&EPA
United States
Environmental Protection
Agency
Environmental Research
Laboratory
Athens GA 3061 3
EPA-600/9-82-015
August 1982
Research and Development
Proceedings of
Stormwater and
Water Quality
Management
Modeling Users
Group Meeting,
25-26 March 1982
-------
EPA 600/9-82-015
August 1982
PROCEEDINGS OF STORMWATER AND WATER QUALITY
MANAGEMENT MODELING USERS GROUP MEETING
25-26 MARCH 1982
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
Mention of trade names or commercial products does not constitute
endorsement or recommendation for use by the U.S. Environmental Protection
Agency. Similarly, publication of studies reporting better results from one
model vis-a-vis others does not constitute endorsement.
-------
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 exchange of
information and ideas between researchers and users, and between the users
themselves. The Stormwater and Water Quality Modeling Users Group, sponsored
joir+.ly by the U.S. Environmental Protection Agency and Environment Canada/On-
tario Ministry of the Environment, was established to provide such a forum.
The group has recently widened its interest to include models other than the
Stormwater Management Model, such as the Hydrologic Simulation Program-FORTRAN,
and other aspects of modeling of water quality in urban and natural waters.
This report, a compendium of papers presented at the March 1982 users group
meeting, is published in the interest of disseminating to a wide audience the
work of group members.
David W. Duttweiler
Di rector
Environmental Research Laboratory
Athens, Georgia
iii
-------
ABSTRACT
This report includes 16 papers on topics related to the development and
application of computer-based mathematical models for water quantity and
quality management presented at the semi-annual meeting of the Joint U.S.-
Canadian Stormwater and Water Quality Modeling Users Group held 25~26 March
1982 in Washington, DC.
Topics covered include a study of selection, calibration and verification
of water quality models in Louisiana and an assessment of measurement uncer-
tainty in the estimation of stream reaeration rates for these models. Cali-
bration of hydrology and sediment transport on small agricultural watersheds
using the Hydrological Simulation Program-FORTRAN is described. Hydrologic
modeling for studies of pollutant loads and transport in large river basins
and the use of continuous simulation model calibration techniques to develop
nonpoint pollution loading factors were described. A verification of a
continuous dissolved oxygen model for a river in Missouri wa.s presented.
State-of-the-art data acquisition techniques in hydrometeorology were discussed,
Mathematical analyses of turbulence in center-feed circular sedimentation
basins and for dynamic model calibration are presented.
Preparing storm designs for urban drainage analysis was presented as
well as studies of pollutant concentrations and loadings in highway runoff.
An application of the EXTRAN model to off-site drainage is described. A
desk-top calculator method for nonpoint source loads evaluation is described.
Modifications to the Q.UAL-1 I model for continuous simulation of fecal coli-
forms is presented. A literature review was done of the relationship between
atmospheric pollution and Stormwater quality.
-------
CONTENTS
Page
FOREWORD lii
ABSTRACT iv
A STUDY OF THE SELECTION, CALIBRATION AND VERIFICATION OF MATHEMATICAL
WATER QUALITY MODELS 1
R.C. Whittemore, J.S. Hovis, and J.J. McKeown; National Council
of the Paper Industry for Air and Stream Improvement
AN ASSESSMENT OF THE MEASUREMENT UNCERTAINTY IN THE ESTIMATION OF
STREAM REAERATION RATE COEFFICIENTS USING DIRECT TRACER
TECHNIQUES 36
J.S. Hovis* R.C. Whittemore*, L.C. Brown*-, and J.J. McKeown*;
*Nationa] Council of the Paper Industry for Air and Stream
Improvement, **Tufts University
CALIBRATION OF HYDROLOGY AND SEDIMENT TRANSPORT ON SMALL AGRICULTURAL
WATERSHEDS USING HSPF 5*»
D.E. Schafer*, D.A. Woodruff*, R.J. Hughto*, and G.K. Young**;
*Camp, Dresser & McKee, Inc., **GKY 6 Associates, Inc.
HYDROLOGIC MODELING FOR STUDIES OF POLLUTANT LOADINGS AND TRANSPORT
IN LARGE RIVER BASINS 69
A. Cavacas, J.P. Hartigan, E. Southerland, and J.A. Friedman;
Northern Virginia Planning District Commission
CONTINUOUS DO RESPONSE PREDICTED USING CSPSS IS VERIFIED FOR SPRING-
FIELD, MISSOURI 90
J.E. Scholl and R.L. Wycoff; CH2M Hill
USE OF CONTINUOUS SIMULATION MODEL CALIBRATION TECHNIQUES TO DEVELOP
NONPOINT POLLUTION LOADING FACTORS 101
J.P. Hartigan, T.F. Quasebarth, and E. Southerland; Northern
Virginia Planning District Commission
HYDROMETEQROLOGICAL DATA ACQUISITION: INNOVATIVE, HIGH-RESOLUTION
PROGRAMMABLE INSTRUMENTATION FOR STORMWATER MANAGEMENT 128
W. James, H. Haro, M.A. Robinson, D. Henry, and R. Kitai;
McMaster University
THE SEPARATION OF BOUNDARY LAYER AND FLOW TURBULENCE OF CENTER-FEED
CIRCULAR SEDIMENTATION BASINS 152
T. Yin; National Capital Park and Planning Commission
-------
DYNAMIC MODEL ADJUSTMENT . . . 162
D. Hoang; The City of Portland, Oregon
AN IMPROVED SURCHARGE COMPUTATION IN EXTRAN 179
J.A. Aldrich and L.A. Roesner; Camp Dresser & McKee, Inc.
PREPARING A DESIGN STORM 191
S.A. McKelvie; Gore & Storrie, Ltd.
A PREDICTIVE MODEL FOR HIGHWAY RUNOFF POLLUTNAT CONCENTRATIONS AND
LOADINGS 210
B.W. Mar and R.G. Horner; University of Washington
CHIMNEY HILL OFF-SITE DRAINAGE STUDY 229
J.M. Normann and E.R. Estes III; MMM1 DESIGN GROUP
DESK-TOP METHODOLOGY FOR NONPOINT SOURCE LOAD EVALUATION 248
A.K. Deb; Roy F. Weston, Inc.
CONTINUOUS SIMULATION OF INSTREAM FECAL COL I FORM BACTERIA 260
A.C. Rowney«, and L.A. Roesner**; *Proctor and Redfern Ltd. and
-•"Camp Dresser and McKee, Inc.
ATMOSPHERIC POLLUTION IN RELATION TO STORM WATER QUALITY MODELING:
LITERATURE REVIEW FOR AN INDUSTRIAL CITY
B. Shivalingaiah and W. James; McMaster University
LIST OF ATTENDEES 296
vl
-------
A STUDY OF THE SELECTION CALIBRATION AND
VERIFICATION OF MATHEMATICAL WATER QUALITY MODELS
By: R.C. Whittemore, PhD, J.S. Hovis,
J.J. McKeown, National Council of the
Paper Industry for Air and Stream
Improvement, Inc., Tufts University,
Medford, Massachusetts 02155
I. INTRODUCTION
The general objective of this study was to examine the
differences among various models when applied to a single
watershed in order to illustrate the limitations inherent in
mathematical water quality modeling. As an outgrowth of this
examination, a procedure is presented for selection, calibration
and verification of such models which uses definitive criteria
to judge model validation. Although these criteria were
developed for this particular study, the procedure is believed
to have general application in cases where waste load must be
allocated to achieve water quality standards for some steady
state condition.
The study objectives may be more specifically stated as
(a) testing the validity of four commonly used water quality models
under increasingly strict calibration and verification criteria,
(b) examining predictive capability of these models for conditions
of varying complexity, and (c) elucidating the dominant issues with
respect to model selection, calibration and verification which are
believed to be of concern in most waste load allocation studies.
The work was conducted on the Ouachita River basin in
southern Arkansas and north central Louisiana with special
cooperation and assistance from the pulp and paper industry
centered in the basin. It is important to note that this study
was not designed to provide a verified model of the Ouachita
River, but rather to fulfill the research objectives in calibra-
tion/verification noted earlier.
This paper will provide an overview of this study by out-
lining the field data collection program, a sensitivity study,
the model selection, calibration, and verification criteria, and
finally model validation. Readers are advised that NCASI Stream
Improvement Technical Bulletins to be released during mid-1982
contain more thorough presentation and development of each of
these topics.
-------
II SITE DESCRIPTION
The Ouachita River has its source in the Ouachita Mountains
in Central Arkansas next to the Oklahoma border. It flows
southeasterly into central Louisiana where it joins the Tensas
River to form the Black River. At the Arkansas/Louisiana state
line, the Ouachita River has a drainage area of 10,835 square
miles. At the confluence with the Tensas River, the Ouachita has
a drainage area of 18,864 square miles.
That portion selected for this study included an approximate
100 mile stretch in north central Louisiana. A generalized
basin map of the study area is presented in Figure 1. The river
passes through the several communities whose total population
approaches 200,000.
»••*(*>•
•u (*^
® - (is) trlbut»rl«i
A-H point source load*
ll.j f,»no'»ll«od mil*
point*
FIGURE 1
at.i U«>
•i.o (
lock end dan
— A IO«.»
-------
The hydrology of this segment is controlled by lock and dam
structures which are operated by the U.S. Army Corps of Engineers.
They are used exclusively for navigational purposes. Typical
summer time dimensions of the Ouachita River are noted as follows:
(1) The average depth ranges from 13 feet at m.p. 106 to
nearly 50 feet at m.p. 0.
(2) The average river width is approximately 460 feet with
a range of 300 to 700.
Ill WATER QUALITY MODEL DESCRIPTIONS
Four water quality models were used in this work to simulate
the hydraulic and water quality characteristics of the Ouachita
River Basin. They include DOSAG (1), QUAL1E (2), QL2SMG (3),
and SNSIM (4). Each has been extensively used by consultants,
state, and federal regulatory groups for water quality assessment
and wasteload allocation (5).
Three of these models have been tested and documented ex-
tensively by NCASI. This documentation included a thorough
analysis of the computer code comprising the model and an ex-
tensive testing of model options. This work uncovered a number
of problems and errors in the computer code which were ultimately
corrected and further documented. The documented models are
published in NCASI Technical Bulletins Nos. 327, 331, and 338 (1-
3).
A comparison of the four models is presented in Table 1.
This table includes both the hydraulic and water quality features
available in each model. Readers are advised to review the
referenced NCASI Technical Bulletins for further details concerning
DOSAG, QUAL1E, and QL2SMG and the literature documentation for
SNSIM (4).
TABLE 1
Model Comparison
Solution Longtltudinal
Model IMtbod lydraullc* Diapereion CBOO HMD teat rat ion SCO
Photo-
aynthcaii
Point
Source Oil tribute!
trature lead* Loada
PQffUfi
OOAL1I
QL2MK
Analytical
lavllcit
finite
difference
Implicit
Unit*
differ* no*
Steady fc
State
Steady Tea
•tat*
Steady T*a
Stat*
lat
ord*r
lat
ord*r
let
order
lat
order
lat
order
Advanced
Nutrient
Cycle
optiona Ho
alloMd
optiona lea
allowed
optiona Tea
al loved
*o
•o
Advanced
Nutrient
Alg«*
rat*
correction
only
rate
correction
only
dynule
heat
option
Steady
Stat*
Steady
State
Steady
Stat*
Ho
Tea
Tea
Coupled
to Algae-
MS IK
Analytical Steady
State
lat
order
lat
order
optiona
al loved
Cycle
Optio
Croaa
rate
connection
Steady
State
Yea
-------
IV MODEL CALIBRATION AND VERIFICATION SURVEY DATA
Four surveys were conducted on the lower Ouachita River
Basin for the purpose of collecting data to calibrate and verify
the four water quality models. The survey dates, locations, and
approximate river flows are noted in Table 2. These surveys
were planned following stable hydraulic and water quality con-
ditions in the basin. No significant long term rain events
preceeded any survey.
The location of the sampling and measurement .locations for
these four surveys is further defined in Table 3 along with the
kinds of measurements made at each. Table 4 presents the format
for the spatial survey data collection at each station.
TABLE 2 SPATIAL SURVEY DATES ON
LOWER OUACHITA RIVER BASIN
*
Approximate flow
Date River Locations (CFS)
7/21-22/80 m.p. 106 to 2.9 3,000
8/17/80 m.p. 106 to 0.0 1,200
9/23/80 m.p. 88.8 to 2.9 1,500
12/3-4/80 m.p. 106 to 2.9 17,000
The survey locations were chosen to coincide with a stream
segmentation process established prior to the spatial surveys.
The segmentation process was based upon the location of tributaries,
point source discharges, significant velocity and depth changes
in the stream, and observed changes in water quality. The velocity
and depth data were taken from the Corps of Engineers HEC-II
model output obtained for the anticipated flows, and the remaining
factors were taken from the historical records and visual observations.
Measurements of stream velocity and depth were made through-
out the basin at different flows for the purpose of verifying
the HEC-II forecasts. The observed depth was approximately 20%
lower than the predicted depth for some stream segments. No
statistical differences between observed and forecasted velocity
were observed.
-------
TABLE 3
SUMMARY OP SPATIAL SURVEY SAMPLING LOCATIONS
River Mile/Location 7/21-22/80
Spatial Survey Date
8/17/80 9/23/80 12/3-4/80
106.0
105.3
Load A
100. e
96.8
Tributary 3
92.5
87.8
82.8
77.8
Tributary 6
74.8
73.0
72.1
Tributary 7
68.4
Tributary 4
Tributary 8
64.8
63.1
57.5
Tributary 9
54.0
48.8
44.8
44.4
Tributary 12
42.4
Lead B
37.4
Tributary 13
32.4
27.6
23.5
16.9
12.8
8.0
2.9
XO
X
XO
XO
XO
XO
XO
XO
XO
XO
XO
X
XO
X
XO
XO
X
XO
X
XO
XO
XO
XO
o
X
X
XO
X
XO
XO
XO
XO
X
XO
XO
X
XO
X
XO
X
XO
X
X
X
X
XO
X
X
X
X
0
XO
X
X
X
XO
XO
X
XO
X
X
X
XO
X
X
X
X
XO
X
X
X
X
XO
X
XO
X
X
XO
X
X
X
X
X
X
XO
X
X
X
XO
XO
XO
XO
XO
X
XO
XO
XO
XO
X
XO
X DO, Temperature
O Ultinate BOD (BODU), NHj-N,
TABLE 4
SPATIAL SURVEY DATA FORMAT
River Mile
Width, Ft.
Depth, Ft.
Headwater, Tributary,
BOD,, Temperature
Time of Day
and Point Source Flow,
West Bank Middepth
Temperature, C
Dissolved Oxygen, mg/1
Midstream Surface
Temperature, C
Dissolved Oxygen, mg/1
PH
Midstream Middepth
Temperature, C
Dissolved Oxygen, mg/1
Conductivity, mho/cm
Sample for BOD
NHj-N,
Midstream Bottom
Temperature, C
Dissolved Oxygen, mg/1
Eastbank Middepth
Temperature, C
Dissolved Oxygen, mg/1
Velocity, ft/sec (at selected stations)
1/4 width middepth
1/2 width middepth
-------
V QUALITY ASSURANCE PROTOCOL
Model calibration and verification required laboratory and
field measurement of a number of physical, chemical, and biological
variables. They are outlined in Table 5 along with appropriate
quality assurance protocol. Quality assurance was integrated
into the field and laboratory measurements to (a) assure the
integrity of the data base and (b) alert the field manager to
analysis or instrumentation problems on a routine basis. The
most frequent problems encountered were related to maintenance
of DO electrode membranes and specific ion probes.
SUMMARY OF QUALITY ASSURANCE PROGRAM
Measurement
Location
Width
Depth
Temperature
Conductivity
pH
Stream Velocity
Dissolved Oxygen
Amoonla
Nitrate
Ultimate BOD
Point Source BOD.
Source/Method Location
Topographic Maps Field
U.S. Army COE Navigation Chart!
Rangefinder, Inc. Model 600 Field
Raytheon FR450W Field
YSI Model 56 DO Meter Field/Lab
YSI Model S-C-T Meter Field
Orion Model 201 Lab
Bendix Model B-10 Ducted Current
Meter Field
Gurley Current Meter
ysi Model 56, Model 57 Field/Lab
Orion Electrode Model 95-10 Lab
Orion Electrode Model 92-0? Lab
Single Bottle with Reaeration Lab
Standard Methods, 14th ed. Lab
Calibration Type
Comparison to known
Comparison to known
National Bureau of
Standards Reference
Calibration
Frequency
Monthly
Monthly
Monthly
Single Reference Buffer per use
Standard Flume
Air Calibration
Wlnkler Met Chemistry
Reference Solution*
and spiked unknowns
Reference Solution,
and spiked unknowns
Glucoae-Cluta*ic Acid
Control
EPA Quality Control
Samples
Monthly
Dally
Weekly
Dally
Dally
Once during
study
Once during
study
VI ESTIMATION OF MAJOR PARAMETERS
The following experimental procedures were used to estimate
major parameter values and their uncertainty. They include
carbonaceous and nitrogenous ultimate BGD's and their reaction
rate coefficients, reaeration rates, sediment oxygen demands,
and photosynthesis related parameters.
A. BIOCHEMICAL OXYGEN DEMAND STUDIES
Long term BOD studies were conducted to (a) estimate the
magnitude of point source loads, and (b) estimate the rate of
river deoxygenation. Both studies were conducted using a single
dilution technique with sample reaeration (6).
-------
The apparatus for these studies is shown in Figure 2. The
strategy was to monitor the dissolved oxygen (DO) concentration
changes and reaerate the contents of the container when the DO
level approached 2 mg/1. Incubation was at 25C for all samples.
Small (25 ml) samples were withdrawn for ammonia and nitrate ion
concentration measurements which were used to define nitrification,
A more detailed discussion of the theory and experience with
this technique is discussed by McKeown, e_t £^L. (6).
MC-HOLE
. RUBIER STOPPER
S TUBE
OKI-HOLE RUIBEH STOPPER
OXYGEN SENSOR
GROUND-CLASS
NECK AND STOPPER
NCAS1 LUC TBfl BO) flPPWATUS
Figure 2
The reaction rates for the carbonaceous and nitrogenous BOD
data are summarized in Table 6 along with an estimate of their
95 percent confidence limits. The reaction rates noted in Table
6 were used as needed in the four water quality models.
It is interesting
aceous BOD of 0.02 per
ally measured in prior
reflects the long term
water samples. Short
resulted in estimates
coefficients. The k
QL2SMG and were obtal
nitrate, and DO data.
to note that the average value for carbon-
day (25C) is lower than has been tradition-
modeling studies. The low value, however,
deoxygenation behavior of these river
term experiments of 20-30 days would have
of lower ultimate BOD and higher rate
and k^n9 values were required only for
d by non-linear regression of the ammonia,
-------
c. S™«ARY OF BOD KINETICS FOR RIVER
D WATER SAMPLES AT 25C
Parameter
"l
kn
H
H
Description
Carbonaceous
BOD Decay
Nitrogenous
BOD Decay
Ammonia
Oxidation to
Nitrite
Nitrite
Oxidation to
Nitrate
Measured Value
(I/day)
0.02
0.13
0.13
0.6
95%
C.L.
(I/day)
+0.01
+0.02
+ 0.02
+ 0.1
Each of the major point source loads were sampled at least
during the summer months for the purpose of determining
their ultimate BOD value. NPDES reporting data which provided 5
day BOD values were also collected from each of the major sources.
It was assumed that the ultimate values obtained from the long
term experiments were constant and represented the point source
for the entire simulation period. Consequently, a ratio of
ultimate to 5 day BOD was established for each load.
It is important that recognition be given to the fact that
ultimate to 5 day ratios are really a companion estimate of the
oxidation rate constant. Because of this correlation, the values
of the ratio and the reaction rate constant are not independent.
If considered independent, selection of one without proper selection
of the other will lead to erroneous application and will produce
considerable bias in the allocation of waste loads.
B. RIVER REAERATION
The river reaeration rates on the lower Ouachita were es-
timated from radiotracer measurements made at two locations.
These studies were performed by Law Engineering Co. of Marietta,
Ga. with NCASI supervision and personnel and are further sum-
marized in Table 7. The standard deviation values presented in
this table are derived from the field data and are discussed in
a companion paper to these proceedings.
8
-------
TABLE 7 SUMMARY OF RADIOTRACER EXPERIMENTS
ON OUACHITA BASIN
Location
in. p.
m. p.
75
195
Segment
Length,
Miles
•
18
Average Average
Approximate Velocity Depth
low' c£s Ft/Sec. Ft
1200
850
0.09
0.40
30
15
95%
Confidence
20(,(l/day) Limits
0.02
0.17
+ 1000
+ 18
A comparison was made between these values and several
empirical equations. It was interesting to note that the
O'Connor-Dobbins equation agreed to within 6% of the radiotracer
measurement at both velocity conditions. This favorable com-
parison was used as the basis for concluding that the O'Connor-
Dobbins equation was appropriate for estimating the reaeration
coefficient of the Ouachita River for similar flow regimes.
This equation, therefore, was used in all four models.
C. SEDIMENT OXYGEN DEMAND
Sediment oxygen demand (SOD) measurements were made on the
Ouachita River Basin following procedures outlined in NCASI
Technical Bulletins No. 317 and 321 (7, 8). The measurements
were made in-situ using the respirometer shown in Figure 3.
Prior research by NCASI has shown that SOD can be a function of
the velocity at the sediment water interface. It was hypothesized
that the rise in turbulence generated by increased velocity was
responsible for increased transport of soluble organic material
across the sediment interface, resulting in increased SOD.
KCL SYRINGE
Oa PROBE \ t DC STIRRER
CONDUCTIVITY
PROBE
ALLENAIR
CYLINDERS
Figure 3
NCASI IN-SITU SO) APPARATUS
-------
The variability in the measured SOD values was assessed with
repeated-measurements. The average and standard deviation were
1.6 gm/m /day and 1.0 gm/m /day (+62%) respectively. Other NCASI
experience has shown that in-situ~SOD measurements exhibit vari-
ability of the order of +44% (8).
D. PHOTOSYNTHESIS
Previous water quality studies of the lower Ouachita River
did not identify algal productivity as a major water quality
process (9). NCASI field work in this area consisted of the
following studies:
(1) Light penetration at 4 locations.
(2) Three diurnal DO profiles at location.
(3) Light-dark bottle at 4 locations.
The light and dark bottle measurements were measured after
a technique developed by NCASI. Standard BOD bottles filled
with river water from a common well mixed sample were suspended
from a triangle rack constructed of 2" x 4" lumber. The bottles
were suspended at depths of 1/2, 1, 2, 4, and 8 feet as shown in
Figure 4. The dark bottles were also standard BOD bottles but
covered with 2 layers of black electrical tape. Six bottles
were suspended at each depth, spaced randomly across the rack.
•IDI VIIW
MO ml BOD
•OTTLM
DDL)
LIGHT/DARK BOTTLE
Figure 4
10
-------
The results of these studies are presented in Table 8. It
was noted that the depth at which the maximum photosynthesis
occurs is approximately 1-2 feet. Net algal respiration occured
below a depth of 2 feet in all cases. The average daily p-r
term in Table 8 of 1.8 mg/l/day applied strictly to the active
photosynthetic zone which was approximately 1 to 4 feet in depth.
This distance was approximately 15% of the basin's average depth
of 25 feet. The average daily p-r term was, therefore, multi-
plied by 0.15 to obtain a cross sectional average of 0.3 mg/l/day
+ 0.1 (95% C.L.). This value was necessary for the SNSIM simula-
tion which required a p-r term for the entire water column. The
uncertainty in p-r of + .1 mg/l/day was later used in the sensiti-
vity study.
Table 8 SUMMARY OF LIGHT-DARK
BOTTLE PHOTOSYNTHESIS EXPERIMENTS
Location
•.p.
•.p.
».p.
•.p.
52
33
73
133
Date
8/24
8/27
8/29
8/28
Initial
DO
(«J/1)
S.I
5.4
6.4
6.5
Dark Bottle
DO
(ng/1)
4.8
4.3
5.5
5.6
(1) Light Bottle DO at Stream
Hater Incubation Depth (mg/1)
Temperature 6' 1' 2' 4' 8'
32C
32C
32C
32C
5.7
5.1
6.8
7.0
5.5
5.6
6.8
7.0
5.4
5.2
6.0
6.1
4.8
4.7
6.2
6.0
4.7
4.7
6.0
6.0
Maximum
Net
p-r, ng/1
0.6
0.2
0.4
0.5
Maximum
Net p-r
per 14 hour
sunlight
day,
ng/ I/day
2.6
0.9
1.7
2.2
Avg.
Standard
Deviation
0.7
(1) All values are an average of 3-6 bottles following 3 1/4 hours
of instrean incubation.
VII PARAMETER SENSITIVITY
A. Mathematical Expression of Sensitivity
Parameter sensitivity in water quality modeling is defined
as the response of an output variable such as DO or BOD to a
change in a single input model parameter. Sensitivity can be
illustrated mathematically using the Streeter-Phelps DO model
shown in Equation 1.
-k,t
_
Vkl
(e
-e
-k9t
2
(Eq. 1)
11
-------
D = DO deficit, mg/1
L = Instream ultimate BOD, mg/1
t = Time, days
k1= Deoxygenation rate parameter, I/day
k2= Reaeration rate parameter, I/day
The sensitivity of the deficit to the parameters k, and k0 is
represented by Equations 2 and 3 respectively. A 2
V —k
K2 Kl
-k t
(D - tLe
(Eq. 2)
do
ak.
