
Sounds, the west by the Suffolk Scarp (an ancient marine terrace), the
north by Albemarle Sound, arid on the south by the Pamlico River. The study
area and proposed plant site are located near the center of the peninsula
within the region of peat deposits. Operations to develop the area began as
early as 1787, with the completion of a canal that drained 10,000 acres for
farming. Drainage and development projects continued in the area with
completion by 1843 of New Lake Canal and Pungo Lake Canal. By the late 1800s
and early 1900s, the area was a thriving farming and lumbering community. It
was not until the early 1970s, however, that the extensive canals and
drainage systems that exist today were developed. Drainage and subsequent
land clearing have opened the area for large-scale farming and livestock
operations.
BASELINE DATA
Intensive meteorological, hydrological, and water quality measurements
were conducted. Figure 2 shows the locations of the water quality and gaging
stations. The water quality model was calibrated primarily by three sets of
data: November 1981, April 1982, and August 1982.
METEOROLOGY
Historical and field meteorological data were collected and analyzed to
characterize the meteorological conditions near the site and support hydrology
and water quality modeling. Historical long-term meteorological data from the
158
-------
NORTH CAROLINA EUUBETH CITY
Figure 1
PROJECT LOCATION MAP
SOURCES: USGS, 1980; ESE, 1982.
-------
Phelps Lake
Figure 2
WATER QUALITY SAMPLING STATIONS AND GAGING STATIONS
SOURCES: USGS, 1980; ESE, 1982.
160
-------
National Climatic Center were available at four stations (Cherry Point,
Elizabeth City, Belhaven, and Aurora, North Carolina) in the study site
vicinity. Approximately 5 years of wind, precipitation, evaporation, solar
radiation, and temperature data were provided by First Colony Farms. An on-
site meteorological station was also maintained during the study period.
Around the study site winds were generally from the southwest or north-
northeast, usually at speeds less than 17 knots; mean annual precipitation
was approximately 50 inches with a maximum and minimum of 75 and 35 inches,
respectively; monthly evaporation rates ranged from 2.5 to 11.0 inches; and
mean temperature was approximately 65°F, with a maximum of 105°F and a minimum
of 5°F.
BATHYMETRY
A bathymetric survey of the Pungo River and adjacent tributaries was
conducted to obtain cross-sectional areas at approximately 0.5-mile intervals,
including thirty-three transects. The width of the river ranged from 66 feet
(ft) in the Pungo River Canal to 1,850 ft at the lower end of the survey
area. Average depths ranged from 5.9 to 11.6 ft.
TIDES
Site-specific tide data were collected at four tide gages for up to
16 days during the November 1981 survey at Stations Bl, PI, CMC2, and CD3.
During the August 1982 survey, three tide gages were operated for 1 month at
Stations Bl, Pi, and CB9. Figure 2 indicates the stations where tide gages
were installed. The data from these tide gages were used to calibrate the
hydraulic model.
National Oceanic and Atmospheric Administration's Tide and Tidal
Current Tables for the eastern coast were examined to obtain historical tide
data for the study site. In addition, the U.S. Army Corps of Engineers (COE)
maintained 24 years of data at the Belhaven tide gage (1957-1982). The COE
data were used to obtain such statistical parameters as mean sea level, mean
high water, and mean low water. The long-term average of the tide range at
Belhaven (0.5 feet) was used as the tidal boundary condition for model
simulation.
Wind set-up effect was noticeable on the water level data. The winds
had two distinct effects on the tides. The first was regional winds acting
generally on the Pamlico Estuary causing long-period variations in tides.
The second wind effect was noticeable but of lesser magnitude. This effect
was due to local winds producing unique short-term variations. These
variations were superimposed on the lunar tides and, with sufficiently strong
winds, would nearly obscure the tidal effect.
CURRENTS AND FLOWS
Two in situ, continuous recording current meters were installed near
stations P2 and P6 (Figure 2) and several instantaneous measurements were made
with an electromagnetic current meter at selected locations throughout the
study area. These current data were used to verify the hydraulic model. A
161
-------
weir was installed in Canal B for a 1-month period to monitor flow. Also,
three types of dye studies were used to determine the flow conditions: (1)
an 8-day continuous injection (dilution study) was conducted in Clark Mill
Creek; (2) a time-of-travel study was conducted on Canal B; and (3) net-drift
studies were conducted at Stations PI and P3. The dye studies indicated the
net flow issuing from Clark Mill Creek was about 6.0 cfs or approximately
0.2 cfs per square mile (mi2) of drainage area during the November study
period.
HYDROLOGY
To compute the hydrologic parameters for the drainage basins in the
study area, the base flow values computed from the available USGS data from
the two nearby basins were used: Herring Run near Washington, NC (16-year
record) and Albemarle Canal near Swindell, NC (3-year record). Each of the
drainage areas has topography, vegetation, land use, and soil cover similar to
the study area. Each of the drainage areas contains numerous man-made canals
which provide drainage and which are similar to canals found in the study
area. Hydrologic characteristics of both watersheds are presented in Table 1.
The Log-Pearson Type II distribution technique was used to compute the base
flow statistics from the Herring Run data under 7Q10, 30Q2, and IQ2 condi-
tions. Only the low-flow discharges for the 1Q2 flow regime could be
calculated for Albemarle Canal since the period of record was not sufficient
to calculate the 7Q10 or 30Q2 flows. The calculated base flow during low-flow
conditions compares closely with the 1Q2 base flow for Herring Run and
Albemarle Canal (0.053 versus 0.051 cfs/mi2). It was assumed that the
statistical analysis of Herring Run data could be applied directly to the
study site.
TABLE 1. HYDROLOGIC CHARACTERISTICS OF COMPARISON WATERSHEDS
Herring Run Albemarle Canal
2 2
Drainage Area: 15 mi 68 mi
Period of Record: 16 years (1965-1981) 3 years (1977-1981)
Average Flow: 10.5 cfs 102 cfs
Low Flows
7Q10 0.7 cfs
30Q2 1.1 cfs
1Q2 0.8 cfs 3.5 cfs
Base Flows
7Q10 0.047 cfs/mi2
30Q2 0.073 cfs/mi2
IQ2 0.053 cfs/mi2 0.051 cfs/mi2
Sources: USGS, 1981; ESE, 1982.
162
-------
WATER QUALITY
Water quality sampling of the Pungo River was conducted in three field
trips conducted on November 4, 1981; Aril 21, 1982; and August 11, 1982. The
water quality sampling stations are shown in Figure 2. During the Novemr
her 19, 1981 and August 11, 1982 sampling trips, morning and afternoon samples
were collected from each station. The water quality parameters include: TSS,
dissolved solids, turbidity, alkilinity, pH, DO, BOD, CBOD, chlorophyll £, 7
dissolved ions, 9 nutrient parameters, and 18 metals.
Sediment Oxygen Demand
On August 12-13, 1982, in situ SOD tests were conducted using benthic
chambers. The chambers were designed to minimize the volume-to-surface area
ratio. Both light and dark chambers were used in order to separate SOD and
photosynthesis. SOD measurements were conducted at Stations P4, P9, and Pll.
Incubation times ranged from 2.5 hours to 5 hours. The SOD rate was computed
by first subtracting out the water column BOD oxidation, and then temperature
correcting to an equivalent 20°C rate. The SOD rates at 20°C range from
0.03 gm/m2/day to 0.19 gm/m2/d. SOD values at these stations are lower than
those reported by Pomeroy and Wiegert (1981) (1) for estuarine sediments and
also lower than values for lacustrine sediments, compiled by Belanger (1981)
(2).
Primary Productivity
Net algal photosynthetic oxygen production can be a significant source of
dissolved oxygen in a system. Estimates of algal photosynthesis and
respiration (P-R) were made using the November 1981 and August 1982 Pungo
River chlorophyll _a, temperature, solar radiation, and secchi depth data, and
generally accepted phytoplankton kinetic rates.
The calculations of algal photosynthesis involved the application of a
light attenuation factor to an ideal algal growth rate and then calculating
the amount of oxygen produced during algal carbon fixation. The light
extinction coefficient was determined from secchi depth measurements. Net
algal oxygen production was greater in the downstream reach in August 1982 and
in the upstream reach in November 1981. These results correspond to the reach
which had the greater phytoplankton biomass (as indicated by chlorophyll a)
for both surveys.
Wastewater Loads
There were eight NPDES permitted dischargers within the model segmenta-
tion. The major discharger is the Belhaven WWTP which was permitted to
have an effluent of 0.5 MGD with 125.0 Ib/day monthly average BOD5-20°C
loading. The only discharge reports relevant to BOD/DO calibration were from
Belhaven WWTP which reported to have had a BODc-20°C loading of 94.0 Ib/day on
November 11, 1981, and 81.9 Ib/day on August 10, 1982. Since there was no
other discharge information available during the field sampling period, the
BOD loadings were assumed to be the discharge limitation stated in the NPDES
permits.
163
-------
MODEL DESCRIPTION AND CALIBRATION
MODEL DESCRIPTION
RECEIV-II is a dynamic, link-node hydrodynamic and water quality model.
The governing equation and numerical scheme is one-dimensional in a strict
sense; however, the model nodes can be linked to construct a two-dimensional
network and can simulate a two-dimensional, vertically integrated system. An
important feature of the model is that it can simulate multiple ocean boundary
conditions. These multiple boundary conditions had to be considered in Pungo
River modeling because the Pungo River is connected to the Albemarle Sound by
the Intracoastal Waterway and the Alligator River. The interaction of the
tidal forcing at Albemarle Sound and Pamilico sound have dynamic effects on
the circulation and tidal flushing.
The quality block of RECEIV-II can simulate the phosphorus cycle,
nitrogen cycle, coliform, chlorophyll a., BOD/DO, salinity, and metal ion
concentrations. Although abundant nutrient and chlorophyll £ data were
collected during the course of the study, only BOD/DO, salinity, and metal ion
were considered because of the high confidence level to be expected. Instead
of solving the coupled equations for nitrate, nitrite, ammonia, phosphorus,
chlorophyll £, and oxygen, the net P-R was estimated by RPW (1982)(3) using
the previously described method. The model input of SOD rates was adjusted to
include the P-R effects using the following formula:
SOD1 = SOD - (P-R) x D
where SOD' = adjusted SOD, and
D = water depth.
The RECEIV-II model uses Churchill's (1962)(4) formula to compute
reaerations rate (Ka). This method, however, could underestimate the
reaeration under low velocity conditions. Therefore, the model was modified
to impose a lower limit, Ka >^ 2/D, suggested by O'Conner (1978X5), for
reaeration coefficient, where Ka is the reaeration rate in day'l, and D is
the water depth in feet.
The quantity block of the RECEIV-II model was calibrated using the tide,
current, flow, and wind data collected in August 1982 and was verified further
using the November 1981 hydrologic and meteorological data. The quality block
of the RECEIV-II model was calibrated using chloride, BOD, and DO data
collected in August 1982, and was verified further using the November 1981
chloride, BOD, and DO data. A dye study conducted in November 1981 was used
to determine the freshwater discharge and to calibrate the quality block of
the model.
Several parameters need to be determined by model calibration. These
are:
1. Manning's roughness coefficients (n),
2'. ocean exchange coefficients,
3. BOD deoxygenation rate,
4. SOD, and
5. net photosynthesis and respiration (P-R) rate.
164
-------
MODEL SEGMENTATION
RECEIV-II model was set up for the Pungo River system using 48 nodes and
49 channels. The segmented model included the Pungo River Canal, Pungo River,
tidally influenced portions of Canal B and Canal D, Clark Mill Creek, the
Intracoastal Waterway and portions of the Alligator River. Figure 3 shows
the locations of all model nodes. The shortest segment in the system was
2000 feet, which prompted choosing a hydraulic time step of 90 seconds to
maintain numerical stability.
HYDRODYNAMIC CALIBRATION
The RECEIV-II model is a quasi-steady dynamic model which assumes the
tide to be periodic and can accept a maximum of three tidal frequency
components. (The term "quasi-steady" is defined as a periodic time varying
condition with constant amplitude and phase lag. The function changes with
time, but the harmonic cycle is repeated.) Since the observed tide data at
all gages were highly irregular, a piecewise calibration technique was used
to simulate the non-periodic wind effects as well as the tidal effects. The
tide record used for the 1982 calibration period was from 00:00 hr on August
29 to 06:00 hr on August 31, and it was divided into three segments:
1. 00:00 hr August 29 to 04:00 hr August 30,
2. 16:00 hr August 29 to 16:00 hr August 30, and
3. 08:00 hr August 30 to 06:00 hr August 31.
There was an overlapping period between segments.
Following the "cold starting" with arbitrary initial conditions, the
water level at nodes and current velocities at channels at the 16th hour Of
the first calibration segment were used as initial conditions for the second
calibration period. Similarly, the water levels and velocities at the 16th
hour of the second calibration segment were used as initial conditions for
the third calibration segment. Different wind conditions were used for each
time segment. This restarting method normally causes some discontinuity
between calibration periods due to numerical truncating error, secondary
dynamic effects, and non-harmonic boundary inputs. However, the results in
the overlapping periods had shown that the effects of discontinuity were
insignificant. The maximum mismatch in the overlapping sections was 0.06 ft.
The simulated water levels were compared with the 1982 tide data, at Belhaven,
PI and CB9, as shown in Figure 4. The simulated current velocities from the
model were compared with current meter data at P2 and P6 as shown in Figure 5.
After the quantity block of the RECEIV-II model was calibrated using
August 1982 hydrologic data, the model was verified using November 1981 data.
The tide record used for the 1981 calibration period was from 20:00 hr
November 2 to 00:00 hr November 4, 1981. The results of the 1981
verification using tide data at Belhaven, PI and CMC2 are shown in Figure 4.
The results of the hydrodynamic calibration showed that the simulated
water level followed closely with the observed tide data at four locations
using two independent sets of data. The maximum difference between the model
simulation and tide data was about 0.1 foot at P2 and smaller at other
locations. The phase of both the simulated tide and current agree closely
165
-------
cr»
7 (5.9 MILES UPSTREAM FROM 13)
.31 (3.4 MILES DOWNSTREAM FROM 30)
iii.(9.0 MILES DOWNSTREAM FROM 30. MOUTH OF PUNGO RIVER)
Lake Mattamuskeet
Figure 3
MODEL SEGMENTATION
SOURCE: ESE, 1982.
-------
10
f
ol L
IO
05
I
KEY
TIDE DATA
MODEL SIMULATION
PI TIDE
CUC2TIDE
I
I
OS
I
05
!
15
10
OS
0 080O WOO 0000
TIME (NOV. 3,1981)
HYDROOYNAMIC CAUBRATION (1981 TIDE)
Figure 4
HYDRODYNAMIC CALIBRATION AND VERIFICATION
SOURCE: ESE, 1982.
I I
8 16
8/29
KEY
TIDE DATA
.MODEL SIMULATION
I I I _!_
BELHAVEN TIDE
I L I I
CBS TIDE
I I I I
8 18
8/30
(1M2)
(1982 TIDE)
PI TIDE
0 8
8/31
-------
KEY
* CURRENT DATA
MODEL SIMULATION
0.5
0.0
P6 CURRENT
in
-0.5
TIME (1982)
Figure 5
HYDRODYNAMIC CALIBRATION (1982 CURRENT)
SOURCE: ESE, 1982.
168
-------
with the data. The error of simulated current velocity was generally larger
than tide because the current meter was located at a single point in channel
cross section, while the simulated current was a cross-sectional average.
The calibrated Manning's n ranges from 0.027 to 0.035.
Dye Calibration
A dye study was conducted by ESE between November 6 and November 14,
1981. The purposes of this dye study were to determine the freshwater
discharge from Canal D and Clark Mill Creek and to ensure that the model could
properly simulate the transport of the discharge plume in the Pungo River
System.
Dye injection with an initial concentration of 4.6 x 10 ppb was begun at
08:41 AM on November 6, 1981, at a rate of approximately 74 ml/min. Automated
Isco samplers, indicated that the dye had stabilized approximately 7 days
after injection began; consequently, the dye study was begun on the eighth day
(November 14, 1981). Dye concentrations were measured at five stations on the
Pungo River and Clark Mill Creek as frequently as possible during a tidal
cycle. Generally, these stations were sampled once every half hour except
when one of the other dye surveys was being conducted. In addition, two
longitudinal dye surveys (morning and afternoon of November 14, 1981) were
conducted on the Pungo River. The in situ photodecay rate of Rhodamin dye was
measured to be 0.0172 day~l during the dye study period, and this value was
used in the model for calibration.
The results of the dye calibration indicated that the freshwater flow
during November 1981 sampling period was 0.22 cfs/mi^. Figure 6 shows the
comparison between mode predicted maximum dye concentration on November 14,
1981 and the measured values.
WATER QUALITY CALIBRATION
The objectives of the salinity calibration were to determine the ocean
exchange coefficients at the boundaries, and to verify that the model can
adequately simulate the diluting effects from the ocean boundary. The results
of chloride calibration showed that the appropriate exchange coefficients were
0.8 at nodes 48 and 0.5 at node 41. Figure 6 also shows the chloride
calibration on a long-term average basis and also for dry and wet seasons.
The Pungo River system was found to be insensitive to the ocean exchange
coefficient especially under low-flow and small tide range conditions.
BOD/DO Calibration
The purpose of BOD/DO calibration is to determine the proper values of
the BOD deoxygenation rate (Kd), SOD rate, P-R rate, and distributed BOD
loading to be used for RECEIV-II simulation. The water quality data collected
in August 1982 were used for calibration, and the November 1981 data were used
for model verification. The calibration was an iterative process in which the
aforementioned parameters were altered until the model simulation represented
a best fit to the observed BOD/DO data. However, the parameter values were
169
-------
KEY
MAXIMUM OBSERVED DYE CONCENTRATION
ON NOVEMBER 14,1981
-MODEL SIMULATED MAXIMUM DYE CONCENTRATION
MODEL CALIBRATION DYE STUDY
s
UJ
IU
200
150
100
SO
10
CLARK MILL CREEK
10 5 0
RIVER MILES
OYE CALIBRATION
30
25
RIVER MILES
PUNQO RIVER
KEY A APRIL, 1885
MODEL Q NOVEMBER, 1981 AM O AUGUST, 1982 AM
"""" SIMULATION NOVEMBER, 1981 PM AUGUST. 1983 PM
10
8
e
10
35
I I
PI
LONG-TERM
(AVERAGE)
(AUQ. 1982)
SEASONAL
I I
30 25
20 15
RIVER MILES
II I I I I I I
10
J I
10
11 12
2O
Figure 6
DYE CALIBRATION AND SALINITY CALIBRATION
NODE
SALINITY CALIBRATION
48
SOURCE: ESE, 1982.
-------
not determined by curve fitting only; they were also supported by the field
data and laboratory analysis.
The carbonaceous BOD (20°C) was measured in the laboratory by incubating
the water sample for 50 days. The ultimate carbonaceous BOD and BOD
deoxygenation rate (Kd) was determined using Barnwell's (1980)(6) nonlinear
least-square method. Figure 7 shows the calculated Kd values at sampling
stations.
According to the characteristics of the estuary, the Pungo River was
divided into three reaches: reach 1, from river mile 23 to 32; reach 2, from
river mile 19 to 23; and reach 3, from river mile 0 to 19. Reaches 1 and 2
were shallow and narrow channels with occasionally turbid water. Reach 3 was
deeper, wider, and less turbid. A Kd (20°C) value of 0.15 dayl was used for
reach 3, and the Kd (20°C) value of 0.12 day1 was used for reaches 1 and 2.
A temperature correction factor of 1.047 was used for Kd.
According to RPW's (1982)(3) analysis of the in situ SOD tests, the SOD
rate in reach 3 was computed to be 0.14 gm/m^/day at 29°C. An SOD rate of
1.0 gm/m2/day was used for reach 2, and an SOD value of 4.0 gm/m^/day
was used for reach 1. The high SOD value was used in the upstream reach
because of the effect of swamps in the upper Pungo River and Pungo River
Canal. The 1981 DO calibration indicated that a temperature correction factor
of 1.08 was appropriate for the SOD rate.
The photosynthesis-respiration rates for August 1982 at 29°C were
computed to be 0.63 mg/L/day for reach 3, and 0.21 mg/L/day for reaches 1 and
2. The DO calibration, using November 1981 data, indicated a temperature
correction factor of 1.055 could be applied to the P-R rate.
The background BOD loading was assumed to be proportional to the water
surface area. The lineal distribution and volumetric distribution of BOD
loading methods were also attempted, but the area distribution seemed to
function better. The 1981 and 1982 BOD calibrations indicated that the
distributed BOD loadings in August 1982 at 29°C were 1.0 gm/m2/day for
reaches 1 and 2 and 2.5 gm/m2/day for reach 3. A temperature correction
factor of 1.059 was used for the BOD distributed loadings. A summary of the
rate constants obtained from BOD/DO calibrations is presented in Table 2.
Figure 8 shows the 1981 and 1982 BOD/DO calibration.
171
-------
KEY
D NOVEMBER 1981, AM
NOVEMBER 1981, PM
O AUGUST 1M2, AM
AUGUST 1982, PM
K* VALUES FOR MODEL INPUT
0.28
024
0.20
0.16
0.12
0.08
0.04
0.00
0
I
35 30 25 20 15
RIVER MILES
I I II I I I I I I
10
P1 23456789
STATION
I I I I I I I I Illllll
10
11 12
Figure 7
BOD DEOXYGENATION RATE (Kd)
SOURCE: ESE, 1982.
172
-------
KEY
O AUGUST. 1M2 AM CROSS-SECTIONAL AVERAGE
AUQUST, 1982 PM CROSS-SECTIONAL AVERAGE
MODEL SIMULATION
1
I
I
38
30
25
20 15
RIVER MILES
J 1 J I I
10
PI
23 4
I i
56 7 89
STATION
10
11 12
NOOK
U>
16
12
35
30
25
I I I
20 15
RIVER MILES
A J L 1 I
1O
11 12
NODE
BOO/00 CALIBRATION (1982 DATA)
Figure 8
BOD/DO CALIBRATION AND VERIFICATION
KEY
0 NOVEMBER. 1M1 AM CflOSMECTlONAL AVERAGE
NOVEMBER, 1981 PM CROSS-SECTIONAL AVERAGE
MODEL SIMULATION
§"
8
I
35
30
25
I L
20 15
RIVER MILES
j I I I I I
10
I
PI 23456789 10
STATION
until i M i i i
11 12
NODE
8
*- ;>
12-
1 1
8-
A ji_-
r
0 I -. I I
35 30 25
II II
PI 23
1 1 II 1 1 1 Illl
I lil ?
xHfl '
i i i
20 15 10
RIVER MILES
1 II ! 1 1 1
456789 10
STATION
[nil ML) I I I I I I
-J-i- \
rr I
i i
5 0
! I I
11 12
I I I
1312345 6jyjj;j^
(101112141919
48
282930 31
NODE
BOO/DO CALIBRATION (1981 DATA)
SOURCE: ESE, 1982.