Vki
-k t
(Lte ^ -D)
(Eq. 3)
In the case of the model expressed by Equation 1, the sen-
sitivity of the oxygen deficit to the parameters k, and k2 is
complex and a function of all variables and parameters in the
model. The sensitivity equations (2 and 3) also illustrate that
there are no apriori assumptions that can be made about one
model parameter being more sensitive than another. The mathemat-
ical representatives of sensitito-ity preclude such assumptions.
B. Significance of Sensitivity Studies
Sensitivity is initially of importance in modeling to define
those parameters that are most sensitive and logically require
emphasis during field parameter estimation work. Consider the
following example in which the importance of SOD is being considered
during a wasteload allocation determination. Historical and
literature sources indicate that the SOD should fall in the range
1-5 gm 02/m day for the basin in question. Further, the measurement
of SOD with an in-situ respirometer would require 1/2 to 1 month
for a two person team to complete. The question is, what would
be the benefit in the wasteload allocation determination as a
result of determining SOD?
12
-------
A sensitivity study could bemused to investigate the impact
of SOD over the range 1-5 gm O2/m day on DO predictions. The
study may reveal that SOD has relatively little impact on stream
DO compared to other stream process and, therefore, should not be
measured. On the other hand, the model may show extreme sensitivity
to SOD and suggest that the work be completed. In this later
case, the field work will become more precise and more fairly
represent this oxygen sink in the model. Proper distribution of
the DO sinks has definite implications on waste load allocation
for if the oxygen sinks can be better defined during calibration,
they won't be lumped together into the stream deoxygenation rate,
V
C. Ouachita River Sensitivity Study
1. Sensitivity^ Study Conditions - In practice sensitivity
is examined by perturbing one parameter from a given value by a
constant amount (i.e. +50%) while the other are kept constant.
The perturbation amount usually represents the amount of un-
certainty or error associated with each parameter.
The sensitivity of each calibrated model is examined relative
to a base case. In each model, the base case is the best cali-
bration of the July survey conditions. These calibrations are
presented in later sections on model selection, calibration and
verification. DOSAG is calibrated for DO and CBOD. QUAL1E,
SNSIM and QL2SMG are calibrated for DO, CBOD and NBOD. The base
case DO varied slightly from model to model because of differences
in individual model conceptualization and input parameters.
The sensitivity in each of the models was examined by perturbing
the input value for each parameter, one at a time, around the
input value for the base case. The perturbation for temperature
is +1°C. The perturbation for reaction rates, algae parameters
and'loads is +50% of the base case value. The perturbation for
hydraulic parameters is +20% of the base case value. The magnitude
of these perturbations is standardized within each group of para-
meters in order to facilitate the comparison of the sensitivity
of similar parameters. "No BOD" means that BOD was not simulated
in the model. In addition, the magnitude of the perturbation
used in each group of parameters represents the relative con-
fidence in the estimation of each parameter type
2. Summary of Sensitivity Study Results - The results of
this sensitivity study are presented graphically in Figures 5 to
8. These results are commented upon in a general sense in the
Tollowing paragraphs. In summary, calibrations of the water
quality models DOSAG, QUAL1E, SNSIM, and QL2SMG for the July
survey conditions have been reviewed for the sensitivity of their
dissolved oxygen predictions to the perturbation of input para-
13
-------
meter values. In general the models showed an increase in sen-
sitivity at the end of the basin over the midpoint, probably
reflecting the increase of travel time in the basin and its affect
on various reaction parameters. In the simplest model, DOSAG,
the two major oxygen reaction rates, k, and k2, displayed the
most sensitivity. As the complexity or the model increased (para-
meters were added) the dissolved oxygen sensitivities evened out
somewhat, and were reduced in magnitude. In QL2SMG, additional
complexity in the form of the algal growth cycle added greatly to
the range of predicted dissolved oxygen values.
While the sensitivity of most reaction parameters remained
roughly equivalent among QUAL1E, SNSIM and QL2SMG, the algae
parameters made the prediction of dissolved oxygen value from
zero to saturation and beyond possible. It should be noted that
the p-r algae term in SNSIM, based on light and dark bottle ex-
periments, showed nowhere near the sensitivity possible with the
QL2SMG algae cycle.
The relationship in water quality models between hydraulic
and water quality parameters was demonstrated. The sensitivity
of the reaction rate and algae parameters was a complex function
of other parameters in the model, particularly hydraulic para-
meters. The simple addition of an SOD term to the model calibra-
tions reversed the relative sensitivities of velocity and depth.
The addition of the QL2SMG algae cycle gave depth an extremely
high sensitivity.
Viewing this sensitivity study from the standpoint of waste
load allocation and model predictive capabilities, several inter-
esting points emerged.
1) The relative sensitivity of point source
loads to DO predictions was often minor
when compared to the relative sensitivity
obtained with some other model parameters.
2) The algae parameters, SOD, and k_
displayed the highest sensitivity.
3) The hydraulic parameters of velocity and
depth confounded the sensitivity of other
model parameters such as k2 and SOU be-
cause they were often related.
This sensitivity study also illustrated some current problems
in the use of water quality models for waste load allocations.
Simple models such as DOSAG show high sensitivity of the CBOD
kinetics to the prediction of D.O. Yet, in complex situations
(like the one modeled herein), the most sensitive parameters are
not always included in simple models. More complex models such
]k
-------
as QL2SMG, however, require the estimation or measurement of
several algae-nutrient cycle parameters as well as SOD. These
parameters which have exhibited high sensitivity are often not
measured in the field when they are modeled. Additional error is
introduced if inappropriate values for these sensitive parameters
are selected.
In short, the most sensitive parameters in many models
were also the least understood parameters. Because of this
lack of understanding and the uncertainty in the parameter
value, model calibration may be little more than a curve
fitting exercise with little hope of achieving the correct
balance between these sensitive parameters. There are many
values for these parameters which may be used together such
that one or two output variables may be calibrated. Yet,
only one of these combinations could be a true mechanistic
representation of the stream. Mechanistic representation
and correct parameter balance can only be tested by model
verification as discussed in the next section.
3.5
r 3.8
2.5-
2.0-
, ,
•Atl CAM '•»
1.0-
0.5-
0.0-
-K
i
» 1C
..»
.,
»••%
»*
JlB.
"A
.-»
"i
_
*—^,
NO
C80O
1
-»*
£iti
••Hi
»*
LOAD
NftOO
i
*•*»
*»•«
0
.*%
*f«*
n*%p—
J *
-«*
-a^l
0
41^
DOSAO SINaiTIVITV
mil* point O.O
Definition
Toaparature,c
Carbonaceous BOD decay, I/day
Nitrogenous BOD deeey, I/day
Rate of reparation, I/day
100 Carbonaceous BOD, ng/1
BOO LOAD Carbonaceous point oourco 6OD,
Nitrogenous BOD, og/1
LOAD Nitrogenous point Oourco BOD, o
Headwater flow, cfa
Tributary 99 Clow, cfa
Velocity, ft/sec
Depth, tt
WD Carbonacaoua BOD not eiAulated
-M* +«*
MATfS LOADS
HYDRAULICS
DOSAG SENSITIVITY-M, P. 0
Figure 5
-------
3.3-
- JOh
-x.
f
9 £
» "n
O
O
H 2.0-
O
m
1.5-
•All CAtg.
1.0-
0.5-
-if
_J
+tF
rn
,,
y
^*
*
^•5
rn
•00
^
,..,
*
HP
CBQD
QUAL1I •EMITIVITV
•»»• polni 0.0
IfjfgUD
ParaMtor n-flnltlon
-•«*
LOAD
CWO
*..»
^2
LOAD
M800
n
0
^J
• iMimzacuro, r
*' Carbanacaoiu BOO Oncay, I/day
"• Dltrooraoui BOD decay, 1/dav
«l ««t« of M.er.tion, l/«Iay
"• Srtlaent o»n«n doand
»M« C*°° Carbonicwnii BOD. n/1
»2
n^i
•"%
..»'"• CarbOMcwo. point Mure. BOO. «/l
«•«> mtro^anoiil BOD, Bj/1
»oo uu HlUo9eno<» point aouroa BOD, «j/l
••*» « Wadwatet film, cf.
*~~1 «• Tributary 19 flow, cf.
„ 1 ' v.loclty, ft/arc
| • Deptll, (t
*"*
HATH tOAO* HYDRAULIC*
OJALJE SENSITIVITY - M.P. 0
Figure 6
4.0-1
* e
— J.5-
a
E
10-
e
e
o
5 W-
O
Z.O-
•Atl CAM.
1.5-
1.0-
OJi-
•KW
+»*
-•»*
T
J
s
••via)
«%
••••
*•
^uV
*••*
•DD
"l
»»» -lA
PHI
MO
CNO
..*0
•»*
^^~
LOW
CBOO
-•«*
LOAD
HMO
^^
e
.^
-t4
*••»
+••«
INBIH Mali ITI WITT
•II* poUt 0.0
TJTTTnrp
Paranfttu Definition
T fHnwratur*, c
H. CubowcamM BOO d««*ri I/day
•« Hitrogtmotu SCO daecv, l/d»r
«, bit* of r*Mr*tlonf l/d*y
*t«A loo Badiamt ootygvit da.wnd /
?
•»*
•
P-* Hat photoaynthwl* nlnm raiplratlos,
•9/1/o.y
c»o Carbonaeaoua BOO, eg/1
"*» ca» i*>* Carbonaoaona point aooro* BOD, eg/I
1 ••oo tfltrogaaou* BOD, *g/l
° I ••09LM* Ultroganoua point aonro* BOD, *g/l
— 1 o, Tributary 19 flov, oil
»»» > velocity, ft/Me
o Davta, ft
•0 CM* C«rbon*c*o« BOO not aUulatad
-««
•ATI* ALOAi LOAOI
SMSIM SBEITIVITY - M.P. 0
Figure 1
16
-------
3.5
3.0H
• UH
" u-
O
O
1 1.5-
.*., eA... K|
0.5-
0.0-J
»M*
•••» r.o
?~C3
£
•'•%
••« ««* M^
QLISMO SENSITIVITV
nil. point O.O
Parameter
Teaperature,. r
Carbonaceoua BOD d«cay, I/day
Rate of amnla decay, I/day
Sedlaent o»yg«i denand
Rate of reaeratlon, I/day
Rate of oaxiBum algal growth, I/day
Rate of algal reaplratlon, I/day
Rate of algal aattllng, I/day
Light extinction coefficient, I/ft
Carbonaceoua BOD, ajg/1
Point aource CBOD load, eg/1
Amonla load, ng/1
Headwater flow, cf.
Tributary 19 flow, efa
Velocity, ft/eec
Depth, ft
Carbonaceous BOD not alauilat.d
X.
•00 LOAD
I) LOAD
• ATI* A LQAf LOADS HYDRAULICS
QL2SMG SENSITIVITY - M.P, 0
Figure 8
VI11 MODEL SELECTION, CALIBRATION AND VERIFICATION
This section addresses the process of selection, calibration
and verification of the four water quality models (DOSAG, QUALlE,
SNSIM, and QL2SMG). It must be stressed that this study presents
this process for a single river basin using a particular method-
ology. The examples given demonstrate many of the problems
commonly encountered in the validation of water quality models.
As a result of examining these problems, a pattern of effective
model selection, calibration, and verification emerges which
should have application to most receiving water situations.
In order to make reasoned decisions as to whether or not a
model is properly selected, calibrated and verified, it is
necessary to establish criteria for the fulfillment of each
modeling step, it is recognized that these criteria may be
initially subjective because it is probable that no two modelers
would view a receiving water situation in exactly the same way.
-------
Nevertheless, establishing a clear set of criteria or adopting
a structured approach to model selection, calibration and verifi-
cation enhances the understanding of the role each of these
steps plays in the development of model predictive capability,
and allows a clearer comparison among models which is the objective
of this study.
A. Model Selection Criteria
The primary basis for choosing criteria which would be used
to select a model was that the criteria must result in a model
that is a "reasonable" representation of the river basin. A
"reasonable" representation is one which models the most im-
portant physical, biological and chemical processes to allow the
prediction of water quality for specific conditions.
Table 9 summarizes the evolution of the model selection
criteria used by NCASI in this study. The first column lists the
various topics explored above. The second column lists the crit-
eria that would have resulted if only the data collected prior to
the NCASI spatial surveys had been used. The third column lists
the criteria which were compatible with NCASI spatial survey data
and the four water quality models selected for comparison. The
fourth column lists the few conditions observed in the basin which
were not included in the selection criteria. For the most part
they represent minor conditions in the basin, and their exclusion
did not affect the predictive capability of the selected models.
However, a few of the conditions (notably the N and algae cycles)
are discussed further in subsequent sections of this report, as
they may play a role in developing model predictive capability.
B. Model Calbration/Verification Criteria
1. General Criteria The first step toward calibration was
described earlier in this report as one in which both model para-
meters and input loads had been measured by appropriate experi-
ments or assigned by sound engineering judgment. The model is
considered calibrated when the resulting model output, such as
stream DO or BOD, matches observed data. The model is considered
verified, on the other hand, when the calibrated model produces
some water quality condition perturbed from the calibration case.
Generally, the perturbed case involves different flow, tempera-
ture, and/or load conditions.
18
-------
TABLE 9
EVOLUTION OF THE MODEL SELECTION CRITERIA
Validation
Topic
Initial Criteria Based On
Data Prior to NCASI
Spatial Surveys
Final Criteria Used in this
Study, Based on NCASI
Survey Data Spatial
Observed Conditions
Excluded from this
Selection Criteria*
Hydraulics
— Loads
Physical, Chemical
and Biological
Processes
Temperature
One-dimensional
steady state
rectangular channel
8 point sources
15 tributaries
non-point sources
CBOD
NBOD
Reaeration
SOD
Direct T Input
T Correction of Rates
One-dimensional
steady state
rectangular channel
6 point sources
7 tributaries
no non-point sources
CBOD
NBOD
Reaeration
SOD
(Algae (p-r)
Direct T Input
T Correction of Rates
Two-dimens iona1
vertically stratified
dynamic
trapezoidal channel
2 point sources
8 tributaries
non-point sources
N-Cycle
Algae-Cycle
Heat Balance
* See text for the significance of these conditions.
-------
This notion of model calibration/verification, which
has been cited in numerous water quality studies during the
past two decades, contains several subjective and arbitrary
features. The first of these was the manner in which parameters
were adjusted during calibration which called upon the modelers
recognition that parameter estimation procedures were subject
to error. The second of these related to the comparison of
model output with observed conditions. The modeler was simply
faced with making a judgment as to when the comparison was
"acceptable". This judgment was necessary for both model cal-
ibration and verification.
A review of state-of-the-art modeling practice has revealed
an interesting dichotomy amongst modelers in their approach to
obtaining an "acceptable" verification. One approach simply
relied upon a visual inspection of the predicted vs observed
model variable and made a subjective judgment about goodness of
fit. A vast majority of current modeling studies fall in this
category. Examples of this approach are presented below:
(1) "The goodness of fit reached for the predicted
variables during calibration of each of the stream
models is illustrated by graph plots of computed and
observed data vs stream distance" (10).
(2) "As shown in the upper panels of Figure X good
agreement is shown between observed and calculated
BOD and DO" (11).
(3) "Verification (calibration) of the DO and the
BOD mathematical models presented elsewhere in this
report has been made for various conditions of flow
and water temperature. Verification (calibration) of
the mathematical models is obtained when good correlation
is evident between the calculated profiles and the
observed data for various conditions of flow, temperature,
and loading, using consistent parameters* (12).
(4) "Inspection of these figures shows that the
calibration on DO and BOD is quite accurate.
Even though the variation in the values measured
at each sampling location was relatively high for both
DO and BOD, the averaged values match reasonably well.
Greater agreement with the averaged data points could
be obtained by adjusting certain parameters. However,
given the variation in the measured data and the fact that
certain parameters such as reaeration, sedimentation
rate, and benthic demand were unknown, it was felt that
a closer fit could not be justified. The calculated
profiles indicate that the proper mechanisms have been
accounted for in the model" (13).
20
-------
The second approach relied upon statistical tests for inter-
preting goodness of fit. O'Connor, et al. (14) examples of
such analyses for a three dimensional eutrophication model of
Lake Ontario.
The verification statistic discussed involved the use of
Student's t distribution to compare model output to observed
data. The results of such analyses were plotted for a general
case in Figure 9. The physical interpretation of the plot in
Figure 9 was that the observed and predicted means were statis-
tically indistinguishable at the 90% confidence level. Inversely,
the modeler was accepting a risk of one chance in ten that the
model mean was statistically different than the observed mean.
It is important to note that another modeler might have chosen
another confidence level where the differences might be significant
a - MODEL MEAN - OBSER VED MEAN
*- REG ION OF NO STA-
TISTICALLY SIGNIFICANT
DIFFERENCE BETWEEN
MODEL AND OBSERVED
MEANS (90% CONFIDENCE
LIMITS)
(After Reference 14)
COMPARISON OF MODEL TO OBSERVED DATA
Figure 9
21
-------
A crude statistical measure of goodness of fit was re-
ferenced in a proprietary modeling study reviewed by NCASI. The
modelers sought to minimize the statistic noted below for DO.
N
> o
Goodness of fit = > (Observed-Predicted)
N
A summary of the use of this statistic during model calibra-
tion and verification is presented in Table 10. Three data sets
(Af Bf C) were available for use. The technique was illustrated
for two different model parameter sets.
Table 10 Summary of Goodness
of Fit (Variance) Statistics
(Proprietary Model Study)
Calibration Verification
Phase Phase
Data Set Data Set Data Set
A B C
Parameter
Set I 1.4561 1.9546 1.5086
Parameter
Set II 1.1797 1.7729 1.5382
Parameter Set II was called the best parameter set because the
variance between model and observed data was lower than other
parameter sets. Furthermore, the verification attempts were
considered acceptable since they produced a variance less than
that obtained during calibration. This implies that the residual
differences between model and observed data were smaller for the
verification phase.
The problem with this methodology lies in the fact that the
goodness of fit or variance for both model calibration and veri-
fication was relatively high. A DO variance (S ) of 1.1797
(Table 10) implied that the mean deviation of the model from the
observed data (\JT~ ) was 1.08 mg/1 DO. This measure of model
predictive capability may not be acceptable for certain model
applications.
22
-------
2. Specific Criteria for NCASI Study - The calibration/
verification criteria used by NCASI for the Ouachita River Basin
modeling work was based upon a structured format. For a model to
be calibrated, its prediction of an output variable (DO, CBOD,
NBOD, etc) had to fall within the 95% confidence range of the
corresponding spatial survey data for 95% of the measured data
points. This was analogous to assuming a risk of 1 time in 20
that the predicted output would not match the observed data with-
in the confidence range specified.
The verification criteria required that the model prediction
of an output variable had to fall within the 95% confidence
limits for 60% of the measured data points. This less stringent
criteria for model verification was based upon the fact that the
95% confidence limits for some of the measured data points was
small (approximately +0.12 mg/1) for DO. Hence, in some cases
the predicted output variable came close to the measured 95%
confidence range without falling inside. A second and more
important point deals with the uncertainty associated with the
algal mechanisms inherent in the models SNSIM and QL2SMG. SNSIM
utilized a gross photosynthesis minus respiration mechanism
whereas QL2SMG used a complex algal growth cycle dependent upon
both light and nutrients. None of the parameters in this latter
mechanism were measured and therefore required estimation.
The NCASI criteria also required that the model output
match all significant trends in the measured data. The assess-
ment of significant trends was made with statistical tests,
thereby avoiding the use of visual observations of trends
when comparing observed data and/or model output.
The three criteria adopted for calibration and verification
were applied to all model outputs, not just DO. Also, these
criteria were selected following an analysis of the field data
and several attempts to calibrate and verify the various models.
Although their format was based on experience, the actual numbers
(i.e. 60% of the values) were selected for this particular study.
If the data had shown 90% of the values to fall within the 95%
confidence level, then the criterion for verification would have
been raised to 90%. As already mentioned, the important point
is that definitive criteria be used to encourage proper consider-
ation of the uncertainty surrounding eventual use of the model
to make water quality forecasts.
23
-------
C. Model Calibration Strategy
A systematic approach to model calibration was applied to the
four water quality models in order to test several aspects of the
model's predictive capabilities. Central to this systematic
approach was the calibration criteria developed in a previous
section. Calibration of the models was attempted using the spat-
ial survey data from 7/21-22/80 (DO, CBOD, and NBOD) as a ref-
erence, even for the models which did not match the selection
criteria. The systematic approach was designed to test the
following aspects of model predictive capability. First, the
ability of a properly selected model (i.e. SNSIM or QL2SMG) to be
calibrated was tested versus an improperly selected model (i.e.
DOSAG or QUAL1E). Here the models were tested to see if it was
possible to arrive at an acceptable representation of the cali-
bration survey data, even when all of the significant processes
in the river were not modeled. To some degree, this was a func-
tion of the confidence range of the parameter estimates and the
survey data being modeled. The effect of model selection on
predictive capability was further examined during verification.
Second, the ability of various state-of-the-art parameter
estimation procedures to define input parameters, and hence lead
to calibration, was examined. The examination focused on four
types of parameters from a previous 208 modeling study which were
largely "textbook" in nature; 2) NCASI "best parameter estimates"
based on field surveys and laboratory data; 3) parameter estimates
calibrated by the perturbation of the "best parameter estimates"
within their 95% confidence ranges; 4) parameter estimates calibra-
ted by the perturbation of the "best parameter estimates" outside
their 95% confidence ranges, but with engineering judgment. The
predictive capability of each type of parameter estimate was
further investigated in the verification section.
Third, the importance of the order and number of survey data
calibrated was investigated. This was done by first calibrating
the models to the spatial survey DO data. DO data had the small-
est confidence range when compared to CBOD and NBOD data. Cali-
bration then proceeded with further spatial survey data (DO and
CBOD; DO, CBOD and NBOD). Then, the process was reversed and
calibration was started with the spatial survey data with the
lowest confidence (largest confidence range), NBOD. The models
were further calibrated with increasing spatial survey data (NBOD
and CBOD; NBOD, CBOD and DO) as will be explained in the dis-
cussion that follows. The effect of these various approaches to
calibration (various orders and numbers of spatial survey data
calibrated) on model predictive capability was further examined
during verification.
-------
The effect of these various aspects of model calibration on
model predictive capability was examined via the four calibration
phases outlined below.
Calibration Phase 1 used the reaction parameters (k, , k_,
k , SOD, algae) from a 208 modeling study of this river Basin to
mBdel the physical and biological processes. The models were
otherwise setup with geometry, hydraulics, and loads as deter-
mined during the NCASI spatial surveys.
Phase 2 was the attempted calibration of the models using
the "best parameter estimates" for all inputs as established by
NCASI field and laboratory studies. The procedures used to
estimate these parameters were reviewed earlier in this report.
Calibration Phase 3 involved the perturbation of the "best
parameter estimates" in order to calibrate with an increasing
amount of spatial survey data. First, the models are calibrated
for the most confident survey data, DO. Then the calibration
proceeded to DO and CBOD; and finally to DO, CBOD and NBOD. In
Phase 3a the perturbations were restricted to the 95% confidence
ranges of all of the input parameters. Phase 3b allowed these
perturbations to fall outside the 95% confidence range, but
within a range which could be justified through engineering
practice. If parameter estimates outside the accepted engineer-
ing judgment range were required to meet the calibration crit-
eria for any particular survey data (DO, CBOD, or NBOD), then
the model was deemed as uncalibrated for that data.
Phase 4 involved essentially the same procedure as Phase 3
except calibration was started with the least confident survey
data, NBOD. From there it proceeded to NBOD and CBOD, and
finally to NBOD, CBOD and DO. Phase 4a allowed perturbations
within the 95% confidence limits of the input parameters, anal-
ogous to Phase 3a. Phase 4b permitted perturbations within the
range of engineering judgment, analogous to Phase 3b. Once
again, any model which could not meet the calibration criteria
for a particular survey data within the constraints of Phase 4b
was deemed uncalibrated for that data.
Table lj^ presents a synopsis of these four calibration
phases^Reference to this table will be useful in interpreting
the calibration/verification summary figures (Figures 10 to 13).
These four figures are a condensation of many individual cali-
bration/verification attempts.
25
-------
PHASE 1
PHASE 2
PHASE 3
b.
PHASE 4
Table 11
Summary of Calibration Phases
208 modeling study parameters
NCASI "best parameter estimates"
based on field and laboratory data.
Calibrated parameter sets, starting
with DO, then DO and CBOD, then
DO, CBOD and NBOD (NH.j).
Parameters held to within 95%
confidence range based on field
and laboratory data.
Parameters held to an accepted
"engineering judgment" range,
outside the 95% confidence range.
As in Phase 3 except calibrated
first with NBOD (NH ), then
NBOD (NH-) and CBOD, then
NBOD (NH, CBOD, and DO.
a. As in 3a.
b. As in 3b.
D. Model Calibration/Verification Summary
Model selection had several effects upon model calibration
and verification. A comparison of DOSAG's calibration (Figure
10) to QL2SMG's calibration (Figure 13) demonstrated this effect.
DOSAG, an improperly selected model, had poor calibration potent-
ial, while QL2SMG, a properly selected model, could be calibrated
in many different ways. The models QUAL1E and SNSIM, on the
other hand, showed similar calibration potentials. This fact
raised an important point about model calibration and its re-
lationship to model selection. SNSIM allows simulation of algal
productivity and QUAL1E does not. However, the 7/21-22/80 survey
data used for calibration did not show much algal productivity.
Thus, the failure of the calibration process to distinguish be-
tween QUALlE and SNSIM was not surprising. It was found to be
important, therefore, to define model selection criteria such
that all major physical, chemical, and biological processes active
during the temporal and spatial regimes of interest are considered,
A single spatial survey was not sufficient in this study to de-
fine these processes.