-------
TABLE 2. RATE CONSTANTS FOR BOD/DO SIMULATION
Parameter
Distributed BOD
loading (gm/m^/day)
SOD (gm/m2/day)
P-R (mg/L/day)
K (day1)
Reaerations (day"1)
*Reach 1 = River mile
Reach 2 = River mile
Reach 3 = River mile
Reach*
1., 2
3
1
2
3
1, 2
3
1, 2
3
23-32.
19-23
0-19
Base Rate
at 20°C
0.59
1.55
2.0
0.5
0.07
0.13
0.39
0.12
0.15
**
Temperature
Correction Factor
1.059
1.059
1.08
1.08
1.08
1.055
1.055
1.047
1.047
1.024
August
1982
(29°C)
1.0
2.6
4.0
1.0
0.14
0.21
0.63
0.18
0.23
November
1981
(17'C)
0.5
1.3
1.59
0.4
0.06
0.11
0.33
0.10
0.13
11.57 x U
0.969
,1.673
' and K > -r-
a - D
Where: U = flow velicity (ft/sec), and
D = water depth (ft).
Source: ESE, 1982.
RESULTS
The 7-day-10-year low flow condition was used as the hydraulic input for
model projection. Sensitivity analysis was conducted to test the effects of
tide range, wind speed and wind direction on the water quality projections.
The results showed that, among these three parameters, tide range is the most
sensitive one. Under a low flow condition, changes in tide amplitude will
cause large changes in tidal prism and consequently the diluting flows.
Figure 9 shows the metal ion dilution ratio under neap tide (0.2 ft), average
tide (0.5 ft), and spring tide (1.35 ft) conditions.
According to the results of quantity simulation, the southwest wind
stress at-water surface will cause water level setup in the upper Pungo River.
The setup at station Pi is about 0.1 ft. This wind surge will build up a head
differential between the two ends of the Introcoastal Waterway, and drive the
water in Pungo River through the Intracoastal Waterway toward Albemarle Sound.
174
-------
KEY
TIDE RANGE - 0.2 «
TIDE RANGE - M «
TIDE RANGE 1.35 ft
C » PREDICTED METAL CONCENTRATION OF PUNGO RIVER
Co - PROPOSED METAL CONCENTRATION OF PLANT DISCHARGE
OOO
35 30
I I
25 20 15
RIVER MILES
I I I I I I LL I
5 0
I I
P1
23 4 56 789
STATION
I I I I I I I IIII HIM Illl I I I i 1
10
11 12
NODE
Figure 9
TIDE EFFECTS ON METAL ION CONCENTRATION
SOURCE: ESE, 1982.
175
-------
The model shows that a southwest wind will cause a flow of 1,560 cfs
(0.55 ft/sec) in the Intracoastal Waterway toward Albemarle Sound. Similarly,
the northeast wind causes a 0.07-ft water level set-down at station PI and a
1,180-cfs flow (0.32 ft/sec) in the Intracoastal Waterway toward the Pungo
River. This wind induced current in the Intracoastal Waterway was
qualitatively verified by field observation. In both cases, the wind will
carry the water from Albemarle Sound or Pamlico Sound toward the confluence of
the Pungo River and the Intracoastal Waterway (node 24). In effect, node 24
becomes an ocean exchange boundary for the upper Pungo River. Therefore, the
wind will increase the dilution ratio in the Pungo River.
Figure 10 shows the effects of wind direction on BOD/DO in the Pungo
River under northeast and southwest (the prevailing wind directions)
conditions. The results show that the northeast wind caused higher BOD
concentration in the Pungo River near the confluence with the Intracoastal
Waterway. The northeast wind tends to carry the water from the Intracoastal
Waterway, which has a high BOD concentration, into the Pungo River, therefore
increasing the BOD near node 24. The simulations of the conservative
substances also indicate that the wind from either direction will reduce the
pollutant concentration in the upper Pungo River.
It is evident that the wind direction has important effects on the water
quality in a multiple ocean boundary system. The wind set-up is effective in
creating net circulation in the estuary while the tidal flushing alone has
gradual effects, especially near the upstream portion of the river. The wind
induced circulation will shift the influence of the ocean boundary further
upstream of the estuary and it will accelerate the tidal exchange.
ACKNOWLEDGEMENT
This study was funded by Kopper, Inc. and conducted by ESE, Inc.,
Richard P. Windfield, and Law Engineering.
176
-------
I '
UI
0
10
KEY
7Q10 PLOWS
TIDE RANGE - 0JS H
PLANT DISCHARGE - 1.6 Cfs
WIND DURATION « 30 days
NE WIND
-------
REFERENCES
1. Pomeroy, L.C, and Wiegert, R.G. 1981. The Ecology of a Salt Marsh.
Springer-Verlag, Publisher.
2. Belanger, T.V. 1981. Benthic Oxygen Demand in Lake Apopka, Florida.
Water Research, 15: 267-274.
3. Winfield, R.P. 1982. Canal B Water Quality Analysis. (Unpublished
report).
4. Churchill, M.A., Elmore, H.L., and Buckingham, R.A. Prediction of
Stream Reaeration Rates. Proc. ASCE, Jour. San. Eng. Div. SA4.
July 1962. p. 1.
5. 0'Conner, D.J. 1978. Fate of Toxic Substances in Natural Waters.
Mathematical Models of Natural Water Systems. Manhattan College
Summer Institute.
6. Barnwell, T.O., Jr. 1980. Least Squares Estimates of BOD Parameters,
ASCE, Div. of Env. Eng., pp. 1187-1202.
178
-------
MODELING ESTUARINE PHYTOPLANKTON-NUTRIENT DYNAMICS USING MICROCOMPUTERS
by: Wu-Seng Lung, PhD, PE
Department of Civil Engineering
University of Virginia
Char lotrtesvi lie, VA 22901
ABSTRACT
A tidally averaged water quality model was used to study the factors
limiting the phytoplankton growth in the James River Estuary in Virginia under
dry weather conditions. The water quality parameters calculated by the model
are CBOD, dissolved oxygen, organic nitrogen, ammonia nitrogen, nitrite plus
nitrate nitrogen, organic phosphorus, inorganic phosphorus, and chlorophyll a..
The model incorporates major physical, chemical, and biological processes
which link these water quality parameters. The kinetics are modeled in each
of the 50 segments in the longitudinal direction of the estuary.
Model calibration analysis using two different data sets collected during
the July (low water slack) and September (high water slack), 1983 was very
successful. The model reproduces these two data sets using a consistent set
of kinetic coefficients. The phytoplankton growth plays a significant role in
contributing to ultimate CBOD through respiration of live phytoplankton as
well as recycling of carbon from algal biomass and in balancing the dissolves
oxygen budget via photosynthesis. More importantly, turbidity plays a key
role in limiting the phytoplankton growth and nutrients (nitrogen and
phosphorus) are not found to be a key limiting factor in most cases.
The execution of this time variable model on an IBM PC (with 256K RAM) is
successful. Further, the use of the 8087 math coprocessor significantly
reduces the run time on the PC. Plotting software developed for this study
generates report quality plots of data and of model results.
179
-------
INTRODUCTION
In the past few years, microcomputer systems have become quite powerful
in terms of their speed and core memory size that they can handle most of
water-quality modeling analyses which were once considered not possible on a
microcomputer. Parallel advancement in microcomputer software has offered
many user-friendly features which significantly improve the efficiency of
running some complicated water quality models. Such improvement in both
hardware and software already have had considerable impacts on wasteload
allocation (WLA) studies as the so-called water quality based approach in
water pollution control calls for sharp increase in the use of water quality
models in WLA and NPDES permit related studies. Thus, easy access to
computing facility and the associated user friendly software is important to
the success of many WLA and other water quality modeling studies. To date,
many large modeling codes such as WASP (Water Quality Analysis and Simulation
Program) have already been installed and run on microcomputer systems
successfully.
This study employed an existing water quality model of the upper James
River Estuary in Virginia to assess the water quality impacts of potential
point source phosphorus control programs in the James River basin (3). This
paper presents the results of model calibration and sensitivity analyses using
the most recent data currently available on the James. An important feature
of this modeling analysis is that the entire modeling analysis was carried out
on a microcomputer system, ranging from data plotting, model running, model
result printing and plotting. Using such a microcomputer system proves to be
much more efficient than using many mainframe computer systems. In addition,
the microcomputer system used in this study has very common configuration and
a quite modest price tag.
APPROACH AND METHODOLOGY
The James River Estuary model (JMSRV), which was developed by Hydro-
science, Inc. (1) for the Virginia State Water Control Board (SWCB), was used
in this modeling study. The model was originally intended for use in
wasteload allocations of BOD loads from municipalities and industries to meet
the dissolved oxygen (DO) standard in the James. The current version installed
on a COMPAQ microcomputer is a 50-segment one-dimensional tidally averaged
model. [See Figure 1 for the model segmentation.] The kinetic structure of
the model is shown in Figure 2. In addition to the BOD/DO kinetics, phyto-
plankton biomass/nutrient dynamics is also incorporated in the model. As
such, the model can be used, in a first-cut analysis, to assess the eutro-
phication potential and the impact of point source phosphorus load reduction.
The JMSRV model was originally calibrated using the data collected in
1976 and 1978 (1). Although the water quality problem of concern at that time
was dissolved oxygen and the emphasis of the modeling analysis therefore was
on the verification of the BOD and DO concentrations, the modeling analysis
also calibrated the kinetic coefficients of phytoplankton-nutrient dynamics in
the upper James River Estuary. In this study, these coefficient values were
first used in the preliminary model calibration with the most recent data
180
-------
*C'
DO
CBOD
D.O.
Wn^
RESPIRATION
ORG.-N
NOZ+ NO,
Wp-
.-P
K,
%-BENTHIC
DEMAND
PHOTOSYNTHESIS
ALGAE
INORG.-P
MORTALITY
Woo1 DISSOLVED OXYGEN OF WASTEWATER
We CBOD WASTE LOADS
Wn:NITROGEN WASTE LOADS
WpsPHOSPHORUS WASTE LOADS
K0s DISSOLVED OXYGEN REAERATION
Kd»DEOXYGENATlON COEFFICIENT
Kn: NITRIFICATION COEFFICIENT
KHns HYDROLYSIS RATE OF ORGANIC NITROGEN
KHp*CONVERSION RATE OF ORGANIC PHOSPHORUS
Kg = GROWTH COEFFICIENT FOR PHYTOPLANKTON
KSo« SETTLING RATE FOR PHYTOPLANKTON
KSfl* SETTLING RATE FOR ORGANIC NITROGEN
KSpa SETTLING RATE FOR ORGANIC PHOSPHORUS
Figure 1. James River Model (JMSRV) Kinetics
(from Hydroscience, 1980)
181
-------
C»
RICHMOND
ft
MOC10N3 C«
MCNHICO ITP
UPPER JAMES RIVER
APPOMATTOX RIVER
ALB.CHEM. (HOPEWELL
HOPEWELL STP
HOPEWELL
N
PHIL. MOM
ALO. CHCN
(CHESTERFIELD)
CHICKAHOMINY
RIVER
Figure 2. James River Model (JMSRV) Segmentation
(from Hydroscience, 1980)
-------
currently available. Such a model calibration effort is necessary to update
the model for changes, if any, in model coefficients (e.g., rate constants,
boundary conditions, loading rates, etc.) and to understand the estuarine
system under existing conditions.
The water quality data collected by the Richmond Regional Planning
District Commission (RRPDC) in the summer of 1983 under the James River Water
Quality Monitoring Program (2) were used in this modeling study. The data
from two slack water surveys were chosen for use in this modeling study. The
summer of 1983 was characterized by a prolonged period of warm temperature and
low flow. The July 28 (low water slack) survey was characterized by the
highest freshwater flow (combined flow near Richmond = 2380 cfs) among all
surveys while the September 20 (high water slack) survey was conducted under a
relatively low flow (combined flow near Richmond = 1140 cfs). The receiving
water quality data and the associated point source monitoring records were
utilized in this study.
The JMSRV model was used to analyze these data related to these two
surveys. In addition, model sensitivity analyses were conducted to fine tune
the model and to identify the factor(s) limiting phytoplankton growth in the
upper James River Estuary.
As indicated earlier, the modeling analysis was conducted using a micro-
computer system. A brief description of the system is included in the Appendix.
MODEL CALIBRATION AND SENSITIVITY ANALYSIS
The JMSRV model was incorporated with the hydrologic and environmental
conditions of the James River associated with the September 20 survey. The
point source loads (CBOD^, organic nitrogen, NH"3, NO +NO , organic phosphorus,
and ortho-phosphorus) shown in Table 1 were also incorporated into the model.
The modeling analysis assumes that the estuarine system is under an
intertidal steady-state condition. In reality> however, steady-state condition
rarely exists, nor does a steady dry weather river flow. In fact, the James
River flow fluctuated widely on a day-to-day basis. To better approximate a
steady water quality condition observed on September 20, it is necessary to
use an average river flow over a period of 7 to 10 days prior to the survey.
An average flow of 1100 cfs was used to best represent the flow near Richmond.
The results of model calibration using the September data are summarized
in Figure 3 for CBOD4Q, NH3, N02+N03, ortho-P, chlorophyll a, and dissolved
oxygen. The calculated ultimate CBOD (CBOD ) concentration are compared with
the observed 40-day CBOD (CBOD,_) data. The long term BOD analysis indicates
complete decay of the organic materials in the river samples by 40 days.
Thus, the measured CBOD,_ values closely represent the ultimate CBOD and can
be compared with the calculated CBOD without serious errors.
183
-------
a. a. a.
20
15
a 10
O
aa
o
5
O
O CJ
E Ol
JC CO
1«
-------
Table 1. Major Wastewater Loadings (Ibs/day) for
September 20, 1983 Survey
Discharger
Richmond
DuPont
Falling Creek
Proctors
Creek
Reynolds
Metals
Am. Tobacco
ICI
Philip Morris
Allied-
Chester
Allied-
Hopewell
Hopewell
CBOD
40
4512.3
202.8
714.7
2602.0
1.8
60.8
31.9
368.4
2480.3
12680.9
8929.1
Org. N
4927.5
230.9
336.0
Total P Org. P Inorg. P
3916.7 2332.9
38.9 9.6
116.1 745.2
208.5 103.4 33.8
3.9
27.2
8.0
27.0
42.9 3.1 61.2
3363.8 2069.0 2349.3
7048.3 5904.6 326.8
2328.4
5.6
502.1
179.3
0.0
1.0
0.7
6.7
2.1
31.2
4.8
267.7
2.3
6.8
1.4
106.2
9.2
80.1
347.7
144.4
2.8
111.2
25.2
2.2
0.7
1.4
39.4
6.1
66.7
205.2
2184.0
2.8
390.9
154.2
0.2
6.2
0.0
66.8
3.1
13.4
142.5
Figure 3 indicates that the model results reproduce the observed trends
of the water quality parameters. The addition of CBOD recycled from phyto-
plankton biomass (the CBOD curve labeled as 'with algal') matches the observed
data reasonably well. Note that the CBOD curve without algal is consistently
below the observed data. Thus, it is important to include the oxygen demand
of decayed phytoplankton biomass in the area where phytoplankton growth is
significant.
The JMSRV model was also applied to analyze the July 28, 1983 data set.
Specific hydrologic (combined flow near Richmond = 2,200 cfs averaged over a
week prior to the survey) and environmental conditions were incorporated along
with the point source loads (Table 2) into the model. The results of model
calibration are summarized in Figure 4. The model calibrations match the
observed data reasonably well. Note that the relatively higher flow in July
slightly reduces the nutrient concentrations while compared with the September
concentrations. However, the chlorophyll a levels remain about the same
between the two surveys. The major difference between the calibration results
of the two data sets is in the saturated growth rate of phytoplankton (2.2/day
for the September condition and 2.0/day for the July condition).
Model sensitivity analyses were designed to vary the model coefficients
to reproduce the data and, therefore, would enable us to better understand the
phytoplankton growth mechanisms. The final product of the sensitivity analysis
is a fine tune model which would reproduce the two data sets using a consistent
set of model coefficients. That is, the results from the sensitivity analysis
would reaffirm the earlier model calibration results. The September survey
data were used in the sensitivity analysis.
185
-------
2r .p.
20
^~N
v.
W*
£ 15-
o
Q 10'
O
CD
a
5
n
CO
a
c
g
f^
o
5
-4
a. a.
CO CO
^ -^
0 CJ
0)
C U
_ o
_ fl
<0
u.
_» . .
Q.
co
_
<0
<0
§
X
a j
/
S t
^
w/algae
7^"^I
/o aTgaeT
2.5
2.0
a? 1.5
0.5
0.0
2.5
02.0
O)
E
'1.5
O
1.0
(N
ifl.5
0.0
110 100 90 80 70 60 50
River Miles from Mouth
110 100 90 80 70 60 50 40
110 100 90 80 70 60 50 40
40
1.0
0.8
a.
o0.4
JT
0.2
0.0
100
80
o
I
L.
_g
o
40
20
12
10
=- a
J, 6
o
Q 4
Q- 0-0-
i| III
CO CO CO
C O O
E 0>
o o
Qi Q.
-------
The model assumed no nitrification in the James River from Richmond to
Hopewell (the first 30 model segments) according to Hydroscience (1). There
are, however, widespread speculations on whether nitrification is occurring in
this section of the James. Nitrifying bacteria data collected in 1983
(1) could not provide a firm answer. Additional field studies to quantify the
growth potential of nitrifiers are underway but their results are not
available at the present time. In this modeling, a number of nitrification
rates ranging from 0.05/day to 0.15/day were incorporated into the model for
segments 1-30 (from Richmond to Hopewell). The results of the sensitivity
analysis are presented in Figure 5. The nitrification rate of 0.05/day (at
20°C) gives the best fit to the data among the rates tested, considering the
reproduction of the NHq, NO 4-NO-, and DO data.
J £ J
Table 2. Major Wastewater Loadings (Ibs/day) for
July 28, 1983 Survey
Discharger
Richmond
DuPont
Falling Creek
Proctors
Creek
Reynolds
Metals
Am. Tobacco
ICI
Philip Morris
Allied-
Chester
Allied-
Hop ewe 11
Hopewell
CBOD
40
Org. N
NH,
N02+N03 Total P
Org. P Inorg. P
5642.1
427.5
1067.2
312.5
3.3
16.3
17.9
485.7
3859.1
16502.0
10347.6
1282.7 3216.5 1379.2
.217.3 0.0 63.1
398.0 328.8 311.5
312.5
45.3
36.2
2314.7
12.6
461.5
156.2
0.7
60.1
8.0
26.7
0.0
14.3
0.6
8.8
0.9
3.1
4.6
351.4
0.6
40.3
1.4
140.0
46.5
3.6
35.7
1163.3 1055.1 1514.9
5046.9 6989.5 429.5
0.0
60.9
322.1
144.7
6.3
109.6
64.3
0.4
17.6
1.4
52.0
0.0
47.3
119.1
2170.0
6.3
351.9
91.9
.2
,7
0.
22.
0.0
88.0
0.0
13.5
203.0
Based on the preceding nitrification analysis, the model was then tested
with different growth rates of phytoplankton ranging from 2.0/day to 3.0/day.
The results of the sensitivity runs are shown in Figure 6. At the lowest
algal growth rate of 2.0/day, nitrogen is shifted from the phytoplankton
biomass to N0_+N0 , resulting in lower chlorophyll a level and slightly lower
dissolved oxygen concentrations in the area between river miles 70 and 90. On
the other hand, a growth rate of 3.0/day produces a chlorophyll j. peak about
85 ug/1. A growth rate of 2.2/day seems to produce a close fit of the
September data (Figure 6).
Hydroscience (1) suggested that the losses of ammonia nitrogen and
ortho-phosphorus between rives miles 90 and 80 may be due to inorganic
nutrient uptake by rooted aquatic weeds in the marshes and oxbows in this
stretch of the river. Since no data is available to confirm this hypothesis,
an empirical approach is taken in this analysis to incorporate a loss rate of
187
-------
25
20
15
a 10
O
CD
a
51
a. a.
8 " °
60). 2
£ C O ID
00
J,
O
o
cs
ifl.5
0.0
(/day)
110 100 90 80 70 60 50
River Miles from Mouth
110 100 90 80 70 60 50 40
110 100 90 80 70 60 50 40
40
1
. o
.
\.
O)
1°
0.
oO
-~
°o
0
.8
.6
.4
.2
0
to
TJ
C
g
r^
U
ce.
I
*
CL
to
^
O
O)
c
1 6
O
Q 4
110 100 90 QQ 70 60 50
40
110 100 90 80 70 60 50 40
110 100 90 80 70 60 50
River Miles from Mouth
40
Freshwater Flow near Richmond: 1100 cfs
Water Temperature: 26°C
Legend:
{
Observed Data
Average and Range
Model Results
Figure 5. Model Sensitivity Results - Nitrification
188
-------
25
110 100 90 80 70 60 50 40
2.5
02.0
Ol
6
1.5
o
z
1.0
i
CN
io.sf
0.0
110 100 90 80 70 60 50 40
River Miles from Mouth
110 100 90 80 70 60 .50 40
110 100 90 80 70 60 50 40
River Miles from Mouth
Freshwater Flow near Richmond: 1100 cfs
Water Temperature: 26°C
Legend:
Observed Doto
Average and Range
Model Results
Figure 6. Model Sensitivity Results - Phytoplankton Growth Rate
189
-------
ortho-phosphorus (0.5 ft/day) into the model. Such a loss rate may include
not only the uptake by aquatic weeds, but also some other mechanisms such as
phosphorus adsorption by sediments. Figure 7 shows that incorporating such a
loss rate brings the calculated ortho-phosphorus concentrations closer to the
data.
DISCUSSIONS
The JMSRV model is now calibrated reasonably well (see Figures 3 and 4)
with a consistent set of model coefficients (see Table 3) using two data sets
from 1983. Additional insights into the estuarine system can be summarized
from the model calibration and sensitivity analysis results. First, the
location of the phytoplankton biomass peak moves up and down the estuary with
the freshwater flow. Between the two water quality surveys, the July
condition (associated with a freshwater of 2,200 cfs in Richmond) produced a
phytoplankton biomass peak near river mile 70. On the other hand, the lower
freshwater flow of 1,100 cfs in the September survey moved the peal upestuary
to river mile 75.
The question of nutrient limitation can be explored from the model
results. Figure 8 shows the degrees of nutrient -limitation (nitrogen and
phosphorus) on phytoplankton growth in July and September, 1983. Both
nitrogen and phosphorus are not limiting the growth rate as the
Michaelis-Menton limitation ratios are close to 1.0 (practically no reduction
in growth rate). Further, light appears to be a major factor in reducing the
growth rate. Figure 9 shows that the specific growth rates of phytoplankton
(/day) are significantly reduced in the turbid water along the estuary in
September, 1983. In an earlier study on the lower James River Estuary,
Neilson and Ferry (4) suggested that factors (other than nutrients), such as
turbidity, mixing, and zooplankton grazing, are likely to control
phytoplankton growth.
SUMMARY AND CONCLUSIONS
The JMSRV model has been calibrated using the most recent data currently
available on the upper James River Estuary. The model sensitivity analyses
indicate that the calculated phytoplankton peak in the upper James River
Estuary is sensitive to the saturated growth rate of phytoplankton. The model
calibration results also indicate that nutrients (nitrogen and phosphorus) are
the key limiting factor for phytoplankton growth. Rather, light or turbidity
is the major limiting factor under existing conditions. A similar finding was
stated by Neilson and Ferry (4) in an earlier study of the lower James River
Estuary.
The modeling analysis was conducted on a microcomputer system. The
execution of the compiled program proves to be very efficient with the help of
a math co-processor 8087. In addition, specially designed software programs
for the microcomputer system were used to generate the output in graphical
form using a HP plotter. The entire operation using the microcomputer system
offers many advantages over the execution on some mainframe systems.
190
-------
25
20
X^
o>
J. 15
o
Q 10
O
00
0
5
n
Q- Q- Q- Q.
to co co co
T3 ^ ^
COO
O <»
E Q) 3
-CCO vj ~ 0 3
{\SL
a m ^~- _, ^^^^^^"~ a
110 100 90 80 70 60 50 40 110 100 90 80 70 60 50 4
° 5 . -^ . . . ion .....
2.0
?
J 1.0
0.5
0 0
T
0
1 I IT
A IT
I *
m - i mm -
IUU
80
\
o
540
^
0
20
n
/X
/ V
T a / JN..
1 / la >«.
T {I / ^X«^^
L^^ 8 0~" a
110 100 90 80 70 60 50 40 110 100 90 80 70 60 50 4
o 5 ...... n . . . .
{..j
x^
o>
E
1.5
o
+ 1.0
CM
io.5
0 0
»-
S
T
L o
\f-
10
Q
\
J. 6
O
"
0
T -«- T
\ t T ? 1 f
M Jj 5 XX ? X
X^LX" X^^X 2
Q 4f
2J
n !
110 100 90 80 70 60 50 40 110 100 90 80 70 60 50 4(
River Miles from Mouth River Miles from Mouth
Freshwater Flow near Richmond: 1100 cfs
Water Temperature: 26°C Legend:
Observed Data
Average and Range
Model Results
Figure 7. Model Sensitivity Results - Nutrient Uptake by Weeds
191
-------
c
"o
1
"-J
0>
o
-«-»
2
o
"8
1
Phosphorus L
1 2 -
1.1 -
1 -
O.9 -
0.8 -
0.7 -
O.6 -
O.5 -
O.4 -
O.3 -
0.2 -
0.1 -
0 -
1
12-
1.1 -
1 _
O.9 -
0.8 -
0.7 -
0.6 -
0.5 -
0.4 -
0.3 -
0.2 -
0.1 -
3 _
1
No Limitation
\
50% Limitation^
Maximum Limitation^
I I i 1
00 80 60
River Miles from Mouth
No Limitation
J "^^^^^^^e-a^-^*^
50? Limitation.
Maximum Limitation-*
i 1 r" 1
00 80 60
River Miles from Mouth
Figure 8. Degree of Nutrient Limitation
192
-------
o
T3
X,
o
O
o
3.5 -
3 -
2.5 -
2 -
1.5 -
1 -
0.5 -
10O
Temperature Adjusted Growth Rate»
Saturated Growth Rate,
Temperature & Light Adjusted Growth Rate
80
River Miles from Mouth
6O
Figure 9. Specific Growth Rate of Phytoplankton
REFERENCES
(1) Hydroscience, Inc., 1980. Water Quality Analysis of the Upper James
River Estuary. Report prepared for the Commonwealth of Virginia State
Water Control Board, 86 p.
(2) Grizzard, T. J. and B. J. Weand, 1984. Water Quality Review and
Analysis: Richmond-Crater James River Water Quality Monitoring Program.
Final Report (1983-1984 Monitoring).
(3) Lung, W. S., 1985. Assessing the Water Quality Benefit of Point Source
Phosphorus Control in the James River Basin. University of Virginia,
Department of Civil Engineering, Technical Report No.
UVA/532533/CE85/101, 50 p.
(4) Neilson, B. J. and P. S. Ferry, 197&. A Water Quality Study of the
Estuarine James River Virginia Institute of Marine Science. Special
Report No. 131, 72 p.
193
-------
Table 3. James River Model Parameters
Kinetics Coefficients (Base e @ 20°C)
Oxygen Transfer
Deoxygenat ion
Nitrification
Hydrolysis - N
- P
Setting - N
- P
- Chi 'a1
Growth
Respiration
Death
Extinc. Coef.
ft/day
I/day
I/day
I/day
I/day
ft/day
ft/day
ft/day
I/day
I/day
I/day
I/meter
Hrs. of Daylight hrs
Benthic Demand gm/m2-day
Stoichiometry & Constants
Temperature
C/CHL Ratio
N/CHL Ratio
P/CHL Ratio
0 /C Ratio
Half. Sat.
Cone. - N
- p
Sat. Light
Avail. Light
mg/yg
mg/yg
mg/yg
mg/mg
mg/1
rag/1
langleys/day
langleys/day
(Segments 1-30)
September Survey
(Segments 1-30)
July Survey
(Segments 31-50)
(Segments 1-30)
(Segments 31-50)
(Segments 1-30)
(Segments 31-50)
3.00
0.10
0.05
0.00
0.15
0.10
0.15
0.05
0.10
0.75
0.75
0.75
2.20
0.10
0.10
1.4 (Segments 1-10)
2.0 (Segments 11-31)
2.3 (Segment 32)
3.0 (Segments 5-6)
12.0 (September)
14.0 (July)
0.5 (Segments 1-30)
1.5 (Segments 31-50)
26.0 (September)
28.0 (July)
0.025
0.007
0.001
2.67
0.005
0.001
300.
600. (September)
450. (July)
194
-------
APPENDIX - MICROCOMPUTER USED FOR THIS STUDY
HARDWARE
The system hardware consists of a COMPAQ (IBM compatible) computer with
the following configuration:
256K RAM
16-bit Intel 8088
Math Coprocessor 8087
« Dual 320K disk drives
Serial port and parallel port
Graphics
The computer comes with a 9" monochrome monitor. In addition, a GEMINI 10X
dot matrix printer is connected with the COMPAQ via the parallel port. A HP
7470A plotter is connected with the computer through the serial port.
SOFTWARE
The following software programs were used in this study:
JMSRV water quality model
FORTRAN 77 compiler which supports 8087 math coprocessor
LOTUS
Specially developed plotting programs used with the HP 7470A plotter
to generate data and model result plots
PERFORMANCE
The source code of the JMSRV model (size about 70K) was compiled using
the Microsoft FORTRAN 77 compiler which has a nice feature of supporting the
8087 math co-processor. The math co-processor, designed to handle float point
executions, is very important in achieving reasonable program execution speed.
In this case, the execution of the compiled program takes about 4 hours to
complete a run (i.e., 120-day time variable run at a 0.5-day time step) on the
COMPAQ prior to the installation of the 8087 math co-processor. Once the 8087
is in place, it takes 11 minutes to complete the same model run. The reduction
of run time in this case makes the use of microcomputers for this study and
many others a very attractive alternative to using the mainframe.
Of course, a typical run like that one usually takes no time on a main-
frame computer. However, it is the other software features which offer a user
friendly environment, making the use of a micro even more appealing. For
example, the LOTUS program can be used to view any model results in the form
of plots and can generate some simple plots using the HP plotter. The final
plots shown in this paper were generated by a specially written program for
the HP plotter. Thus, a number of key water quality constituents can be
presented on a'single plot to tell the whole story.
195
-------
APPLICATION OF FACTOR ANALYSIS TO MANAGEMENT
OF IMPOUNDMENT WATER QUALITY
by: Curtis D. Pollman
Robert E. Dickinson
Environmental Science and Engineering, Inc
Gainesville, Florida 32602
ABSTRACT
An inherent problem in monitoring studies of water quality is reducing
the dimensionality of the data so that a determination of processes or factors
controlling water quality can be made. One approach is principal components
analysis (PCA) which constructs from the data a few important factors from all
the measured variables by taking advantage of intra-variable correlations.
PCA was used to derive water quality factors from over 40 chemical parameters
measured in a series of reclaimed phosphate pit lakes located in central
Florida. The relationship between the extracted factors and reclaimed lake
morphometry and hydrology were evaluated through multiple linear regression.
Extracted factors generally fall within a well-defined typology, i.e.
lake trophic state factor, inorganic sediment factor, inorganic nitrogen
factor, etc. The strongest correlation with the modeled morphometric and
hydrologic variables was observed for the factor reflecting the major
dissolved constituents in the water column, followed by the trophic state
factor. Multiple regression analysis suggests that water quality in reclaimed
lakes is largely controlled by in-lake processes. For example, the lake
trophic factor was negatively correlated with maximum lake depth, indicating
that sedimentation of detritus as a nutrient sink is important in regulating
trophic state. In fact, overall data variability appeared most strongly
influenced by the two depth variables, maximum and average depth, followed by
lake volume and volume development. Lake hydrology was somewhat less
important as was lake orientation to prevailing winds.
INTRODUCTION
In October 1981, the Florida Institute of Phosphate Research (FIPR)
sponsored an extensive research program to study 12 artificial lakes created
196
-------
during the course of reclamation of mined phosphate pits in central Florida.
Data on the physical, chemical, and biological conditions of each lake were
collected during a 1-year sampling program.
One aspect of the data analysis program was to examine the influence of
lake morphometric and hydrologic features on lake water quality. An analytic
procedure, principal components analysis (PGA), was used to create "factors"
that would reduce the dimensionality of the data. These factors were then
correlated with morphometric and hydrologic parameters to find influential
parameters.
The purpose of the PCA ultimately was the creation of lake design
criteria. Given controllable features such as mean depth, maximum depth, and
shoreline configuration, how should reclaimed lakes be designed to maximize
lake water quality? This analysis in turn can be applied to any manmade lake
such as stormwater detention ponds and water supply reservoirs.
PRINCIPAL COMPONENTS ANALYSIS
The FIPR reclaimed lake study had 80 measured water quality and
biological variables. The large number of variables would render the
determination of how lake morphometric and hydrologic characteristics affect
lake water quality and biology a tedious task without the aid of a simplifying
analytic procedure. A PCA constructs a few important "factors" from the many
variables by using the intra-correlations of the variables to reduce the
dimensionality of the problem. The resulting factors, each a linear
combination of all the variables, are uncorrelated with one another, and a few
factors often account for most of the variation of the entire set of
variables.
The meaning of a particular factor is determined by considering those
variables that "load" highly on the factor. The variable loadings may be
interpreted as correlation coefficients. They range from -1 to 1, with -1
indicating perfect negative correlation, 1 a perfect positive correlation, and
0 no correlation.
The variables were first transformed by using the In transform and
controlling both seasonal (sampling trip) and spatial (station) variability
before using the principal component analysis. Seasonal and spatial
variability were controlled by modeling each variable as a function of season
and station and calculating the least-squares residual. The least-squares
residuals are the differences between the actual observation and the mean for
the corresponding season-station combination.
The variables were divided into chemical and biological subsets, and
separate PCA's were performed for the set of chemical or abiotic variables and
the set of biological or biotic variables. This separation of analysis groups
aided in the interpretation of the factors. Only the results of the abiotic
PCA will be discussed in this paper.
The PCA was then performed on the transformed residuals by employing an
iterative process. The extracted factors were evaluated at each iteration,
197
-------
and all variables which did not contribute strongly to any factor were
assumed to be relatively minor components and removed from further
iterations. This process was continued until only contributory variables
were retained.
The factors from the PCA were subsequently subjected to stepwise
multiple linear regression with a set of morphometric/hydrologic parameters.
The selected parameters were features that can be reasonably controlled
during the course of reclamation design and encompassed all aspects of lake
morphometry and hydrology. The parameter list consisted of:
o AREA = total lake surface area (ha)
o FLTOT = hydraulic loading rate (m/yr)
o TIME = hydraulic residence time (yr)
o SDI » shoreline development index
o VDI = volume development index
o VOL = total lake volume (nr*)
o WSIA = annual wind stress index (orientation parameter)
o Z = mean lake depth (m)
o ZMAX = maximum lake depth (m)
The stepwise multiple regression approach was used as a model-building
process to determine which combinations of independent parameters we^e highly
correlated with water quality factors. The procedure used each of the water
quality factors (determined by PCA) as a dependent variable and then added
independent parameters to the model in their order of importance as
explanatory parameters. Parameters were added as long as their contribution
to the model was statistically significant at the alpha = 0.15 level. The
stepwise model building procedure acted as a filter to screen the most
important independent parameters. It was not intended to provide a predictive
model but to indicate strong associations.
The results of PCA on total analytical concentrations of the various
chemical parameters are presented in Table 1, which is restricted to those
loadings exceeding or equal to 0.50. Individual eigenvalues and the percent
of the total variance explained by each factor are also included in the
table. Over 50 percent of the total variance in the data set was explained
by the first four factors, with Factors 1 and 2 accounting for 33.3 percent.
Factor 1 is related to the inorganic constituency of sediments in the
reclaimed lakes, with sediment-associated chromium (SCr) and total phosphorus
(Total P) loading most highly on the factor (r - 0.93 and 0.87). Sediment
moisture content, which is indicative of the energy state or depositional'
environment prevailing at the sediment-water interface (1,2, and 3), also
loaded relatively highly on Factor 1, although the correlation was weaker (r »
0.56).
The second factor reflects primarily major dissolved components in the
water column. Total dissolved solids (TDS) and conductivity are indicative
of the ionic strength of the water column and are strongly correlated with
Factor 2 (r » 0.88 and 0.74, respectively). Calcium (Ca) and magnesium (Mg)
dominate the factor with alkalinity and, to a lesser extent, fluoride (F)
associated with Factor 2 as the counter ions. Sulfate (SO^), which is the
198
-------
TABLE 1. EIGENVALUES, IERCENT VARIANCE EXPLAINED, AND LOADINGS FOR PCA FOR RECLAIMED .LAKE
CHEMISTRY PARAMETER
Variable
Moisture
SCa*
SMg
SK
SBa
STP
SCr
SH>
SCd
SeRa226
SSr90
pH
Conductivity
TD3
Ca
Mg
Alkalinity
F
TOC
Turbidity
Secchi
Total P
Total N
Factors
123456789 10
0.59 0.56
0.61
0.56
0.82
0.84
0.87
0.93
0.82
0.77
0.82
0.75
0.59
0.74
0.88
0.94
0.97
0.76
0.64
0.58
0.67
-0.84
0.65
0.61
-0.63
STCC
SIGN
SCI
Cd
S%
SCP
K
Ha 226
SS04
0.86
0.77
0.57
0.62
0.78
-0.73
0.78
0.68
-0.71
0.76
0.69
-0.54
199
-------
TABIE 1. EKZNVALUES, PERCENT VARIANCE EXPLAINED, AND HIDINGS FCR PCA FOR RECLAIMED IAKE
^_ CHEMISTRY PARAMETER. (CONTINUED. PAGE 2 OF 2)
Factors
Variable 123456789 10
SAs 0.88
SPb 0-76
NH3 0.72
NC~ 0.85
SSe O-88
D02
^ 0.76
Eigenvalue 7.44 6.55 4.19 3.50 3.24 2.88 2.50 1.82 1.50 1.33
Percent
Variance
Cumulative
Variance
"Variables
17
17
preceded by an
.7 15.6 10.0 8.3 7.7 6.
.7 33.3 43.3 51.6 59.3 66.
S are for sediment data (i.e., SCa =
9 6.0 4.3 3.6
2 72.1 76.5 80.0
1 sediment calcium) .
3.2
83.2
predominant anion on an average basis for the 12 reclaimed lakes, loaded
rather weakly on Factor 2 despite its close correlation with Ca (r » 0.54).
The third factor extracted from the principal component analysis is
indicative of trophic state and'is similar to trophic state factors derived
by Shannon (4) and Preston (5) for other Florida lakes. The nutrient forms,
total nitrogen (Total N) and Total P, as well as parameters indicative of
light transparency (i.e., turbidity and secchi disk transparency) are
associated with this factor.
The strongest loading on Factor 3 is Secchi disk transparency, which
loads negatively on the facCor and implies reduced-light transparencies in
enriched or eutrophied reclaimed lakes. More difficult to explain is the
association of SO^ with this factor. A priori considerations suggest
that variability in 804 levels would be" most closely associated with a
factor related to ionic distribution and weathering (i.e., Factor 2). The
correlation of 804 with Factor 3 may reflect 804 reduction and
depletion in hypolimnetic waters, a process (6) recently demonstrated to be
rather significant in the hypolimnion of a productive, softwater lake.
The fourth factor, which accounts for 8.3 percent of the total variance
in the data, is essentially an organic matter deposition factor and relates
200
-------
those variables that tend to be most influenced by the accumulation of
sedimentary organic matter. The highest loading is for sedimentary total
organic carbon (STOC), followed by sedimentary total organic nitrogen (STON).
Also associated with Factor 4 is interstitial ammonia (SNH3) which builds
up in the pore water as a direct consequence of catabolic processes within the
sediments.
It is interesting to observe that sediment moisture content loads in a
positive sense for Factor 4 as well as Factor 1, which relates inorganic
sediment variables. Previous studies have demonstrated a close relationship
between sedimentary water content and organic content (7, 8, and 3); however,
sediment moisture content also reflects selective size sorting processes and
increases with decreasing particle size due to interparticle repulsion (9).
This implies that the sedimentary accumulation of trace elements identified
in Factor 1 is the result of adsorption to clay particles, which generally
carry a net negative charge and have larger surface area:mass ratios than
coarser particles.
Of the remaining factors that exceed the minimum .eigenvalue threshold
level of 1.0, Factors 8 and 10 are of the most interest. Factor 8 is
essentially an inorganic nitrogen factor and comprises ammonia (NH3) plus
nitrate- and nitrite-nitrogen (NOX) forms. Factor 10 is mainly the result
of the positive correlation of dissolved oxygen and the negative correlation
of dissolved radium-226 (Ra-226). Factors 7 and 9 result from the correlation
of sedimentary arsenic (SAs) and lead (SPb) (Factor 7) and sedimentary
selenium (SSe) (Factor 9). The remaining factors (Factors 5 and 6) are
complex and difficult to interpret with respect to underlying causes.
ABIOTIC FACTORS AND LAKE MORPHOMETRY
The relationship between morphometric features and reclaimed lake
chemistry was investigated using stepwise multiple linear regression; the
abiotic factors were the dependent variables, and the lake morphometric and
hydrologic parameters were the independent variables (see Table 2 for
results).
From a trophic state perspective, the most interesting model is
represented by the relationship between Factor 3 and the lake morphometry/
hydrology. The parameters FLTOT, VDI, Z, and ZMAX were all found to be
significantly correlated (P <0.0001) with the trophic state factor (Factor 3).
Inclusion of other variables in the regression equation did not sufficiently
improve the coefficient of determination (r^ « 0.50) to justify increasing
the complexity of the model. Inspection of calculated F values for each
variable indicates that lake maximum depth and volume development index exert
the greatest influence on the factor. Of these two, the most important
variable in the relationship is ZMAX, which was negatively related to the
factor. The relationship between the trophic state factor and ZMAX agrees
with established limnological principles, viz., all other factors being equal,
overall lake nutrient levels should decrease with increasing depth because of
the reduced rate of nutrient recycling across the sediment water interface.
Furthermore, sediment focusing may occur in lakes with deep holes, effectively
removing detrital material from interacting with the trophogenic zone of the
water column.
201
-------
The positive correlation of mean depth with the trophic state factor
conflicts with the influence of maximum depth and further illustrates the
relative importance of deep holes on reclaimed lake chemical dynamics. Within
a particular class of lakes, internal loading of nutrients in response to
disturbances at the sediment-water interface created by wind-driven
circulation and other processes tends to diminish in importance as mean depth
increases. This, of course, reflects the greater amount of energy that must
be applied at the lake .surface for wind-induced wave energy to extend to the
bottom in deeper lakes. The net effect of the sediments as an ultimate sink
for nutients, therefore, increases with mean depth (12, 13).
The fact that reclaimed lakes with greater mean depths tend to have
increasing values for the trophic state factor is indicative of the extremely
shallow nature of these lakes and suggests that other factors beyond lake
hydrodynamics and simple increases in assimilative capacity are associated
with changes in mean depth. Mean depth for the 12 lakes ranged from 1.8 to
only 5.4 meters (m). Chapra (14) indicates that the effects of sediment
resuspension is probably significant in lakes with a mean depth of 10 m or
less. It seems likely, therefore, that sediments throughout the basins for
each reclaimed lake are periodically disturbed and resuspended, with
deposition and removal occurring only in deep holes. "The resultant effect of
deep holes would be increased water clarity because of lower turbidity levels
and decreased rates of algal productivity.
Within the depth range of the reclaimed lakes surveyed, it is apparent
that increasing "z has virtually no distinguishable effect on internal
loading processes. The positive effect of mean depth on the trophic state
factor instead suggests that penetration of the lake basin into the
phosphatic bedrock underlying the surficial unconsolidated layer of sands
contributes to the trophic state of reclaimed lakes. Reclaimed lakes with
.shallower mean depths included in this study generally were constructed by
filling the initial mine pit with a relatively larger volume of overburden;
consequently, these lakes are less influenced by residual tailings and clay.
Hydraulic loading rate (FLTOT) also affects the trophic state factor in a
negative Sense, although the relatively low F value indicates that the role of
this parameter is not as important as the physical structure of reclaimed
lakes. The negative relationship between FLTOT and the trophic state factor
may simply reflect the effects of dilution on nutrient inputs. In other
words, considering two lakes with similar external mass loads of Total N and
Total P, the lake with the larger hydraulic loading rate correspondingly will
have reduced input nutrient concentrations and will be characterized by a
lower trophic state as well. This relationship also may imply that internal
recycling or loading is important in maintaining high rates of productivity in
reclaimed lakes and that the relative importance of internal processes in
controlling trophic state diminishes as the hydraulic loading rate increases.
Lake trophic state is also related to Factor 8, which reflects inorganic
nitrogen concentrations. Because of the high levels of phosphate typical of
reclaimed phosphate pit lakes, nitrogen availability may be hypothesized to
limit primary production. A substantial'portion of the dispersion in the
inorganic nitrogen factor can be accounted for by the morphometric/hydrologic
202
-------
variables FLTOT, TIME, and Z. .Multiple linear regression of the factor
against these variables yields r^ equal to 0.42 (p<0.0001) (Table 2).