26
-------
The effect of the parameter estimation procedure on model
predictive capability was also illustrated in Figures 10 to 13.
Several important effects were noted. First, Phase 1 (parameters
used in Phase 1 were largely textbook values) provided the least
exact predictions of the observed conditions.
Second, the effect of parameter uncertainty on model cal-
ibration and verification was explored. Both Phase 1 and Phase 2
failed to calibrate and verify because in both cases, the input
parameter values were treated as absolute numbers. This rigidity
in input parameter specification did not recognize that every
measurement had an inherent uncertainty. Failure to recognize,
and quantify this uncertainty was partly responsible for the poor-
model predictions shown in Phase 1 and 2 of this study.
On the other side of this issue was the use of parameter
values outside their established confidence ranges. A comparison
of the "a" and "b" type calibrations and verifications of Phase 3
and 4 demonstrated this effect. Parameter values in Phase 3a and
4a calibrations and verifications adhered to their 95% confidence
limits, while Phase 3b and 4b allowed values outside that range
where they could be justified. In general, it was possible to
calibrate more survey data with "b" type input data, but pre-
dictive capability was not enhanced. In examining Figures 10 to
13, the only place where significant predictive capability was
gained using "b" parameter values was with QL2SMG. In that case,
it was not a calibration parameter that was outside its 95% con-
fidence range, but rather the verification flow for the 8/17/80
survey. The use of a flow outside the 95% confidence range was
actually a compensation for a known violation of the basic steady-
state modeling assumption (during the 8/17/80 survey). Incidently,
in the case of SNSIM and QUAL1E, the use of "b" parameter values
during calibration actually reduced the predictive capability of
the model.
It was found that parameter uncertainty must be applied in
moderation during -the perturbation of parameter values for cali-
bration and verification. Mathematical water quality modeling
parameters are neither single values nor excessively broad ranges.
State-of-the-art parameter estimation procedures must involve
field and laboratory measurements of important environmental
processes to establish the uncertainty estimates in parameter
values. If the model is appropriate for the natural system being
represented, then parameter values outside the experimentally or
theoretically determined confidence range should never be needed.
In order to verify if an output variable was correctly modeled,
it was first necessary to calibrate the model to predict the
specific variable. In other words, a model which was calibrated
for only NBOD was not expected to predict DO for some future con-
dition. Several of the model calibrations demonstrated this point
27
-------
(See Figures 10, 11 and 12). For example, the calibration of
QL2SMG for all outputs (DO, CBOD, and NH_) enhanced the pre-
dictive capability over partial calibrations (NH~ and CBOD only)
as shown in Figure 13.
Finally, the order of calibration of output variables did
not have any measurable effect on predictive capability. If the
calibration of complex water quality models is an iterative
process, as was the case in this study, the order of calibration
has little meaning. However, if calibration and verification
are kept independent, as is the traditional case, then the order
of calibration may be important in certain cases.
An additional point was that calibration and a verification
of a model was not the final test of model predictive capability.
•For example, both SNSIM and QL2SMG were calibrated and verified,
yet QL2SMG possessed predictive capabilities that SNSIM lacked.
This was due to the different style of modeling the algal pro-
cesses in these two models. SNSIM modeled algae (p-r) and nitro-
gen (NBOD) as gross, unrelated processes. QL2SMG, on the other
hand, modeled algae and nitrogen as a mechanistic nutrient-algal
growth cycle. This cycle was complete with interdependencies
between the processes. Although the QL2SMG nutrient-algal growth
cycle was not a thorough representation of algae and nitrogen in
the aquatic environment, the cycle did establish the important
relationships between the natural processes. This allowed various
model outputs to vary as the input environmental conditions changed"
This mechanistic interrelationship between the processes in QL2SMG
gave this model more predictive capability when compared with the
gross modeling processes used in SNSIM. In summary, QL2SMG could
be used to predict conditions that SNSIM could not.
This suggested that selection criteria may need to in-
corporate factors beyond a "simple" representation of all the
processes important to the river basin. In this case, in order
to ensure maximum predictive capability, the selection criteria
should have included a requirement for a mechanistic representation
of the important processes occurring in the river.
An important process can be considered as one which may
change significantly over the range of conditions to be predicted.
In this study, such mechanistic representations would have re-
quired a host of additional parameters to be measured or estimated.
As shown in the sensitivity study, QL2SMG was extremely sensitive
to many of these parameters (notably maximum growth rate, res-
piration rate, light extinction coefficient, and depth). Thus
the selection requirement of mechanistic representations for
processes like algae adds to the complexity of the modeling pro-
cess.
28
-------
DOSAG SUMMARY
CALIBRATION
VERIFICATION
7/21-22/80 B/17/BO B/23/BO 12/3-4/80
CALIBRATION
PHASE DO CBOD NBOD DO pp QQ BOD
phaaa 1
phaaa 2
phaaa 3a
phase 3b
pha»« 4a
phaaa 4b
O O O D D H D
o 3 o a a n n
• O O D D • a
O O
O
O
o a
O D
• a
• a
D •
D D
• D
a D
D
D
D
3
O
CALIBRATION LECEHD
Matches 95% or »ort of calibration data.
Calibration criteria it »et.
Matches >SO% but ^95% of calibration data
Matches <50» of calibration data
VERIFICATION LEGEND
•Hatches 60% or Bore of verification data.
Verification criteria it met.
lal Matches > 30% but< 60% of verification data.
I I Matches<30% of verification data.
DOSAG CALIBRATICMERIFICATION SUM^RY
Figure 10
29
-------
QUAL1E SUMMARY
CALIBRATION
VERIFICATION
7/21-22/80 B/I7/BO 9/23/BO 12/3-4/80
CALIBRATION
PHASE DO CBOD NBOD DO DO DO BOD
phas* 1
phase 2
phase 3a
o o o n n a n
d 3 O D D D D
• o o n n • a
• • O D D • •
Ph...3b • • • n n a •
,h...4. o o o n n n n
ph...4b • • • D D a •
3
O
CALIBRATION LEGEMD
Hatchet 951 or mot* of calibration data.
Calibration criteria i» Bet.
Matches >SOt but<9Sl of calibration data
Hatches <50% of calibration data
VERIFICATION LEGEND
Matches C0% or aore of verification data.
Verification criteria ii met.
Matches > 301 but< 60% of verification data.
Matches<301 of verification data.
QUAUE CALIBRATION/VERIFICATION
Figure 11
30
-------
SNSIM SUMMARY
CALIBRATION
VERIFICATION
S/17/BO 9/23/BO 12/3-4/80
O
CALIBRATION
PHASE
phaaa 1
phaaa 2
phaaa 3a
phaaa 3b
phaaa 4a
phaaa 4b 0 9
CALIBRATION LECEHP
Hatch** 95% or more of calibration data,
Calibration criteria i* »*t.
Match** >50% but <95% of calibration data
Match** 301 but< 60% of verification data.
LJ Matches < 301 of verification data.
SNSIM CALIBRATIOWERIFICATION SUWARY
Figure 12
-------
QL2SMG SUMMARY
O
CALIBRATION
VERIFICATION
7/21-22/80 8/17/80 8/23/80 12/3-4/80
CALIBRATION .. ^ _^ ^.^
PHASE DO CBOD NBOD DO DO DO BOD
phase 1
phase 2
phase 3a
phase 3b
phase 4a
O O O D D D D
O D D a D
o a
• a
o •
•
D
a
3
a
phese4b
3
D a
CALIBRATION LEGEND
Natch** 95% or «or* of calibration data.
Calibration criteria 1* Mat.
Natcbat>SO% but <95% of calibration data
Match** <50% of calibration data
VERIFICATION LEGEND
•/ Natch** 60% or *»r* of verification data.
/ Verification criteria i* B*t.
Uaf Natch** > 30% but<60% of verification data.
I—I Natch**<30% of verification data.
QL2SMG CALIBRATIOWERIFICATION SUWARY
Figure 13
32
-------
IX CONCLUSIONS
The following conclusions have been divided into the topical
areas of selection, parameter estimation, and model calibration-
verification technique.
MODEL SELECTION ISSUES
(1) Model documentation is an important precursor to the
development of a specific water quality model, and can be a most
significant issue.
(2) It is important to define model selection criteria
such that all major physical, chemical, and biological
processes active during the temporal and spatial regimes of
interest are considered. It is possible, for example, that a
specific stream process may not be significant during model
calibration but become dominant in a verification attempt based
upon a later stream survey. This problem was encountered in
this model study with a,lgal productivity becoming more significant
in the verification surveys of August 17 and September 23 than
in the July 21-22 calibration survey.
(2a) Model development should emphasize a
selection phase which should precede the
calibration-verification phase.
(2b) Model selection should not be based upon a
single temporal or spatial survey of stream
water quality.
(3) The selection of an applicable water quality model is
enhanced by a careful mechanistic representation of stream processes
The model DOSAG, for example, could not be verified because it
did not include sediment oxygen demand or photosynthetic oxygen
production effects. QL2SMG which included these stream processes
was verified and, therefore, had greater predictive accuracy.
PARAMETER ESTIMATION ISSUES
(4) It was not possible to calibrate and verify a water
quality model in this study that contained lumped parameters. A
lumped parameter was defined as one which jointly represented
more than one stream process such as instream BOD and sediment
oxygen demand.
33
-------
(5) It was important to measure all major model parameters
to establish measurement uncertainty.
(6) Water quality models based upon measured parameters
are more readily verified and, hence, provide more predictive
capability.
(7) The model calibration phase should serve to refine
measured parameter estimates within their range of uncertainty
and not be solely used to assess their magnitude.
(8) The state-of-the-art procedures used in this study for
estimating stream deoxygenation, reaeration, and SOD were ex-
tremely useful in model calibration.
(9) The measurement of the parameters associated with
nutrient interactions and algal productivity limited the extent
to which these processes could be modeled and, therefore, limited
the predictive capability of the resultant model.
(10) It was possible to calibrate and in certain cases
verify the models SNSIM and QL2SMG with parameter values within
the measured 95% confidence limits for the lower Ouachita.
(11) The use of textbook parameter values for stream de-
oxygenation, reaeration, and SOD did not result in calibrated
models.
MODEL CALIBRATION-VERIFICATION ISSUES
(12) It is important to establish calibration and veri-
fication criteria in model development. The following were
defined for this study:
(a) Model calibration was achieved when the
model output predicted within the 95% confidence
range of 95% or more of the variables' measured
output values.
(b) Model verification was achieved when
the model output predicted within the 95%
confidence range of 60% or more of the variables'
measured output values.
(13) It is important to calibrate and verify as many out-
put variables as possible. DO, BOD, and a nitrogen species were
examples used in this study.
-------
(14) Model calibration and verification based upon only
one variable such as DO did not result in verification for other
variables. It was necessary to calibrate and verify based upon
DO, BOD, and a nitrogen species. Predictive capability was
enhanced when all three variables were verified simultaneously.
(15) The order of model calibration-verification was not
important in this study. Those models calibrated in the order
(1) DO, (2) DO and BOD, and (3) DO, BOD, and nitrogen species
provided the same predictive capability as those calibrated in
the reverse order (1) nitrogen species, (2) nitrogen species and
BOD, and (3) nitrogen species, BOD, and DO.
(16) A procedure for model calibration is likely to be
situation dependent. Sensitivity analysis is a useful approach
to illustrate which measured parameters to refine.
(17) Sensitivity was observed to be situation and model
dependent and a function of stream location, point source loads,
temperature, and the magnitude of individual parameter values.
(18) The most sensitive parameters in the four models
investigated were those which were often estimated through cali-
bration and not measurement in past studies. SOD, reaeration,
and algal productivity are three examples.
X. REFERENCES
1 . "A Review of the Mathematical Hater Quality
Model DOSKG and Guidance for Its Use,*
NCASI Stream Improvement Bulletin No. 327, NY, NY.
Oct., 1979.
2. «A Review of the Mathematical Hater Ouality
QUAL1B and Guidance for Ita Dae,* NCASI Stream
Improvement Bulletin No. 331, NY, NY,
April, 19(0.
3. *A Review of the Mathematical Model Oual-II and
Guidance for It* U«e," NCASI Stream Improvement
Bulletin No. 338, NY, NY, Oct., 1980.
4. Bralter, R.E., et al., •Documentation for
SNSIM.* US EPA,"Region II, 26 Federal Plata,
NY, NY, March, 1971.
c 'Use of Mathematical Model! in the Development
* of Areawlde 208 Plane,' NCASI Stream Improvement
Bulletin NO. 32i, NY, NY, Sept., 1*79.
C HcKeown, J., et al., "Ultimate BOD Estimation
In Receiving Alter Quality Modeling,' NCASI
Central Lakes States Regional Meeting.
Chicago, 111., June, 1980.
7. *Interfaclal Velocity Effects on the Measurement
of SOD.* NCASI Stream Improvement Bulletin No. 317,
m, NY. November, 1971.
•Further Studies of SOD and Its Measurement
Variability,* NCASI Stream Improvement
Bulletin No. 321, NY, NY, March, 1979.
•Ouachita River Basin Hater Quality Management
"nai
10. ?KY ?Ji'i •?" &£*• Wit" ou-H'y «<*»i of
the Illinois Rlverissln,1 uses. Little Rock.
Arkansas, March, 1*80.
11. 'upper Mississippi River 208 Grant Hater
Quality Modeling Study* Bydrosclence, Inc.
Jan., 1979*
12. O'Connor, D.J., -Hater Quality Analysis of
The Mohawk River Bridge Canal," NY State Dept.
of Heelth, July, 1968.
13. Naddel, N.N., et al., "A Hater Quality Model for
the South Platle VFver Basin,* Pacific Northwest
Lab. Batelle, Richland, Naah., April, 1974.
14. O'Connor, D.J., •Verification Analysis of
Lake Ontario and Rochester Embayment 3D
Entrophlcatlon Modela,* EPA-600/3-79-094,
Aug. 1979.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
35
-------
An Assessment of the Measurement Uncertainty in
the Estimation of Stream Reaeration Rate Coefficients
Using Direct Tracer Techniques
By: J.S. Hovis , R.C. Whittemore, EhD ,
L.C. Brown, PhD , J.J. McKeown
I. INTRODUCTION
All experimental measurements have an inherent uncertainty
associated with the measurement technique. The purpose of this
paper is to elucidate the major sources of error in the direct
tracer methods of reaeration rate coefficient (k2) measurement.
The magnitude of several of the component errors is estimated for
direct tracer measurements by both the hydrocarbon and radiotracer
methods. Examples from the literature and measurements made on
the Ouachita River Basin by NCASI, Law Engineering Testing Co. of
Marietta, Georgia, and the USGS are used. Estimates of total
measurement error are made, and related to the component sources
of error.
The utility and necessity of error quantification becomes
clear when the role of k~ in water quality management is examined.
The reaeration rate is one of the most important parameters in
the dissolved oxygen budget; it has a significant impact in most
water quality management decisions, particularly waste load alloca-
tion.
Experimental measurements do not always advance the state of
knowledge about the process under study. When the measurement
error is greater than the bounds which could be placed on the
quantity in question prior to the experiment, then the measurement
may be of little use. For example, it is possible, particularly
in low rate situations, to obtain a measured k,, value that has
greater error than would result from the use of an appropriate
empirical equation. This situation will be demonstrated in this
1. Research Engineer, National Council of the
Paper Industry for Air and Stream/Improvement, Inc.
(NCASI), Northeast Regional Center, Tufts University,
Medford, Massachusetts 02155
2. Professor, Department of Civil Engineering, Tufts
University
3. Regional Manager, NCASI, Northeast Regional Center
36
-------
report. The impact of this finding is that the use of the direct
tracer methodologies for k_ measurement is probably not justified
in some situations. Some guidelines for conducting direct tracer
measurements such that measured rates will have sufficient pre-
cision to justify the expense of the measurements are given.
II. BACKGROUNDf TRACER TECHNIQUES AND STATISTICS
It is assumed that the reader is familiar with the radiotracer
and the hydrocarbon tracer techniques. As it is beyond the scope
of this paper to review those techniques, the reader is directed
to the work of Tsivoglou (1) and Tsivoglou, et al. (2) for a
description of the radiotracer technique, and to the work of
Rathbun, et al. (3, 4) for a description of the hydrocarbon tech-
nique. In addition, it is assumed that the reader will possess a
general knowledge of statistical principles, particularly those
related to error structure, experimental measurement uncertainty,
precision, accuracy, and their contributions to total error.
Reference to a general statistics text such as Mandel (5) should
provide sufficient background.
One statistical concept that will be introduced in detail
here is the concept of propagation of error. Serth, et al. (6)
presented an excellent discussion of the propagation of experi-
mental error through mathematical relationships. Random error,
experimental precision, may be propagated through a generalized
mathematical relationship using Equation 1.
'a[A,n]
(Eq. 1)
A + a and B + b are error bounds
for x, and xl
Systematic error, experimental accuracy, may be propagated through
a generalized mathematical relationship using Equation 2.
sgn[ABJab ± [a|e|
(Eq. 2)
sgn (AB) denotes the algebraic
sign of the product AB
37
-------
The propagation of component random and systematic errors results
in an assessment of total error.
The concept of error propagation is important to reaeration
measurement for two reasons. First, the calculation of the re-
aeration rate from gas tracer data involves several mathematical
relationships. In order to completely assess the measurement
uncertainty of the technique, any experimental error, either
random or systematic, must be properly propagated through these
calculations. Second, it is important to recognize that mathe-
matical water quality models are equations. In a mathematical
water quality model with an analytical solution it is possible to
apply Equations 1 and 2 to the entire set of parameters and thus
determine the impact of parameter uncertainty on the water quality
model output. Although such an exercise is beyond the scope of
this paper, the recognition of the possibility of such a calculation
will help one understand the significance of the results presented
herein.
III. THE ERROR COMPONENTS OF REAERATION RATES AS
MEASURED BY DIRECT TRACER METHODS
A. General Discussion
Contributions to the error of measured reaeration rates come
from many parts of the measurement method. It is the purpose of
this section to enumerate the various error sources in the two
techniques. Then several examples quantifying some of these
component errors are given. These component errors will be related
to the overall measurement uncertainty examined later in this
paper (Section IV.).
The total error of direct tracer reaeration is broken down
into three separate areas for examination. First are errors
generated by the selection of a particular protocol. Second are
errors generated within the data analysis, particularly those
related to the reaeration model which is fit to the data, and to
the parameter estimation technique employed. Third are the random
experimental errors inherent in any measurement methodology.
Each of these areas is examined in detail below.
B. Protocol Selection Errors
Protocol selection errors are errors which are generated by
the selection of a particular measurement protocol. These errors
are related to the accuracy of the methodology. Among the accuracy
questions which arise are those concerning sampling, the nature
of the gas tracer, the nature of the conservative tracer, the
hydraulics of the test site, the gas tracer to oxygen transfer
ratio, and the temperature correction coefficient, 0.
38
-------
The sampling methodology employed by both techniques is
recognized not to be state-of-the-art dissolved gas sampling (1,
2, 3). It is possible with either technique to inaccurately
represent the instream gas transfer through either improper
tracer release or sampling location. Velten (7) demonstrated that
even "proper" sample handling and shipping resulted in some loss
of tracer gas in transit. Such a loss could affect the measured
rate.
The nature of Kr-85 appears to make it an ideal gas tracer
because it is inert and easily detected by liquid scintillation
counting. Ethylene and propane are not ideal gas tracers.
Abeles (8) has reviewed the biological activity of ethylene.
Dissolved ethylene gas may be consumed by microorganisms. In
addition Swimmerton and Lamontague (9) have measured very small
natural concentrations of the hydrocarbon tracers in ocean
waters.
Among the conservative tracers, tritiated water is an ideal
dispersion tracer for water. It behaves identically to the water
in which it is mixed. Rhodamine WT and other fluorescent dyes,
however, are not ideal conservative tracers. Smart and Laidlaw
(10) in their review of fluorescent dye tracers quantified many
of the potential losses of such tracers in natural systems
including photochemical decay, bleaching, pH effects, adsorbtion.
An example of the effect of loss of conservative tracer on the
measured reaeration coefficient is presented below.
The hydraulics of the test site enter the accuracy
considerations from two perpectives. First, there is a concern
over whether or not the path taken by the tracer slug is an
accurate representation of the entire stream cross section.
Second, there is the question of the accuracy of the hydraulic
steady state. As will be explained in Section III. D., in low
reaeration situations the tracer must be followed for quite a
length of time in order to ensure a precise measurement. Over
the time lengths in question (often multiple days), the accuracy
of the hydraulic steady state could limit the measurement
accuracy.
The accuracy of the ratio between the gas tracers and oxygen
transfer rates is well established both experimentally and
theoretically (1, 11, 12). However, it should be noted that no
experimental measurements of this critical ratio have been made
at low transfer rates (<1.0, I/day, base e).
The accuracy of the temperature correction coefficient, G,
has not been established. The multitude of values currently in
use demonstrates this fact (13). Indeed, the values used in the
two techniques are different;, 0 = 1.022 for the radiotracer
technique (1), 9 = 1.024 for the hydrocarbon technique (14).
39
-------
Consider this example of the effect of protocol selection
errors on reaeration rate measurement accuracy. The results of a
hypothetical hydrocarbon tracer study are presented in Table 1.
A single reach with three sampling stations, and triplicate deter-
minations of the tracer materials for each station were assumed.
The reaeration rate was calculated from these data using the two
parameter log transformation model, with a linear least-squares
parameter estimation scheme (discussed in Section III. C.). In
addition to the original data, the reaeration rate was then re-
calculated from modified data representing a 10% and 20% loss of
the conservative tracer relative to the original hypothetical
data. This conservative tracer loss was calculated by assuming
that the total loss had occured by Station 3 (See Table 1) , The
conservative tracer data for Station 3 was reduced by the total
percent dye loss. The conservative tracer data for Stations 1
and 2 were reduced by a fraction of the percent dye loss based on
the relative time to the peak as compared to Station 3. The
resulting calculated reaeration rates, with 95% confidence limits
(95% C.L.) are presented in Table 2. The 95% confidence limits
presented in this example reflect only the error associated with
the linear regression of the log transformed data. Error sources
in the calculation of k2 from k (ethylene) were not considered.
As can be seen from Table 2 , non-conservative characteristics of
the conservative tracer can seriously affect the accuracy of the
reaeration rate calculation.
C. Parameter Estimation and Reaeration Model Errors
Errors associated with the reaeration model and parameter
estimation raise some complex questions concerning the correct
way to fit the models to experimental data. The details of these
issues are beyond the scope of this paper, however references
presented in the discussion below should provide ample detail for
the interested reader.
There are many models for the desorption of the tracer gas
with time (and hence the absorption of oxygen with time). All of
the models used to estimate the reaeration rate are based on the
basic form given by Camp (15) in Equation 3 .
= k (Coo - C) (Eq. 3)
k = transfer rate
t - time
C = concentration of gas at time, t
GO, = concentration of gas at infinite time
ko
-------
4:
The integrated form of Equation 3 is presented in Equation
-oo
- C = (Coo - CQ) e
-kt
(Eq. 4)
C = concentration of gas at zero time
A common, logarithmic transformation of Equation 4 is shown in
Equation 5;
In (Co, - C) = In (C
co
-kt
(Eq. 5)
TABLE 1
Hypothetical Hydrocarbon Tracer Study Data
Peak Concentration (^g/1)
Station Ethylene
1 50.4
49.5
51.6
2 26.7
26.9
26.3
3 17.5
17.3
17.2
(pg/D
Dye
12.7
12.6
12.6
9.5
9.2
9.3
7.6
7.3
7.3
Ratio
Ethylene /Dye
3.97
3.93
4.10
2.81
2.92
2.83
2.30
2.37
2.23
In Ratio
1.379
1.369
1.411
1.033
1.072
1.037
6.833
0.863
0.802
Time to Peak
(days)
0.125
0.194
Oi292
TABLE 2
Calculated Reaeration Rates from Hypothetical
Hydrocarbon Tracer Study with Loss of
Conservation Tracer During the Study
Amount of
Conservative Tracer
(Dye) Lost
None (1>
10%
20% <2>
Calculated Reaeration
Rate with 95% C.L., n - 9
(I/day, 20C, base e)
3.7 + 6.8
3.1 jh 0.9
2.8 + 0.9
(1) Data as in Table 1
(2) % loss relative to data
in Table 1, lost over
0.292 days.
-------
Equations 4 and 5 are the models most commonly used to fit gas
tracer data. Gas transfer parameters may be obtained from the
data using Equation 4 via either a two or three parameter non-
linear regression. A two parameter, linear regression analysis
is used to obtain gas transfer parameters when Equation 5 is
applied. The two parameter methods have assumed values at C^ .
Both the hydrocarbon and the radiotracer technique often
employ Equation 5 and fit the data from two sampling stations
(upstream and downstream) to obtain two parameters (C and k)(l,
3). This is a trivial case of the linear regression ?it using
Equation 5. Because two data values are used to fit a two para-
meter model, there are no degrees of freedom left from which to
estimate the error in the parameters.
Rathbun e_t al. (12) discussed three separate estimation
methods for extensive hydrocarbon tracer data. Attempts were
made to use two and three parameter non-linear estimations with
Equation 4 as well as a two parameter linear estimation with
Equation 5. Rathbun1s final choice for the analysis of hydrocarbon
tracer data was a two parameter non-linear regression with Equa-
tion 4. The three parameter model was rejected because C^ could
not be precisely estimated from the data. The linear, two para-
meter model was not used because the non-linear, two parameter
model appeared to give a better fit of the data.
There is a process analagous to the direct tracer measurement
of k2 for which parameter estimation techniques have been care-
fully studied. This process is the unsteady-state, clean water
test for wastewater treatment system oxygen transfer. Because
both oxygen transfer and tracer gas transfer are governed by
Equation 3, the parameter estimation problems are similar. Refer-
ence to the extensive literature on parameter estimation for the
oxygen transfer test sheds some light on the proper method of
parameter estimation in direct tracer measurements.
Boyle, et al. (16), Brown (17) and Stenstrom, et al. (18)
extensively discuss the parameter estimation issues forThe oxygen
transfer test. Stenstrom et al. state that it is the opinion of
all the investigators cited, and of the ASCE sub-committee on
Oxygen Transfer Standards that the three parameter, non-linear
estimation is the prefered method of analysis. If non-linear
estimation is not possible, then a two parameter, linear estimation
with an iterative estimation of C^ should be used. This iterative
Coo estimate involves multiple applications of the linear
regression analysis with various estimates of €„, until a minimum
residual sum of squares is found.
The discussions by Boyle et al. (16), Brown (17) and
Stenstrom et al. (18) concerning the choice among the various
42
-------
parameter estimation possibilities primarily focus on an analysis
of the residuals from the regression. As observed by Rathbun
(12), the non-linear techniques (both two and three parameter)
resulted in lower and more uniformly distributed residuals for
the analysis of direct tracer data. However, the two parameter
technique is inappropriate unless an iterative procedure is used
to estimate C^ , the third parameter. Because of the expense and
difficulty in applying a non-linear regression analysis, the
linear regression of log transformed data has been recommended as
the technique of choice for the two parameter estimation. The
complications of an iterative, non-linear regression analysis do
not seem justified.
It should be noted that neither of the direct tracer
methods currently employ the recommended parameter estimation
procedures. When the linear, two parameter estimation is used, C
is assumed rather than estimated by iterative methods. When non-
linear programming is applied, only a two parameter estimate is
supported by the data. The probable reason in both cases for the
inability to estimate Coo is insufficient data collection in the
vicinity of equilibrium.
The issue of the data requirements for proper parameter
estimation is an important one. Guidelines have been proposed to
the ASCE concerning the proper protocol for data collection to
allow the estimation of K a via the clean water unsteady-state
test. These guidelines are presented here because they are also
believed to be applicable to the direct tracer measurement of k_.
The tracer experiment should be followed for a time period
not less than 4/k2 (98% gas transfer). In that period there are
three critical regions for parameter estimation, the areas near
zero time, around l/k_, and approaching infinite time. In order
to ensure that each or these regions is appropriately represented
it is recommended that a minimum of 10 data points be collected
during each tracer study. Those data points should be
distributed with two-thirds between zero and 2/k2, evenly spaced;
and one-third between 2/k2 and 4/k2, evenly spaced. An
experimental protocol for the direct tracer measurement which
includes this amount of data would certainly enhance one's
ability to estimate the reaeration rate.
On the practical side, tracking the tracers to 4/k2 may be
problematic. For a k2 of 0.1 (l/dayf base e), the recommended
guidelines require the experiment to be conducted for a time of
40 days. Such a time period for testing is clearly not possible.
Dilution of the tracers by dispersion can severely attenuate
their concentrations. There are practical restrictions on the
amount of tracer which can be "instantaneously" released (as well
as legal restrictions in the radiotracer case). It is generally
wise to make the appropriate dilution/dispersion calculation
-------
prior to the tracer release. An insufficient tracer release, or
tracking of the tracers for less than the recommended length of
time, will result in an increase in the error of the estimated
K,- •
The level of error associated with the improper application
of the parameter estimation techniques is difficult to quantify.
Much of the error hinges on the quantity and quality of the direct
tracer data collected. In the oxygen transfer work, however, the
error has been quantified, and has often been found to be sub-
stantial (16, 17).
D. Random Experimental Error
Random experimental error is an inherent part of any measure-
ment procedure. The component errors associated with random
experimental error all contribute to the precision of a measurement
technique. Often one or more of these errors will limit the
applicability of a measurement technique due to the error's mag-
nitude.
Random experimental errors enter the uncertainty calculations
for reaeration measurements from several sources. Each of the
analytical methods (liquid scintillation counting, gas chromotography,
fluorometry) has an inherent measurement precision. Cohen, et al.
(19) presented data from which the liquid scintillation counting
precision has been ascertained. In the range of values commonly
encountered in the radiotracer technique, the liquid scintillation
counting has a precision of +1% to +4% per channel. Shultz, et
al. (20) analyzed data from the gas chromatographic procedure
used in the hydrocarbon technique to determine its precision.
Schultz found that the gas chromatography technique had a co-
efficient of variation in the range of ±1.3% to +6.9% of the
measured tracer gas concentration. ~
The precision of several of the constants used in the cal-
culation of k_ from the tracer rate coefficient has been presented
in the literature. Tsivoglou (1, 2) found the ratio between the
krypton-85 transfer rate (k ) and k2 to be 0.83 + 0.04 (s.d.).
Rathbun, et al. (12) presented the precision of the ratio between
the ethylene transfer rate (k ) and k2/ and the ratio between the
propane transfer rate (k ) ana k2« k : k2 was found to have a
precision of 1.15 + 0.023 (95% C.L.); k : k2 was found to have a
precision of 1.39 + 0.028 (95% C.L.). Tsivoglou (1) also pre-
sented the precision of the temperature correction coefficient 0
as 1.022 +0.004 (s.d.). Because there are many values for this
coefficient presented in the literature, their precision may be
of little consequence relative to the question of the accuracy of
0. In order to demonstrate how these various experimental pre-
cisions can limit the applicability of a measurement procedure,
an example calculation using the radiotracer method is presented
below.
-------
As mentioned before, the liquid scintillation counting tech-
nique has a precision in the range of +1 to +4% (95% C.L.) for
each counting channel (see Cohen, et al_. , 197. Applying the
random error propagation formula, Equation 1, to the calculation
method presented in Cohen et al. (19), this precision translates
to a precision of about +7% (95% C.L.) in the precision of the
krypton (Kr) to tritium JT) ratio at each sampling station. This
results in a precision of +9% (95% C.L.) in the ratio between the
Kr : T at the downstream sampling point and the Kr : T at the
upstream sampling point. k is then calculated from the sampling
point ratios via a regression analysis. However, because the
model, Equation 3, is exponential, the precision from upstream to
downstream does not directly translate to the precision of k .
The precision of k is dependent on the magnitude of the ratio
between (Kr : T) downstream and (Kr : T) upstream. This effect
is demonstrated in Table 3.
Table 3 Theoretical Precision of k
Kr : T downstream % of Tracer Gas Precision of
Kr : T upstream Lost k
KIT
0.1 90% + 4%
0.5 50% +13%
0.9 10% +90%
Since the ratio between the downstream and upstream (Kr : T)
is directly related to the amount of tracer gas lost from the
stream during the study, then the precision of k is also related
to the amount of tracer gas lost. The amount of Eracer gas lost
from the stream is dependent on two things, the magnitude of k
and the length of time between the upstream and downstream sampling
points. Because k^ is related to k» via a constant of known
precision (0.83 + 5.04, 95% C.L.), it is possible, from the theo-
retical precision of k , to determine the length of time a Kr-85
tracer study must be followed in order to achieve a desired pre-
cision, given an approximate knowledge of the oxygen transfer
coefficient. Some results of this calculation are presented in
Table 4 (21). Ultimately, the precision of the Kr-85 technique
is limited by ones ability to follow the dye tracer to detect the
tracer slug, and by the precision of the 0.83 k., : k0 ratio.
K IT Z
-------
Table 4 Length of Time a Kr-85 Tracer Study
Must Be Followed to Achieve a Desired
Precision Based on Theoretical Considerations
Approximate k? Time Necessary Time Necessary
to Be Measures to Achieve^ to Achieve+
(I/day, base e) Precision - 50% Precision - 10%
1.0 3.9 hours 22. hours
0.5 7.8 hours 1.8 days
O.i 1.6 days 9.2 days
0.05 3.2 days 18. days
IV. ESTIMATION OF UNCERTAINTY IN ACTUAL
TRACER MEASUREMENTS
A. General Discussion
The preceeding review of the error components of the direct
tracer measurement of reaeration rates has raised some important
questions concerning measurement uncertainty. Because of the
questions which have been raised, an assessment of the
measurement uncertainty of the entire direct tracer measurement
process is warranted. To accomplish this assessment, tracer
studies described in the literature as well as studies conducted
on the Ouachita River Basin in Arkansas and Louisiana by NCASI,
Law Engineering and the USGS will be reviewed.
There are several ways to assess the uncertainty of a measure-
ment technique. If replicate measurements are available (pre-
ferably three or more) for controlled experimental conditions,
then an examination of the variability among the replicates can
lead to an understanding of the precision of the measurement. In
order to isolate the sources of the random error, it is important
to know all the conditions of the experiment, as well as how well
they were controlled. Control of experimental conditions is
particularly difficult in environmental measurements. Nevertheless/
an estimate of the experimental precision of both the radiotracer
and the hydrocarbon techniques at moderate to high transfer rates
(k9 > 1.0, I/day, base e) can be made from published replication
experiments. This analysis is presented in Section IV. B.
-------
A second method of evaluating the uncertainty of a measure-
ment technique is to start with the raw data and conduct the
entire parameter estimation procedure, statistically analyzing
the confidence limits during each calculation step. This pro-
cedure was followed for data collected from three radiotracer
studies conducted by NCASI and Law Engineering on the Ouachita
River Basin, as well as for one hydrocarbon study conducted by
the USGS in the same basin. These studies involved reaeration
rates in the range of 0.02 - 1.0 (I/day, base e). The results of
the measurement uncertainty for these studies based in parameter
estimation statistics and error propagation through the calculations
are presented in Section IV. C.
The evaluation of the measurement uncertainty of the direct
tracer methods from the literature and the Ouachita River Basin
studies is primarily an analysis of precision. The accuracy of
the direct tracer methods is more difficult to assess because the
techniques represent the current state-of-the-art in k~ measure-
ment. Currently, by definition, they are accurate. Tfiis is not
to say that there is no bias in the direct tracer measurement
techniques. The sources of bias enumerated in Section III. B.
illustrate that measurement bias should be a concern when these
techniques are used. However, without a standard for comparison,
the quantification of the accuracy of these techniques is not
possible.
B. Replication Studies From the Literature
A major replication study was conducted early in the testing
of the radiotracer technique on the Jackson River near Coving ton,
Virginia (1). In that study a k2 of the magnitude of 1.5 (I/day,
base e) was measured for 14 subreaches from seven tracer releases
spread out over 14 days. Between two and three replicate deter-
minations were made for each subreach. During the course of the
study, the hydraulic stability of the stream was rather good; the
flow varied +10% over the 14 days. Among the 14 sets of replicate
determinations the coefficient of variation for the replicates
ranged from + 1.0% to +25.8%. The mean coefficient of variation
was +11.9%. It should be noted that these coefficients of varia-
tion~represent uncertainty in both the measured value and in the
hydraulic stability of the stream segment during the replication
study.
There is limited data available for the assessment of
uncertainty in the hydrocarbon technique literature. However, one
study designed to compare the hydrocarbon technique and the
radiotracer technique does provide some replicate information
(14). This study was conducted on two small streams in Wisconsin.
In each stream a double gas tracer dump (ethylene and propane) was
made and the reaeration rate was computed via two different
parameter estimation methods for each of the tracer gases. This
-------
gave a total of four possible replicate k~ determinations for each
reach. A total of four reaches were studied. It should be noted
that because the replicate determinations for each reach were
calculated from a single tracer release the question of hydraulic
stability is not as serious in this study as it was in the Jackson
River study previously discussed. The coefficient of variation
among the three or four replicates for each reach ranged from
+4.3% to +13.8%; the average coefficient of variation was
about +7.5%. The magnitude of k measured in this study was
about 7.5 (I/day, base e). The variability calculated from this
hydrocarbon work can be attributed to random experimental error
and the variability introduced by the two different parameter
estimation techniques.
C. Ouachita River Basin Studies
Replication of direct tracer studies in order to determine
the measurement uncertainty is rarely possible because of the
cost of such studies. Reaeration rate uncertainty estimates must
often be based on a single tracer study. Fortunately, statistical
methods may be applied to the regression analyses used to estimate
the gas transfer rate from direct tracer data allowing one to
estimate the confidence range of estimated parameters. The un-
certainty thus calculated may then be propagated through the
calculation necessary to compute k2, giving an estimate of the
error in k*. Such calculations have been made in radiotracer
data collected by NCASI and Law Engineering and on hydrocarbon
data collected by the USGS, on the Ouachita River Basin in the
summer of 1980. All data were analyzed using the two parameter,
linear regression model based on Equation 5. The results of this
uncertainty analysis are discussed below.
As part of NCASI's MWQM program, three radiotracer studies
were conducted in the Ouachita River Basin by NCASI and Law Engi-
neering (21). The reaeration coefficients for a total of seven
stream reaches were estimated from these tracer studies Figure
_! presents an example of the data collected with the results of
the linear, log transformation parameter estimates. The 95%
confidence limits from the parameter estimation statistics for
each of the seven reaeration coefficients were propagated through
the conversion calculations from k (T, I/day) to K, (20C, 1/dav)
(via the ratio = 0.83 + 0.04 and 0Kr= 1.022 +0.01). Table 5 presents
the estimated reaeration coefficients for the Ouachita River
Basin studies with their confidence limits. Also presented in
Table 5 is the theoretical precision of the measured k based on
the propagation of the random error associated with liquid scinti-
llation counting. A discussion of these theoretical calculations
may be found in Section III. D. A comparison of the 95% confidence
limits (95% C.L.) with the theoretical precision calculations
implied that the precision of the liquid scintillation counting
technique was a ma^or component of the measurement uncertainty of
-------
the reaeration coefficients estimated in this study. It also
appears that the two reaeration coefficient measurements with low
gas loss (0.02 +1000% and 0.13 +800%, I/day, base e, 95% C.L.)
are of limited utility because of the magnitude of their uncertainty.
The five tracer measurements with greater than 25% gas loss all
display acceptable error.
— Slope = k
Kr
2
O
(J
2
O
u
5
3
K
t-
O
H
o.
>
K
(33 Cj= 0.18 ±0.02
(I/DAY. BASE e. 95% C.L.)
0.2
B-H stations
OBSERVED KrTT RATIO WITH MEAN
O.8
1.2
1.6
2.0
TIME (days)
FIGURE 1
-------
Table 5 Uncertainty of Reaeration Coefficient, k_, as Measured by
The Radiotracer Technique, and Comparison With Theoretical
Precision of The Radiotracer Technique Based on Gas Loss
Measured
95% Confidence
Study
Area
OUACHITA RIVER
Northeastern La.
TRIBUTARY
OUACHITA RIVER
Southeastern Ark.
TRIBUTARY
TRIBUTARY
k9, 20C Range of k. % Tracer Gas
(I/day, base e) (as a % of k,) Lost During Study
0.02
0.13
0.17
+ 1000%
(2)
0.34 to 0.37 +
0.91 to 1.06 +
800%
18%
14%
6%
2.8
29
28 to 30
54 to 89
Theoretical Precision
of k. Based on Tracer
Gas Loss (as a % of k2)(l)
+ 200%
+ 300%
+ 27%
+ 30% to + 25%
+ 12% to + 4%
(1) See explanation in Section III. D.
(2) Technique may have had interference due to partial withdrawal of
tracer material by power plant intake.
At the same time that NCASI and Law Engineering conducted
the radiotracer study on the Ouachita River in Southeastern
Arkansas, the USGS also conducted a hydrocarbon study on the same
river reach. NCASI has attempted to quantify the uncertainty of
this hydrocarbon measurement at the low transfer rate/ and to
compare the radiotracer and the hydrocarbon measurements of the
reaeration coefficients for that river reach (22). Unfortunately,
sampling problems and limited data collection severely restricted
the information available to estimate the hydrocarbon transfer
rate. The final value for k_, 20C computed by NCASI using several
assumptions about the data was 0.44 +0.98 (l/dayf base e, 95%
C.L.). The error from this calculation did not compare favorably
with the radiotracer error (0.17 +0.03, I/day, base e, 95% C.L.).
The primary reason for the discrepancy in error between the
two techniques was probably the sampling and data collection
differences. The radiotracer reaeration coefficient estimate was
based on 24 data points distributed over 8 sampling locations.
The hydrocarbon reaeration coefficient estimate was based on
three data points from three sampling locations, one of which was
highly suspect. This comparison between the two levels of sampling
and data collection clearly illustrates the importance of the
quantity and quality of data in reaeration coefficient parameter
estimation (see Section III. C.). The eight sampling station
data collection scheme (radiotracer work) with triplicate deter-
minations at each station (24 data points) resulted in an estimate
of k, to +18% (95% C.L.). The three sampling station data (hydrocarbon
workT with one highly suspect data point (3 data points) resulted
in an estimate of k to +230% (95% C.L.).
50
-------
V. SUMMARY
The quantification of the uncertainty of direct tracer
measurements of reaeration coefficients is important for several
reasons. Reaeration rates play an important role in dissolved
oxygen budget calculations used in water quality management
decisions. Because of the cost of direct tracer measurements, a
cost/benefit analysis comparing the expected knowledge to be
gained from the tracer study to the tracer study cost is usually
in order.
The error components in the direct tracer measurements fall
into three broad classes.
1) Protocol selection errors (accuracy considerations)
include sampling methodology and location, the
nature of the gas conservative tracers, the accuracy
of 9 and the tracer gas to oxygen transfer ratios,
and consideration of the hydraulic state of the testing
site.
2) Parameter estimation and reaeration model errors
involve both the estimation technique and the data
requirements. The preferred estimation technique is
a three parameter, non-linear regression estimation
k, C and CQO • A second estimation method is a two
parameter, linear estimation on log transformed data,
where C and k are fit directly and Cm is iteratively
estimatid. The data requirements for these estimations
procedures is a minimum of 10 data points, two-thirds
evenly spared between 0 and 2/k_, one-third evenly spaced
between 2A2 and 4/k2.
3) Random experimental errors (precision considerations)
include the precision of the analytical methodologies
(liquid scintillation counting, gas chromatography,
fluorometry), the precision of the transfer rate ratios
and the precision of 0.
The significance of the propagation of random error through
the reaeration coefficient calculation was demonstrated via an
analysis of the effect of the amount of tracer gas lost on the
reaeration coefficient's theoretical precision. In addition,
theoretical calculations showing the length of time a radiotracer
study must be followed in order to achieve a desired precision in
the measured rate were presented.
Actual direct tracer reaeration measurements were analyzed
for uncertainty. Errors in the radiotracer technique measurements
ranged from +1.0% to +1000% (the former from replication studies
at a moderate rate, the latter from a single low rate measurement
on the Ouachita River). Errors in the hydrocarbon technique
51
-------
measurements ranged from +4.3% to +230% (the former from the
Wisconsin replication study, the latter from a single measurement
on the Ouachita River). The relationship between the amount of
tracer gas lost and the measurement uncertainty of the Ouachita
River Basin radiotracer measurements was found to parallel the
theoretical precision calculations.
The large uncertainty associated with several of the direct
tracer measurements has raised serious questions about the utility
of these measurements in low transfer environments. The uncertainty
of the hydrocarbon tracer measurement from poor and insufficient
sampling and of the radiotracer measurement from the inherent
analytical precision demonstrates the need to calculate potential
errors prior to conducting such studies. If it is impossible to
improve the experimental protocol to reduce these errors to an
acceptable range, then consideration should be given to not conducting
the studies. In such situations empirical equations may produce
reaeration coefficient estimates having comparible precision to
those from the direct tracer measurements.
VI. REFERENCES
1. Tsivoglou, E.C., "Tracer Measurements of Stream
Reaeration," FWPCA, Dept. of the Interior,
Washington, D.C. (1967).
2. Tsivoglou, E.G., Cohen, J.B., Shaerer, S.D., and
Godsil, P.J., "Tracer Measurements of
Atmospheric Reaeration. II. Field Studies,"
JWPCF 4£:285 (1968).
3. Rathbun, R.E., Shultz, D.J., and Stephens, D.W.
"Preliminary Experiments With a Modified
Tracer Technique for Measuring Stream
Reaeration Coefficients," USGS Open File Report
75-256, Bay St. Louis, Miss. (1975).
4. Rathbun, R.E., Shultz, D.J., Stephens, D.W.,
and Tai, D.Y. "Experimental Modeling of
the Oxygen Absorption Characteristics of Streams
and Rivers," International Association for Hydraulic
Research, 17th Congress, 1^A61 (1977).
5. Mandel, J., The Statistical Analysis of Experimental
Data, Interscience Publishers, New York, N.V. (1964),
6. Serth, R.W., Hughes, T.W., Opferkuch, R.E.,
and Eimutis, E.C. "Analysis of Uncertainty -
Principles and Applications, "USEPA - 600/2-78-004U
(1978).
7. Velton, R.J., "Laboratory Procedures," Symposium
on Direct Tracer Measurement of the Reaeration
Capacity of Streams and Estuaries, Water Pollution
Control Research Series, 116050, USEPA (1972).
8. Abeles, F.B., Ethylene in Plant Biology, Academic
Press, New York, N.Y. : 302 (1973).
52
-------
9. Swinnerton, J.W. and Lamoutague, R.A., "Oceanic
Distribution of Low-molecular-weight Hydrocarbons;
Baseline Measurements, EST 8,7:657 (1974).
10. Smart, P.L., and Laidlaw, I.M.S., "An Evaluation
of Some Fluorescent Dyes for Water Tracing,"
Water Resources Research 13,1:15 (1977).
11. Tsivoglou, E.G., O'Connell, R.L., Walter, C.N.,
Godsil, P.J. and Logsdon, G.S., "Tracer
Measurements of Atmospheric Reaeration - I.
Laboratory Studies," JWPCF 37,10:1343 (1965).
12. Rathbun, R.E., Stephens, D.W., Shultz, D.J.,
and Tai, D.Y., "Laboratory Studies of Gas
Tracers for Reaeration," JEEP, ASCE 104,EE2;215
(1978).
13. Zison, S.W., Mills, W.B., Deimer, D., and
Chou, C.W., "Rates, Constants, and Kinetic
Formulations in Surface Water Quality Modeling,"
USEPA - 600/3-78-105 (1978).
14. Rathbun, R.E. and Grant, R.S., "Comparison of the
Radiotracer and Modified Techniques for Measurement
of Stream Reaeration Coefficients," USGS Water-
Resources Investigations 78-68 (1978).
15. Camp, T.R., Water and Its Impurities, 4th Printing,
Reinhold Publishing Co., New York, N.Y. (1968).
16. Boyle, W.C., Berthouex, P.M. and Rooney, T.C. "Pitfalls
in Parameter Estimation for Oxygen Transfer Data,"
JEEP, ASCE 100,EE2;391 (1974).
17. Brown, L.C., "Oxygen Transfer Parameter Estimation,"
Proceedings, Workshop Toward an Oxygen Transfer
Standard, USEPA - 600/9-78-621 (1979).
18. Stenstrom, M.K., Brown, L.C., and Hwang, H.J., "Oxygen
Transfer Parameter Estimation," JEEP, ASCE 107,EE2, :
379 (1981).
19. Cohen, J.B., Setsen, J.L., Kelley, W.D., and
Shearer, S.D., Jr. "Determination of H and
Kr in Aqueous Samples by Liquid Scintillation
Techniques," Tanlanta 15:233 (1968).
20. Schultz, D.J., Pankow, J.F., Tai, D.Y., Stephens,
D.W., and Rathbun, R.E. "Determination, Storage,
and Preservation of Low Molecular Weight Hydrocarbon
Gases in Aqueous Solution," Jour. Research U.S.
Geol. Survey 4,2;247 (1976).
21. Whittemore, R.C. and Hovis, J.S., "A Review of
Reaeration Capacity Estimation and Its Measurement
Uncertainty," NCASI Technical Bulletin, New York,
N.Y. (pending, 1982).
22. Whittemore, R.C. and Hovis, J.S., "A Comparison of
Reaeration Estimation Techniques for the Ouachita
River Basin," NCASI Technical Bulletin, New York,
N.Y. (pending, 1982).
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
53
-------
CALIBRATION OF HYDROLOGY AND
SEDIMENT TRANSPORT ON SMALL AGRICULTURAL
WATERSHEDS USING HSPF
By
David E. Schafer 1
David A. Woodruff 1
Richard J. Hughto2
G. K. Young3
INTRODUCTION
The ability to accurately predict the hydrologic response and sediment
movement on agricultural watersheds through mathematical simulation modeling
can play an active role in the development of sound agricultural management
practices. Specifically, it provides the agricultural scientist with a
valuable yet inexpensive means to assess non-point source runoff potential and
its resulting sediment transport capacity. Coupling this with knowledge
pertaining to the physio-chemical characteristics and transport mechanisims
associated with agricultural chemicals provides a basis for evaluating
receiving water quality impacts.
This paper discusses the capabilities of the Environmental Protection
Agency's Hydrologic Simulation Program-Fortran (HSPF) for modeling runoff and
sediment transport on three small agricultural watersheds: Mississippi 802;
Oklahoma C-4, and; Oklahama C-5. Following a brief description of each a
general calibration methodology is outlined and simulation results are
presented. Direct comparisons of hydrologic and sediment simulation results
are made utilizing monthly and annual totals, double mass ananlyses, monthly
standard deviations and coefficients of variation. Fi-nally, relationships
among physically based parameter value estimats are investigated.
WATERSHED DESCRIPTIONS
Watershed 802, located in the Mississippi Delta Region, covers 15.5 ha.,
and is comprised of a Sharkey Silty Clay. The land has been mechanically
formed to a slope of 0.2 percent and planted with one meter row spacing. The
drainage pattern, designed to direct runoff via turn-rows or shallow
V-ditches, has been equipped with instrumentation for flow monitoring and
sampling. During the 1974 and 1975 calibration period, cotton was the sole
crop grown and harvested.