The sign of the coefficients in the multiple regression model indicates
that in-lake nitrogen concentrations increase with the hydraulic loading rate
and residence time (Table 2), The relationship of the hydraulic loading rate,
FLTOT, in the model strongly suggests that inorganic nitrogen levels in
reclaimed lakes result primarily from allochthonous sources. In addition, the
positive relationship of residence time in the model for the inorganic
nitrogen factor suggests that reclaimed lake water quality is in a
transitional state. This should not be construed that, ceteris paribus, lakes
with long residence times will necessarily have higher inorganic nitrogen
concentrations than rapidly flushed lakes. Immediately after the formation
and inundation of an artificial basin, a pulsed release of nitrogen usually
accompanies the decomposition of submerged vegetative matter (16). At this
point, in-lake nitrogen concentrations exceed levels supported by external
inputs and the lake will approach a lower, steady-state condition at a rate
inversely related to the hydraulic residence time. Once reclaimed lakes
mature and approach steady-state conditions, lakes with longer residence times
will characteristically have lower nutrient levels because of the
nonconservative behavior of nitrogen and phosphorus. Biological uptake of
nitrogen and phosphorus results in a net loss of these substances from the
water column to the sediments. Apparent settling velocities are on the order
of 5.5 and 8.5 meters per year (m/y) for nitrogen and phosphorus, respectively
(13); thus, net removal of nutrients increases with lake detention time.
Multiple regression analysis indicates that ionic content or salinity in
reclaimed lakes, as represented by Factor 2 (Table 2), is largely controlled
by in-lake processes. F statistics for this model, which account for 77
percent of the dispersion in the salinity factor, indicate that significance
in the model is primarily attributable to mean depth and the wind stress
index. The positive correlation of mean depth with the salinity factor
suggests that penetration of the lake basin into the calcareous bedrock
underlying the surficial unconsolidated layer of sands controls the ionic
character of the water column. This is supported by inspection of the
components that comprise the salinity factor. This factor is dominated by
the positive associations of Mg (r = 0.97) and Ca (r = 0.94), which are
derived primarily from the bedrock. Furthermore, cation balances show that
the reclaimed lakes contain primarily Ca and, to a lesser extent, Mg. Sodium
(Na), which is derived principally from atmospheric inputs, and potassium (K)
are considerably less prevalent. It is important to note that unlike
inorganic nitrogen, the ionic constituency of reclaimed lakes is not
controlled to an appreciable extent by hydraulic residence time.
Somewhat more difficult to interpret is the influence of maximum depth
on variability in the salinity factor. In the desired statistical model
(Table 2), the effect of ZMAX is negativeopposite in sign to mean depth.
This dichotomy in effect by these two parameters was also observed for the
trophic state factor and is apparently indicative of the importance of deep
holes as traps or sinks for detrital or easily weathered material. Thus,
assuming that weathering within the lake basin dictates the ionic composition
of reclaimed lakes, it follows that the wind stress index is positively
203
-------
TABLE 2. MULTIPLE LINEAR REGRESSION RELATIONSHIPS BETWEEN ABIOTK PGA FACTORS AND LAKE
Factor
Factor Characteristic ND. Significant Morphologic Features
Sediment Inorganic
Constituents
Water Cblunn Ionic
Content
Trophic State
Indicators
Sediment Organic tetter
Deposition
Water Cblunn Inorganic
Nitrogen
Dissolved Ckygen/
Radiun-226
10
1.26 - 0.151 (ZMAX)
8.65 - 9.41 (VDI) + 1.82 (WSIA) -i- 3.43 (Z)
- 1.25 (ZMAX)
0.14
0.77
9.86 - 0.0930 (FLTOT) - 7.04 (VDI) + 2.08 (Z) 0.50
- 0.924 (ZMAX)
-1.34 + 0.814 (VDI) + 0.00274 (VOL)
- -1.33 + 0.291 (FLTOT) + 1.11 (TIME)
- 0.359 (Z)
-0.272 + 0.0672 (FLTOT)
0.42
0.37
0.04
Where:
ZMAX = Maximun lake depth,
VDI = volune development index (bottom configuration),
WSIA = annual wind stress index (lake orientation to prevailing wLnds),
Z - mean depth,
FLTOT = annual loading rate.
VOL = lake volune, and
TIME » hydraulic residence time,
correlated with the salinity factor. WSIA is a shape factor that weights the
configuration and orientation of a lake with the seasonal distribution of
wind vectors and, independent of lake area, estimates the relative quantity
of energy transferred to the water column because of wind-induced mixing.
The effect of VDI on the salinity factor is difficult to interpret in a
manner consistent with other parameters in the model. VDI is a crude
indication of the overall bottom configuration and represents the ratio of
lake volume to that of a cone with basal surface area (corresponding to lake
surface area) and height ZMAX. Consequently, VDI reduces to:
VDI
(3)
204
-------
The relationship of VDI in the stepwise regression model implies that
for reclaimed lakes, a conical basin configuration (i.e., low VDI) results in
higher ionic content than more developed, teacup-shaped basins. This aspect
of the model may reflect greater amounts of calcareous bedrock exposure in
poorly developed lakes.
SUMMARY
From a water quality perspective, two factors were separated for further
evaluation. One factor represents lake trophic state indicators (Total N,
Total P) and another represented water column inorganic nitrogen
concentration. Maximum lake depth was negatively correlated with trophic
state indicators and was .also the most important morphometric relationship.
Interpretation of this relationship would indicate that as maximum depth
increased, concentrations of indicators declined. Sedimentation of detrital
material may be the process involved in removing nutrients from the
trophogenic or biologically active zone. In contrast, mean depth was
positively correlated with trophic state indicators. This relationship may be
confusing at first; however, it does present a situation where deep holes can
be a vital component of lake design while keeping the overall mean depth in
line with natural or unmined lakes. Sediments throughout the basins for each
reclaimed lake are periodically disturbed and resuspended, with net deposition
and removal occurring only in deep holes. Irregular bottom contours,
consisting of intermittent deep holes, require greater wind energy to induce
mixing than smooth bottom lakes. Since reclaimed phosphate pit lakes are
either mesotrophic or eutrophic, the reduction of surplus nutrients by
sedimentation to deep water zones with minimal recirculation potential can be
a major factor in lake design.
The inorganic nitrogen factor provided indications that reclaimed lake
water quality is in a transitional state, dependent upon inorganic nitrogen
from allochthonous sources. This factor was negatively correlated with lake
mean depth. Since phosphorus is in sufficient supply within these lakes,
nitrogen supply is critical to maintain high productivity levels. Increasing
mean depth would remove nitrogen from the trophogenic zone. This factor was
also strongly correlated with lake residence time (+) and hydraulic loading
rate (+).
The strongest statistical relationship was with the water column ionic
content factor. These results indicated that penetration of the lake basin
into the calcareous bedrock underlying the surficial unconsolidated layer of
sands controls the ionic character of the water column. Both mean depth and
orientation to prevailing wind stress were primary lake design features
relative to water column ionic strength. Maximum depth was also significant,
indicating the importance of deep holes as traps or sinks for detrital or
easily weathered material. Since ionic content is governed by internal
processes, lake orientation to wind can aid in redistribution and
recirculation of ionic constituents.
The overall effect of a particular morphometric or hydrologic design
variable can be evaluated relative to other variables by considering its
significance in each factor relationship (as determined by its F value), the
205
-------
amount of overall variability accounted for by each factor, as well as the
strength or explanatory power of each model given by the coefficient of
determination.
This approach indicates overall data variability appears most strongly
influenced by the two depth variables, ZMAX and Z, followed by lake volume
and volume development. Lake hydrology embodied as the hydraulic loading
rate and residence time was somewhat less important, as was lake orientation
(WSIA). The remaining morphometric/hydrologic variables were not
particularly effective in accounting for overall variability in reclaimed
lakes.
A final word of caution is perhaps in order. Implications of cause and
effect have been intended to account for the correlations between the various
morphometric/hydrologic variables and the chemical attributes of the
reclaimed lakes. For example, supporting mechanisms have been hypothesized
to account for the statistical relationship between ZMAX and the lake trophic
state and salinity factors. However, correlations between variables do not
confirm cause and effect, and the possibility that ZMAX may be correlated
with some other (unmeasured) parameter that is essentially the true cause of
the observed effect cannot be discounted. Further res'earch is necessary to
confirm the cause-and-effeet implication.
ACKNOWLEDGEMENTS
The work described in this paper was funded by the Florida Institute of
Phsophate Research (18). We wish to thank the project manager at ESE, Oliver
C. Boody, IV, for his excellent support in the course of this project. The
data analyses were performed using the Statistical Analysis System (19, 20).
REFERENCES
1. Hakanson, L. Bottom dynamics in lakes. Hydrobiologia 91:9-22, 1982.
2. Hakanson, L. Lake bottom dynamics and morphometry: the dynamic ratio.
Water Resources Research 18:1444-1450» 1982.
3. Pollman, C.D. Internal loading in shallow lakes. Ph.D. Dissertation.
University of Florida, Gainesville, 1983.
4, Shannon, E.E. Eutrophication-trophic state relationships in north and
central Florida lakes. Ph.D. Dissertation. University of Florida,
Gainesville, 1970.
5. Preston, S.D. Numerical analyses of a eutrophic lake chain.
M.S. Thesis. University of Florida, Gainesville, 1983.
206
-------
6. Kelley, C.A., Rudd, J.M.W., Cook, R.B., and Schindler, D.W. The
potential importance of bacterial processes in regulating rate of lake
acidification. Limnol. Oceanogr. 27:868-882.
7. Nissan, J.D. Sediment-water nutrient dynamics. M.S. Thesis. University
of Florida, Gainesville, 1975.
8. Pollman, C.D. Sediment characterization and nutrient dynamics of
selected real estate canals. M.S. Thesis. University of Florida,
Gainesville, 1977.
9. Berner, R.A. Principles of chemical sedimentatology. McGraw Hill, New
York, 1971.
10. U.S. Army Coastal Engineering Research Center. Shore Protection Manual
Vol. I. 1977.
11. Mortimer, C.H. Lake hydrodynamics. Mitt. Int. Ver. Limnol. 20:124-197,
1974.
12. Vollenweider, R.A. Input-output models with special reference to the
phosphorus loading concept in limnology. Schweiz. Z. Hydrol., 37:53-84,
1975.
13. Baker, L.A., Brezonik, P.L., and Kratzer, C.R. Nutrient loading-trophic
state refationships in Florida lakes. Publ. No. 56, Florida Water
Resources Research Center, University of Florida, Gainesville, Florida,
1981. 126 pp.
14. Chapra, S.C. A budget model accounting for the positional availability
of phosphorus in lakes. Water Res. 16:205-209, 1982.
15. Wetzel, R.G. Limnology. W.B. Saunders Co., Philadelphia, Pennsylvania,
1975.
16. Ostrofsky, M.L. Trophic changes in reservoirs: an hypothesis using
phosphorus budget models, jnt. Revenue Ges. Hydrobiol. 63:481-499,
1978,
17. Sonzogni, W.C., Larsen, D.P., Maleug, K.W.,, and Schuldt, M.D. Use of
large submerged chambers to measure sediment-water interactions. Water
Res. 11:461-464, 1977.
18. Boody, O.C., Pollman, C.D., Tourtellotte, G.H., Dickinson, R.E., and
Arcuri, A.N. Ecological considerations of reclaimed lakes in central
Florida's phosphate region. Florida Institute of Phosphate Research,
1984.
19. SAS Institute. SAS Users Guide: Basics. SAS North Carolina, 1982.
20. SAS Institute. SAS Users Guide: Statistics. Gary, North Carolina,
1982.
207
-------
THE APPLICATION OF QUAL-II TO AID RESOURCE
ALLOCATION ON THE RIVER BLACKWATER, ENGLAND.
presented at the
Stormwater and Water Quality Model Users Group Meeting,
University of Florida, Gainesville, January 31 - February 1,
1985
by
Bob Crabtree, Ian Cluckie, Paul Crockett and Chris Forster;
Department of Civil Engineering, University of Birmingham,
Birmingham B15 2TT U.K.
ABSTRACT
The River Loddon is a tributary of the River Thames. It has a
deterioration in water quality downstream of the confluence with its major
tributary, the River Blackwater. The Blackwater catchment is densely
populated and the river is of poor quality due to a high proportion of
treated effluent. Qual-II has been applied to this system as a forecasting
tool to investigate any improvements in the quality of the River Loddon
that may be achieved by altering the allocation of effluent treatment
resources within the Blackwater Catchment. In particular, the effects of
a proposal to extend an existing effluent treatment works were evaluated.
INTRODUCTION
The daily domestic consumption of water in England and Wales is some
16 x 106m3. The majority of this is eventually discharged to inland rivers
and streams, the total length of which is some 38 000 km. In addition,
about 20 x 106m3 is abstracted directly from these rivers by industry,
used and then returned. Some of these waters will have become contaminated
during their use and it is therefore necessary for them to undergo treatment
before they can be discharged. Failure to do this can result in serious
changes in the chemical composition of the river water, causing deteriora-
tion in the river ecology possibly coupled with a significant degree of
oxygen depletion.
Rivers within the U.K. have to accept a significant load of oxygen
demanding pollutants. However, as long as the load on any particular reach
208
-------
is not excessive, the river can assimilate it. Discharges of this type
which have occurred on a regular, continuous basis for sometime are likely
to have caused a change in the downstream ecology, relative to the time
before the discharge. As long as the load is not increased, this "new"
downstream ecology will be stable and, in a lot of cases, will be acceptable.
However there is little point in knowing what pollutants are present, either
in an effluent discharge or in the river itself, or knowing what effect
those pollutants may have on the stream biota, if their addition to natural
waters cannot be controlled. Controlled that is, in such a way as to reduce
their impact to a level acceptable for all the uses required for the river,
which may include transport, amenity, and potable and industrial water
supply, as well as waste disposal. This is particularly true if the river
is to be used to its maximum economic potential by paying full regard to
to its ability for self'purification. Models, properly applied, can help
to optimise the use of rivers.
River quality models are commonly used to forecast dissolved oxygen
levels (DO) at a point in a river system resulting from the interaction of
the physical and biological assimilative processes of oxygen demand. These
models can then be used to predict the effects of proposed water quality
control strategies on water quality and use. Modelling can therefore be
used as an aid to quality management decision making and resource alloca-
tion to achieve designated levels of quality, termed river quality objec-
tives (RQO) in the U.K.
There are two general types of model that have been applied to river
systems; stochastic and deterministic. Seven out of the ten U.K. Regional
Water Authorities (RWA) have at some time used a form of deterministic
model for predicting river water quality (Russell and Pescod, 1981) or
assessing the impact of effluent discharges on downstream river quality.
The models used are typically river specific derivatives of the Water
Resource Centre river quality model, developed by Knowles and Wakeford
(1978). This evaluates downstream quality changes of conservative and
non-conservative substances in a reach by reach mass-balance basis and
incorporates the effects of mixing, dispersion and biological processes.
It is no longer widely used and it is generally only applied to specific
problem rivers. Compared with North America, the.use of deterministic
river quality models for water quality management in the U.K. is in its
infancy. The RWA control all aspects of flood control, water supply,
drainage, sewerage, effluent treatment and river pollution management.
They are large, catchment-based public bodies which are charged with con-
ducting their activities to achieve a satisfactory balance between the
quality of the services provided and the resulting costs borne by the
consumer. The main dischargers of sewage effluent to the rivers are the RWA
who operate the sewage works. These treat domestic sewage and much of the
industrial trade effluent which is allowed to be discharged to the public
sewers. Therefore the main method of controlling river quality is by
controlling the discharges of effluent to the river. This can lead to a
situation, depending on local circumstances, where the RWA is both a
"poacher" and a "gamekeeper", causing and managing pollution.
209
-------
Historically, effluent discharge standards (consents) have been a
legally binding value and have tended to be based on the 1912 Royal
Commission standard of 30 mgl * suspended solids (SS) and 20 mgl"1 BOD,
without due consideration of the effect on the receiving water course.
This ignores the fact that the 30:20 standard was based on the quality of
effluent produced by a well operated biological filtration plant and
assumed that the discharge was to a water course with a dilution factor of
at least 8:1. The use of a blanket 30:20 discharge standard may result in
either:
1) a consent which is unnecessarily stringent in some circumstances
or not stringent enough in others; or
2) a consent which is inconsistent with existing treatment facilities
or with the cost of improvement.
To avoid either river pollution or undue capital cost, the National
Water Council (1977) recommended that effluent discharge consents should
take account of local circumstances and should be based on the desired
downstream river quality, the RQO. While discharge consent conditions
would still be fixed figures which could not legally be exceeded, it was
not expected that RWA or private individuals would launch prosecutions
where occasional samples exceeded the consent conditions. This was with
the proviso that samples exceeding the consent were not greater than the
variability to be expected from a well managed system.
In 1977 the assessment of river and effluent quality in the U.K. was
placed in a probabilistic framework. A new classification of river water
quality, shown in simplified form in Table 1, was adopted. This was
based on quality criteria (RQO) in terms of 95 percentile class limits.
The 95 percentile is a sample based statistic and is the value of the
river quality that could be expected to be achieved in 95 percent of
samples taken. The use of probabilistic class limiting criteria for river
quality and effluent discharges reflects the underlying stochastic nature
of river quality and flow processes and permits the statistical treatment
of quality data within such a probabilistic framework (Cluckie and Forster,
1982).
In 1978, the National Water Council (NWC, 1978) made recommendations
to ease the implementation of the 1974 Control of Pollution Act, part II
whereby there would be greater public accountability by the RWA for the
management of pollution control. The recommendations allowed the resetting
of consents to suit local environment, financial and technical conditions
prior to the implementation of the new act.
The new consents are set in terms of interim consents (to maintain
present levels of river quality) and long term consents (to achieve the
designated RQO) . The long term consents can be set by optimising river
pollution control by whole catchment baaed river quality models in which,
for a river system subdivided by RQO, full use of river self-purification
could be made between effluent discharges. Therefore the correct consents
210
-------
TABLE 1.
River
Class
1A
SIMPLIFIED NATIONAL WATER COUNCIL (1977) CLASSIFICATION OF RIVER
QUALITY
IB
Class limiting quality
criteria (95 percentile)
DO greater than 80% saturation
BOD not greater than 3 mgl l
Ammonia not greater than 0.4
mg 1
DO greater than 60% saturation
BOD not greater than 5 mgl
Ammonia not greater than 0.9
mgjf '
DO greater than 40% saturation
BOD not greater than 9 mgl :
DO greater than 10% saturation
BOD not greater than 17 mgl 1
not likely to be anaerobic
inferior to Class 3
Comments
Suitable for potable supply
game fisheries and high
amenity value.
less high quality than 1A but
visible evidence of pollution
should be absent.
Moderate amenity value, capable
of supporting coarse fisheries.
Visibly polluted, and likely
to cause nuisance.
grossly polluted and likely to
cause nuisance.
are required to manage a river system to avoid undue treatment costs or
downstream pollution with the risk of prosecution. The simplest method
for calculating consents is a simple mass balance of conservative sub-
stances, which assumes instantaneous mixing of a single effluent entering a
river, shown in Figure 1.
This has the form:
C =
Qi
where Q - 5% upstream flow (95% exceedence flow)
C = mean concentration upstream of discharge
o
Q = mean effluent flow
C effluent discharge consent
i
C - 95% downstream class limiting concentration
2
211
-------
RIVER,
SEWAGE
TREATMENT
WORKS
UPSTREAM
Figure 1.
DOWNSTREAM
Simple mass balance mixing for an effluent discharge
River Thames
Sandhurst S.T.W.
Swallowfield
gauging stn.
KEY
o Sampling point
Sampling point /
gauging station
Sewa ge -1 real ment
works
Camberley S.T.W.
Figure 2 Location of
effluent discharges and river
sampling points on the Loddon
- Blackwater river system.
"ornborough S.T.W i
Farnborough ^Ash Yale S.T.W.
gauging stn.;
'Aldershot military SIVY"'
Atdershot town S.T.W. )
9
212
-------
There has been much critisism of this equation. At best it results in
a conservative value of GI which is an unknown statistic, not a 95 percentile
value, as recommended, for consent setting. Statistical mixing techniques
(Cluckie and Forster, 1982) can be used to give a more robust statistical
value, but this can only be applied to single point discharges and takes no
account of the self-purification processes acting in the river. To do this
for a number of discharges on a river system some form of catchment based
river quality model is required. The purpose of this study is to apply
QUAL-II to a pollution control problem on a U.K. river in an attempt to
optimise the allocation of sewage treatment facilities on that river system.
THE BLACKWATER-LODDON RIVER SYSTEM
The river Loddon is a tributary of the river Thames, with its catchment
area to the south west of London. The river Blackwater is the main tribu-
tary of the Loddon, shown in Figure 2. Above its confluence with the
Blackwater the Loddon has a quality class IB, but class 2 below the conflu-
ence (RQO = IB) due to a high BOD load from the river Blackwater (Table 2).
TABLE 2 RIVER QUALITY FLOW AT THE CONFLUENCE OF THE RIVERS LODDON AND
BLACKWATER. 1981 - 1982
Mean Flow D.O. BOD NHif-N NOa-N CLASS*
nr's'1 mgl"1 mgi"1 mgl'1 mgl'1
Loddon1 2.44 10.0 1.6 0.1 7.5 IB
Blackwater 3.38 9.2 2.4 0.3 6.8 2
Loddon2 5.82 8.9 2.2 0.2 7.2 2
1 above confluence
2 below confluence
* National Water Council 197 / River Quality Classification
results as means for the two year period.
Class limiting criterion is BOD
The quality of the Blackwater is the problem in the system. The river
Blackwater flows through an urbanised catchment area underlain by Tertiary
clays and sands, and has a naturally poor water quality and is designated
as quality class 2 (RQO = 2). The headwaters of the Blackwater contain
several old, overloaded sewage treatment works and the river carries a
high proportion of treated sewage effluent, which can be as high as 70% at
times of low flow.
213
-------
TABLE 3 RIVER QUALITY OBJECTIVES AND ACHIEVEMENTS 1976 - 82
River Quality Class
Reach Objective Achievement
length (km)
River Loddon
Source to Basingstoke S.T.W.
Basingstoke S.T.W. to River
Blackwater
Blackwater to Confluence
Thames
7.8
18.2
19.2
1A
IB
IB
IB
IB
River Blackwater
Source to Aldershot S.T.W
Aldershot S.T.W. to Cove Brook
Cove Brook to Confluence Loddon
4.5
11.7
19.5
2
2
3
3
Table 3 shows that despite a RQO of class 2, the Blackwater below
Farnborough sewage works is a class 3 river until the effluent load is
diluted below the confluence with the R. Whitewater, which is class IB.
This current situation, with a large portion of the Blackwater in class 3,
is itself undesirable for Thames Water Authority (TWA). However, its
effect on the Loddon, shown in Figure 3, is the major problem as this
source could be used for potable water supply.