*Water Resources Engineer, Camp, Dresser & McKee, Inc., Boston, MA
2Senior Water Resources Engineer, Camp, Dresser & McKee, Inc., Boston, MA
President, GKY & Associates, Inc., Springfield, VA
-------
Oklahoma Watersheds C-4 and C-5 are representative of cropland of the
Central Great Plains. C-4 is 12.1 ha. in size and has also been planted with
cotton. C-5, situated approximately one kilometer from C-4, covers an area of
5.2 ha., and has been continuously planted with winter wheat. Man-made ridges
form the watershed boundaries and each has been graded and smoothed to a slope
of 0.3 percent. Soil types are similar and are characteristic of those found
in alluvial bottom land deposits of clay and silt loams. The two-year
calibration period selected for these watersheds extends from January, 1973,
through December, 1974.
Annual tillage operations performed on each watershed are considered
typical with respect to crop type and location. For Watershed 802 these
included shredding cotton stalks after harvest, disking and forming rows in
late winter or early spring, followed by herbicide application, planting,
cultivation, and application of pesticides during the growing season. In
addition to the above, the cotton crop on C-4 was irrigated during the 1974
cropping season. On C-5, dryland wheat cultivational practices included disk
harrowing in the summer, followed by planting in early winter and harvest in
late spring.
Field Data
Climatological data, used as the driving force for model simulation,
consisted of 15 minute precipitation volumes and daily evapotranspiration
measurements. The rainfall data for Watershed C-4 were adjusted to include
irrigation applications on appropriate dates. Model calibration data
consisted of monthly total runoff volumes and sediment mass.
Annual precipitation totals for the calibration periods were 1732 mm and
1467 mm for Watershed 802 and 1023 mm and 714 mm for Watersheds C-4 and C-5.
Compared with 30 year average annual values, Watershed 802 received 38% and
17% above normal rainfall; Watersheds C-4 and C-5 received 28% above normal
rainfall in 1973 followed by 10% below normal rainfall in 1974. This is of
particular interest with respect to the calibration of Watershed 802, since,
significant deviations from average annual conditions during both hydrologic
and sediment calibration can have significant impact, should the model be used
for prediction. There is a general need, therefore, to verify the calibration
variables using precipitation data that approach average conditions over the
simulation period prior to utilization- of calibrated parameter values in a
predictive mode.
CALIBRATION PROCEDURE
Hydrologic and sediment simulation was conducted utilizing the PWATER and
SEDMNT sections of the PERLND module of HSPF. The PWATER algorithms in HSPF
are designed to continuously simulate the hydrologic processes occuring on a
watershed. For the purposes of this calibration, the major component of
interest is total overland flow. The SEDMNT algorithms in HSPF simulate the
production and removal of detached fines. Since sediment simulation is
55
-------
dependent upon the results of the hydrology simulation, model calibration was
carried out in two stages, which involved calibration of PWATER parameters
followed by the SEDMNT parameter calibration on each watershed.
For each stage of the calibration, an eight-step procedure was followed:
1. Identify and compile appropriate watershed and field
measurement data;
2. Determine appropriate target values and time scales for
calibration based on available field data;
3. Select initial values of section module parameters based on
reported watershed characteristics;
4. Identify good results from previously calibrated modules;
5. Make an initial simulation run based on selected parameter
values, compare model output to field measurements, and
establish a base-case;
6. Identify the set of primary calibration variables;
7. Make a series of model simulations adjusting, as
appropriate, the primary calibration parameter values from
the base-case to minimize the error between simulated
results and measured data, and;
8. Interpret the results of final calibration and sensitivity
analyses to facilitate model verification and application.
Hydrologic Calibration
Initial estimates for values of the PWATER parameters for each site were
developed from a review of available field data, the HSPF User's Manual (1),
guidelines provided in the ARM User's Manaul (2), previous algorithm
calibration experience (3) and several site related publications
(4,5,6,7,8,9,10,11). A preliminary screening of measured monthly
precipitation and runoff data revealed that the hydrologic response of each
watershed can be reasonably described on a monthly basis, however, no
dlscernable trends were apparent from the data with respect to the annual
cropping cycle.
Calibrated parameter estimates for the PWATER module are listed in Table
1. Simulation results for each watershed are presented graphically 1n Figures
1, 2, and 3. As shown by the dashed lines 1n the doulble mass analysis .plots,
overall model performance for the PWATER module was good in each case. Early
periods of undersimulation resulted in underestimation of the period of record
totals, however, monthly values generally follow a linear trend with a slope
parellel to the line of one to one correspondence of measured versus simulated
56
-------
TALBE 1
SUMMARY OF CALIBRATED PWATER PARAMETER ESTIMATES
Parameter
802
.C-4
C-5
NBLKS
LZSN
INFILT
LSUR
SLSUR
KVARY
AGWRC
FOREST
PETMIN
PETMAX
INFEXP
INFILD
DEEPFR
BASETP
AGWETP
INTFW
IRC
CEPS
SURS
UZS
IFWS
LZS
AGWS
GUVS
UZSN
CEPSC
NSUR
LZETP
1
2.0
0.005
500
0.002
0 **
1.0 **
0 **
35 **
40 **
2.0
1.0
1.0 **
0 **
0 **
0.7
0.01
0.001 *
0.001 *
0.15
0.001 *
2.0
0.001 **
0.001 **
0.05-0.30
0.05-0.25
0.20
0.05-0.40
1
2.5
0.15
650
0.003
0 **
1.0 **
0 **
35 **
40 **
2.0
2.0
1.0 **
0 **
0 **
0.6
0.01
0.001 *
0.001 *
0.3
0.001 *
3.0
0.001 **
0.001 **
0.05-0.30
0.10-0.25
0.15-0.25
0.10-0.25
1
2.5
tol&
300
0.003
t) **
i.o **
o **
35 **
4*0 **
2.0
2.0
ud **
0 **
o **
0.6
0.01
o.ooi *
o.ooi *
0.20
0.001 *
s.d
d.odi **
d.ooi **
0.15-0.40
0.10-0.20
0.15-0.20
o.id-o.2d
* Model default values required for program execution.
** Model-default values required for progarm execution but not a/pplicable
to the Oklahoma Watershed application of HSPF.
57
-------
U
N
-
i
N
M
M
300 -
240
20
60
0
n
FIGURE 1
HYDROLOGIC CALIBRATION RESULTS
FOR WATERSHED 802
ill
I I
MAR MAY JUL SEP NOV JAN MAR MAY JUL SEP NOV JAN
APR JUN AUG OCT DEC FEB APR JUN AUG OCT DEC FEB
MEASURED
SIMULATED
MONTH Cl974-I 976)
Line of one-to-one correspondence
Line parallel to line ol one-to-ont correipoodanct
200
400 600 BOO IOOO
MEASURED RUNOFF (millimeleri)
I40O 1600
-------
100
I
1
•..'
F
i
m
80
40 ~
20
0
FIGURE 2
HYDROLOGIC CALIBRATION RESULTS
FOR WATERSHED C-4
Jb
JAN MAR MAY JUL SEP NOV JAN MAR MAY JUL SEP NOV
FEB APR JUN AUG OCT DEC FEB APR JUN AUG OCT DEC
MEASURED
SIMULATED MONTHS Cl973-I 974)
T
Une of one-to-oni corriipondenca
Line parallel lO'llMOf OM-to-OM correspondervc*
80 120 160 200
MEASURED RUNOFF (mlllimettri)
240
280
59
-------
F
'•:
F
I
m
80
60
40-
FIGURE 3
HYDROLOGIC CALIBRATION RESULTS
FOR WATERSHED C-5
In
I I I I I I I I I I I I I Mil! I
JAN MAR MAY JUL SEP NOV JAN MAR MAY JUL SEP NOV
FEB APR JUN AUG OCT DEC FEB APR JUN AUG OCT DEC
MEASURED
SIMULATED
Z50
MONTHS CI973-I974)
Lift* of on«-to-on» corre»pond«nct
S ^Lina parallel to lin* of on»-«o-ont corrtipondtnc*
40
BO 120 160
MEASURED RUNOFF (mllllmtUri)
ZOO
240
60
-------
results.
For Watershed 802, poor simulation during May and June, 1974 is
attributed to extreme hydrologic conditions occuring on the watershed.
Precipitation measured 365 mm and 237 mm, respectively, compared with thirty
year monthly mean values of 120 mm and 77 mm. Runoff for these months
measured 266 mm (73% of rainfall) and 159 mm (67%) whereas simulation totals
were 204 mm (56%) and 120 mm (51%). These discrepancies accounted for
approximately 80% of the total annual error in 1974.
The slightly more erratic behavior of the model in simulating runoff for
the Oklahoma watersheds is primarily attributed to the characteristics of
major runoff producing storm events. Review of the precipitation data
revealed that the Oklahoma sites were frequently subjected to storms of high
intensity and short duration. Simulation of these events using HSPF, however,
is extremely difficult due to the non-convergent nature of the model's
algorithims for cases of intense precipitation. This was found to be
particularly true for small agricultural watersheds where annual tillage
practices can drastically impact hydrologic response on a storm level of
detail.
For each calibration, parameter sensitivity runs indicated that annual
results were most sensitive to the values of INFILT, the infiltration capacity
coefficient, and UZSN, the upper zone nominal storage capacity parameter.
More pronounced, however, was the impact that distributed monthly UZSN values
had upon the annual distribution of runoff. Direct variation of UZSN on a
monthly basis to reflect the effects that soil type, tillage practices and
antecedent soil moisture conditions have on soil moisture field capacity
allows for a better representaton of annual variations in the physical system
simulated. Parameters not found to significantly impact simulation results
included: NBLKS, LSUR, SLSUR, INFEXP, INFILD, IRC, and NSUR.
Statistical comparison of simulation results can be made utilizing
computed coefficients of variation, monthly runoff totals, and monthly
standard deviations. For the six watershed years simulated, period of record
coefficients of variation were 0.30, 0.58, and 0.71 for Watersheds 802, C-4,
and C-5. Corresponding mean monthly runoff totals were 61 mm, 12 mm, and 10
mm, respectively. Average monthly standard deviations* of simulated versus
measured runoff were 18.3 mm for Watershed 802, 7.0 mm for C-4, and 7.1 mm for
C-5. These results indicate that, although the best overall simulation was
achieved on Watershed 802, simulation on the Oklahoma sites was also good.
Sediment Calibration
Calibration of parameters in the SEDMNT module of HSPF followed the
general procedure previously outlined. From step four, since simulation
results of PWATER serve as direct input to SEDMNT, particular attention was
focused on the possibility of error transfer between modules. Testing this
hypothesis involved an examination of the natural association between runoff
and sediment yield for each watershed under cropped and follow conditions.
Through logarithimic linear regression, correlation coefficients were
61
-------
calculated as 0.92 and 0.91 for Watershed 802, 0.84 and 0.94 for C-4 and 0.38
and 0.98 for C-5. These correlations imply that a good hydrologic simulation
is an important pre-requsite to an accurate simulation of sediment removal.
All analyses performed, however, indicated that PWATER results would not
significantly limit SEDMNT parameter calibration.
The calibrated SEDMNT parameter estimates for each watershed are listed
in Table 2. Since tillage operations during each cropping season were
frequent, the models' "special actions" feature was utilized to reset the
storage of detatched fines (DETS) to an amount representative of newly tilled
soil. All major activities occuring on each watershed were assumed to affect
the supply of detached fines to the same degree, therefore, a single value for
DETS was used at each soil disturbance. Adjustment of this parameter in an
effort to simulate peak sediment production months facilitated calibration of
remaining variables.
Following establishment of the value for DETS, sediment parameters KSER
and JSER, the coefficient and exponent in the transport equation, were
determined to be most sensitive during and immediatly after months in which
tillage operations occured. During periods when tillage did not occur,
simulation results were sensitive to the values of detachment algorithim
parameters. Gully erosion was considered negligible in each watershed.
Sediment simulation results for each watershed are graphically
illustrated in Figures 4, 5, and 6. The double mass analysis plots for
sediment indicate, as did the hydrology plots, that undersimulation in the
initial months resulted in underestimation of the two year totals. Overall
simulation for months in which a substantial amount of sediment was measured
appear to ge good. This fact is shown by observing that those months for
which the cumulative mass curves are increasing significantly, the trend is
toward a line with a slope approximately equal to one. The erratic behavior
exhibited during periods in which very little sediment was recorded reveals
that equally distributed over and undersimulation of the data occured. Given
the relative magnitudes of these values, this erratic behavior is not
considered significant.
Sediment simulation results for each watershed were, however, found to be
closely related to hydrologic simulation results. This is best illustrated
through review of the monthly totals in bar graph form. The correspondence
between months in which over or undersimulation occured for both hydrologic
and sediment results existed in approximately 80% of the months in which there
were measurable differencees in their respective totals.
Coefficients of variation for the measured versus simulated sediment
results were 0.69, 0.97, and 1.08 for Watersheds 802, C-4 and C-5.
Corresponding mean monthly sediment yields were 1.65, 0.47 and 0.08
tonnes/ha., and monthly standard deviations were 1.14, 0.46, and 0.09
tonnes/ha., respectively.
The results of these calibrations indicate that parameter values in the
sediment .simulation algorithms of HSPF, although empirically derived, are
related for watersheds having similar characteristics. For Watersheds C-4 and
62
-------
TABLE 2
SUMMARY OF CALIBRATED SEDMNT PARAMETER ESTIMATES
802
0
0
0.1
2.5
1.0
1.8
0.10
0.10
0.10
0.10
0.10
0.50
0.90
0.90
0.90
0.90
0.50
0.20
:re) 1.0
1 0.05
C-4
0
0
0.05
2.0
0.5
1.6
0.10
0.10
0.10
0.10
0.30
0.60
0.90
0.90
0.90
0.90
0.80
0.50
5.0
0.05
C-5
0
0
0.05
2.0
1.0
3.0
0.90
0.90
0.90
0.90
0.90
0.90
0.20
0.10
0.10
0.10
0.50
0.90
5.0
0.05
Parameter
KGER*
JGER*
KRER
JRER
KSER
JSER
COVERM:
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
DETS**(ton/acre)
AFFIX (day"1)
*Gully erosion was assumed to be zero.
** The reset value for detached fines storage following tillage
operations.
63
-------
T
T
12
FIGURE 4
SEDIMENT CALIBRATION RESULTS
FOR WATERSHED 802
Jn L J In In
I
1LJ .1
i i i i i i i i i i i i i i i I i i
MAR MAY JUL SEP NOV JAN MAR MAY JUL SEP NOV JAN
APR JUN AUG OCT DEC FEB APR JUN AUG OCT DEC FEB
MEASURED
SIMULATED
MONTH Cl974-I 976)
Line of ons-to-one correspondence
Mora. 1975
Line parallel 1o line ol one-to-one cixrespondenci
50
MEASURED SEDIMENT (tonnes/hedare)
-------
.
E
D
;
I
N
;
o
N
:i
E
.
/
H
A
L
FIGURE 5
SEDIMENT CALIBRATION RESULTS
FOR WATERSHED C-4
II
I I I I ! I I I I I I I I i I II MIT
JAN MAR MAY JUL SEP NOV JAN MAR MAY JUL SEP NOV
FEB APR JUN AUG OCT DEC FEB APR JUN AUG OCT DEC
MEASURED
SIMULATED
MONTHS Cl973-1974)
Lin* of on«-to-on« corritpondanc*
I Z
MEASURED SEDIMENT (tonnei/hectar.)
-------
I .25
I
M
N
:
.
N
I
0
N
N
i
.
H
'
0.75"
0.5
0.25"
0
FIGURE 6
SEDIMENT CALIBRATION RESULTS
FOR WATERSHED C-5
In
In
JAN MAR MAY JUL SEP NOV JAN MAR MAY JUL SEP NOV
FEB APR JUN AUG OCT DEC FEB APR JUN AUG OCT DEC
MEASURED
SIMULATED
MONTHS (1973-1974)
Lini of ont-to-one correiponoence
45678
MEASURED SEDIMENT (tonnej/hectari)
66
10
11
12
-------
C-5 parameters KRER, JRER, DETS and AFFIX have identical values. This would
be expected since the values of these parameters are primarily related to the
credibility and detachability of the specific soil type and land surface
conditions, and soils in the two watersheds are nearly the same. Differences
in the monthly distributions of COVERM are a function of the differences in
the growing seasons.
Values for parameters KSER and JSER were expected to vary over the year,
as these are transport capacity coefficients and exponents, and should vary as
crop cover varies over the growing season. These parameters represent
approximations as to the relationships between overland flow intensity and
sediment transport capacity. Within year differences are expected, since the
algorithms attempt to combine the effects of slope, overland flow length,
sediment particle size, and alternative agricultural management practices into
a single relationship. Since several watersheds and crop characteristics are
incorporated into these parameter value estimates, numerical differences in
parameter values cannot be attributed to a single factor.
CONCLUSIONS
The results of HSPF calibration on the Oklahoma watersheds were not as
good as those obtained in the Mississippi watershed calibration. However, the
ability of the model to simulate the hydrologic response and sediment removal
from the watersheds is considered reasonably good. Large percent differences
between measured and simulated annual totals resulted from poor simulation of
relatively few months in which measured values were extremely high, this is
particularly evident during the initial months of simulation in which initial
conditions may be partially responsible for poor simulation. Overall good
model performance is best illustrated by the double mass analysis plots.
These show that, aside from undersimulation during model start up, parallel
lines can be drawn to the lines of one-to-one correspondence of measured and
simulated values.
Sediment simulation errors on Oklahoma Watersheds C-4-and C-5 were found
to be significantly related to errors in their respective hydrology
calibrations. Months during which sediment was oversimulated (or
undersimulated) normally corresponded with similar results for hydrology.
Difficulties in the hydrologic calibration on both watersheds is presumed to
stem from the inability of the model to consistently simulate storms of high
intensity and short duration. Calibrated parameter value estimates which
physically should be transferable between these watersheds were found to
provide good simulation results for each. In addition, correlation was found
between calibrated parameter extimates for the Mississippi site for those
parameters that are physically based, and have similar characteristics.
HSPF appears to be suitable for simulating runoff and sediment from
agricultural watersheds. However, the model should be verified on each of
these watersheds before being used as a predictive tool. An independent data
set for verification would provide a test for the calibrated parameter values,
and further analyze the model's applicability to these particular watersheds
and to agricultural watersheds in general. For the Oklahoma sites calibration
67
-------
over a two year period with one above average and one below average rainfall
year provides a good first test for the model, but is not a substitute for
independent verification.
REFERENCES
1. Johanson, R.C., Imhoff, J.C., and Davis, H.H. Jr., "User's
Manual for Hydrological Simulation Program-Fortran (HSPF)," U.S.
Environmental Research Laboratory, Publication No. EPA-600/9-80
-015, Athens, Georgia, 1980
2. Hydrocomp, Inc., ARM User's Manaul, Palo Alto, California, 1977.
3. GKY & Associates, Inc., "Calibration and Testing of Agricultural
Runoff Management (ARM) Model and Non-Point Source Pollutant
Loading Model," Alexandria, Virginia, 1979.
4. McDowell, L.L., Personal Communication and Correspondence, 1981.
5. McDowell, L.L., Willis, G.H., Murphree, C.E., Southwichk, L.M.,
and Smith, S., "Toxaphene and Sediment Yields in Runoff from a
Mississippi Delta Watershed." Journal of Environmental Quality,
10:120-125, 1981.
6. Murphree, C.E., Muchler, C.K., and McDowell, L.L., Sediment
Yields from a Mississippi Delta Watershed," In Proc. 3rd Fed.
Interagency Sediment Conference, Denver, CO, March 22-25, 1976.
7. Willis, G.H., McDowell, L.L., Parr, J.F., Murphree, C.E.,
"Pesticide Concentrations and Yields in Runoff and Sediment from
a Mississippi Delta Watershed" In Proc. of the 3rd Fed.
Interagency Sedimentation Conference, Denver, CO, March 22-25, 1976.
8. Menzel, R.G., Rhoades, E.D., Olness, A.E., and Smith,.S.J.,
"Variability of Annual Nutrient and Sediment Discharges in Runoff
from Oklahoma Cropland and Rangeland," Journal of Environmental
Quality. Vol. 7, No. 3, 1978
9. Nicks, A.D., Gander, G.A., Frere, M.H., Menzel, R.G:."Evaluation
of Chemical Transport Models on Range and Cropland Watersheds,"
Presented at the Summer Joint Meeting of the American Society of
Agricultural Engineers and the Canadian Society of Agricultural
Engineering, held at Winnepeg, Manitoba, June 24-27, 1979.
10. Olness, A., Smith, S.J., Rhoades, E.D., and Menzel, R.G.,
"Nutrient and Sediment Discharge from Agricultural Watersheds
in Oklahoma," Journal of Environmental Quality. Vol. 4, No. 3, 1975.
11. Rhoades, E.D., Welch, N.H., and Coleman, G.A., "Sediment-Yield
Characteristics from Unit Source Watersheds," In Present and
Prospective Technology for Predicting Sediment Yield and Sources,
U.S.D.A. ARS Southern Regional Bulletin, ARS-5-40, 1974.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
68
-------
HYDROLOGIC MODELING FOR STUDIES OF POLLUTANT LOADINGS
AND TRANSPORT IN LARGE RIVER BASINS
By Alan Cavacas,1 John P. Hartigan,2 Elizabeth Southerland,3
and John A. Friedman
Introduction
The Chesapeake Bay, located in eastern Maryland and Virginia, is one of
the largest and most economically important of the 850 estuaries that ring
the United States. The Bay is approximately 300 mi long with 13,000 mi of
shoreline and a surface area of 4,300 sg mi. In order to determine the
sources of eutrophication problems in the upper Bay and major tidal
tributaries and to formulate appropriate control strategies, the U.S.
Environmental Protection Agency (EPA) Chesapeake Bay Program funded the
development of three computer models to represent the fluvial and estuarine
sections of the Chesapeake Bay system. The River Basin Model, which is the
subject of this paper, simulates streamflow and transport of point and
nonpoint source pollution loadings in the 64,000 sq mi drainage area of the
Bay (see Figure 1). The Major Tidal Tributary Model serves as the interface
between the Basin Model and the Bay Model by simulating pollutant transport
through the Potomac, James, Rappahannock, and York estuaries. These models
are linked to the two-dimensional Main Bay Model which simulates
vertically-averaged water quality impacts of pollutant loadings delivered to
the Chesapeake Bay. A flow chart which outlines the relationships among the
three models of the Chesapeake Bay system is shown in Figure 2. By
operating the three computer models in series, management agencies can
evaluate Baywide impacts of regional water quality management strategies in
terms of the frequency of violations of water quality criteria/standards for
various beneficial uses (i.e., fisheries habitat, recreation).
This paper describes the calibration/verification of the hydrology
component of the River Basin Model. The continuous simulation hydrologic
model was linked with a water quality model for studies of nonpoint
pollution loadings and the transport of nutrient and organic loadings in
tributary watersheds of Chesapeake Bay. The River Basin Model represents
the Susquehanna, Potomac, and James river basins as well as 33 other river
basins which contribute freshwater inflows, nonpoint pollution loadings, and
wastewater discharges to the Bay. The Model was used by the EPA Chesapeake
Bay Program for assessments of management strategies for nonpoint pollution
and wastewater treatment. This paper focuses on the hydrology/hydraulics
modeling requirements for the river basin water quality modeling studies.
Resources Engineer, Northern Virginia Planning District Commission
(NVPDC), 7630 Little River Turnpike, Annandale, Virginia 22003.
Director, Engineering-Planning Division, NVPDC
•^Environmental Engineer, NVPDC
Resources Engineer, NVPDC
69
-------
LOCATION MAP
Bay
Drainage AreaJI
LEGEND
SUB-BASIN
RIVER BASIN
SUSQUEHANNA
RIVER
--*•—CHANNEL
• GAGE
POTOMAC
RIVER
JAMES
RIVER
Figure 1. Map of Chesapeake Bay Basin Showing
Calibration/Verification Gages and
Sub-basin/Channel Network
WATERSHED
MODEL
POINT \
SOURCES /
nvi
ING
WATER MODEL
TIDAL
TRIBUTARY
ft MAIN BAY
MODEL
METEOROLOGIC
INPUT
NFS
WASHOFF
BASIN
HYDROLOGY
POLLUTANT
TRANSPORT,
DECAY, ft
TRANSFORMATION
STREAMFLOW
ROUTING
RECEIVING
WATER
QUALITY
HYDRO-
DYNAMICS
Figure 2. Flow Chart Showing Models
Included in Chesapeake Bay
Model Package
-------
Modeling Framework
The principal functions of the River Basin Model are to use meteorologic
records to calculate the streamflow and nonpoint pollution loadings in the
Bay's 64,000 sq mi drainage area (see Figure 1) and to simulate the
transport of point source and nonpoint pollution to the Bay's estuarine
system. In other words, the water quality problem to be addressed is the
transport of streamflow and pollutant loadings to the Bay and its tidal
tributaries, rather than localized receiving water quality problems (e.g.,
dissolved oxygen sag) within the fluvial river system.