The analysis of the sewage treatment works performance data, tabu-
lated in Table 4, shows that with the exception of Camberley, all the
works achieve compliance with their consents. However these are short
term or interim consents, set to match the current works performance, not
the desired downstream RQO. The consents for Ashvale, Farnborough
Camberley and Sandhurst all indicate poor performance due to overloading
and delapidation of the treatment plants. An additional problem for
water quality management on the Blackwater is that of Aldershot Military
sewage treatment works. This works is operated by the Government, it has
no consent and its discharge quality and quantity is unknown. A simple
mass balance suggests that the mean flow is around 7 Ml per day, with an
operational performance equivalent to a consent of 30/20/5.
To improve the quality of the Blackwater and the Loddon to achieve
214
-------
CAMBERLEY
R . BLACKWATER
\_
3
SANDHURST
FARNBOROUGH
2
*ASH VALE
ALDERSHOT
MIL!TARY
ALDERSHOT
Scale
mm
0 5km
Sewage Treatment Works
1 B-River Quality Class .
Figure 3 Current river quality classification for the Loddon-
Blackwater river system, based on mean annual data, 1973 - 1981
215
-------
TABLE 4 RIVER LODDON - BLACKWATER SYSTEM: SEWAGE TREATMENT WORKS PERFORMANCE
Works Stream Consent % Compliance
SS/BOD/AMM with consent
Basingstoke Loddon 15/10/10 96
Aldershot Blackwater 20/12/3 95
Ashvale Blackwater 40/12/15 99
Farnborough Blackwater 40/20/15 96
Camberley Blackwater 20/25/5 92
Sandhurst Blackwater 25/9/12 95
the designated RQO, TWA have proposed two improved management schemes for
sewage treatment on the river Blackwater. The two schemes are designed for
the river system tp achieve its RQOs with larger flows resulting from the
increased urban and industrial development predicted for 2020 A.D. Each of
the schemes, shown in Figure 4-has three phases of development to spread
capital expenditure. Both assume no changes at Aldershot Town and Aldershot
Military STW, with present performance and flows maintained. A capital
expenditure programme at Ashvale is already committed and this will take
flows upto 9.0 Ml/day with a consent of 10/10/3 by using a new Carrousel
treatment system (Forster, 1980).
The ultimate aim of scheme B was to close Farnborough, Camberley and
Sandhurst works and build a complete new works at a new site. This
proposal has been abandoned due to high cost. Therefore quality modelling
has been limited to forecasting the effects of Scheme A. This proposed the
eventual closure of Farnborough and Sandhurst and the diversion of flows to
Camberley. At Camberley, land is available for expansion at the present
site. In phase III the works would be expanded to treat the increased
flow but only to present levels of performance. The present works is over-
loaded and produces a poor quality effluent.
WATER QUALITY MODEL APPLICATION
Water quality problems on the river Blackwater were originally invest-
igated by Casapieri et al (1978) using a version of the W.R.C. river quality
model (Knowles and Wakeford, 1978). This, "Blackwater" model, is river
specific and was developed for TWA to predict the possible effects of
population growth. The deterministic model is one dimensional and operates
in a steady state mode. A plug-flow of water passing down the river is
simulated by splitting the river into reaches, at each inflow a new reach
is initiated. Non-point sources must be entered as point inflows, thereby
216
-------
Figure 4. Thames Water Authority effluent treatment management proposals
Sewage Effluent Flows to the River Blackwaler (Ml day7J raean)
WORKS
BASELINE
FLOW
PROPOSAL A
PHASE 1
II
II I
PROPOSALS
PHASE I
11 | III
ALDER5HOT
TOWN
ALDERSHOT
MILITARY
ASH VALE
FARNBOROUGH
10-5
7-0
7'4
7-5
10-5
7-0
9-0
8-A
ID'S
7-0
9-0
X
10'5
7-0
9-0
X
10-5
7-0
9-0
fl-A
10-5
7-0
9-0
X-
105
7-0
90
X
CAMBERLEY
SANDHURST
16 A
11-8
22-0
16-1
41-2
K-2
55-4
X
13-2
13-9
13-2
K-7
X
X
ATMOSPHERIC
AERATION
WEIR
AERATION
o
< t
u.
cc
2
g
c
5
0
II
o:
£
1
AMMONIA
nitrosomonos
NITRITE
nitrobacter
NITRATE
organic carbon
NITROGEN
/
Y
G
F
N
/ \
X^
\
BENTHIC
DEMAND
.,
CARBONACEOUS
BOD
Figure 5.
PLANT
RESPIRATION
AND
PHOTOSYNTHISIS
actions of the W.R.C. Blackwater
river quality model.
217
-------
starting a new reach. The water quality constituents simulated are DO, BOD,
ammonia (N1U-N) and nitrate (N03-N). Longitudinal dispersion and river
temperature are not modelled. The rate of nitrification is controlled by
the concentration of Nitrosomonas bacteria. The model input requires user -
calculated hydraulic relationships for each reach, for example, depth, time
of travel and channel cross-section. Also, relative plant density for each
reach must be supplied. The model is based on fieldwork derived rate con-
stants and constants. Mathematical solution is by numerical integration
of complete differential equations. The main interactions of the model are
shown schematically in Figure 5. The model has not been used since the
initial study and TWA have now implemented a general purpose stochastic
river quality model.
As a long term planning tool, the Blackwater model was run using
triannual mean data to predict mean annual river conditions. In this study
the NCASI (1980) version of QUAL-II SEMCOG (EPA, 1981) was applied to the
Loddon-Blackwater system using the same calibration data (1973-1975
triannual mean) as used for the Blackwater model. Wherever possible, the
same values for constants and rate constants were used in both models. The
main differences between the two models are shown in Table 5.
TABLE 5 WATER QUALITY MODEL COMPARISON
1) W.R.C. Blackwater Model:
Deterministic, steady state
input: 17 rate constants
15 physical constants and parameters values
output: Dissolved Oxygen, BOD, Ammonia, Nitrate,
- Specific to River Blackwater.
2) U.S. E.P.A. Qual-II Model:
Deterministic, steady state/pseudo dynamic
input: 28 constants and river identification parameters
16 parameter values
output: 8 non conservative substances
3 conservative substances
+ temperature
I
- Generally applicable to any river system
218
-------
QUAL-II is a non specific model and has a complex representation of bio-
chemical interactions. The main differences in the biochemical modelling
by the two models are:
1) the use of algae by QUAL-II and plant density by the Blackwater model,
as biological factors influencing DO concentrations.
2) the Blackwater model uses a single stage nitrification mechanism
controlled by the concentration of Nitrosomonas bacteria. QUAL-II uses a
two stage feedback mechanism of algal growth to influence the conversion
rates from amonia to nitrite and nitrite to nitrite.
The parameter estimates used in applying QUAL-II are shown in Table 6.
The Blackwater has significant weed growth during the summer and it was
possible that this might cause problems with QUAL-II. QUAL-II is also
usually used to predict specific situations, not general forecasts based
on means as input data.
TABLE 6 PARAMETER ESTIMATES USED IN BLACKWATER/LODDON CATCHMENT SIMULATION
Parameters
Value/Comments
Reaeration Rate
Longitudinal Dispersion
Decay/Oxidation Rates (1 day
BOD-U
Ammonia N
Nitrate N
Algal Parameters
Maximum Growth Rate
Respiration Rate
Settling Velocity
Chlorophyll Content
P Content
N Content
Light Extinction
Half-Saturation Constants
Algal P Uptake
Algal N Uptake
Tsivoglou and Wallace, 1972 reach coeff 1.3
Elder (1959)
0.15-0.25
0.35
1.0
2.5 1 day/"1
.1 1 day i
.15 m day"1
.050 yg Chl-a mg"1 Algae
.012 mg P mg"1 Algae
.085 mg N mg"1 Algae
38.0 m2 g"1Chl-a
.04 mg m_
.3 mg m
-3
Benthic Oxygen Fluxes gm"2 day"1 - 0.0
Both models were calibrated using triannual data for 1973-1975 and then
verified and recalibrated using data for 1975-1978. They were then used to
219
-------
predict mean annual quality for 1981 and the results were compared to the
observed mean river quality. The comparisons for DO, BOD and NH^-N are
shown in Figure 6. To assess the goodness-of-fit of the models to the
observed data, the weighted root mean square error (WRMSE) statistic
(Anglian Water Authority, 1979) was assessed for each of the four modelled
variables, where:-
N
2
WRMSE = <> Wi (0± - Ci)
i = 1
N
0. is the observed concentration at the i sampling
point
C. is the calculated concentration
N is the total number of sampling points
W is a weight ing factor, the reciprocal of the standard
deviation of the observed concentrations.
The WRMSE statistic is non-dimensional and allows a comparison of
results for different variables. The smaller the value of the WRMSE than
statistic then the closer is the fit between observed and calculated results.
Table 7 shows a comparison of the results.
TABLE 7 WRMSE STATISTIC VALUES
Variable Model
QUAL-II Blackwater
DO 0.47 0.80
BOD 0.35 0.49
NHit-N 0.42 0.41
N03-N 0.82 1.23
from 1981 mean annual results.
Neither model gave a perfect prediction. The results for QUAL-II showed
that BOD and NHi»-N were the 'best' modelled variables, followed by DO and
NOs-N. In selecting a model for any specific situation, it is important to
use the simplest model that will yield adequate results, that is, fitting
the model to the problem, not the problem to the model. In this study the
two models differ markedly in their complexity. However the simpler
Blackwater model produces poorer results than QUAL-II. The major difficulty
in applying either model is the availability of sufficient hydraulic data.
220
-------
11
DO
(mg I'1)
RIVER BLACKWATER
1981 Mean Values
8LACKWATER
/".,-QUAL II
/ .//OBSERVED
12 24
Distance (km)
BOO
(mg I1)
RIVER BLACKWATER
1981 Mean Values
QUAL II
2-4
1-6
NH^-N
(mg I'1)
0-8
OBSERVED
BLACKWATER
12 24
Distance (km)
RIVER BLACKWATER
1981 Mean Values
BLACKWATER
-QUAL II
OBSERVED
12 24
Distance (km)
Figure 6. Comparison of Water Quality Model predictions
221
-------
Also only routine monitoring quality data was available.
Despite these data problems, the comparison of the models showed that
QUAL-II could be applied to the Blackwater-Loddon system to predict mean
annual river quality conditions.
The QUAL-II model was then recalibrated using 1981 data and reverified
against 1982 data. The observed and predicted results are shown in Figure.
7. Although these were better than for the model comparison, NHi*-N was still
the worst modelled variable. The predicted mean values for 1982, shown in
Figure 8, indicate a slight improvement in the river quality since the 1970s.
DO is not a class limiting parameter, however the river Blackwater is in
class 2 above its confluence with the river Whitewater due to high values
for BOD and NHi^-N. There are no class 3 reaches on the Blackwater but the
river Loddon is still class 2, not IB below its confluence with the
Blackwater. BOD is the limiting parameter.
FORECASTING THE EFFECTS OF CHANGED TREATMENT STRATEGIES
In modelling the effects of the Scheme A proposals for changes in the
effluent treatment strategy on the river Blackwater, shown in Figure 4, it
was decided to attempt to obtain results within the probabalistic framework
of the consents and RQO. To do this QUAL-II was run on a worst case basis.
Effluent discharges were set to mean flows and the quality of each effluent
was taken as its 95 percentile (i.e. consent) quality. 95 percentile values
for river quality and 95 percent exceedence river flows were modelled. While
it is recognised that this worst case modelling procedure will give results
that are not 95 percentile values, the results will be highly conservative
and will probably overestimate the true 95 percentile values. Therefore
they can be used as guidelines for assessing the predicted river quality in
terms of the 95 percentile class limiting criteria used to test for
compliance with the set RQO.standards. This is an initial crude attempt
to use QUAL-II with stochastic input variables to predict river quality
within the probabalistic framework of consents and RQO used in the U.K.
The results of modelling the worst case situation for phase I of the
scheme are shown in Figures 9 and 10. In this phase, effluent discharges
would be increased but there would be no changes to the treatment facilities.
The results showed that the quality of the Blackwater and Loddon would
deteriorate. The main limiting variable would become NHi,-N. Parts of the
Blackwater would become class 3. This would be worse than the present
situation, therefore phase I proposals are not acceptable. Similarly
phase II was also assumed to be unacceptable and was not modelled. Phase
III proposals involve closing Farnborough and Sandhurst works and diverting
the flows to an enlarged works at Camberley. However, this works would
maintain its present consent. The results of this proposal would be worse
than those for phase I, as shown in Figures 9 and 10. Phase III proposals
cause a high input of BOD and NHit-N at a single point in the Blackwater
resulting in class 3 conditions downstream.
The only possible solution to the problem was to reduce the pollutant
222
-------
LJ
X
o .
UJ8-
o -
CO
OBSERVED DO
PREDICTED DO
msTfl« ,K»f ";" " *
to
cr
0)
tfl
3
C
CO
T3
0)
o
CD -
88.19 3>.M 7..M ig.
DISTflNCE (KM)
OBSERVED BOO.flTU
PREDICTED BOD.HTU
1 1 1
13.M t.M
1
i.a
OBSERVED NHV.N
PREDICTED NHV.N
~i 1 1 1 r~
M.« a.a S..M te.n 12.*
DISTflNCE (KM)
a
OJ
CD-
CM
«X)
CO O
o) 6
> ,
^2 CO
o
M
C CD
Q) >
0) -H
CU
C
0)
3»
O &
co A:
H O
5rt J2
CO iH
O. PP
e i
o a
CJ O
-a
o
r^* i 1
-------
ro
ro
W
M
pi
O
sr
O
rt
(D
l-t
CO
0)
rt
(6
&
o>
O
l-t
0)
f
O
a.
a-
o
3
1982 MEflN WflTER QURLITY
9TERDY STflTE
X
_+
«
<>
FLOW(CUMECS)
DO
BOD.flTU
NH^-.N
NQ3.N
2
f'e.K)
88.
1 - 1 - 1 - 1 - 1
80.08 2V. M 18.00
DISTflNCE (KM)
«0(_J
LJ
- z:
-8§
s
12.80
6.80
0.00
-------
X .
o
DSJ
UU
O
LOS-
* ,
V
/r\
w. n 36.ee
N.W.C. RIVER QUALITY CLASSIFICATION
95 percentile class limiting criteria - DO
1A
IB
-X PROPOSRL fl PHflSE i
-+ PROPOSflL fl PHRSE! 11
-<>-- PROPOSflL fl PHRSEIII UPG
is.ra a..ea is.et
DISTflNCE (KM)
12.00 6.B3
a.ze
N.W.C RiVER QUALITY CLASSIFICATION
95 percentile class limiting criteria BOD
X PROPOSflL fl PHRSEI
- PROPOSflL fl PHHSEI 11
o PROPOSOL R PHRSE I 11 UPC
w.ea w.az ae.aa ae.ua T..BB is.ra is.ee fi.ea
DISTflNCE (KMJ
a.ee
Figure 9. Predicted worst case effects of effluent management
schemes on river quality -DO and BOD
225
-------
y
PROPOSRL R PHflSEI
PROP09RL R PHflSEI
PROPOSRL R PHRSEI
2
UPC
DISTRNCE (KM)
N.W.C. RIVER QUALITY CLASSIFICATION
95 percentile class limiting criteria NH^-N
Figure 10. Predicted worst case effects of effluent
management schemes on river quality NHi»-N
226
-------
load entering the river Blackwater. This could not be done using the
existing treatment facilities and the scheme A proposals. Scheme B, to
build a complete new works at a new location, was not feasable. Therefore
a new scheme had to be proposed. Land for expansion is available at
Camberley and pollutants from Camberley have time to be assimilated in the
river before entering the river Loddon. Therefore it was decided to base
the new proposals on upgrading the existing plant at Camberley. Farnborough
and Sandhurst would be closed and the flows diverted to Camberley. At
Camberley, 25% of the existing flow, and the diverted flow plus the future
forecasted flow increment would be treated by a new Carrousel activated
sludge plant. This would enable the present biological filtration plant to
be retained and to perform satisfactorily without overloading. The new
Camberley plant would be similar to the new system under construction at
Ashvale and would be expected to produce a similar strength effluent
(Forster, 1980). The new works would have a consent of 10/10/3 instead of
the present consent of 20/25/5.
The results of modelling this proposal (scheme A, phase III with
upgrading) as a worst case situation are shown in Figures 9, 10 and 11.
With this proposal all the RQO could be complied with. The river Blackwater
would become class 2 along all its length. Ammonia would be the class
limiting variable, due to the poor effluent dilution. The main problem in
the system would be overcome, the river Loddon would maintain its class IB
quality below the confluence with the Blackwater shown in Figure 12.
Thames Water Authority is now reconsidering the proposals for changes
in the management of waste treatment facilities on the Blackwater-Loddon
system in view of the effects predicted by applying QUAL-II.
REFERENCES
Anglian Water Authority, 1979. The Bedford Ouse Study. Final Report.
Casapieri, P. Fox, T.M. Owens, P. and Thompson, G.D. 1978. "A mathe-
matical deterministic river quality model. Part 2, Use in evaluating
the water quality management of the Blackwater catchment" Water
12, 1154-1161.
Cluckie, I.D. and Forster, C.F. 1982. "Observations on a statistical
approach to the setting of discharge consent conditions". Environment^
Technology Letters 3. 111-116
Elder, J.W. 1959. "The dispersion of marked fluid in turbulent shear
flow", Journal of Fluid Mechanics,. 5 544-560
Environmental Protection Agency, 1981. Users manual for stream quality
model (QUAL-II) EPA-600/9-81-015 Athens '
227
-------
STEflOY STRTE
FlOUtCUKECSJ
--- X --- DO
--- + -- BQD.flTU
CE
ID
C3
<> --- N03.N
. 1^ I I I I II I I I I I I 1 I I I
*a.ea V2.ee ss.aa sa.ea T..88 is.ea 12.ea e.u i.u
DISTRNCE (KM)
UJ
21
Ls
Aldershot
Aldershot
Military
Cove
Brook
R.Whitewater
R.Thames
Ashvale Camberley
R.Loddon
N.W.C. RIVER QUALITY CLASSIFICATION
95 percentile class limiting criteria
BOD NHAN
Figure 11. QUAL-II simulation of worst case river quality for
the optimum effluent treatment management scheme
228
-------
ALDER5HOT
MIL!TARY
ALDERSHOT
Scale
^M
0 5km
Sewage Treatment Works
1 B-River Quality Class .
Figure 12. Predicted river quality classification for the Loddon-
Blackwater river system with the optimum effluent treatment scheme.
229
-------
Forster, C.F. 1980. "A comparison of the performances achieved by the
Carrousel and the Mammoth rotor versions of the oxidation ditch".
Environmental Technology Letters 1. 366-375
Knowles, C. and Wakeford, A.C. 1978. "A mathematical deterministic
river quality model. Part 1. emulation and description". Water
Research 2, 1149-1153
National Council of the Paper Industry for Air and Stream Improvement
(NCASI) 1981. "A review of the mathematical water quality model QUAL-II
and guidance for its use". Stream improvement technical bulletin. 338
National Water Council, 1977. "River Water Quality - the next stage".
National Water Council London.
National Water Council, 1978 "Review of discharge consent conditions-
consultation paper". National Water Council London
Russell, K.D. and Pescod, M.B. 1981. "The use of mathematical modelling
as a decision making tool in the Water Authorities of England and Wales
for Water Quality Management". Water Pollution Control. 80 No. 3.
390-397
ACKNOWLEDGEMENTS
This study was supported by funds from the U.K. Science and Engineering
Research Council and the Natural Environmental Research Council. The data
and model for the Loddon-Blackwater system were supplied by Thames Water
Authority. Additional travel funds were provided by The Royal Society.
230
-------
TECHNIQUES AND SOFTWARE FOR RESERVOIR EUTROPHICATION ASSESSMENT
by: William W. Walker, Jr.
Environmental Engineer
1127 Lowell Road
Concord, Massachusetts 01742
ABSTRACT
Results of a research project on empirical techniques for predicting
eutrophication and related water quality conditions in reservoirs are
summarized. The research is documented in a series of reports describing data
base development(1), preliminary model testing(2), and model refinements(3).
Using an extensive reservoir data base, previous models based primarily upon
data from phosphorus-limited, northern, natural lakes have been tested and
modified to improve their performance and generality in reservoir
environments. The revised models account for algal growth limitation by
phosphorus, nitrogen, light, and flushing rate. By considering nutrient
sedimentation and transport mechanisms in a mass-balance framework, spatial
variations in nutrients and related water quality conditions can be simulated.
A manual and three computer programs have been developed to facilitate
implementation of the models (4). The paper presents an overview of the
research and software.
INTRODUCTION
Eutrophication can be defined as the nutrient enrichment of water bodies
leading to an excessive production of organic materials by algae and/or
aquatic plants. This process has several direct and indirect impacts on
reservoir water quality and uses for water supply and recreation. Valid,
practical assessment techniques for eutrophication are required to support
reservoir water quality management efforts.
A four-phase research project has been undertaken to develop empirical
modeling approaches for reservoir applications (1,2,34). The first phase
involved the compilation and statistical summary of a data base describing
morphometry, hydrology, and water quality conditions in 299 Corps of Engineer
(CE) reservoirs (1). The second phase involved preliminary screening of
existing models based upon the CE data base and an extensive list of model
formulations compiled from the literature (2). The third phase involved model
231
-------
refinements and additional testing based upon independent lake and reservoir
data sets (3). The fourth phase involved the development of an applications
manual and supporting computer programs (4). Following are descriptions and
illustrations of model structures and supporting software.
EMPIRICAL MODEL EVOLUTION
Eutrophication models can be broadly classified as theoretical or
empirical. The former generally involve direct simulation of physical,
chemical, and biological processes superimposed upon a simulation of reservoir
hydrodynamics. The latter are based upon mass-balance and limiting-nutrient
concepts; they relate average eutrophication symptoms to external nutrient
loadings, hydrology, and, morphometry using statistical models derived from a
groups of lakes and/or reservoirs. Within their application ranges, empirical
methods offer certain advantages in terms of simplicity and limited data
requirements. Early empirical models developed by Dillon and Rigler(5),
Vollenweider(6), and several others were based primarily upon data from
northern lake data sets; their applicability to reservoirs is questionable
because of lake/reservoir and regional differences in characteristics which
influence nutrient dynamics, including morphometry, hydrology, and
sedimentation (7).
As summarized by Reckhow and Chapra (8), empirical approaches generally
involve the linkage of two types of models:
(1) Nutrient Balance Models relate pool or discharge nutrient levels to
external nutrient loadings, morphometry, and hydrology.
(2) Eutrophication Response Models describe relationships among
eutrophication indicators within the pool, including nutrient levels,
chlorophyll-a, and transparency.
Generally, models of each type must be employed to relate external nutrient
loadings to lake or reservoir water quality responses. In the absence of
loading information, however, application of eutrophication response models
alone can provide useful diagnostic information on existing water quality
conditions and controlling factors.