The temporal dimensions of the River Basin modeling framework are as
follows: (a) simulations of weekly or monthly pollutant loadings delivered
to the Bay's estuarine system are more important than simulations of
hour-to-hour changes in water quality within the tributary rivers; (b) since
one would not expect all point source and nonpoint pollution loadings to
reach the Bay due to physical, chemical, and biological processes during
channel transport, the River Basin Model must be capable of accounting for
pollutant degradation enroute to the Bay; (c) simulations of long-term
records of streamflow and pollutant loadings which reach the tidal
tributaries and the Main Bay are necessary for analyses of the frequency of
water quality criteria violations in the Bay system; (d) in order to
simulate long-term records of streamflow and nonpoint pollution loadings,
the River Basin Model should be capable of accounting for long-term changes
in watershed state variables (e.g., soil moisture, vegetative cover, soil
disturbance, and pollutant loading factors); (e) since most computer models
require short-term rainfall intensity data in order to accurately calculate
the amount of rainfall which does not infiltrate into the soil profile and
soil erosion due to the kinetic energy of raindrops, precipitation records
should be input at intervals of one hour or less; (f) since localized water
quality impacts in the river basins are not the principal focus of the
modeling study, idealized channel reaches with relatively high travel times
(e.g., 1-3 days) and lengthy computation time-steps (e.g., 12 hrs) for flow
routing and water quality processes can be used to ensure that computer
costs for long-term simulations do not become prohibitive? and (g) April 1st
through October 31st is the critical period for studies of eutrophication
management in the Bay system.
The spatial dimensions of the River Basin modeling framework are as
follows: (a) sub-basin size is limited by homogeneity of hydrologic
characteristics (e.g., soils, geology), and by areal variations in rainfall;
(b) since the Model is not intended for flooding simulations, accurate
projections of flood elevations and hydrograph timing are not critical,
thereby permitting a network consisting of relatively large sub-basins
(e.g., 1,000-2,000 sq mi) and relatively long channel reaches (e.g., 30-50
mi) with idealized trapezoidal cross-sections; and (c) since localized
receiving water quality problems are not addressed by the modeling study, a
one-dimensional model assuming completely mixed conditions in each reach can
be used.
71
-------
Description of River Basin Model
Based upon the temporal and spatial dimensions of the required modeling
framework, software which had previously been applied to portions of the
Chesapeake Bay Basin was selected to serve as the River Basin Model. The
selected software (1) is a predecessor of EPA's HSPF model (2) with a
hydrologic submodel based upon a modified version of the Stanford Watershed
Model (J3).
The Model is executed with hourly rainfall and daily evaporation records
to calculate the amount of rainfall converted to runoff and to provide a
continuous accounting of the water balance on the land surface and within
several idealized storage compartments in the soil profile. During storm
periods, hourly rainfall is distributed between surface runoff and soil
moisture storage compartments based upon adjusted infiltration rates and the
nominal storage capacities assigned to different sections of the soil
profile. Between rainstorms, water storage in soil moisture zones is
depleted by mechanisms such as evapotranspiration and subsurface recharge of
streams, thereby freeing up soil moisture storage capacity for rainfall
inputs from the next storm. Streamflow transport is handled with a form of
kinematic wave routing, while pollutant transport out of a given channel
reach into a downstream channel reach is based upon advection. A
twelve-hour routing interval was used for hydrologic simulations.
The selected software has been used for several modeling studies in the
metropolitan Washington, D.C. region. It has been applied most extensively
to the 580 sq mi Occoquan River Basin of Northern Virginia for studies
ranging from nonpoint pollution management assessments (±,5) to evaluations
of advanced wastewater treatment (AWT) needs (6) .
Description of Model Networks and Input Datasets
Sub-basin Data. The delineation of the sub-basin network was based upon
an analysis of geographic variations in the following characteristics listed
in the order of consideration: (1) physiographic province; (2) topography;
(3) hydrologic soil group; and (4) total water holding capacity of soil.
Map overlays (1:500,000 scale) were developed for each dataset for purposes
of hydrologic segment delineation. In the case of datasets 2, 3, and 4, the
overlays display the predominant or average characteristic in a 100 sq mi
grid cell network which is used to manage and aggregate basinwide physical
features data (i.e., each cell has dimensions of 10 mi x 10 mi).
The 100 sq mi grid size was felt to provide a reasonable level of detail
for a 64,000 sq mi basin. The authors' previous modeling experiences in the
Northern Virginia region suggested that localized variations in physical
features tend to have a relatively insignificant effect on the development
of lumped parameter model datasets for sub-basins on the order of a few
hundred square miles. Confidence in the representativeness of the 100 sq mi
grid cell datasets for the Chesapeake Bay Basin was gained through
sensitivity studies based upon county soils maps and 1:24,000 scale
topographic maps.
72
-------
Due to the size of the Chesapeake Bay Basin, assessment of soils
characteristics were based upon soil association characterizations carried
out on a state-by-state basis. The distribution of the Chesapeake Bay
Basin's 135 soil associations among the five states in the drainage area is
as follows: Virginia: 54; Maryland: 27; West Virginia: 12; Pennsylvania:
34; and New York: 8. Each soil association consists of one-to-four soil
series listed in order of dominance. Composite characteristics for each
association are typically developed by weighting the characteristics of
individual soil series according to the fraction of the association that is
typically attributed to a series. Unfortunately, statewide data on the
fraction of each association that is attributed to each series is available
only for Pennsylvania and Northern Virginia. The only source of such data
in other sections of the basin are the soil surveys prepared by the
individual counties in each state. Based on a review of typical soil series
distributions attributed to Pennsylvania and Northern Virginia soil
associations, the following relative distributions were assumed to analyze
soil association characteristics throughout the Chesapeake Bay Basin:
Soil Series Order
1st Series
2nd Series
3rd Series
4th Series
Total Number of Soil Series in Association
Two Soil Series Three Soil Series Pour Soil Series
60%
40%
60%
30%
10%
50%
30%
10%
10%
The "soil series order" refers to the hierarchy implicit in the soil
association name. These assumed distributions were used to weight the
hydrologic characteristics of each series to derive a composite
characteristic for each soil association.
Average values of the following characteristics were derived for each
soil association by applying,the assumed distributions to the soil series
data presented in State SCS-5 reports: permeability; hydrologic soil group;
total water holding capacity; soil texture; soil depth; and erodibility
factor (K). Using the 100 sq mi grid cell network, 1:500,000 scale map
overlays showing the average value for the predominant soil association in
each grid cell were developed for each of these characteristics.
The first step in the delineation of homogeneous hydrologic segments was
to overlay the physiographic province map with the grid map of average
slopes. Areas with relatively uniform slopes within physiographic provinces
were selected as the first segment approximation. The next step was to
overlay this intermediate segment network with the grid maps of hydrologic
soil group and total water holding capacity. More detailed segments were
derived from these last two overlays since they delineated areas with
relatively similar infiltration rates and soil moisture storage capacities.
Particular attention was given to defining more detailed segments in the
Coastal Plain where surface runoff and nonpoint pollution loads have a
greater chance of reaching the Bay's estuarine system due to the relatively
short travel times. This analysis of basinwide physical features resulted
73
-------
in the delineation of a preliminary hydrologic unit network consisting of 23
hydrologic segments.
This preliminary segment network was further refined by analyzing areal
variations in rainfall patterns within each segment. National Weather
Service (NWS) tapes with hourly/daily raingage records for the period
1966-1978 were used for the modeling study since this period included a good
mix of wet, dry, and average years. A total of 93 raingages were included
in this study (see Figure 3). Statistics such as mean annual volume,
•standard deviation, and coefficient of variation were calculated for each
raingage and compared with surrounding gages to identify groupings with
similar characteristics. Raingage groupings were further refined through
intercorrelation analyses based upon daily rainfall totals. Basinwide
isohyetal maps and Thiessen polygons constructed for the final raingage
groupings were used to further subdivide seven hydrologic segments,
resulting in the final network of 30 segments shown in Figure 3. In
general, the hydrologic segment boundaries correspond to physiographic
boundaries as might be expected for a generalized network representing a
64,000 sq mi basin. Segments 1-10 represent Coastal Plain province areas,
segments 11-15 represent Piedmont province areas, segments 16 and 17
represent the Blue Ridge and Great Valley province, segments 18-24 represent
Appalachian Ridge and Valley province areas, and segments 25-30 represent
Appalachian Plateau province areas. Average soils characteristics and
topography data used in the hydrologic submodel were tabulated for each
segment by weighting the values stored in the grid cell dataset.
Since the hydrologic segments often overlapped river basin boundaries, a
network of sub-basins was delineated to represent each river basin. In
order to maximize homogeneity, the sub-basins were sized to ensure that the
majority of the drainage area was located within a single hydrologic
segment. In addition, an effort was made to maintain a bankfull channel
travel time on the order of 24-72 hrs in establishing the outflow of each
sub-basin. As shown in Figure 1, the resulting sub-basin network for the
major river basins consists of 34 sub-basins on the order of 1,000-2,000 sq
mi. In addition, 29 Coastal Plain watersheds are included in the Model
network to provide detailed representations of direct inflows to the major
tidal tributaries and embayments of the Main Bay. The hydrologic
characteristics assigned to the segments traversed by each sub-basin were
weighted for input to the hydrologic submodel.
In light of the size of the study area, sophisticated remote-sensing
techniques offered the only feasible method for defining land cover data for
each sub-basin. Existing land use summaries for each sub-basin are based
upon interpretations of LANDSAT satellite/images from the period 1977-1979
with state-of-the-art software available at the Goddard Space Flight Center
in Greenbelt, Maryland. A total of 15 LANDSAT scenes were required to cover
the entire Chesapeake Bay Basin.
Meteorologic Data. To produce a single hourly rainfall record for each
hydrologic segment for the period 1966-1978, the Thiessen polygon method
was used. Special software was designed to produce an hourly rainfall
record which was based on a representative area-weighted daily rainfall
-------
Mil tl
LEGEND
• Hourly Ralngages
• Dally Ralngages
— — — Hydrologlc Segment
Figure 3. Map of Hydrologic Segments
Showing Distribution of
Hourly and Daily Raingages
Table 1
Listing of Streamgages Used for Calibration/Verification
Uses G AGE
NAME
MAP
KEY* NUMBER
A 02035000 James River at Cartersvllle, VA
B 01668000 Rappahannock River near Fredericksburg, VA
C 01613000 Potomac River at Hancock, MD
D 01636500 Shenandoah River at Millville, WV
E 01646500 Potomac River near Washington, D.C.
F 01551500 West Branch Susquehanna River at Williamsport, PA
G 01S36500 Susquehanna River at Hilkes-Barre, PA
H 01567000 Juniata River at Newport, PA
I 01576000 Susquehanna River at Marietta, PA
J 01578310 Susquehanna River at Conowingo, MD
K 02042500 Chickahominy River at Providence Forge, VA
L 01674000 Mattaponi River at Bowling Green, VA
M 01674500 Mattaponi River at Buelahville, VA
N ^ASTERN SHORE GAGES (SUM OF FLOWS)
01491000 Choptank River near Greensboro, MD
01487000 Nanticoke River near Bridgeville, DE
01485000 Pocomoke River near Willards, MD
DRAINAGE AREA
(SO MI)
6,257
1,596
4,073
3,040
11,560
5,682
9,960
3,354
25,990
27,100
248
257
€01
249
(113)
(75)
(61)
locations are shown in Figure 1
-------
volume, preserved hourly rainfall intensities, and compensated for missing
records. Other daily meteorologic datasets (e.g., evaporation, air
temperature, solar radiation, wind speed) required for the hydrologic and
water quality submodels were derived for eight meteorologic regions. Eight
regions for non-precipitation data were felt to be adequate because the
gages for these records are fewer in number than the rainfall gages, areal
variations in these meteorologic indicators tend to be easier to
characterize, and hydrologic reponses in the river basins tend to be more
sensitive to month-to-month fluctuations in these data in comparison with
the day-to-day fluctuations in rainfall which are so important.
Channel Data. A single idealized trapezoidal channel with constant
cross-sectional geometry is assigned to each sub-basin. To facilitate the
development of idealized cross-sections, all channels are terminated at
U.S.G.S. streamgaging stations where data on channel geometry is available
from the table (Form 9-207) used to construct the stage-discharge
relationship for the gage. As indicated above, idealized channel length was
established in conjunction with the determination of sub-basin size in order
to maintain bankfull travel times on the order of 1 to 3 days for each
channel reach. Flood plain slopes for each idealized channel were
determined from 1:24,000 scale topographic maps for the sub-basin. A sketch
of the Basin Model's channel reach system is shown in Figure 1. It consists
of 28 idealized channels with lengths ranging from 25 to 190 mi and seven
reservoirs. The idealized channel system is restricted to the main stems of
the major river basins, since the focus of Basin Model applications is the
transport of pollutant loadings to the Chesapeake Bay rather than localized
receiving water problems. For purposes of this pollutant transport study,
it was assumed that the time lag and pollutant transformations achieved by
channel storage in minor tributaries and small Coastal Plain watersheds is
relatively insignificant.
Since the idealized channel reaches begin and end at U.S.G.S.
streamgaging stations, the channel invert elevations reported for each
streamgage were assigned to the respective channels for gradient
calculations. Data on channel roughness coefficients was collected from
state and regional agencies which had performed local flood insurance
studies in the Chesapeake Bay Basin. An average value was derived for each
idealized channel reach based upon the arithmetic means of roughness
coefficient values at several representative cross-sections.
The following major reservoirs are represented by the Basin Model as
single-layer lakes: Lake Anna (York River Basin); Lake Chesdin (Appomattox
River Basin); Raystown Reservoir (Juniata River Basin)j the two Patuxent
River reservoirs; the Lake Aidred/Lake Clarke reservoir system (Susquehanna
River Basin); and Conowingo Reservoir (Susquehanna River Basin). In the
case of the two Patuxent reservoirs and the Lake Aldred/Lake Clarke system,
the two reservoirs in series are combined into a single idealized
impoundment with appropriate aggregate characteristics. For the Conowingo
hydroelectric reservoir at the mouth of the Susquehanna River, a separate
operating rule computer program (2> is used to calculate daily spills from
simulated daily streamflows entering the reservoir.
76
-------
Hydrology Calibration/Verification
Methodology. As indicated in Table 1 and Figure 1, a total of 14
streamgages were used for hydrology calibration/verification. The period
April 1971 through October 1976 was used for model calibration, since this
period included a good mixture of relatively wet, dry, and average years.
Model verification was based upon the periods April 1966 through June 1970
and November 1976 through December 1978, which were generally somewhat drier
than the calibration period. During calibration, the models were operated
with meteorologic records for the entire 5.75-yr period and a single set of
parameter values. Based on comparisons of simulated and observed
streamflows, the most sensitive parameter values were iteratively adjusted
to establish the final parameter sets. After acceptable agreement was
achieved on a seasonal and annual basis, simulated and observed daily
streamflows were compared for storm events to set hydrograph shape factors
and for dry weather flow periods to set baseflow recession constants.
Following calibration, the models were operated for the 6.4 year
verification period without any adjustment to the calibrated parameter
values to determine how well the models represented conditions different
from the calibration period.
Since only a very small percentage of the Chesapeake Bay Basin is
urbanized, calibration activities focused on soils parameters that determine
an undeveloped area's hydrologic characteristics. Due to the distribution
of raingages and streamgages, it was not possible to calibrate the Basin
Model for every major watershed in the 64,000 sg mi basin. Therefore, one
objective of hydrology calibration was to derive relationships between river
basin physical features and model parameter values which could be applied to
major watersheds that could not be calibrated separately. In other words,
rather than indiscriminately adjusting the Model's parameter values to
produce the best possible comparisons between simulated and observed
streamflows at each streamgage, hydrology calibration focused on developing
parameter estimation methods that could be applied to ungaged watersheds.
This approach has previously been used to calibrate hydrologic models of
several watersheds in the Northern Virginia portion of the Chesapeake Bay
Basin (£,.9,10).
LZSN (soil moisture storage capacity) and INFIL (infiltration rate) are
the most important parameters for simulations of annual streamflow volumes.
An increase in LZSN will increase the storage of water in the idealized
lower zone of the soil, thereby lowering seasonal and annual streamflows by
increasing the depletion of soil moisture through evapotranspiration and by
reducing the frequency of saturated soil conditions. An increase in INFIL
will likewise lower annual streamflows by reducing direct runoff due to
higher infiltration and increasing soil moisture depletion through
evapotranspiration. INFIL is most often used to modify seasonal streamflows
after LZSN and a reasonable range of INFIL values which achieve acceptable
annual streamflows have been identified. As has been the case in previous
studies (.8r.9f.10r.ll)' LZSN in the Chesapeake Bay Basin was found to generally
be directly related to average total water holding capacity and depth of
soil above the restrictive layer. Calibrated LZSN values ranged from
approximately 2-4 in. in the Appalachian Plateau and mountainous areas of
77
-------
the Appalachian Ridge and Valley to approximately 6-8 in. in the Coastal
Plain, lower Piedmont, and portions of the Appalachian Ridge and Valley.
INPIL was found to be related to indicators of surface runoff potential such
as hydrologic soil group, soil permeability, and soil texture. Calibrated
INPIL values ranged from approximately 0.01 to 0.015 in/hr in the
Appalachian Plateau and sections of the Coastal Plain and lower Piedmont
with "D" hydrologic soil groups to 0.06 to 0.075 in/hr in sections of the
Coastal Plain with "B" hydrologic soil groups. K3 is an evapotranspiration
index that is generally set at reasonable levels to reflect vegetative
cover, and then held constant while LZSN and INFIL are calibrated. K3
values on the order of 0.3-0.45 were used throughout the Chesapeake Bay
Basin, with the higher values typically associated with areas characterized
by high forest cover to account for the effect of vegetative cover on
evapotranspiration. Although it is not as sensitive a parameter as either
LZSN or INFIL, UZSN (i.e., moisture storage near the soil surface most
closely related to depression storage) can have some effect on seasonal and
annual streamflow volumes because of its impact on individual storm events.
UZSN is most often related to the calibrated LZSN value and in the
Chesapeake Bay Basin was typically found to be 5%-15% of LZSN. After
acceptable agreement between simulated and observed streamflows was achieved
on an annual and seasonal basis, the interflow coefficient INTER was
adjusted to redistribute streamflows between surface runoff and subsurface
flows in order to match the shapes of the recession limbs of observed
hydrographs. Relatively low INTER values assign a higher percentage of
streamflow to surface runoff, thereby resulting in higher peak flows and
steep recession limbs. Like previous studies (_8,J),J.£), Chesapeake Bay Basin
calibration results indicated that INTER could be related to average land
slopes, with values ranging from 1.0 in relatively flat Coastal Plain areas
to 1.5 in mountainous areas. After INTER had been set, the baseflow
recession coefficient (KK24) was fine tuned to improve the agreement between
simulated and observed dry weather flows.
Results. Calibration runs were terminated for each streamgage when it
was determined that: (a) the differences between simulated and observed
streamflows could not be improved with further parameter adjustments, and
(b) a sufficient number of model runs had been completed to develop
reasonable regional parameter sets. The hydrology parameter sets produced
by these calibration runs are quite reasonable based upon some previous
continuous simulation modeling studies in the Chesapeake Basin (1,9,10,12)
and other literature values (L3,1.4,jL5).
Comparisons of simulated and observed streamflow data based upon
streamflow volumes (annual and seasonal), daily streamflow time series,
correlation coefficients, and daily flow-duration plots (period of record
and seasonal) generally indicate very good calibration and verification
results. Table 2 summarizes comparisons of simulated and observed annual
streamflow volumes for 10 major streamgages. As may be seen, differences
between simulated and observed annual volumes are typically within the range
of observation errors associated with meteorologic and streamflow data
collection activities. In general, the greatest differences between
simulated and observed streamflow were associated with winter periods and
drought periods. Winter periods, which were typically somewhat
78
-------
Table 2
t-
Mean Ratio of Simulated to Observed Water Year o
Streamflow Volumes and Number of Years with £
Ratio between 0.8 and 1.2
5 10
(i.e., No. of Vrs./Total No. of Yrs.) £ ai 0
STREJVKAGE
SIMULATED : OBSERVED WY VOI HMFS.
CAUBMIIM PEMM1 VERIFICATION PERIOD
SUSOJENAHNA R.
AT MtUES-BAKKE, PA.
WEST DR. SUSWEHAMM R.
AI MIUIAHSPORT, PA.
JtHIATA R,
AT HEKPORT, PA.
SUSgUEMAIBA R.
AI mRlEIIA, PA.
SUSOHEHAMA R.
AT COKOKIN60, IV.
JAMES R.
Al CARTERSVlllt, VA.
RAPPAHANKOCK R.
KEAR FREDERICKSBUR6. VA.
POTOMAC R.
AT HABCOCK, F3,
StffBWJDOflH R.
AT fllUVILLE, S.VA.
POTWWCR.
NEAR MASH..O.C.
• - • _-_--.-—-=-••---f_? ••w-y.'uij >i_no iu^i iva >i.niuv
1.10 10.13 5/b 1.07 10.27 5/5
11,560 0.9X 10.06 6/6 0.911 10.10 W5
99.9
LESCND
SIMULATED
-— RECORDED
1971-1976
(a)
8.0OO 1^000 24JOOO 32,000
DAILY STREAMFLOW (CF3)
LE8END
SIMULATED
RECORDED
1966-1970
AND
1976 - 1978
e,OOO 16,000 24,000 32.0OO
DAILY STREAMFLOW (CFS)
Figure 4.
40POO
40^)00
Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: James River at Cartersville,
VA (6,257 sq mi)
-------
LEOEMO
— SIMULATED
•— RECORDED
1971 -1976
(a)
4,000 8,000 I2JOOO 16,000 2Q0OO
DAILY STREAMFLOW (CFS)
6000 B000 18000 24000
DAILY STREAMFLOW (CFS)
30000
oo
o
LEOEND
SIMULATED
— RECORDED
1966 -1970
AND
1976 - 1978
600O I20OO I800O 24000
DAILY STREAMFLOW (CFS)
3O0OO
9*9
3 90
u.
a ro
IS TO
| eo
~ 50
u 4O
P 30
O 20
»-
3 10
u
oc 0.1
LE6ENO
SIMULATED
RECORDED
1966 -1970
AND
1976 - 1978
(b)
4,000
DAILY
8.0OO 12,000 I6.OOO
STREAMFLOW (CFS)
20£00
Figure 5. Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: Potomac River at Hancock,
MD (4,073 sq mi)
Figure 6. Flow-Duration Curves for
Calibration (a) and Verification {b)
Periods: Shenando'ah River at
Millville, W.VA. (3,040 sq mi)
-------
o
u
99L9
80
u
5
5 60
Al
U
I 40
i-
20
u
K
oj
LE9ENO
SIMULATED
RECORDED
1971-1976
(a)
15,000 30,000 45,000 60jOOO 75,000
DAILY STREAMFLOW (CFS)
LEGEND
SIMULATED
RECORDED
1971-1976
(a)
8,000 12,000 I6JOOO
STREAMFLOW (CFS)
20,000
oo
o
5
Al
99.9
90
8O
TO
60
50
4O
30
ZO
10
O.I
LE9ENO
SIMULATED
RECORDED
1966 -1970
AND
1976 - 1978
(b)
15,000 30,000 45,000 6OPOO
DAILY STREAMFLOW (CFS)
75000
Figure 7.
Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: Potomac River near
Washington, D.C. (11,560 sq mi)
LEGEND
SIMULATED
RECORDED
1966 -1970
AND
1976 - 1978
(b)
i6jooo 20,000
Figure 8.
DAILY STREAMFLOW (CFS)
Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: -Rappahannock River at
Fredericksburg , VA (1,596 sq mi)
-------
LESEND
SIMULATED
RECORDED
1971 -1976
(a)
ICjOOO 20,000 30,000 40,000
DAILY STHEAMFLOW (CFS)
3QOOO
LEOENb
SIMULATED
RECORDED
0.0
1971 -1976
(a)
W000 40000 60,000 80.00O
DAILY STREAMFLOW (CFS)
lOOyOOO
oo
LEOENO
SIMULATED
RECORDED
0.0
1966 -1970
AND
1976 - 1978
(b)
10,000 20,000 30,000 40£OO
DAILY STREAMFLOW (CFS)
30,000
LE9ENO
SIMULATED
RECORDED
1966 - 1970
AND
1976 - 1978
0.0
(b)
ZOpOO 4O0OO 6OJOOO 80.OOO
DAILY STREAMFLOW (CFS)
IOCMDOO
Figure 9. Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: West Branch Susquehanna
River at Williamsport, PA
(5,682 sq mi)
Figure 10. Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: Susquehanna River at
Wilkes-Barre, PA (9,960 sq mi)
-------
SIMULATED
— RECORDED
99.9
8 90
o
in
80
70
5 60
I W
u 40
*- 30
« 20
£ 10
u
e o.l
u
0.0
4000 8000 12,000 16000
DAILY STREAMFLOW (CFS)
20000
0.0
LEGEND
— SIMULATED
RECORDED
1971 -1976
(a)
30000 60000 90000 120000
DAILY STREAMFLOW (CFS)
190000
CO
LE8END
SIMULATED
RECORDED
1966 -1970
AND
1976 - 1978
(b)
4000 B00O 12,000 16000
DAILY STREAMFLOW (CFS)
LEGEND
SIMULATED
RECORDED
u
' ^
20000 w
1966-1970
AND
1976 - 1978
(b)
30,000 60,000 90000 I20JOOO
DAILY STREAMFLOW (CFS)
150000
Figure 11. Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: Juniata River at Newport,
PA (3,354 sq mi)
Figure 12. Flow-Duration Curves for
Calibration (a) and Verification (b)
Periods: Susquehanna River at
Marietta, PA (25,990 sq mi)
-------
undersimulated by the Model/ tend to be characterized by frozen ground
conditions which are not simulated by the River Basin Model as well as the
highest raingage errors due to freezing conditions. Streamflow errors
during drought periods, which were typically oversimulated by the River
Basin Model, can be attributed in large part to the tendency of the
Thiessen-weighting procedure to exaggerate the areal distribution of
localized storms which tend to be more significant during droughts.
Comparisons of simulated and observed daily flow-duration curves are
shown in Figures 4 through 12 for both calibration and verification periods
at nine major streamgages. As may be seen, the agreement between simulated
and observed curves is typically quite good, with goodness-of-fit for the
verification period typically almost as good as for the calibration period.
In general, the calibration period exhibits better agreement for low flow
periods than does the verification period, which was characterized by
drought periods that presented difficulties with the development of a
representative mean segment rainfall record.
As another indication of goodness-of-fit, Table 3 summarizes statistics
on the correlation between simulated and observed weekly streamflows.
Correlation coefficients are based upon weekly streamflows because
weekly-to-monthly flows were of greatest interest for the pollutant
transport study. As may be seen in Table 3, correlation coefficients for
the calibration period are somewhat higher than for the verification period,
although coefficients for both periods are within acceptable ranges. The
gages at the mouths of the three largest river basins are characterized by
the highest correlation coefficients.
Comparisons of simulated and observed hydrographs for Hurricane Agnes
(late June 1972) and Tropical Storm Eloise (late September 1975) reveal that
the Model appears to handle these relatively infrequent events rather well
at some gages. Sample comparisons are presented in Figure 13
for the gages on the West Branch Susquehanna River at Williamsport,
Pennsylvania and on the Potomac River at Hancock, Maryland. It is felt that
the River Basin Model handled infrequent storm events better than some
earlier models (_1,J3,JLO) of smaller watersheds in large part because the
rainfall during these storms tends to be rather uniformly distributed over
sub-basins with areas on the order of 2,000 sq mi.
In summary, both calibration and verification results are very good
particlarly in light of the very large rainfall segments. Most of the
remaining error can probably be attributed to factors such as frozen ground
conditions which were not explicitly represented by the Model and to errors
in the mean segment rainfall record due to localized rainstorms, low
raingage densities in some areas, missing rainfall records, and freezing
conditions.
Use of Hydrologic Model to Identify Production Run Periods
Introduction. In the case of low flow, high flow, and long-term
assessments, model executions focus on impacts during the spring, summer,
and fall when water temperatures are high enough to result in the most
-------
£240000
b.
"210,000 -
O 180.000
a 30,000
Figure 13.
LEGEND
SIMULATED
RECORDED
108,000
96000
84,000
72000
60,000
48000
36,000
24000
I2.0OO
0
LEGEND
SIMULATED
RECORDED
-HURRICANE
A6NES
JUNE 1972
JULY 1972
00
SEPTEMBER 1975 "OCT. I97S MAY 1972 JUNE 1972 JULY 1972
(a) "- -- (b) (c)
Comparison of Simulated and Recorded Hydrographs: (a) and (b) West Br. Susquehanna
River at Williamsport, PA; and (c) Potomac River at Hancock, MD
POTOMAC RIVER NEAR WASHINGTON, D.C. (11,560 SQ. Ml.)
SUMMATION OF RECORDED DAILY FLOWS'
SUSQUEHANNA, POTOMAC, AND JAMES (43,807 SQ. Ml.)
LC8END
1971
1974
1978
1966'1978
APRIL I - OCTOBER 31
- OCTOBER 31
0,1 100 200 300 400 500 6O0 7O0 800 900 993 O.I IO0 200 300 4O0 600 6O0 700 800 90.0 993
PERCENT OF TIME £ INDICATED FLOW PERCENT OF TIME £ INDICATED FLOW
Figure 14. Sample Flow-Duration Curves for Recorded Daily Streamflow: Typical Year Analysis
-------
Table 3
Correlation Coefficients for Weekly Streamflows at Major River Gages
CORRELATION COEFFICIENT
(WEEKLY STREAMFLOWS)
CALIBRATION
STREAMGAGE PERIOD
James River at Cartersville, VA
Potomac River at Hancock, MO
Shenandoah River at Millville, HV
Potomac River near Washington, D.C.
Rappahannock River near Fredericksburg, VA
West Branch Susquehanna River at Williarasport, PA
Susquehanna River at Wilkes-Barre, PA
Juniata River at Newport, PA
GtiaraiAhanna Biwar Ht Marietta, PA
0.94
0.91
0.92
0.95
0.85
0.85
0.88
0.86
0.92
VERIFICATION
PERIOD
0.84
0.78
0.74
0.80
0.81
0.71
0.74
0.81
0.80
critical water quality problems within the Bay's estuarine system.
Following calibration/verification, the hydrologic model was used to
identify the periods to be used for production runs to evaluate water
quality management strategies.
Analyses of Long-Term Water Quality Impacts. It is technically possible
to execute the Bay Model package with many years of rainfall and streamflow
records in order to assess long-term water quality impacts of a particular
management strategy. However, since the computer costs for long-term
continuous simulations can be very high/ some investigations have relied
upon a short period (e.g., a full year; several seasons) which can serve as
a less expensive surrogate for the multi-year period of interest
(.1».4».16>.17) • This shorter "typical" period should be characterized by
streamflow and water quality statistics that are reasonably close to the
period of record for the meteorologic data which would be used for a
long-term simulation. Production runs of the Model package with
meteorologic records for the typical period are then assumed to produce
streamflow and water quality statistics which approximate the statistics
that would result from a model production run covering the entire period.
In terms of streamflow statistics, this typical period can also be referred
to as a period of "average wetness" since its flow-duration curve and total
streamflow volumes will most closely approximate long-term conditions.
Since the use of a typical period permits simulations of long-term water
quality impacts at a reasonable cost, this approach was selected for
management studies with the Bay Model package.
A fall-line monitoring study by the U.S. Geological Survey (18) had
previously demonstrated a positive relationship between daily streamflow and
pollutant loadings at the mouths of the Susquehanna, Potomac, and James
rivers. The USGS study produced regression equations relating streamflow
and pollutant loadings which can be used to produce loading-frequency
relationships from a daily flow-duration curve. Therefore, a year
characterized by a daily flow-duration curve that approximates the
streamgage's long-term flow-duration curve is also characterized by
86
-------
loading-frequency relationships which approximate long-term loading
statistics. Daily flow-duration relationships for gages at the mouths of
the Susquehanna, Potomac, and James rivers were screened to identify a
typical year for water quality management studies. A composite
flow-duration curve based upon the summation of daily flows at the three
river mouth gages was also included in the screening study.
Because the available meteorologic record for River Basin Model studies
covered the period 1966-1978, the selection of the typical period was based
on analyses of the daily streamflow statistics associated with April through
October of each year in this thirteen-year period. The year with
April-October streamflow statistics which come closest to the statistics for
the full thirteen-year period can be designated as a typical year for
assessments of long-term impacts with the Bay Model package. Simulated
flow-duration curves for individual years were plotted with the simulated
flow-duration curve for 1966-1977 and the assessment of similarity in
distribution was based upon visual inspection. Sample flow-duration curves
are shown in Figure 14. Since 197'4 exhibits better agreement with
the 1966-1978 curves for the Susquehanna, Potomac, and "summation" gages, it
was selected over 1976 as the most typical year.
Analyses of "Worst Case" Point Source Impacts. As suggested above,
Model executions for a "dry year" with an extended low streamflow period can
be expected to provide the greatest insights into the Baywide impacts of
wastewater treatment strategies for the Chesapeake Bay Basin. Because it is
characterized by the highest frequency of extreme low flows in the two
largest river basins and the lowest overall volumes for the sum of
Susquehanna, Potomac, and James basin streamflows, April-October 1966 was
designated for dry year production runs of the Bay Model package.
Analyses of "Worst Case" Nonpoint Source Impacts, Model operations for
a "wet year" characterized by relatively high streamflows will provide the
greatest insights into "worst case" impacts of nonpoint pollution. A review
of simulated flow-duration curves for the Susquehanna, Potomac, and James
basins and the summation dataset indicates that 1972 is the wettest year for
all four gages, with 1975 the second wettest. Since it was felt that a
period which includes an event as rare as Hurricane Agnes (June 1972) may
not be an agfiropriate "design condition" for basinwide assessments of
nonpoint pollution controls, the second wettest year, 1975, was selected for
wet year production runs.
Further Evaluations of Selected Periods. The decision to restrict the
rainfall records to the period 1966-1978 was one based primarily on costs in
an effort to keep the budget for the acquisition and analysis of NWS
rainfall tapes from becoming prohibitive. Since NWS hourly rainfall records
for the Chesapeake Bay Basin recording gages are currently available for 17
additional years (i.e., 1949-1965), it was necessary to screen observed
April-October flow-duration curves for each year in the period 1949-1978 to
demonstrate that the selepted years were the most appropriate for Model
production runs.
87
-------
For the typical year comparisons, the flow-duration curves for the
period 1966-1978 were found to adequately approximate the curves for the
full 30-year period (1949-1978) covered by NWS hourly rainfall records,
indicating that selections of a typical year based on comparisons with the
shorter period were appropriate. Further, 1974 exhibited as good an
agreement with the 1949-1978 period as did the only two years (1950 and
1961) in the earlier period which merited consideration as a typical year.
Similar reevaluations reinforced the selections of 1966 and 1975 as the
design dry year and wet year, respectively.
Acknowledgements
The work described herein was funded through a Cooperative Agreement
with the U.S. Environmental Protection Agency's Chesapeake Bay Program
(CPB). The, assistance of James T. Smullen, EPA Project Officer, and Alan M.
Lumb of the U.S. Geological Survey is gratefully acknowledged.
References
1. Hydrocomp, Inc., "The Occoquan Computer Model: Calibration,
Verification, and User's Manual," prepared for Northern Virginia
Planning District Commission, Falls Church, VA, May 1978.
2. Johanson, R.C., Imhoff, J.C., and Davis, H.H., "Users Manual for
Hydrological Simulation Program - FORTRAN (HSPF)," EPA-600/9-80-015,
U.S. Environmental Protection Agency, Athens, 6A, April 1980.
3. Crawford, N.H., and Linsley, R.K., "Digital Simulation in Hydrology:
Stanford Watershed Model IV," Dept. of Civil Engineering Technical
Report 39, Stanford University, Stanford, CA, 1966.
4. Hartigan, J.P., et al., "Areawide and Local Framework for Urban Nonpoint
Pollution Management in Northern Virginia," Stormwater Management
Alternatives, Tourbier, J.T. and Westmacott, R., eds., University of
Delaware, DE, April 1980, pp. 211-245.
5. Bonuccelli, H.A., Hartigan, J.P., and Biggers, D.J., "Computer Modeling
for Watershed Management in Northern Virginia," Proceedings of
Stormwater Management Model User's Group Conference; January 10-11,
1980, EPA 600/9-80-017, U.S. Environmental Protection Agency,
Washington, D.C., March 1980, pp. 17-40.
6. Northern Virginia Planning District Commission (NVPDC), "Follow-up
Assessments of Alternate AWT Operating Rules with Occoquan Basin
Computer Model," prepared for Camp, Dresser, McKee, Inc., consultants to
Virginia State Water Control Board, Richmond, VA, February 1980.
7. Arbruster, J.T., "Flow Routing in the Susquehanna River Basin: Part I -
Effects of Raystown Lake on the Low-Flow Frequency Characteristics of
the Juniata and Lower Susquehanna Rivers, Pa.," U.S. Geological Survey
Water Resources Investigations 77-12, USGS, Harrisburg, PA, April 1977.
-------
8. Hartigan, J.P., et al. , "Calibration of Urban Nonpoint Pollution Loading
Models," Proceedings of ASCE Hydraulics Division Specialty Conference on
Verification of Mathematical and Physical Models in Hydraulic
Engineering, ASCE, New York, NY, August 1978, pp. 363-372.
9. NVPDC, "Water Quality Modeling Study of Goose Creek, Broad Run, and
Sugarland Run Watersheds," Falls Church, VA, June 1980.
10. NVPDC, "Modeling Study of Nonpoint Pollution Loadings from Potomac
Enbayment Watersheds, * Final Report prepared for Virginia State Water
Control Board, Richmond, VA, March 1981.
11. Lumb, A.M. and James, L.D., "Runoff Files for Flood Hydrograph
Simulation," Journal of the Hydraulics Division, ASCE, Vol. 102, No.
HY10, October 1976, pp. 1515-1531.
12. CHoM-Hill, "Stormwater Management: A Comprehensive Study of the Muddy
Branch and Seneca Creek Watersheds," prepared for Montgomery County (MD)
Planning Board, Silver Spring, MD, April 1975.
13. Donigian, A.S. and Crawford, N.H., "Modeling Nonpoint Pollution from the
Land Surface," EPA-600/3-76-083, U.S. Environmental Protection Agency,
Environmental Research Laboratory, Athens, GA, July 1976.
14. Hydrocomp, Inc. Hydrocomp Simulation Programming; Hydrology Simulation
Operations Manual, Palo Alto, CA, January 1976.
15. Lumb, A.M., "UROSO4: Urban Flood Simulation Model, Part 1:
Documentation and Users Manual," School of Civil Engineering, Georgia
Institute of Technology, Atlanta, GA, March 1975.
16. Huber, W. , "Discussion Remarks," Proceedings of Seminar on Design Storm
Concept, Ecole Polytechnique de Montreal, Montreal, Quebec, 1979.
17. Souther land, E. , "A Continuous Simulation Modeling Approach to Nonpoint
Pollution Management," Dissertation presented to Virginia Polytechnic
Institute & State University, Blacksburg, VA, in December 1981, in
partial fulfillment of the requirements for the degree of Doctor of
Philosophy in Environmental Sciences and Engineering.
18. U.S. Geological Survey, "Water Quality of the Three Major Tributaries to
the Chesapeake Bay, January 1979-April 1981," prepared for USEPA
Chesapeake Bay Program, November 1981.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
89
-------
CONTINUOUS DO RESPONSE PREDICTED
USING CSPSS IS VERIFIED FOR
SPRINGFIELD, MISSOURI1
By: James E. Scholl2, P.E.
and
Ronald L. Wycoff2, P.E.
ABSTRACT:
Springfield, Missouri was one of 15 receiving water sites studied for the
1978 Needs Survey to estimate the cost associated with controlling storm-
water-induced pollution nationwide. Receiving water impacts on the James
River near Springfield were evaluated using the Continuous Stormwater
Pollution Simulation System (CSPSS), which is documented in a user's
manual by Wycoff and Mara (1979). The simulation of James River DO concen-
trations was calibrated to match continuous in-stream DO measurements
collected by the USGS between 1974 and 1977. This calibrated CSPSS model
was then used to simulate the impact of the recently completed (November
1977) AWT facility in Springfield and to quantify the impact of alternatives
to control pollution fron urban stormwater runoff.
Since 1954, six major fishkills have occurred on the James River after
storm events. At the outset of this project, the cause of those major
fishkills was suspected to be urban stormwater pollution. However, as
reported in a paper by Scholl and Wycoff (1981), the analysis, using CSPSS
indicated that, the poor quality secondary effluent of the WWTP, rather
than urban stormwater pollution, was probably the major cause of these
fishkills. It is the purpose of this paper to present information which
verifies the conclusions reported in the paper by Scholl and Wycoff (1981).
The verification was obtained by developing a cumulative duration curve
for DO concentrations observed at the calibration point on the James River
between 1979 and 1980 (post-AWT conditions).
INTRODUCTION
In light of the significant costs associated with water pollution control
projects, it is imperative that the potential for perceptible improvements
in the use of a receiving water body be evaluted prior to committing
public monies for such projects. The potential for perceptible improve-
1Presented at the Stormwater and Water Quality Users Group
Meeting, in Washington, D.C., March 25-26, 1982.
2Water Resources Engineer, CH2M HILL; P.O. Box 1647,
Gainesville, Florida 32602. (904) 377-2442.
90
-------
ments in receiving water use can be determined using water quality simu-
lation if sufficient field data are available.
This paper summarizes the application of the Continuous Stormwater Pollution
Simulation System (CSPSS) to the James River near Springfield, Missouri.
The CSPSS is a water quality planning model developed for the Facilities
Requirements Division of EPA. It was possible to calibrate CSPSS to
observed dissolved oxygen (DO) concentrations in the James River using
continuous water quality data collected by the U.S. Geological Survey
(USGS). After the CSPSS was calibrated, the DO response was predicted for
proposed AWT facilities and possible treatment of urban stormwater.
Details related to DO calibration and prediction were published in a paper
by Scholl and Wycoff (1981). Recent continuous DO data for the James
River provide a unique opportunity to determine if the continuous DO
responses predicted in 1978 using the CSPSS were in fact reasonable.
The discussion which follows includes brief descriptions of the Springfield
study site, CSPSS, and results of the model calibration and verification
for the James River.
SITE DESCRIPTION
The Springfield study area as shown on Figure 1 is located in the Missouri
Ozark Plateau province of the White River basin. The City lies on an
east-west ridge which divides two major watersheds in Missouri. Surface
drainage north of this east-west divide flows into the Osage and Missouri
River basins, and to the south of this divide drainage is into the James
and White River basins. The James River is the major receiving water body
of concern in this paper.
Incorporated in 1846, Springfield experienced little growth until westward
railroad expansions occurred during the 1870'.s. In 1970, the population
was estimated to be 145,000, with an incorporated area of 62.2 square
miles. Improvements to storm drainage and wastewater collection facilities
have not kept pace with recent urban growth, resulting in serious flooding
and pollution problems. For example, six major fishfills have occurred in
the James River downstream from Springfield since 1954. In November of
1977, construction of AWT facilities was completed at the City's Southwest
Wastewater Treatment Plant (WWTP).
A review of water quality planning performed for Springfield's AWT facility,
which cost approximately $41.6 million (ENR approximately 2410), revealed
shortfalls in the water quality planning process. The most apparent had
to do with establishing a cause and effect relationship between pollutant
loads from the Springfield urban area and observed fishkills on the James
River. Since such a relationship was not clearly established, the potential
for eliminating fishkills by constructing AWT facilities was not known.
In fact, it was hypothesized by some investigators that only by capturing
and treating urban stormwater would future fishkills be eliminated.
Five of the six USGS gauging stations within the study area are located as
shown on Figure 1. The sixth station (No. 07050700) is located on the
91
-------
James River upstream from Lake Springfield. The types of data collected
at these stations are identified in Table 1. Continuous DO concentrations
and temperature were monitored at three of these stations (No. 07052100,
07052160, and 07052250) beginning in 1972. Collection of this continuous
data was terminated in 1980 at Station No. 07052250 (Frazier Bridge).
MODEL DESCRIPTION
The CSPSS was developed specifically for use in the receiving water impact
portion of the 1978 and 1980 Needs Surveys. A User's Manual by Wycoff and
Mara (1979) documents the model's theoretical basis and data requirements.
The CSPSS is a computer-based probabilistic simulation model that generates
long-term synthetic records of: (1) rainfall, (2) runoff, (3) runoff
quality, (4) upstream receiving streamflow, (5) excess sewer sytem infil-
tration,, (6) dry-weather WWTP discharges, and (7) receiving water quality
response. In addition, the CSPSS accounts for storage and treatment of
urban runoff. Pollutants considered are: (1) biochemical oxygen demand
(BOD), (2) total kjeldahl nitrogen (TKN), (3) suspended solids (SS),
(4) lead (Pb), and (5) fecal coliform. The recently added fecal coliform
simulation capability is currently operational and will be documented in
an updated version of the user's manual scheduled for completion in 1982.
Receiving water responses can be simulated for DO concentrations, SS
concentrations, total and dissolved lead concentrations, and fecal coliform
organisms.
MODEL CALIBRATION
The continuous DO simulation for the James River was calibrated in 1978
for pre-AWT conditions (Scholl and Wycoff, 1981). This calibration was
based primarily on continuous DO records collected by the USGS at Frazier
Bridge Gauging Station No. 07052250 (see Figure 1) during water years 1974
through 1977. A summary of calibrated values for selected receiving water
variables on the James River is provided in Table 2. The slope model as
proposed by Tsivoglou and Neal (1976) was used to estimate a preliminiary
value of K2, the reaeration coefficient, for DO calibration. After reason-
able values of all other input data were established, K2 was adjusted
until the predicted DO cumulative duration curve closely matched the
duration curve observed at Frazier Bridge. The energy dissipation model,
as proposed by Tsivoglou and Neal (1976) was then used to confirm that the
calibrated K2 value was in fact reasonable. The calibrated pre-AWT
cumulative duration curve simulated by CSPSS and the observed pre-AWT
(1974-1977) curve are shown on Figure 2. The agreement between these two
curves is generally good, especially in the range of 0 to 6 rag/1.
Statistical characteristics of the pre-AWT calibration are presented in
Table 3.
To simulate the probable impact of AWT effluent on DO concentrations in
the James River, the calibrated CSPSS data set was modified to include AWT
effluent characteristics and a reduced benthic oxygen demand of 0.5 g 02/
m2-day. The cumulative duration DO curve for this post-AWT simulation is
shown on Figure 3. To simulate the probable impact of urban stormwater
treatment after AWT facilities are operating in Springfield, the post-AWT
92
-------
CSPSS data set was modified to include storage and treatment of urban
stormwater. The results of these simulations did not alter the resulting
cumulative duration curves enough to justify plotting a line separate from
the post-AWT line shown on Figure 3.
MODEL VERIFICATION
In the fall of 1981, additional DO data for the Frazier Bridge calibration
point became available for water years 1978, 1979, and 1980. Data for the
last 2 years (1979 and 1980) were used to construct an observed post-AWT
cumulative duration DO curve. Records for water year 1978 were not used
since this was considered a transition year between pre-AWT and post-AWT
conditions. These additional data provide a unique opportunity to compare
post-AWT conditions predicted by the 1978 CSPSS simulations (Scholl and
Wycoff, 1981) to actual post-AWT conditions observed during 1979 and 1980.
The observed post-AWT cumulative duration curve is shown on Figure 3 along
with the predicted post-AWT curve. This data verifies that the predicted
DO curve for post-AWT conditions is reasonably close to the observed DO
curve, especially in the range of 0 to 9 mg/1. Statistical characteristics
of the post-AWT verification are presented in Table 3.
A probable cause of disagreement between predicted and observed post-AWT
cumulative duration curves above 9 mg/1, is the low monthly mean tempera-
tures observed during January and February of 1979 and 1980. For the
calibrated CSPSS simulation, a mean monthly temperature of 11.0°C was used
for both January and February. Observed monthly mean temperatures for the
period 1979-1980 were 4.8 and 6.5°C for January and February, respectively.
Thus, the predicted post-AWT cumulative duration curve does not account
for the lower temperatures and higher DO concentrations observed during
1979 and 1980.
Comparing the pre- and post-AWT cumulative duration curves on Figure 3 at
a DO concentration of 5.0 mg/1 indicates that the observed decrease in the
frequency of exceedance was 44.5 percent. The decrease predicted by CSPSS
was 42 percent. In addition, the area between observed pre- and post-AWT
curves indicates that the average increase in DO concentrations at Frazier
Bridge was approximately 3.30 mg/1. The average increase predicted by
CSPSS was approximately 2.90 mg/1.
Based on this analysis, it appears logical to conclude that urban stormwater
was not the cause of low DO and subsequent fishkills on the James River
Rather, it appears that the poor quality secondary effluent of the WWTP
was the likely cause of low-DO fishkills.
CONCLUSIONS
1. Continuous rainfall/runoff/receiving water quality simulation such as
provided by the CSPSS and other available models can be a valuable
tool in AWT planning and can quantify the water quality response of
alternative pollution control strategies.
2. Continuous problem-oriented water quality monitoring programs are
93
-------
extremely valuable aids to the decision-making process and should be
encouraged. Such programs can provide a sound basis for the optimum
investment of public monies.
ACKNOWLEDGEMENTS
This work was fully funded by the Facility Requirements Division of the
EPA as part of the 1982 Needs Survey, Contract No. 68-01-5890. James A.
Chamblee was the project officer.
REFERENCES
1. Scholl, J.E., and Wycoff, R.L. 1981. "Continuous DO Simulation at
Springfield, Missouri," Journal of the Environmental Engineering
Division, ASCE, Vol. 107, No. EE1, Proc. Paper 16021, pp 69-82.
2. Tsivolgou, E.G., and Neal, L.A. 1976. "Tracer Measurement of
Reaeration: III. Predicting the Reaeration Capacity of Inland
Streams," Journal of the Water Pollution Control Federation,
Vol. 48, No. 12, pp 2669-2689.
3. Wycoff, R.L., and Mara, M.J. 1979. "Continuous Stormwater Pollution
Simulation System—User's Manual," EPA-430/9-79-004, U.S. Environmental
Protection Agency, Washington, D.C.
-------
Table 1
USGS GAUGING STATIONS ON WILSONS CREEK AND JAMES RIVER
VD
Location
Wilsons Creek near Springfield, MO
Wilsons Creek near Battlefield, MO
James
James
James
James
River near Springfield, MO
River near Nixa, MO
River near Wilsons Creek, MO
River near Boaz, MO
Station No.
07052100
07052160
07050700
07050750
07051600
07052250
Period of
Record
9-Years
(1972-81)
11-Years
(1968-70)
(1972-81)
26 -Years
(1955-81)
12-Years
(1966-75)
(1977-80)
14-Years
(1967-81)
8-Years
(1972-80)
Drainage Area
(mi*)
31.4
55.0
246.0
273.0
329. Oa
462.0
Type of Data Collected
Flow, continuous DO concen-
trations, and temperature
Flow, continuous DO concen-
trations, and temperature
Flow only
Daily water temperature
only. Discontinued 1980.
Monthly water quality only
Flow, continuous DO concen-
trations, and temperature.
Discontinued in 1980.
Estimated drainage area from USGS quadrangle map.