Lake nutrient-balance models have generally evolved from a simplistic
"black-box" representation which treats the impoundment as a continuous
stirred-tank reactor at steady-state and the sedimentation of phosphorus as a
first-order reaction (Figure 1). Using mass-balance data from groups of
lakes, the sedimentation terms are empirically calibrated for predicting
spatially- and temporally-averaged conditions. Model inputs can be expressed
in three terms:
(1) Inflow Total Phosphorus Concentration (nutrient supply factor);
(2) Mean Depth (morphometric factor);
(3) Hydraulic Residence Time (hydrologic factor).
232
-------
Response models (Figure 2) typically consist of bivariate regression equations
relating each pair of response measurements (e.g., phosphorus/chlorophyll,
chlorophyll/transparency, etc.). Phosphorus is assumed to control algal
growth and other eutrophication-related water quality conditions.
INFLOW-* ^U OUTFLOW
t
NET SEDIMENTATION
Figure 1. Lake Model Segmentation
INFLOW TOTAL Ps^^
MEAN DEPTH -^D~TOTAL P'
HYDflAULIC RESIDENCE TIME
Figure 2. Lake Eutrophication Model Network
In adapting this modeling approach for use in reservoirs (2,3),
modifications have been designed to account for the following:
(1) Effects of nonlinear sedimentation kinetics; a second order kinetic
model appears to be more general than a first-order model both for
predicting among-reservoir, spatially-averaged variations and for
predicting within-reservoir, spatial variations in total phosphorus and
total nitrogen;
(2) Effects of inflow nutrient partitioning (dissolved vs. particulate or
organic vs. inorganic); because of differences in biological
availability and sedimentation rates, reservoir responses appear to be
much more sensitive the ortho-phosphorus loading component than to the
non-ortho (total - ortho) component;
(3) Effects of seasonal variations in nutrient loadings, morphometry, and
hydrology; pool water quality conditions are related more directly to
seasonal than to annual nutrient balances in impoundments with
relatively high flushing rates;
(4) Effects of algal growth limitation by phosphorus, nitrogen, light, and
flushing rate; simple phosphorus/chlorophyll-a relationships are of
limited use in reservoirs because nitrogen, light, and/or flushing rate
may also regulate algal growth, depending depending upon site-specific
conditions;
233
-------
(5) Effects of spatial variations in nutrients and related variables;
spatial variability in trophic state indicators is significant in many
reservoirs (in some cases, spanning from oligotrophic to hypereutrophic
conditions, Figure 3). and predictions of "average" conditions are of
limited use.
2.00
1.75
i 1.50
I
0 1.25
o
«i
O
z
O
P 0.75
en
0.50
0.25
0.00
'
1
I
1
1
.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.62.8 3.0
INFLOW TOTAL P
Figure 3. Ranges in Station-Mean Chlorophyll-a (PPB) vs. Reservoir
Inflow Total Phosphorus Concentration (PPB), LOG10 Scales
To permit simulation of spatial variations, nutrient balance models are
implemented in a spatially segmented framework which accounts for advection,
dispersion, and sedimentation (Figure 4). While each segment is modeled as
vertically mixed, the methodology is applicable to stratified systems because
the sedimentation rate formulations have been empirically calibrated to data
from a wide variety of reservoir types, including well-mixed and vertically
stratified systems. The revised model network (Figure 5) is designed to
improve generality (vs. Figure 2) by incorporating additional independent
variables and controlling factors found to be important in model testing
(2,3). When the network is applied to predict spatially-averaged conditions
in 40 CE reservoirs, coefficients of determination (R-squared values) are 91%
for Total P, 88% for Total N, 80% for Chlorophyll-a, 86% for Secchi Depth, and
92% for Hypolimnetic Oxygen Depletion Rate. Model structures and coefficients
have been tested against several independent lake and reservoir data sets.
INFLOW-
t t AOVECTION t t
*
t
^
olo
J
DISPERSION
J.
^
1
-OUTFLOW
NET SEDIMENTATION
Figure 4. Reservoir Model Segmentation
234
-------
MEAN HYPOLIMNETIC DEPTH
INFLOW TOTAL P.
INFLOW ORTHO P-
MEAN TOTAL DEPTH.;
HYO. RESIDENCE TIME<
INFLOW TOTAL N-
tNFLOW INORGANIC N'
SUMMER FLUSHING RATE:
MEAN DEPTH OF
MIXED LAYER-
NON-ALGAL TURBIDITY'
RESERVOIR
TOTAL P
HYPOLIMNETIC 0?
DEPLETION RATE
METALIMNETIC 0>
DEPLETION RATE
SECCHI
ORGANIC N
TOTAL P ORTHO P
Figure 5. Reservoir Eutrophication Model Network
APPLICATIONS MANUAL AND SOFTWARE
Figure 6 depicts basic steps involved in applying the methodology, as
described in the applications manual (4). Three computer programs have been
written to assist at various stages of the analysis. The functions of these
programs are outlined below:
(1) FLUX: estimation of tributary mass discharges (loadings) from
concentration data and continuous flow records;
(2) PROFILE: display and reduction of pool water quality data;
(3) BATHTUB: implementation of nutrient-balance and eutrophication-response
models in a spatially segmented hydraulic network.
Each program is written in FORTRAN-66, a language which is highly
transportable among computer systems. The basic structure and functions of
each program are outlined below.
PROBLEM DEFINITION
I
DATA COMPILATION
I
DATA REDUCTION
MODEL IMPLEMENTATION
TRIBUTARIES
FLUX
POOL
PROFILE
BATHTUB
Figure 6. Assessment Pathways
FLUX is an interactive program for estimating loadings or mass discharges
passing a tributary monitoring station over a given period (Figure 7). The
loading estimates can be used in formulating reservoir nutrient balances over
annual or seasonal averaging periods appropriate for application of empirical
eutrophication models. The function of the program is to interpret water
quality and flow information derived from intermittent grab or event sampling
to estimate mean (or total) loading over the complete flow record between two
dates.
235
-------
DATA
ENTRY
DATA DIAGNOSTIC
LISTING PLOTS
DATA
STRATIFICATION
ONLINE
DOCUMENTATION
(HELP)
RESIDUALS
ANALYSIS
LOADING
CALCULATION
Figure 7. FLUX Schematic
Since the appropriate loading calculation method depends partially upon
the concentration/flow/seasonal dynamics characteristic of a given station and
component and upon the sampling program design, five alternative calculation
methods are provided. An option to stratify the samples into groups based
upon flow and/or date is also included. In many cases, stratifying the sample
increases accuracy and reduces potential biases in loading estimates (9,10).
The variances of the estimated mean loadings are calculated to provide
relative indications of error. A variety of graphic and statistical
diagnostics are included to assist the user in evaluating data adequacy and in
selecting the most appropriate calculation method and stratification scheme
for each loading estimate. The program can also be used to improve the
efficiencies of monitoring programs designed to provide data for calculating
loadings and reservoir mass balances by optimizing sampling effort among flow
strata.
PROFILE is an interactive designed to assist in the analysis and
reduction of pool water quality measurements (Figure 8). The user supplies a
data file containing basic information on the morphometry of the reservoir,
monitoring station locations, surface elevation record, and water quality
monitoring data referenced by station, date, and depth.
OATA
OATA INVENTORY DATA
LISTING i TRANSFORMATION
DATA
ENTRY
ONLINE
DOCUMENTATION
(HELP)
OXYGEN
DEPLETION
CALCULATIONS
PROFILE
MAIN
PROGRAM
DATA
WINDOW
DATA
DISPLAY
SURFACE
WATEfl QUALITY
SUMMARY
Figure 8. PROFILE Schematic
236
-------
PROFILE'S functions are in three general areas:
(1) display of concentrations as a function of elevation, location, and/or
date, using a variety of one-, two:- and three-dimensional formats;
(2) robust calculation of mixed-layer summary statistics and standard errors
in a two-way table format (spatial x temporal);
(3) calculation of hypolimnetic and metalimnetic oxygen depletion rates from
temperature and oxygen profiles;
Given adequate pool monitoring data from a particular reservoir, PROFILE
assists the user in developing an appreciation for spatial and temporal
variability in the reservoir. This may lead to refinements in monitoring
program design. Summary statistics calculated in a uniform manner
characterize reservoir trophic status and can be compared with predictions of
nutrient-balance and eutrophication-response models in subsequent modeling
steps.
BATHTUB facilitates application of empirical models to morphometrically
complex reservoirs. The program performs water and nutrient balance
calculations in a steady-state, spatially-segmented hydraulic network which
accounts for advective transport, diffusive transport, and nutrient
sedimentation (Figure 4). Eutrophication-related water quality conditions are
predicted using empirical relationships tested for reservoir applications
(Figure 5). As indicated in Figure 6, applications of BATHTUB normally follow
use of FLUX for reducing tributary monitoring data and use of PROFILE for
reducing pool monitoring data, although use of the data-reduction programs is
optional if independent estimates of tributary loadings and/or average pool
water quality conditions are used.
To reflect data limitations or other sources of uncertainty, key inputs
to the model (flows, loadings, morphometry, observed pool concentrations,
etc ) can be specified in probabalistic terms (mean and coefficient of
variation) Outputs are expressed in terms of a mean value and coefficient of
variation for each mass balance term and response variable. .Output
coefficients of variation are based upon a first-order error analysis (11)
which accounts for input variable uncertainty and inherent model error (Figure
9).
INPUT
I
MODEL CORE-*.
I )
ERROR ANALYSIS^
I
OUTPUT
Figure 9. BATHTUB Schematic
237
-------
Potential applications of BATHTUB can be broadly outlined in the
following categories:
(1) Diagnostic:
formulation of water and nutrient balances, including identification
and ranking of potential error sources;
ranking of trophic state indicators in relation to user-defined
reservoir groups and/or the CE reservoir data base;
identification of factors controlling algal production;
(2) Predictive:
assessing impacts of changes in water and/or nutrient loadings;
assessing impacts of changes in mean pool level or morphometry;
predicting long-term-average conditions in a new reservoir;
estimating loadings consistent with water quality objectives.
BATHTUB operates in a batch model (non-interactive) and generates output in
ten optional formats, as appropriate for specific applications. The water
balances are expressed as a system of simultaneous linear equations which is
solved via matrix inversion to estimate the advective outflow from each model
segment. The mass balances are expressed as systems of simultaneous nonlinear
equations which are solved iteratively via Newton's Method (12).
Through appropriate configuration of model segments, BATHTUB can be
applied to a wide range of reservoir morphometries and management problems.
Possible segmentation schemes include:
(A) Single Reservoir, Spatially Averaged
(B) Single Reservoir, Spatially Segmented
(C) Partial Reservoir or Embayment, Spatially Segmented
(D) Single Reservoir, Spatially Averaged, Multiple Load Scenario
(E) Collection of Reservoirs, Spatially Averaged
(F) Network of Reservoirs, Spatially Averaged
Segments can be modeled independently or linked in a one-dimensional, branched
network. Multiple external sources and/or withdrawals can be specified for
each segment. Within certain limitations, combinations of the above schemes
are also possible.
Typical model output for Segmentation Schemes B-F is illustrated in
Figure 10. Observed phosphorus concentrations in these plots are based upon
mixed-layer, growing-season measurements in the reservoir or lake pool.
Estimated concentrations are calculated from external phosphorus loadings
using empirical second-order sedimentation models developed for CE reservoirs
(3). The program generates similar plots for each response variable. Schemes
A-C deal with single reservoirs or embayments. In Schemes D and E, model
segments are run "in parallel" to permit simulation of alternative loading
scenarios or independent reservoirs, respectively. Scheme E is particularly
useful for obtaining regional perspectives on trophic status and load/response
relationships in a collection of lakes or reservoirs. Scheme F permits
routing of water and nutrients through a network of lakes or reservoirs, each
of which may be spatially segmented.
233
-------
(B)
LAIS RAT 80BBARD
(C)
(D)
am
i mi
\m
TOTAL P HG/H
lg.17 27.10 45.41 7t.U 127.57 213.82 158.17
is:! -^=*
«.?
8.8
ST ALBAIS BAY, LAIB C B A II P L A I B
TOTAL P RG/H *
17.14 25.16 JS.fi M.M 79.84 117.29 172.32
S.i
MABBBACB RB8BRVOIR
TOTAL t BO/«
4.44 t.12 8.96 U.77 18.15 25.60 3t.t7
m
S:J
'3
t BUM. ILMR 0» 1177
lift! lift
\m 7:$-^
*
(E)
BORTIBOTOB DISTRICT, COB
TOTAL P NG/H
7.41 13.34 24.03 41.28 71.94 140.17 2S1.BO
(F)
OR B At L At S<
TOTAL P NG/M
^.81 3.34 6.18 U.14 20.90 11.50 70.92
i:!
ii:8
1:!
g:i
if:!
Figure 10. Observed (0) and Estimated (E) 90% Confidence Ranges
for Mean Total Phosphorus Concentrations Illustrating
Various Types of BATHTUB Segmentation Schemes
CONCLUSIONS
Results of a recent research project on empirical modeling techniques for
reservoir eutrophication have been summarized. Detailed documentation of
model origins, structures, and limitations can be found in the project reports
(1,2,3). The applications manual and software (4) will be generally available
in mid-1985.
REFERENCES
1. Walker, W.W., "Empirical Methods for Predicting Eutrophication in
Impoundments; Report 1, Phase I: Data Base Development", prepared for
Office of the Chief, Army Corps of Engineers, Washington, D.C., monitored
by Environmental Laboratory, USAE Waterways Experiment Station,
Vicksburg, Mississippi, Technical Report E-81-9, May 1981.
239
-------
2. Walker, W.W., "Empirical Methods for Predicting Eutrophication in
Impoundments; Report 2, Phase II: Model Testing", prepared for Office of
the Chief, Army Corps of Engineers, Washington, D.C., monitored by
Environmental Laboratory, USAE Waterways Experiment Station, Vicksburg,
Mississippi, Technical Report E-81-9, September 1982.
3. Walker, W.W., "Empirical Methods for Predicting Eutrophication in
Impoundments; Report 3: Model Refinements", prepared for Office of the
Chief, Army Corps of Engineers, Washington, D.C., monitored by
Environmental Laboratory, USAE Waterways Experiment Station, Vicksburg,
Mississippi, Technical Report E-81-9, June 1984.
4. Walker, W.W., "Empirical Methods for Predicting Eutrophication in
Impoundments; Report 4: Applications Manual", prepared for Office of the
Chief, Army Corps of Engineers, Washington, D.C., monitored by
Environmental Laboratory, USAE Waterways Experiment Station, Vicksburg,
Mississippi, Technical Report E-81-9, Draft March 1985.
5. Dillon, P.J. and F.H. Rigler, "The Phosphorus-Chlorophyll Relationship in
Lakes", Limnology and Oceanography. Vol. 19, No. 4, pp. 767-773, 1974.
6. Vollenweider, R.A., "Advances in Defining Critical Loading Levels for
Phosphorus in Lake Eutrophication", Mem. 1st. Ital. Idrobiol. , Vol. 33,
pp. 53-83, 1976.
7. Thorton, K.W., R.H. Kennedy, J.H. Carrol, W.W. Walker, R.C. Gunkel, and
S. Ashby, "Reservoir Sedimentation and Water Quality An Heuristic
Model", presented at the ASCE Symposium on Surface Water Impoundments,
Minneapolis, Minnesota, 2-5 June 1980.
8. Reckhow, K.H. and S.C. Chapra, Engineering Approaches for Lake Management
Volume 1; Data Analysis and Empirical Modeling. Butterworth Publishers,
Boston, 1983.
9. Bodo B., and I.E. Unny, "Sampling Strtegies for Mass-Discharge
Estimation", Journal of the Environmental Engineering Division. American
Society of Civil Engineers. Vol. 109, No. 4, pp. 812-829, August 1983.
10. Bodo B., and T.E. Unny, "Errata: Sampling Strtegies for Mass-Discharge
Estimation", Journal of the Environmental Engineering Division. American
Society of Civil Engineers. Vol. 110, No. 4, pp. 867-870, August 1984.
11. Walker, W.W., "A Sensitivity and Error Analysis Framework for Lake
Eutrophication Modeling", Water Resources _B_ul_l_et_in. Vol. 18, No. 1, pp.
53-61, February 1982.
12. Burden, R.L., J.D. Faires, and A.C. Reynolds, Numerical Analysis.
Prindle, Weber, and Schmidt, Publishers, Boston, Massachusetts, Second
Edition, 1981.
240
-------
MODELING OF PHOSPHORUS CONCENTRATIONS FROM DIFFUSE SOURCES
by: D.J. Andrews
Marshall Macklin Monaghan Limited
Don Mills, Ontario
Canada M3B 2Y1
K.K.S. Bhatia
National Institute of Hydrology
Roorkee (U.P.)
India - 247 667
E.A. McBean
Department of Civil Engineering
University of Waterloo
Waterloo, Ontario
Canada N2L 3G1
ABSTRACT
The primary nutrients which control the trophic state of water bodies are
nitrogen, phosphorus and carbon. Of these, phosphorus is recognised as the
most limiting and most easily controlled because a primary source of phosphoru
is frequently from domestic sewage. However, control of this particular
nutrient involves high costs and even if the phosphorus input is controlled
from point sources, a significant phosphorus presence may remain, as contri-
buted by the uncontrolled diffuse sources. The focus of the paper is to link
phosphorus concentration data with commonly-measured watershed characteristics
such as daily flow. The data bases used in the study are, four watersheds of
the Grand River (for a period of three years and nine months). The regression
models for the most part, were based on flow or parameters which can be
derived from flow. The dependent variables include a) total phosphorus,
b) filtered reactive phosphorus concentration and c) total dissolved phosphor-
us concentration. The regression models were further based on seasonality.
Useful regression models were produced on all the watersheds studied for the
total phosphorus form of species. However, for other species the results were
not very conclusive. The models had a tendency to overpredict low to median
concentrations and to underpredict high concentrations. Conclusive results
were obtained on the use of seasonal models.
241
-------
INTRODUCTION
In recent years a lot of attention has been directed toward the
eutrophication of water bodies (1). Such studies have led to the identifica-
tion of primary nutrients - nitrogen, phosphorus and carbon in controlling the
trophic state of lakes (2). Phosphorus is normally present in most natural
waters in relatively low concentrations. Because it is essential to the plant
growth, it has become the focus of attention in the entire issue of
eutrophication which has led to a trend in wastewater treatment facilities to
control output of this nutrient, often at very high costs. The possibility
exists, however, that even with complete elimination of point source
phosphorus, there will be sufficient phosphorus contributed through
uncontrolled, diffuse sources to provide an adequate nutrient basis for the
eutrophication process.
A primary difficulty in characterizing the diffuse sources is caused by
their variability, both spatially and temporally. Efforts aimed at
characterizing the phosphorus contribution have therefore taken the form of
agricultural plot studies.or flux studies (3),(A). The agricultural plot
studies, while able to give typical concentrations found in runoff from the
test sites, do not account for the transport phenomena and therefore cannot
specify the ultimate magnitude and time of delivery to the major water body.
The flux studies, though overcoming this difficulty, are normally watershed-
specific. Ideally, to properly estimate diffuse source loading, some form of
model is required which links phosphorus concentrations with a commonly
measured watershed parameter such as daily flow.
The purpose of this paper is to describe the success in formulating
regression models, based on data collected from intensively measured water-
sheds. The regression models for various forms of phosphorus were formulated
using flow (or variables derived from flow), various seasons and time. The
models were developed for the Grand River Basin. In a related paper (5),
these models are used in the St. John River Basin.
DATA USED IN THE STUDY
The data used for developing the regression models were for the Grand
River (Figure 1). The data were provided by the Ontario Ministry of the
Environment, covering the period from March 1974 to December 1977, for some
40 stations (6). Table 1 gives the major features of the various sampling
sites (four in number).
METHODOLOGY
After a critical review of the models available for phosphorus modelling
(7),(8),(9), it was observed that
a. Different models are required for different species of phosphorus
(even within the same watershed) because of relative effects of each
species on the environment and also because of runoff-dilution effects;
242
-------
Figure 1. Sampling stations - Grand River
b. It is useful to include various phenomena in models such as seasonal
variation, effect of antecedent conditions, and, physical effects of
rainfall-runoff intensity;
c. Where possible, the effect of point sources should be minimized; and,
d. A primary aim should be towards the formulation of models that will
predict the shape of the phosphorus profile.
VARIABLE DELINEATION
The only variable which is measured frequently, and which may be readily
simulated in most watersheds, is flow. Consequently, this study has focussed
primarily on parameters which may be derived from the flow records. Four
major groups of variables have been derived from the flow records. They
include:
243
-------
TABLE 1. DETAILS OF VARIOUS GAUGING SITES
Site
No.
1
2
3
4
Name of
Sice
Canagagigue Creek
-near Floradale
Nith River
-near Canning
Speed River
-near Guelph
Grand River
-near West Montrose
Code
No.
AG-4
GR-20
UL-3
UL-21
Location
Recording gauge
02 GA 036
Recording gauge
02 GA 010
Recording gauge
02 GA 015
Recording gauge
02 GA 034
Drainage
Area
(Miles2)
6.9
398
229
451
Length of
Data Set
(No. of
Samples)
After Red.
256
183
217
232
Length of
Data Set
(Period)
Feb. 19, 1975
-Dec. 13, 1977
Feb. 25, 1975
-Dec. 19, 1977
Feb. 4, 1975
-Dec. 15, 1977
Feb. 19, 19 75
-Dec. 23, 1977
Average
Flow
(cfs)
8.5
447
213
483
1. Flow;
2.
3. State variables; and,
4. Time.
Each of these four groups are described briefly here:
1. Flow: The instantaneous flow at the time of measurement(Q.) was included
in order to determine the direct dependence between flow and concentration.
The square of variable Qi (O^SQ) and natural logarithm of Q^LNO^) were
included to detect possible curvature which might exist in this
relationship. The possibility of rising phosphorus concentrations
lag the rising flow levels was tested using Ql (flow lagged by one day)
and the effect of normalizing the flow variable was considered using a
variable VARQ defined as
n -2
VARQ - (Q± - Q) * Z (Qi - Q) /n
where n is sample size and Q is the average flow over the n samples.
2. Antecedent Flow: Under this category, the following three variables
were included to fulfill the above-given functions:
AQ = Antecedent flow
J-3
= E (ADF)./3 where
244
-------
(ADF). is the average daily flow recorded j days prior to the
measurement.
AQSQ = square of variable AQ and
LNAQ = natural logarithm of the variable AQ.
3. State Variables: These variables were included to take into
consideration certain aspects of flow regime not readily quantifiable in
numeric terms. The state variable group consisted of two dummy variable:
and a single numeric value as given below:
Dl - dummy variable, normally given a value of zero, but set to one
during periods of rapidly rising flow (Dl = 1 if Qi+1 > 1.15 Q^
D2 - dummy variable, normally given a value of zero but set to one
during periods of rapidly declining flow.
QAQ - a numeric value defined as quotient of the variable Q and AQ
(Q/AQ).
The dummy variables are included so that some differentiation may be
made between similar magnitudes of flow, when recorded on opposite limbs
of the same hydrograph. The variable QAQ (not strictly a state
variable) is included to consider the sharpness of the rise and fall of
the flow regime at specific points of time.
4. Time: This was included to give some indication of the temporal effects
on phosphorus concentration.
No effort was made to create dummy variables to indicate season as there
was sufficient data to run individual seasonal regressions. Variables
were created, however, in an effort to ascertain the most important
portion of each season. These variables, designated as DAY1, DAY2 and
DAYS are represented schematically in Figure 2 and represent various
linear combinations of the days within each season.
In addition, two temporal variables were defined which were not bounded
by seasons, these being
T - time in days from the last recorded major peak flow; and,
DAY - consecutive day number set to zero for January 1, 1975.
The variable T was included in order to determine the degree of
dependence existing between phosphorus concentration and the accumulation
period This concept is similar to one used in various other simulation
models'(SWMM, STORM). The variable DAY was included to consider for any
trend in the level of concentration over the full study period.
The sixteen variables defined above comprised the complete set of
independent variables employed.
5. Dependent variables include:
TP- total phosphorus concentration which includes orthophosphates,
condensed phosphates and originally bounded phosphorus;
FRP: filtered reactive phosphate (phosphorus which passes through
1-2 ym filters); and,
TDP: total dissolved phosphorus (phosphorus which passes through a
0.45 urn filter).
245
-------
901-
DAY I 45
0
MARCH APRIL
MAY
DAY 2
DAYS
Figure 2. Definition of variables Dayl, Day2, Day3.
Also utilized were the natural logarithms of TP, FRP and TDP. The
natural logarithm was represented by LN. The units of concentrations
were mg/2. and flow was measured in cubic foot per second.
SEASONAL BREAKDOWN
Regression analyses were performed in each case on the full data set.
Six seasonal subsets as listed below were used:
1. R - Overall regression on data.
2. NSR - Non spring delineation of data, observations from March,
April, and May of each year deleted.
3. WR - Winter regression, only data from December, January, and
February included.
4. SR - Spring regression, only data from March, April, May included.
5. SUR - Summer regression, only data from June, July and August
included.
6. FR - Fall regression, only data from September, October, and
November included.
246
-------
Numerals appended to the regression type indicate the sequence of the
regressions (i.e. WR4 would designate the fourth regression completed, using
the winter data set).
REGRESSION ANALYSIS
Regression analyses were performed using a computerized format developed
for this study. The main control program, 'REGMASTER' utilized a number of
sub-programs created to provide an additional data manipulation to ensure
proper input to library regression routines (10). A flow diagram showing the
procedure used in REGMASTER is shown in Figure 3.
Formation of the matrix of
dependent variables: TP;
FRP; TDP; LHTP; LNFRP; LNTDP
Selection of dependent
variables to be used in
present run
Formation of a vector of
required independent
variables (Seasonally-based)
Formation of the
Regression matrix
this*
Y /the first Reg?_
^for this seasonal
^yp?x
Set up revised vector
of independent
variables
Stepwise Regression] | Standard Regression |
| Calculate Residuals|
three
regressions been
comleted
"Have all
^seasonal options
been satisfied
Designate next
seasonal set of
data
Figure 3. Flow diagram for REGMASTER
247
-------
Preliminary regression analysis on all watersheds showed a consistent
lack of significance in a number of variables outlined earlier. For this
reason, all time variables, the flow variables Ql and VARQ, and the dummy
variable D2 were excluded from the standardized regression analysis.
Regressions were run for each seasonal subset in each watershed using the
remaining independent variables (Q, QSQ, LNQ, AQ, AQSQ, LNAQ, Dl, QAQ) and the
dependent variables (TD, FRP, TOP). Where some of these variables proved to
be insignificant, they were eliminated and the regression was re-run.
Insignificance of a variable was defined by a 't' statistic less than
2
t(0.995,n) and a contribution to the total R of less than 0.015. In some
cases, important combinations of variables were missed by the computerized
regression format, which was limited to a maximum of three runs in any
particular season. Hence, careful attention was paid to the results of the
computerized regression and auxiliary regressions were performed whenever a
promising group of variables appeared to have been omitted. Following
completion of the regression Analysis, plots of predictions and residuals
were completed. Visual analyses were performed to end the analysis (5),(10).
RESULTS AND DISCUSSIONS
1. Regressions for the Site AG-4
Six regressions (one for each seasonal breakdown) were selected from
completed regressions. The details of pertinent statistics and various
alternatives studied (like outliers, etc.) are discussed elsewhere (5). The
selected regressions (based on best statistics) for total phosphorus, filtered
reactive phosphorus and total dissolved phosphorus are given in Tables 2(a),
2(b) and 2(c) respectively. Table 2(b) and 2(c) also give the statistics.
TABLE 2(a). SELECTED REGRESSIONS -TP
REGRESSION TYPE
EQUATION
Overall (R2)
Non Spring (NSR3)
Winter (WR5)
Spring (SR4)
Simmer (SUR4)
Fall (PR6)
Y 7.30xlO~2 + 3.21xlO"3 AQ - 3.51xlO"2LNAQ-1.22xlO"5AQSQ
(t - 4.8) t - -7.2) (t - -3.5)
4 7.37xlO~2 Dl + 2.24 x 10~3Q
(t - 4.7)
Y .0825 + .0075 AQ
(t - 7.3)
(t 12.6)
Y " .0816 + .0030 Q - .0177 LNAQ
(t - 6.8) (t - 3.6)
Y - .04105 + .00245 Q + .07140 Dl
(t - 14.2) (t - 2.8)
Y - .10717 + .02389 AQSQ + .01231 QAQ - .08234 AQ
(t - 47) (t - 2.8) (t - -3.9)
Y - .10097 -I- .0072Q - .0547 LNAQ
(t - 4.6) (t - 4.6)
243
-------
TABLE 2(b). SELECTED REGRESSIONS - FRP
Regression
Type
Overall
(R4)
ton Spring
(NSR3)
Winter
(WR5)
Spring
(SR5)
Summer
(SUR2)
Fall
(FR3)
Equation
Y- .02401 + .000537Q + .03781D1
(t - 7.3) (t " 4.4)
Y- .03418 + .02084LNQ - .01674 LNAQ
(t - 6.5) (t - 5.1)
Y- . 04016- . 01310LNAQ+. 00256Q-2 . 01xlO~5QSQ
(f -4.5) t«4.7) (t - 3.3)
Y- .01673 + .00056Q + .04734 Dl
(t- 6.1) (t« 3.5)
Y" .0032 - .00761 LNAQ + .00347 AQSQ
(t- -4.0) (t - 6.3)
Y- .04133 + .02848LNQ - .02822 LNAQ
(t- 3.6) (t - -3.5)
SSR
.181
.048
.013
.167
.0048
.038
ssu
.445
.148
.026
.261
.0031
.090
+/-
Ratio
.53
.75
.43
.41
.75
.52
F
51.5
21.2
10.5
37.7
19.6
7.4
R2
.289
.243
.327
.390
.611
.298
TABLE 2(c). SELECTED REGRESSIONS - TDP
Regression
Type
Overall
-------
In general, the regressions produced for AG-4 predict the measured
concentrations adequately. Peaks and valleys associated with extremes in
concentration are predicted well, especially during periods of high flow.
During the periods of near constant low concentrations, the shape of the
measured profile is, in general, well simulated by the regressions, but the
magnitude of the prediction usually exceeds that measured. As the lower
concentrations are the most frequent, there is therefore, a tendency to
overpredict on a day-to-day basis. It should be noted that no lag effect was
noticeable on a consistent basis when comparing simulated profiles to the
measured profiles. It is felt that absence of this may be due to low
magnitude of flow associated with this watershed.
While the regressions created for the FRP and TDP forms were in all cases
significant, the statistics associated with each equation were not impressive.
To compare the developed models with models consisting of sample mean, the
models were constructed for each species based on the sample mean and a 99%
confidence interval. This hypothesis concluded that the models developed
earlier were much superior to the mean type of models.
2. Regressions for the Site GR-20
The regression analyses for the Nith River (GR-20) followed a format
similar to that used in the study of the Canagagigue Creek watershed (AG-4).
Those regressions which proved to be the best of each seasonal type are shown in
Tables 3(a), 3(b) and 3(c) for the three types of species of phosphorus. The
detailed statistics are given elsewhere (5).
TABLE 3(a). SELECTED REGRESSIONS - TP
REGRESSION TYPE
EQUATION
Overall (R2)
Non Spring (NSR3)
Winter (WR2)
Spring (SR2)
Summer (SUR3)
Fall (FR4)
.0327 + .000105Q
(t = 27.4)
Y =
.00105 + .00021Q - 3.49x10 QSQ
(t=12.1) (t= -7.2)
-.32394 + .07347 LNQ
(t=15.2)
.0330 + .15996 Dl x 8.773xlO"5Q
(t=3.0) (t=10.2)
.02991 -I- 1.036 x 10~6QSQ
(t = 6.0)
.00536 + .00023 Q - .06535 Dl
(t=14.3) (t= -4.4)
250
-------
TABLE 3(b). SELECTED REGRESSIONS - FRP
REGRESSION TYPE
Overall (R2)
Non Spring
(NSR2)
Winter (WR2)
Spring (SR6)
*Sunnner (SUR4)
Fall (FR5)
EQUATION
Y - -.06606
y = -.084 -
+ 4.06
(t -
Y = -.0912
Y = -.12537
Y = -.1654
Y = -.01801
- 2.3875 x 10~9
(t = 05.3)
3.48 x 10"3 QAQ
(t = -5.1)
x 10"5 Q
8.0)
+ 2.09 x 10~2LNQ
(t = 4.3)
+ .02632 LNQ
(t = 9.3)
- .0153D1 - .0084
(t = -1.8) (t-
QSQ + 2.7296 x 10~5
(t = 4.7)
- 1.23 x 10~2 Dl + 1
(t = -2.7)
- 3.91 x 10~3 QAQ 4-
(t = -6.0)
QAQ + .0376 LNQ
-1.32) (t = 2.62)
- 3.469 x 10"2 Dl + 1.467 x 10~4Q -
(t- -5.6) (t = 7.2)
0+ 1.3989xlO~2LNQ
(t =3.7)
.81 x 10~2 LNQ
(t = 5.3)
3.39 x 10~5 Q
(t = 6.4)
4.572 x 10~8QSQ
(t = -3.2)
*Regression is not significant, nor are any of the independent variables. Removal
of any variable results in a reduction of the significance of the other variables.
TABLE 3(c). SELECTED REGRESSIONS - TDP
REGRESSION TYPE
EQUATION
Overall (R2)
Non Spring(NSR2)
Winter (WR3)
Spring (SR3)
Summer (SUR4)
Fall (FR2)
Y - -.16327 + 3.5757 x 10~2LNQ - 6.6067 x 10~10 QSQ
(t - 14.7) (t - -3.8)
Y - -.2004 - 3.563 x 10~3QAQ + 4.424 x 10~2 LNQ
(t - -3.8)
(t = 11.5)
- 1.834 x 10~2 Dl + 7.198 x 10" QSQ
(t - -2.9)
(t - 5.3)
Y - -.2898 + .0636 LNQ - .0051,QAQ
(t = 8.2) (t. - -3.1)
Y - -.1065 + .0251 LNQ
(t - 7.0)
,-7
Y - .00328 +5.0399 x 10 QSQ,
(t - 7.8)
Y - .4020 + 6.904 x 10~4Q - 4.19 x 10~2D1 - 2.36 x 10"7 QSQ
(t - 6.9) (t - -5.7) (t - -5.7)
- 2.69 x 10~2 QAQ - 7.95 x 10"2LNQ - 1.35 x 10"2 LNAQ
(t - -4.5)
(t -4.1)
(t = -3.4)
251
-------
The statistics showed very encouraging results with the exception of the
summer period as all values of F exceeded 100 and R values exceeded 0.7. The
seasonally aggregated model produced the best residual statistics. The
statistics achieved for the filtered reactive phosphorus were, in general,
inferior to those calculated for the total phosphorus form. Still the model
R2 is the best suited for this form of species. The results achieved for
total dissolved phosphorus were found to be better than FRP species.
3. Regressions for the Site UL-3
Significant regressions were obtained for all model types with the
exception of the fall model, in the case of total phosphorus. Table 4
gives the 'best' regressions for the TP species.
TABLE 4. SELECTED REGRESSIONS - TP
REGRESSION TYPE
Overall (R2)
Non Spring (NSR2)
Winter (WR4)
Spring (SR3)
Summer (SUR4)
Fall (FR3)
EQUATION
Y - -.2661 + 2.756 x 10 QSQ + .2436 QAQ - 6.348 x 10"4 Q
(t - 18.6) (t - 13.5) (t = -11.3)
4 -7
+ 5.333 x 10 AQ - 2.028 x 10 AQSQ -I- .1540 LNAQ - .1415 LNQ
(t " 6.0) (t - -5.4) (5 - 5.3) (t - 4.9)
Y -.1265 + .1334 QAQ 4 .1200 LNAQ - .1162 LNQ
(t " 10.5) (t - 7.0) (t - 6.8)
Y - -.0467 + .0625 QAQ + 9.6 x 10"5 AQ
(t - 7.6) (t - 4.4)
Y - -.1785 + 3.378 x 10~7 QSQ - 9.146 x 10"4 Q + .2198 QAQ
(t - 17.8) (t - -11.6) (t - 12.4)
+ 8.552 x 10~4 QSQ - 2.840 x 10"7 AQSQ
(t - 8.5)
(t - -6.7)
.0274 + 8.753 x 10"7 QSQ - 6.510 x 10~7 AQSQ
(t 12.2)
Y - .00658 + .01618 QAQ
(t 1.9)**
(t - -3.4)
"Regression is not significant, nor is the independent variable
The statistics achieved, varied greatly from one regression to the next.
In general, the statistics appeared to be excellent for the overall and the
spring regressions. The winter, non-spring and summer regressions showed only
adequate statistics by comparison. In the cases of FRP and TDP, due to
extreme skew in the data, it is felt that use of the overall regression
analysis proved to be futile (5), and regression would produce misleading
results. It is recommended, therefore, that where necessary, prediction of
FRP and TDP within this watershed should be accomplished through the use of
their respective mean concentrations.
252
-------
4. Regressions for the Site UL-21
It is noteworthy that this station was the only one selected on the main
stem of the Grand River. While it drains predominantly agricultural land it
lies downstream of the sewage treatment plant at Fergus and Elora. In
addition, the Grand River is controlled above the sampling site at the Shand
Dam. This site was chosen in order to ascertain difficulties which might
arise in areas not dominated by agricultural landuse.
Significant regressions were achieved for all seasonal types with the
exception of winter data subset. Table 5 gives the best regression models for
TP. In general, for other forms of phosphorous, the regressions produced in
this watershed were fairly weak and highly oriented towards relatively small
number of major events, which may be due to the controlling Shand Dam.
Efforts were made to discern the effect of reservoir by using differencing
techniques on data collected at sampling sites, just below the Shand Dam.
Unfortunately, no firm conclusions were possible due to inconsistency in the
sampling dates between the two sites. It is felt that the 'best' possible
model available would be based simply on a series -of seasonal means. In
general, it is felt that any attempt to predict these species within this
watershed, based upon information presently available, would lead to
misleading and inaccurate results.
TABLE 5. SELECTED REGRESSIONS - TP
REGRESSION TYPE
Overall (R4)
Non Spring (NSR2)
Winter** (WR4)
pring (SR2)
Summer (SUR3)
Fall (FR3)
EQUATION
Y .44167 + 2.303 x 10~4 Q + 5.865 x 102QAQ - 9.045 x 10~2LNQ
(t - 5.3) (t - 6.3) (t - -4.7)
o _o
- 1.944 x 10 QSQ - 7.623 x 10
(t " -4.1) (t -3.4)
AQSQ
Y - .1484 - .1965 LNQ + .1008 QAQ + .1619 LNAQ + .0817 Dl
(t - -3.6) (t « 4.4) (t 3.0) (t 2.9)
+ 3.465 x 10"
(t - 7.7)
QSQ
Y " .5369 + .00027 Q - .1134 LNQ + .0136 LNAQ
(t - 1.3) (t -.7) (t - .1)
Y " .0027 + .0064 QAQ + 2.187 x 10" . Q
(t - 7.6) (t - 4.0)
(t - -4.2)
,-4
Y - .3798 + .1718 Dl - .0694 LNQ + 2.45 x 10 " Q
(t 5.0) (t " -2.8) (t " 9.7)
Y " -1.589 + 7.008 x 10"7 QSQ - 1.343 x 10"3 Q + .3045 LNAQ
(t " 4.8) '- ~ '" '" ~ ' ""
(t - 3.7)
- 8.798 x 10"8 AQSQ + .2595 QAQ
(t - -3.3)
(t - 5.9)
**Regres8ion Not Significant; No Variables Significant.
253
-------
5. Comparison of Selected Regression Analyses
The models selected for the prediction of TP in each of the four Grand
River watersheds are shown in Table 6. Associated statistics are shown in
Table 7. No general form may be associated from the models selected, although
the basic flow variable (Q) appears to be of predominant importance in most
of the regressions. The state variable QAQ and the logarithmic transform
of the basic antecedent flow variable (LNAQ) also appear frequently in the
selected models. It does unquestionably show that the models developed cannot
be transformed between watersheds within the Grand River system without some
sort of normalization procedure (5).
TABLE 6. SELECTED MODELS - GRAND RIVER BASIN (TOTAL PHOSPHORUS)
Watershed
AG-4
*AG-4
GR-20
UL-3
UL-21
Applicable
Season
Spring
Summer
Fall
Winter
All
All
All
All
Selected Models
[TP] = .03618 + .0028 Q
ITP] = .10717 + .0239 AQSQ + .0123 QAQ - .0823 AQ
ITP] = .08905 + .00716 Q - .04971 LNAQ
[TP] = .08164 + .00296 Q - .01766 LNAQ
ITP] = .08077 - .02073 LNAQ + .00317 Q
[TP] = .00327 + .000105 Q
[TP] --.18378 + .17807 QAQ - .15947 LNQ + .16631 LNAQ
[TP] =-.01747 + .07155 QAQ + 2.83 x 10~5 Q
*Not intially selected, included here due to General Form Prevalent in Other
Watersheds
N.B. Variables are listed in order of significance
TABLE 7. MODEL STATISTICS - GRAND RIVER BASIN (TOTAL PHOSPHORUS)
Watershed
AG-4
*AG-4
GR-20
UL-3
UL-21
Season
Spring
Summer
Fall
Winter
All
All
All
All
N SSR SSU R2 F +/-
Ratio
118 2.02 0.43 .822 535 .63
28 0.068 0.043 .610 12.5 1.00
37 0.19 0.14 .575 23 .54
69 0.095 0.134 .415 23 .60
252 2.31 0.86 .729 334 .60
183 5.81 1.40 .806 752 .43
166 0.19 0.11 .630 92 .59
231 2.02 1.55 .565 149 .65
Residual Statistics
% of Max Min
AMR Mean Res. Res.
.043 38 .284 -.127
.031 30 .110 -.055
.049 46 .190 -.089
.027 32 .271 -.049
.041 40 .263 -.144
.043 39 .704 -.340
.015 38 .133 -.087
.042 46 .620 -.219
254
-------
CONCLUSIONS
This work was initially aimed at extending the ability to predict
phosphorus concentration for three different sub-species - Total Phosphorus
(TP), Filtered Reactive Phosphorus (FRP), and Total Dissolved Phosphorus
(TDP). Useful regression models were produced on all the Grand River water-
sheds studied, for the TP form. Attempts to form useful models for the FRP
and TDP forms met with variable success however, and it must be concluded
that, in general, the simulation of these species was not possible using the
available techniques. All conclusions, therefore, are directed towards the
TP specie.
In general, the models relied heavily on the standard flow variable (Q)
and flow curvature forms (QSQ and LNQ). In seasonally-based regressions the
variable QAQ and LNAQ proved useful. The presence of QAQ in the model
indicates the importance of hydrograph shape while the presence of LNAQ
indicates a relationship between concentration and antecedent conditions. All
models showed a tendency to overpredict low to median concentrations (to a
minor extent) and underpredict higher concentrations (to a greater extent).
This may be due to skew, evident in most of the data sets and to the
mathematical techniques used in multiple regression analysis. In most of the
cases, seasonally-based models proved to be better than non-seasonal models.
It is noteworthy that the predicted shape of the phosphorus profile
corresponded well in all cases to the measured profile. The models therefore
show a good potential with respect to predictive ability. Further study may
show the identification of parameters which will help to reduce the error
present in the prediction of extreme values.
ACKNOWLEDGEMENT
Dr. K.K.S. Bhatia was supported by a UNESCO Fellowship under the Project
UNDP/IND/74/045 and is grateful to Dr. Satish Chandra, Director, National
Institute of Hydrology, Roorkee, India, for constant encouragement.
REFERENCES
1. Bowen, D.H.M. The great phosphorus controversy. Environmental Science
and Technology. Vol. 4. 1970. pp. 725-726.
2. Cahill, T.H., Imperato, P., and Verhoff, F.H. Evaluation of phosphorus
dynamics in a watershed. American Society of Civil Engineers, Journal
of Environmental Engineering Division, 100, EE2, April 1974. pp. 439-458.
255
-------
3. McComas,M. R, et al. A comparison of phosphorus and water contributions
by rainfall and snowmelt in Northeast Ohio. Water Resources Bulletin
12:3, 1976. pp. 519-527.
4. Gearheart, R.A. A eutrophication model of White River Basin above Beaver
Reservoir in Northwest Arkansas. University of Arkansas Water Resources
Research Centre Project No. B-012-ARK, Publication No. 15, August 1973.
5. Andrews, D.J., Bhatia, K.K.S., and McBean, E.A. Regression modeling of
diffuse phosphorus loadings. In preparation for submission to American
Society of Civil Engineers. Journal of Environmental Engineering
Division.
6. Ministry of the Environment, Canada. Water quality data pollution from
land use activities reference group (PLURAG) study 1975, 1976 and 1977.
Water Quality Data Series. Vol. XIII. 1977. p. 500.
7. Bouldin,D.R. et al. Transport in streams. In; Keith Porter (ed.),
Nitrogen and Phosphorus: Food Production, Waste and the Environment.
Ann Arbor, Michigan, 1975.
8. McBean, E.A. and Gorrie, J.E. Non point source contributions to water
quality problems. In; Proceedings of the 10th Canadian Symposium. 1975:
Water Pollution Research Canada, 1975. pp. 142-150.
9. Haith, D.A. Land use and water quality in New York rivers. American
Society of Civil Engineers. Journal of Environmental Engineering
Division, 102, EE1, February 1976. pp. 1-15.
10. Andrews, D.J. Modelling of phosphorus concentration from diffuse
sources. M.A.Sc. Thesis, University of Waterloo, 1979. p. 375.
256
-------
A MODEL FOR ASSESSING THE COST-EFFECTIVENESS OF AGRICULTURAL
BMP IMPLEMENTATION PROGRAMS ON TWO FLORIDA BASINS1/
by: C.D. Heatwole, Graduate Assistant
A.B. Bottcher, Associate Professor
L.B. Baldwin, Associate Professor
Agricultural Engineering Dept. - Univ. of Florida
ABSTRACT
A model was developed to evaluate the cost-effectiveness of alternative
BMP implementation schemes on two agricultural basins in Florida. The model
selectively applies the desired BMPs throughout the basin,
associated costs, and predicts the water quality improvement
nitrogen and phosphorus). Fifteen BMP scenarios were
prioritizing BMPs for implementation in these basins. Applying the maximum
level of BMPs is estimated to cost around $1.2 million (annually), while the
four most cost-effective BMPs would cost only one quarter as much, yet provide
approximately 90% of the water quality improvement.
estimates the
(reductions in
evaluated to aid in
INTRODUCTION
The Lower Kissimmee River, LKR, basin (2010 sq km) and the Taylor Creek-
Nubbin Slough, TCNS, basin (498 sq km) lie in the "flatwoods". region of
Florida, an area characterized by very flat sandy soils and a high water table
that fluctuates from the surface to 2 meters deep. Ranching and dairying are
the primary land uses of these basins. In 1980 the Taylor Creek watershed was
69% improved pasture, 16% forest and range, 3% citrus, 4% urban, and 8% mis-
cellaneous (2). The Nubbin Slough watershed has similar land use, while the
larger Lower Kissimmee River basin is less developed (more unimproved pasture,
forest and range instead of improved pasture).
These basins lie on the north side of Lake Okeechobee and contribute
about 35% of the water flowing into the lake. However, they also discharge a
1
/ This research was supported by the U.S. Army Corps of Engineers,
Jacksonville, Fla.
257
-------
disproportionate amount of nutrients into the lake, accounting for 49% of the
phosphorus, and 31% of the nitrogen annually added to the lake waters (7).
This 31% contribution of nitrogen, though lower than the water contribution,
is still proportionally higher than loads coming from neighboring sandy
watersheds north of the Lake. The organic soils south of the Lake. The
organic soils south of the Lake contribute disproportionality high amounts of
nitrogent making the number only appear low. Reduction of these nutrient
loads is being tackled at the upland source areas through the use of
appropriate best management practices (BMPs). A list of BMPs applicable to
Florida conditions has been compiled (4), but quantifying the effects of BMPs
is a much more difficult task.
Detailed hydrologic-water quality models are the primary means of evalua-
ting BMP effectiveness (3). However, these models generally do not directly
consider the economics of the different BMP alternatives. The goal of this
project has been to develop a model for the LKR and TCNS basins whereby the
cost-effectiveness of different BMP implementation scenarios can be evaluated
on a basin scale.
MODELING APPROACH
The U.S. Army Corps of Engineers has developed an extensive geographic
database of the LKR and TCNS basins. The two basins were divided (200 m by
250 m grid) into 50,165 five hectare (12.35 acre) cells. Data for each cell
includes basin and sub-basin codes, soil group, hydro!ogic soil group
(A/B/C/D), presence of a stream in the cell, elevation (for flood plain
cells), and land use for the years 1980, 2010, and 2030. Of this, hydrologic
soil group, land use, and stream location data were used in this study. This
'cell1 was the basic land unit used by the model for applying BMPs and in
water quality modeling.
The objective of the model is to provide a tool for evaluating: different
BMPs, different levels of application of a given BMP (e.g. 1 cm impoundment
versus 2 cm impoundment), and various combinations of BMPs. The model has two
major components. The first section of the model takes the desired BMP sce-
nario, applies the BMPs to each applicable cell in the basin, and summarizes
the costs of the 'applied' BMPs. The second component is a water quality
model that predicts the average annual nutrient loading of nitrogen (N) and
phosphorus (P) from the basin. This is compared against a base-line nutrient
loading (predicted by the model for the 'no BMP' condition) to give the ex-
pected water quality improvement in kilograms of N and P reduced per year due
to the BMPs. Model outputs for each BMP scheme considered are: water quality
improvement, BMP costs, and cost effectiveness in terms of improvement per
dollar cost.
BMP SELECTION AND COSTS
BMPs for the LKR and TCNS basins must be oriented to: keeping livestock
away from drainageways, dispersing wastes for soil assimilation and plant
uptake, proper fertilization and water management, and impounding runoff for
nutrient attenuation. Specific BMPs used in this study were: fencing streams
and wetlands that border pasture, constructing runoff detention basins, and
impounding runoff from dairy barn lots for application to pasture and crop
land.
258
-------
BMPs were assigned by considering the land use, cattle density (for
pasture land), hydrologic soil group, and the distance of the cell from a
stream or wetland. Cells neighboring a stream cell would be expected to
deliver greater amounts of N and P to streams than those not bordering the
stream. Thus BMP application levels were decreased with increasing distance
of a cell from a stream or wetland. For fencing of pasture from wetlands
decreasing amounts of fencing were applied as the distance of the adjoininq
wetland from a stream cell increased.
For cells identified as 'dairy1, a method for estimating cattle density
was needed. The number of milking cows at each dairy was known, but addi-
tional breakdown of the 'dairy1 land into pasture or hayland was not avail-
able. Dairy cells were divided into three groups of decreasing cattle densi-
ty ('cow pasture', 'dry cow pasture', and 'hayland1), based on distance from
the barn cell. Cattle density was then computed as a function of the assigned
dairy land use group and the size of the milking herd.
The cost of installing the BMPs was obtained by multiplying unit costs by
the total amount of each BMP applied. Unit costs were 1984 estimates obtained
from contractors and the USDA-SCS. Annual maintenance costs were also esti-
mated. These costs were amortized over the expected life of the BMP to obtain
an 'annual' cost. This annual cost estimate was used in the cost-effective-
ness calculations.
WATER QUALITY MODEL
The water quality model is a combination of two sub-models. The first
sub-model is a modified version of the CREAMS model (CREAMS-WT) which predicts
nutrient and water yield from each individual cell. These results are then
passed to the second sub-model, BASIN (developed by the authors), which inte-
grates the cell results over the entire basin. The CREAMS model, developed by
USDA-ARS (10), predicts runoff, erosion, and nutrient yield (nitrogen and
phosphorus) for field-sized areas. A description of the modifications made
for the CREAMS-WT sub-model will follow. The BASIN sub-model takes the nutri-
ent predictions for each cell and predicts nutrient delivery to the edge of
the nearest stream and at selected sub-watershed outlets in the basins. This
modeling approach utilized all the information available at the cell level and
enabled prediction for the entire basin.
THE CELL SUB-MODEL: CREAMS-WT
Objectives in the design of CREAMS (Chemicals, Runoff and Erosion from
Agricultural Management Systems) were that the model should: 1) simulate major
physical processes that control water balance, erosion, sediment yield, and
movement of plant nutrients and pesticides, 2) use physically based parameters
that can reflect changes in management systems, 3) be computationally effi-
cientoperate on a daily time step, and 4) be field scale, since this is the
common base for BMP selection (6). Because the parameters of the model are
physically based, they can be estimated from site visits, maps, county soil
survey reports, and the CREAMS manual. An additional advantage of physically
based parameters is that the need for calibration is minimized.
259
-------
The CREAMS-WT version of CREAMS was developed by Heatwole et al. (9) for
the South Florida flatwoods. CREAMS-WT has the added ability to follow the
fluctuating water table which strongly influences the hydrologic processes of
this area. This improves the model conceptually, and yields better estimates
of annual runoff and evapotranspiration.
To assure that the simulated results represented a long term average,
simulations were run using 20 years of actual weather data. The first two
years of the simulation were used to assure stable initial conditions, with
the final 18 years used to obtain an average annual N and P load for each
cell.
The simulation of BMPs with CREAMS-WT involves changing the model para-
meters (such as curve number and soil properties) to reflect changes the new
practice would cause. One of the most important BMPs in this study, fencing,
is not reflected directly in any of the CREAMS-WT parameters. Fencing was
modeled by distributing animal waste (which CREAMS-WT considers as a 'ferti-
lizer' application) between the pasture and the wetland or stream which it
borders. The fraction of waste distributed to each depends on the degree of
fencing.
THE BASIN SUB-MODEL: BASIN
The functions of the BASIN sub-model are: 1) Model the effects of BMP's
(detention and retention basins) that cannot be simulated in CREAMS-WT, 2)
Provide background loading for forests, native range, and other non-agricul-
tural land use, 3) Attenuate the nutrient loads from cells to compute edge-of-
stream loadings, 4) Compute nutrient loads due to cattle in wetlands (CREAMS-
WT does not handle wetlands), and then attenuate these loads based on flow
distance through the wetland to a stream, and 5) Attenuate nutrient loads
from edge of stream to sub-watershed outlets.
The most important parameters for this model are the attenuation factors
needed for the functions mentioned above. These were obtained by summarizing
the results of many nutrient studies in controlled marshes and natural wet-
lands in South Florida (5).
MODEL CALIBRATION
In the water quality model, CREAMS-WT and BASIN are both designed to use
physically based parameters thus minimizing the need for calibration. Very
little calibration was actually needed, as comparison of model output with
available observed data showed good correspondence for the initial parameter
values chosen. The CREAMS-WT hydrology section was the primary component of
the water quality model that was calibrated. Parameters were adjusted so that
predicted average annual evapotranspiration and runoff would match the average
annual values determined from water balance studies (1,12). Other non-cali-
brated parameters for the CREAMS-WT and BASIN sub-models were estimated from
the previously mentioned sources and from personal observations of the area.
260
-------
MODEL VERIFICATION
Verification of the model output was limited by the amount of available
data. Comparison was made at two levels: the cell predictions by CREAMS-WT,
and on a larger scale, the predictions of the BASIN sub-model. On these two
levels, both runoff and nutrient yield were compared with measured data.
Since limited data were used for calibration of runoff prediction on the cell
scale, additional verification could not be done for this aspect of the
model. On a watershed scale, estimates of the long term average annual runoff
of 26.9 to 30 cm (1,12) compared favorably with the the basin-wide 27.8 cm
average predicted by this model.
Nutrient concentrations and loads predicted by CREAMS-WT were checked
against data from three pasture sites (8,11), and the output of the BASIN sub-
model was compared with data from the Taylor Creek watershed (2). In each
case, the predicted concentrations and loads fall within the range of observed
values. Considering the scope and goal of the model and the expected vari-
ability in the measured data, the model predictions were considered accept-
able.
RESULTS AND DISCUSSION
Fifteen different BMP scenarios were evaluated and are listed in Table 1
along with their resulting cost-effectiveness. In addition to the scenarios
in Table 1, a 'do nothing' scenario was also simulated as the 'base line' for
comparative evaluation of the other BMP scenarios. Water quality improvement
was defined as the reduction in the average annual nitrogen and phosphorus
load. Cost-effectiveness was then computed by dividing nutrient load reduc-
tion by the annual cost of the BMP scenario, giving units of kilograms reduced
per dollar per year.
A variety of individual BMPs and combinations of BMPs are included in
Table 1 as applied to both the LKR and TCNS basins. Because of the interac-
tion between BMPs, the results from individual BMP evaluations cannot be added
to estimate the response of combinations of BMPs. To determine the response
of a particular group of BMPs, that combination of BMPs must be evaluated
independently.
Several interesting relationships can be observed in Table 1. First, as
the level of BMP application increases, the cost-effectiveness decreases
(scenarios 1-3 and 13-15). Second, fencing is one of the more cost-effective
practices (and also turns out to be the largest contributor to reducing the
total nutrient load from the basins). Third, pasture impoundment turned out
to be the least cost-effective of the practices evaluated. Forth, cost-
effectiveness values for the TCNS basin are generally higher than the LKR
basin, due in part to the higher intensity of land use in the TCNS basin. The
higher nutrient loads will generally result in a higher percentage reduction
thus higher cost-effectiveness, even though the remaining nutrient load may
still be higher.
Since the streams are the primary pathways of mass transport of nutri-
ents, the proximity of a cell to a stream has a dramatic impact on the actual
261
-------
Table 1. BMP scenarios and their cost-effectiveness for the Taylor Creek-
Nubbtn Slough (TCNS) and Lower Kissimmee River (LKR) basins.
Scenario Description
COST EFFECTIVENESS
LKR TCNS
Nitrogen Phos. Nitrogen Phos.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
All BMPs applied - high rate
" - medium rate
" - low rate
Dairy pasture runoff impoundment
" barn lot impoundment
" pasture fencing
Beef pasture impoundment
All dairy BMPs (barn and pasture)
Fencing of all beef pastures
Beef pasture fencing:
" - Intensively managed pasture
" - Improved pasture
" - Semi -improved pasture
Row crop and citrus impoundment:
" ~ ni9n rate
" - moderate rate
" - low rate
0.42
0.54
1.10
0.06
2.00
1.90
0.01
0.19
1.40
6.90
1.50
0.43
0.39
0.63
1.90
v*y
0.11
0.14
0.26
0.03
0.47
0.51
0.02
0.04
0.33
1.70
0.36
0.11
0.18
0.30
0.86
I/I
0.67
0.75
1.10
0.04
3.60
2.60
0.02
1.40
1.70
6.90
1.90
1.90
0.08
0.13
0.18
0.15
0.17
0.25
0.02
0.24
0.55
0.02
0.33
0.38
1.50
0.41
0.40
0.06
0.10
0.14
nutrient delivery to a stream and therefore its potential nutrient reduc-
tion. Thus high nutrient producing activities, such as dairy barns, will have
much less impact on water quality if located in 'upland' areas away from
streams. The apparent reason for the relative ineffectiveness of pasture
impoundment (scenarios 4 and 7) is that they were not used to treat nutrient
loads in the streams or wetlands. The difference between the basins in row
crop and citrus runoff impoundment effectiveness is likely due to the location
of those areas relative to streams.
Using the cost-effectiveness data, the individual BMPs can be ranked.
Considering both basins together (not a simple average of Table 1 values
because of the differing areas involved), the BMPs in order of decreasing
effectiveness are:
1. Fence dairy cows from streams and wetlands.
2. Fence intensively managed beef pasture and dry cow pasture from
streams and wetlands.
3. Retain runoff from dairy barn holding lots and distribute on low den-
sity pasture or crop land.
4. Fence beef cattle on improved unirrigated pasture from streams and
wetlands.
262
-------
5. Impound (detain for nutrient attenuation) runoff from dairy cow
pastures.
6. Impound runoff from intensively managed beef pasture and dry cow
pasture.
7. Impound runoff from citrus and row crop land.
8. Impound runoff from unirrigated improved beef pasture.
The first four BMPs, if fully implemented could potentially reduce load-
ings at the edge-of-stream by 60% for N and 60% for P in TCNS, and 35% for N
and 40% for P in the LKR basin. Note that these reductions cannot be directly
projected to the basin outlet. The actual percent reduction further down-
stream would be less because of the natural background nutrient concentra-
tions. Implementing the maximum BMP application (scenario 1) in both basins
would have an initial cost of $5.2 million, with an annual cost of $1.2 mil-
lion. In comparison, the four most cost-effective BMPs provide about 90% of
the maximum load reductions found in scenario 1, but at an annual cost of only
$0.3 million and an initial cost of $1.3 million. The benefit of this type of
analysis is apparent.
An interesting observation, and one important to an actual implementation
program, is the decrease in cost-effectiveness with increasing levels of
application of a particular BMP. Even though a BMP may be the most cost-
effective at one level of analysis, it may not be desirable to apply only that
BMP in greater and greater amounts to meet some water quality goal. Rather, a
point may be reached where a different BMP will become more cost-effective
than continued use of the original BMP.
The model's assignment of BMPs to a particular cell is not to be taken as
a field guide for actual implementation of a BMP program. The model only
indicates the criteria for BMP application and an estimate of the total amount
of BMPs applied in the basin. This will, however, be a very helpful guide for
field personnel who ultimately must determine the proper location of the BMPs.
The model described in this article has been demonstrated to be an effec-
tive tool for evaluating various BMP applications in the LKR and TCNS ba-
sins. For a given BMP implementation program, the relative water quality
improvement can be determined, and the cost of installing and maintaining
those BMPs estimated. The cost-effectiveness of this particular program can
then be compared with other BMP options. However, a final decision on a BMP
strategy must depend on more than just cost-effectiveness. A specified level
of improvement in water quality may be required, and also, there will
generally be a limit on the available funding. The model provides an estimate
of these three factors and can thus aid in the selection of the most cost
effective BMP implementation program for these basins, taking into
consideration water quality goals and financial constraints.
263
-------
REFERENCES CITED
1. Allen, L. H., Jr., W. G. Knisel and P. Yates. 1982. Evapotranspiration,
rainfall, and water yield in southern Florida research watersheds. Soil
Crop Sci. Soc. Fla. Proc. 41:127-139.
2. Allen, L. H., Jr., J. M. Ruddell, G. H. Ritter, F. E. Davis and P.
Yates. 1982. Land use effects on Taylor Creek water quality. In E. G.
Kruse, C. R. Burdick and Y. A. Yousef [eds.] Environmentally Sound
Water and Soil Management. Amer. Soc. of Civil Eng., New York, NY. pp.
67-77.
3. Beasley, D. B., L. F. Muggins, and E. J. Monke. 1982. Modeling sediment
yields from agricultural watersheds. J. Soil Water Cons. 37(2):113-
117.
4. Bottcher, A. B., and L. B. Baldwin. 1983. BMP Selector - General
Guide for Selecting Agricultural Water Quality Practices. IFAS
Brochure SP-15. Univ. of Fla., Gainesville, Fla.
5. Bottcher,'A. B., L. B. Baldwin, C. D. Heatwole, and L. W. Miller. 1984.
Cost effectiveness analysis of BMP alternatives for the Taylor Creek-
Nubbin Slough and Lower Kissimmee River Basins. Final Report to the US
Army Corps of Engineers, Jacksonville, FL.
6. DelVecchio, J. R. and W. G. Knisel. 1982. Application of a field-scale
nonpoint pollution model. In E. G. Kruse, C. R. Burdick and Y. A.
Yousef [eds.] Environmentally Sound Water and Soil Management. Amer.
Soc. of Civil Eng., New York, NY. pp. 227-236.
7. Federico, A. C., K. G. Dickson, C. R. Kratzer, and F. E. Davis. 1981.
Lake Okeechobee water quality studies and eutrophication assessment.
South Fla. Water Manag. Dist. Tech. Pub. 81-2. West Palm Beach,
Fla. 90 pp.
8. Goldstein, A. L. 1982. Utilization of a freshwater marsh to treat
rainfall runoff from upland pasturelands. In: Proc. IFAS Conf. on
Nonpoint Pollution Control Technology. Gainesville, Fla. pp. 106-125.
9. Heatwole, C. D., J. Capece, A. B. Bottcher, and K. L. Campbell. 1984.
Modeling the hydrology of flat, high-water-table watersheds. In: Proc.
Hydrology Days, 1984. Front Range Branch, A.G.U. Fort Collins, CO.
pp. 1-12.
10. Knisel, W. G., [ed.] 1980. CREAMS: A Field-Scale Model for Chemicals,
Runoff, and Erosion from Agricultural Management Systems. U.S.D.A.-
S.E.A. Conserv. Res. Rep. No. 26.
264
-------
11. Ritter, G. J., and L. H. Allen, Jr. 1982. Taylor Creek Headwaters
Project Phase I Report - Water Quality. South Fla. Water Manag. D1st.
Tech. Pub. 82-8. West Palm Beach, Fla. 140 pp.
12. Yates, P., L. H. Allen, Jr., W. G. Knlsel, and J. M. Sheridan. 1982.
Channel modification effects on Taylor Creek watershed. In E. G.
Kruse, C. R. Burdick and Y. A. Yousef [eds.] Environmentally Sound
Water and Soil Management. Amer. Soc. of Civil Eng., New York, NY. pp.
78-86. P
265
-------
ATTENDEES
Name
L.B. Baldwin
Tom Barnwell
B.A. Benedict
K.K.S. Bhatia
Del Bottcher
Linfield Brown
Kenneth Campbell
J.C. Capece
M-S Cheng
Ivan B. Chow
B.A. Christensen
Bob Crabtree
L.J. Danek
Robert Dickinson
Forrest Dierberg
Henry Dorzback
Jean Dorzback
Wayne Downs
Ann Dudek
Gary Elwell
Efi Poufoula-Georgiou
W.E.Fulong, Jr.
Richard Bigney
Representing
University of Florida
EPA
University of Florida
Indian National Institute
of Hydrology
University of Florida
Tufts University
University of Florida
University of Florida
MD-National Capital Parks &
Planning Commission
Applied Technology &
Management, Inc.
University of Florida
University of Birmingham, U.K.
University of Florida
University of Florida
Florida Institute of Technology
University of Central Flordia
University of Central Florida
University of Florida
Hazen & Sawyer
HKM Associates
University of Florida
University of Florida
Camp, Dresser and McKee, Inc.
266
-------
Bernard L. Golding
Philip Gronstal
James P. Heaney
John Hendrix
Nancy Hicks
Wayne C. Huber
William James
Kazuaki Kawamoto
Frank Kerino
Vlademir Krejci
Howard Litwack
W-S Lung
Ed MeBean
Sharon Oakes
W.N. Pandorf
Curt Pollman
R.L. Powell
Leila Rhue
Mark Robinson
Ash ok Shahane
Ed Sharp
Charles Swallows
M .P. Timpe
Hilburn Engineering
Dallas Water Utilities
University of Florida
University of Florida
University of Florida
McMaster University
Tokyo Metropolitan
Sewage Works
Swiss Federal Institute of
Water Resources
State of New Jersey, Dept.
of Water Resources
University of Virginia
University of Waterloo
City of Gainesville
Environmental Science
and Engineering
Environmental Science and
Engineering
MD-National Capital Parks &
Planning Commission
CH2M-H111, Inc.
McMaster University
FL. Dept. of Agriculture
City of Winnipeg
Geoscience
Water & Air Research
267
-------
William Tucker
S. Udhiri
A. Visvanatha
William Walker, Jr.
Marty Wanielista
Ray Wiles
John T. Westgate
Ronald Wycof f
Yousef A. Yousef
Environmental Sciences
and Engineering
MD-National Capital Parks &
Planning Commission
Underwood-McLellan Ltd.
Environmental Consultant
University of Central Florida
Geoscience
Seaburn & Robertson Engineering
CH2M-Hill, Inc.
University of Central Florida
S3t-m/10Ml
268
------- |