-------
Table 2
CALIBRATED JAMES RIVER RECEIVING WATER VARIABLES
Calibrated
Receiving Water Variable Value
Mean annual upstream flow9, cfs 325.0
Carbonaceous waste decay rate, Klf day 1 1.00
for stormwater and upstream flow
Carbonaceous waste decay rate, Kx, day 1 1.42
for WWTP effluent
Atmospheric reaeration rate, K2, day 1 3.06
Nitrogenous waste decay rate, K3, day 1 0.30
Sediment oxygen demand, g oxygen 1.00
mz-day
Maximum monthly temperature , °C 30.50
Mean upstream BOD , mg/1 1.89
Mean upstream TKN3, mg/1 0.34
Mean upstream DO deficit , mg/1 1.34
aObserved daily flow values for a representative 5-year
period were input to CSPSS.
Monthly mean values were input to CSPSS.
96
-------
Table 3
STATISTICAL CHARACTERISTICS OF SIMULATED
PRE-AWT AND POST-AWT CUMULATIVE DO CONCENTRATIONS
Statistical Parameter
Number of data points
Average error, % of time equal
Pre-AWT
Value
11
2.08
Post-AWT
Value
12
5.17
to or less than indicated DO
concentration
Relative error, % of time equal 0.03 0.08
to or less than indicated DO
concentration
Standard error of estimate, % of 4.35 11.76
time equal to or less than
indicated DO concentration
Coefficient of variation 0.07 0.19
Linear regression intercept, % of 0.27 4.00
time equal to or less than
indicated DO concentration
Linear regression slope 0.96 0.86
Linear regression correlation 0.995 0.975
coefficient
97
-------
Urban Basin 1
Urban Basin 2
Southwest
lA/astewater
Treatment
Plant
Springfield
City Limits
Spring .
7-1- — -/—._
Lake
Springfield
Wilson's Creek
Battlefield
National Park
MileO i
1 y, 0 1 mile
Scale in miles
07052250--USGS
• Gauging
Station
FIGURE 1. Map of Springfield, Missouri, Study Area (1 mile = 1,607 m).
98
-------
cr
UJ
§3
ll
II
QJ r™"1
—J
o
100
90
80
70
60
50
40
30
20
*
10
Calibrated
CSPSS
Simulation
jtl\
Observed Pre-AWT
(1974-1977)
10 11 12 13
DO Concentration, mg/l
FIGURE 2. Calibration of DO on James River at Frazier Bridge.
99
-------
01 23456
8 9 10 11 12 13 14 15
DO Concentration, mg/l
FIGURE 3. Calibration and verification of continuous DO response on the
James River near Springfield, Missouri.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of 'the
Agency and no official endorsement should be inferred.
I 00
-------
USE OF CONTINUOUS SIMULATION MODEL CALIBRATION
TECHNIQUES TO DEVELOP NONPOINT POLLUTION LOADING FACTORS
By John P. Hartigan1, Thomas F. Quasebarth2,
and Elizabeth Southerland^
Introduction
This paper describes the derivation of nonpoint pollution loading
factors for a River Basin Model (_!) of the 64/000 sq mi drainage area of
Chesapeake Bay (see Figure 1). The purpose of the River Basin modeling
study was to compare the delivery of point source and nonpoint pollution
loadings of nitrogen and phosphorus to the Bay's estuarine system and to
study locational differences in pollutant contributions. The River Basin
Model was also used to evaluate alternate eutrophication management
strategies for the Bay's estuarine system.
To develop nonpoint pollution loading factors for rural-agricultural and
urban land use categories, a version (2) of USEPA's NFS model (3) was
calibrated to several test watersheds characterized by relatively
homogeneous hydrologic characteristics and a single land use. A total of 25
test watersheds were monitored from late 1979 through mid-1981 under a $2.5
million study funded by the EPA Chesapeake Bay Program (_4). An earlier
twelve-month monitoring study (1976-1977) of 16 urban test watersheds in the
Virginia suburbs of Washington, D.C. produced the majority of the urban
nonpoint pollution loading data (2). This paper focuses on the development
of rural-agricultural loading factors and the verification of 1976-1977
urban loading factors described elsewhere (2) through the calibration of the
NPS model to the EPA Chesapeake Bay Program test watersheds. River Basin
Model simulations which verified the nonpoint pollution loading factors are
also described.
Test Watershed Monitoring Studies
In most cases, the test watershed sites covered only one of the
following land use categories: forest; pasture; high tillage cropland; low
tillage cropland; and urban residential. Since urban land uses only
represent approximately 3% of the Chesapeake Bay drainage area, the test
watershed studies focused on rural-agricultural land uses.
As shown in Figure 1, the test watersheds monitored under the EPA
Chesapeake Bay Program were located in Coastal Plain and Piedmont river
basins in the vicinity of the Bay's estuarine system. A summary of the test
watershed monitoring sites is presented below:
^Director, Engineering-Planning Division, Northern Virginia Planning
District Commission (NVPDC), 7630 Little River Turnpike, Annandale, VA 22003
Environmental Planner-Engineer, NVPDC
Environmental Engineer, NVPDC
101
-------
RIVER BASIN
LOCATION
Occoquan River Northern Virginia
Ware River
Pequea Creek
Patuxent River
Chester River
Southeastern Va.
Lancaster, Pa.
Maryland
Maryland
NO. OF
SITES
I
I
;
9
INVESTIGATOR
Virginia Polytechnic Institute
& State University
Va. Institute of Marine Science
U.S. Geological Survey
State of Maryland
State of Maryland
Each test watershed site was equipped with a flowmeter, automatic
sampler, and was served by a continuous recording raingage located within or
nearby the site. Either a natural (e.g., ephemeral stream) or artificial
(e.g., H-flume, Parshall flume) drainage control was typically used to
establish stage-discharge relationships at the outlet of each watershed.
The sampling interval was generally automatically initiated by the flowmeter
at the start of a runoff event. For studies which relied upon
flow-composite sampling methods, the flowmeter activated the sampler at
preselected increments of runoff volume. For studies which relied upon
LOCATION MAP
"CHESAPEAKE
BAY
Figure 1,
• TEST
WATERSHED
Map of Chesapeake Bay Basin
Showing Locations of River
Basins with Test Watersheds:
Pequea Creek (A), Chester
River (B), Patuxent River
(C), Occoquan River (D),
and Ware River (E)
102
-------
sequential-discrete sampling methods (i.e., collection of discrete samples
at numerous points along the runoff hydrograph), the flowmeter activated the
sampler at preselected stage increments. At the larger sites which
exhibited dry weather flow, baseflow samples were periodically collected.
Runoff and baseflow samples were analyzed for plant nutrients and total
suspended solids, with periodic analyses for organics.
At the time the test watershed monitoring studies were designed and
implemented, a work program for data management and model calibration had
not yet been developed, although a follow-up modeling study was under
consideration. The absence of a modeling study work program at the start of
the monitoring studies tended to significantly complicate data
reduction/management activities during the modeling effort and to reduce the
amount of monitoring data that was suitable for model calibration studies.
The final section of this paper discusses specific problems and
recommendations for coordinating future test watershed monitoring and
modeling investigations.
Data Reduction/Management Requirements for Model Calibration
Introduction. The test watershed monitoring investigators reduced the
data required to characterize the runoff pollution loadings from each runoff
and baseflow sample. Investigators who relied upon flow-composite sampling
techniques reported mean flow rate and mean concentration data for the
runoff or baseflow sampling interval, while investigators who relied upon
sequential-discrete sampling methods reported instantaneous flow rate and
instantaneous concentration for each runoff or baseflow sample. All other
meteorologic and hydrologic data required for model calibration typically
had to be reduced by the modeling investigator.
Meteorologic Data. Since the test watersheds were relatively small, a
15-minute time step was required for the continuous rainfall record to
ensure that the rainfall interval was not significantly greater than the
watershed's time of concentration. Drum raingage stripcharts used at the
Pequea Creek and Ware River basin sites were manually reduced, while other
rainfall stripcharts were reduced with a Numonics digitizer equipped with
software to create files with the appropriate time-step. Approximately
1.0-1.5 yrs of stripchart record was reduced for the modeling studies.
Since the monitoring investigators were only required to report and analyze
water quality monitoring data, raingage maintenance appeared to receive the
lowest priority of all equipment checks and considerable gaps were found in
the onsite records. During periods when the sampling station was shut down
due to a breakdown of either the flowmeter or automatic sampler, the
raingage was sometimes shut down until water.quality sampling was resumed.
Since a continuous simulation model requires rainfall records covering the
periods between monitored runoff events in order to calculate antecedent
soil moisture and sediment accumulations for each monitored storm, missing
rainfall records for periods when the onsite raingage was shut down had to
be constructed from a nearby raingage. Separate software was developed to
create a continuous rainfall file for input to the NPS model.
103
-------
The other meteorologic input file required for the NFS model calibration
study is a time series of daily potential evapotranspiration for the test
watershed monitoring period. A continuous potential ET record was developed
for each river based upon meteorologic data at the nearest NWS station.
Hydrologic Data. Since it is advisable to use a long-term runoff record
to calibrate a continuous simulation model, the reported flow records for
runoff events with water quality samples were expanded to include all flow
records collected during the monitoring period. Although all monitoring
sites were equipped with continuous recording flowmeters, the Pequea Creek
test watersheds were the only sites with a complete daily streamflow
record. The flow records at some sites were restricted to those runoff
events with water quality samples. Stripcharts with additional flow records
were reduced with the same digitizer software used to construct rainfall
records.
Since the quality control programs of most monitoring investigators
concentrated on laboratory analyses, the modeling investigator was required
to perform most of the quality assurance checks on the hydrometeorologic
datasets. The digitized rainfall and runoff records were integrated for
each storm event and rainfall and runoff volumes were compared to identify
potential water balance problems due to such factors as backwater, an
incorrect stage-discharge relationship or an error in the reported flowmeter
setting. Flow stripcharts for monitored storms were also checked for
"flat-top" hydrographs which indicated that the maximum stage exceeded the
full-scale setting of the flowmeter. In cases of spurious rainfall/runoff
ratios or "flat-top" hydrographs, the runoff and water quality data were
deleted from the observed dataset for model calibration.
Water Quality Data. Software was developed to calculate total storm
loads from datasets with mean or instantaneous flow rates and concentrations
for each runoff sample. For test watersheds with flow-composite samples,
the product of mean flow rate and mean concentration was multiplied by the
storm duration to calculate total load. For test watersheds with sequential
discrete samples, the time series of instantaneous loading rates (i.e.,
product of instantaneous flow rate and concentration) was numerically
integrated between the first and last sample time to calculate total loads
for each storm. Mean storm concentrations were also calculated for the
sequential discrete sampling datasets. Dry weather flow concentration
statistics were calculated for test watersheds with significant amounts of
baseflow.
Test Watershed Data. Land use and drainage area data was typically
based upon maps, drawings, and tables compiled by the monitoring
investigator, which were checked through site inspections and with available
aerial photographs and topographic maps. At one test watershed where a
check of rainfall/runoff ratios revealed a serious water balance problem,
the authors performed a plane-table survey which produced a significant
increase in the drainage area and more reasonable rainfall/runoff
relationships. For urban test watersheds, percent imperviousness was
determined by planimetering aerial photographs or site plans.
-------
Soils characteristics for each test watershed were derived from county
soil series maps and surveys. The predominant hydrologic soil group was
determined and average values of permeability, total water holding capacity,
and erodibility were calculated for use in deriving hydrologic and nonpoint
pollution model parameters.
Average overland flow slope was typically reported by the monitoring
investigator. Reported values were checked with 1:24,000 scale maps of the
test watershed and surrounding areas.
For cropland sites, data on monthly vegetative cover and the timing and
extent of tillage activities are required to accurately model soil loss.
Based upon discussions with local SCS staff and information on the timing
and extent of harvest and tillage operations at the test watershed, a time
series of monthly ground cover was derived for each cropland site. Since
the NFS model does not permit the user to alter ground cover time series
from year-to-year, two different input datasets were sometimes required to
model cropland watersheds with more than one year of monitoring data due to
changes in harvest and or tillage dates from one year to the next. The
monthly ground cover time series and tillage parameter values were refined
during model calibration to help achieve acceptable agreement between
simulated and observed sediment loadings.
Although the monitoring investigators made an effort to select test
watersheds which included only one land use, finding an acceptable catchment
with a single land use was not always possible. For test watersheds with a
mixed land use pattern, aerial photographs of the test watershed were
checked to determine the character and extent of any secondary land uses.
In cases where nonpoint pollution loading rates from the secondary land use
were significantly higher than the loading rates for the primary land use,
the complexity of the model calibration study was significantly greater. An
example is the forestland test watershed (Site #2) in the Pequea Creek basin
which included a high-tillage cropping site (approximately 13% of drainage
area) that contributed the majority of the nonpoint pollution loads during
major storm events. In such a case, loading factors for the secondary land
use had to be set based on model calibration for a similar test watershed
prior to model calibration at the mixed land use site. Also, the monitoring
dataset had to be screened to identify those storm events (e.g., minor
runoff events in the case of Pequea #2) which are least likely to be
characterized by major nonpoint pollution loading contributions from the
secondary land use.
Description of NFS Model
Introduction. NFS is a continuous simulation model which includes
hydrologic and nonpoint pollution loading submodels that are executed
sequentially. A flow chart summarizing the processes represented by the
model is shown in Figure 2.
105
-------
in
3
-------
Nonpoint Pollution Submodel. The nonpoint pollution loading submodel of
the modified NFS model (.2) operates on rainfall intensity records and on the
hydrologic submodel's output of surface runoff and subsurface flow records.
Unlike other options available in the HSPF package, such as the ARM model
(2)i the NPS model treats all constituents conservatively and does not
transport pollutants between the idealized soil moisture storage
compartments.
For cropland, the model assumes that sediment generation and washoff are
the driving forces for loadings of all pollutants. Cropland loadings of
sediment, which are calculated from 15-minute rainfall records with a soil
loss algorithm related to the Universal Soil Loss Equation, are assigned
sediment "potency factors" (i.e., ratio of pollutant mass to sediment mass)
to calculate loadings of other pollutants. For urban and pasture land uses,
nonpoint pollution washoff algorithms relate the washoff of accumulated
pollutant loads to the simulated runoff rate in each time-step. Accumulated
pollutant loads at the start of a rainstorm are calculated from the "daily
pollutant accumulation rates" (Ibs/ac/day) assigned to each land use
classification to represent the buildup of pollutants on the land surface
and in the atmosphere (i.e., air pollution). For the forestland category,
pollutant loading calculations are based upon soil loss/potency factors as
well as daily pollutant accumulations, with the former more prominent during
periods of low leaf cover and the latter more prominent during periods of
high leaf cover.
Given the state-of-the-art of nonpoint pollution loading models, loading
factors such as sediment potency factors and daily pollutant accumulation
rates are probably best viewed as empirical factors which can provide a
reasonable approximation of a land use's nonpoint pollution loading
potential, much like the C coefficient in the "rational formula" is viewed
as an empirical factor that relates rainfall intensity to peak runoff. The
model has the capability to use monthly variations in pollutant loading
factors. This feature permits a representation of variations in the
pollutant loading potential of cropland areas due to such factors as
fertilizer/manure applications, crop harvest, etc. Subsurface flow loadings
based upon user-specified concentrations are added to hourly runoff
pollution loadings and delivered to the outlet of the test watershed.
Test Watershed Model Calibration
Introduction. Not all of the test watersheds were characterized by
sufficient land use homogeneity and hydrometeorologic data to permit NPS
model calibration. Sufficient hydrometeorologic and water quality data was
available to calibrate the NPS model to 11 of the 12 acceptable sites in the
Occoquan River, Ware River, and Pequea Creek basins. A summary of the sites
with monitoring records suitable for model calibration is shown in Table 1.
Due to the late start of the Maryland test watershed studies, insufficient
hydrometeorologic data was available for model calibration of the single
land use sites in the Patuxent and Chester basin sites. Therefore, analyses
of monitoring data from the Maryland test watersheds were restricted to
standard statistical tests. Following model calibration studies of the
Virginia and Pennsylvania test watersheds, linear regression analyses of
107
-------
monitoring datasets were used to relate the calibrated loading factors to
the Maryland test watershed data.
Table 1
SUMMARY OF MODELED TEST WATERSHED CHARACTERISTICS AND HYDROLOGY CALIBRATION RESULTS
REGRESSIONS OF SIMULATED AND OBSERVED FLOW VOLUMES
LAND USE/SITE
AREA
(acres)
MONITORED STORMS
N
A. HIGH TILLAGE CROPLAND
A. PEQUEA #3 115.2
B. WARE #7 16.2
B. LOW TILLAGE CROPLAND
A. OCCOQUAN #2 26.6
B. OCCOQUAN #10 25.8
C. PASTURE
15
7
8
7
SLOPE
0.76
0.72
0.98
1.03
DAILY STREAMFLOWS
SI SLOPE R2
0.88
0.99
0.98
0.99
492a
0.98
0.70
A. OCCOQUAN #1
B. OCCOQUAN #5
D. FOREST
A. PEQUEA #2
B. OCCOQUAN #9
C. WARE #8
E. RESIDENTIAL
A. PEQUEA #4
B. WARE #5
31.3
18.8
128.0
75.8
17.4
147.2
6.2
5
5
18
7
9
26
17
0.81
1.07
0.69
1.11
1.15
0.86
0.80
0.95
0.90
0.62
0.95
0.97
0.98
0.92
222E
0.7
0.79
374C
0.96
0.84
aMay 23, 1979 - September 26, 1980
''May 23, 1979 - December 31, 1979
cMay 23, 1979 - May 31, 1980
For most test watersheds, the nonpoint pollution monitoring records were
not extensive enough to permit subdividing the dataset into separate
calibration periods. Consequently, the entire test watershed monitoring
dataset was used for NFS model calibration. The calibrated NFS loading
factors were verified through applications to mixed land use river basins in
the Chesapeake Bay drainage area. Even in the absence of verification in
the Chesapeake Bay river basins, it is felt that the risk of producing
biased calibration results from the test watershed modeling studies are
significantly reduced by the use of continuous simulation calibration
techniques which involve long-term simulations and parameter adjustments
that are not keyed to individual storm events.
108
-------
Hydrology Calibration. Due to the relatively small size of the test
watersheds, subsurface flows were often not detectable at the monitoring
stations and therefore, baseflow and interflow components of runoff had to
be suppressed in hydrologic model calibrations for most sites. Typically,
baseflow and interflow were only included in models of forested watersheds
where the dry weather flow component represented a significant fraction of
monitored flows.
For most test watersheds, hydrologic calibration focused on achieving
acceptable agreement between simulated and observed storm volumes. Each
test watershed model was iteratively executed with a continuous rainfall
record which bracketed the 1-2 year monitoring period, and agreement with
monitored flows was checked for each model parameter set. A 3-6 month
antecedent rainfall period was used for most test watersheds to minimize the
impacts of the assumed soil moisture conditions at the start of the
simulation period. Scatterplots and simple linear regressions of simulated
and observed runoff volumes were generated for each calibration run to guide
parameter adjustments. Sample comparisons of the final simulated and
observed runoff volumes for monitored storm events are shown in Figure 3.
Based on the goodness-of-fit regression statistics for monitored storms
presented in Table 1, it was concluded that acceptable hydrology calibration
had been achieved at the 11 test watersheds. For the three Pequea Creek
test watersheds, continuous -daily flow records were also available for
calibration. As shown in Table 1, acceptable regression statistics were
achieved for all three Pequea Creek test watersheds. The lower slope
term for the forested watershed (Pequea 12) is probably due in large part to
the underlying Conestoga Valley limestone formations which appeared to
contribute subsurface flows that originated outside the drainage area.
Nonpoint Pollution Loading Calibration. After achieving an acceptable
hydrologic calibration, nonpoint pollution loading factors were calibrated
for each test watershed by iteratively executing the NPS model with
continuous meteorologic records for the entire monitoring.period and
checking model projections for monitored storms.. The calibration of
nonpoint pollution loading factors for total P and total N focused on the
agreement of simulated loadings with monitored storms which exhibited
acceptable hydrologic simulations. Goodness-of-fit evaluations were based
upon conventional and nonparametric statistical analyses.
For urban land uses, daily pollutant accumulation rates developed by the
earlier study (2) for pervious and impervious fractions were tested
with the Chesapeake Bay Program monitoring data. The calibration techniques
used to derive separate loading factors for the pervious and impervious
fractions are described elsewhere (.2) • These urban loading factors were
held constant in the models for Pequea 14 and Ware 15 to see how well they
represented loadings in different regions under different meteorologic
conditions. As was the case in the previous urban modeling study, the daily
pollutant accumulation rates were held constant from month to month.
For cropland test watersheds, monthly variations in sediment potency
factors were required to account for such factors as fertilizer/manure
applications and crop harvest, with the higher potencies generally
109
-------
NO TILLAGE CROPLAND'
OCCOQUAN 2 (26.6 AC)
Ijj 60000
3
40,000
§80,000
flC
m
R8 • 0.96
SLOPE'0.98
tojooo 40,000 eouooo
OBSERVED RUNOFF VOLUME (CU FT)
MEDIUM DENSITY SINGLE FAMILY RESIDENTIAL
t 6<*>oo- PEQUEA 4 (147.2 AC)
3
~ 30*000
u
40JOOO
80(000
1 20.000
(e
S
3
IQ.OOO
0
R2-0.98
SLOPE « 0.86
20,000 4Q000 «0|000
OBSERVED RUNOFF VOLUME (CM FT)
u
MINIMUM TILLAGE CROPLAND*
OCCOQUAN 10 (25.8 AC)
R* * 0.99
SLOPE* 1.03
§
0 £0,000 4O.OOO 6OQOO
OBSERVED RUNOFF VOLUME ICU FT)
J.ARGE-LOT SINGLE FAMILY RESIDENTIAL'
WARE 5 (6.2 AC)
Re«0.92
SLOPE' 0.80
'O 20JOOO 40,000 60,000
OBSERVED RUNOFF VOLUME {CM FT)
Figure 3. Sample Regressions of Simulated and Observed Runoff Volumes for Monitored
Storm Events: Occoquan #2, Occoquan #10, Pequea #4, and Ware #5
-------
associated with the months characterized by highest percentages of
vegetative cover and vice versa. The monthly distribution of potency
factors is established during model calibration by deriving the upper limit
for the summer months of high ground cover and the lower limit for the
winter months of low ground cover. For forestland test watersheds, monthly
variations in sediment potency factors and daily pollutant accumulation
rates were required to account for the variations in ground cover and leaf
litter. Table 2 illustrates the relationship between monthly potency factor
and monthly ground cover for cropland and forest land uses. For pasture
test watersheds, monthly variations in sediment potency factors were
generally not required to achieve an acceptable calibration.
The conventional goodness-of-fit evaluations included scatterplots and
linear regressions of simulated and observed storm loads and comparisons of
simulated and observed volume-weighted mean concentrations for the entire
monitoring period. The simulated volume-weighted mean concentration was
calculated by summing the loads and runoff volumes for all storm events
within the simulation period/ including storms which were not covered by the
monitoring study. The protocol for NFS loading factor adjustment after each
calibration placed greater emphasis on the agreement of volume-weighted mean
concentrations, since long-term loading trends were felt to provide the best
indication of the need for and direction of further loading factor
adjustments. In other words, whenever a parameter adjustment decision
involved choosing between improving volume-weighted mean concentration vs.
improving the storm load scatterplots and linear regression statistics, the
former usually governed.
Comparisons of simulated and observed volume-weighted mean
concentrations for the calibrated NFS loading factors are shown in Table 3.
As may be seen, the ratios of simulated to observed mean concentrations
typically fell within the range 0.75-1.25 which is comparable to the typical
errors inherent in hydrometeorologic gaging and laboratory analyses. Thus,
Table 3 indicates that the calibrated NFS loading factors provide a good
representation of long-term nonpoint pollution loads per unit volume during
the monitoring period.
Some of the better scatterplots and regression results for storm loads
are shown in Figure 4. As may be seen, agreement between simulated and
observed storm loads was quite good for selected sites. However, regression
statistics for several other test watersheds were insufficient to
demonstrate goodness-of-fit for storm loads. It is felt that much of the
difficulty in achieving acceptable regression statistics for storm load
comparisons can be attributed to the lower power of conventional normal
statistics for evaluations of small sample sizes characterized by skewed
(i.e., non-normal) distributions.
To provide a visual check of agreement between simulated and observed
frequency distributions, box and whisker plots were developed for the
simulated and observed storm load datasets for each test watershed. As
illustrated in Figure 5, the box and whisker plot displays the 25th, 50th,
and 75th percentile values in the frequency distribution as well as the
upper and lower extremes. The departure of the 25th and 75th percentile
111
-------
N>
Nonpoint
Table 2
Pollution Loading Factors Applied to Chesapeake Bay Drainage Area:
Monthly Distributions for Potomac and James River Basins
LAND USE/PARAMETER
JAN
FEE
MAR
APR
MAY
A. FOREST
1. GROUND COVER (%)
2. SEDIMENT POTENCY : TOTAL N (%)
3. SEDIMENT POTENCY: TOTAL P <%)
B. HIGH-TILLAGE CROPLAND
1. GROUND COVER (%)
2. SEDIMENT POTENCYi TOTAL M (%)
3. SEDIHEHT POTENCY: TOTAL P (%)
C. LOW-TILLAGE CROPLAND
1. GROUHD COVER (%)
1
2. SEDIMENT POTENCY i TOTAL M (%)
3. SEDIMENT POTENCY. TOTAL P {%)
O. PASTURE
1. GROOM) COVER (%>
2. SEDIMENT POTENCY: TOTAL N (%)
3. SEDIMENT POTENCY: TOTAL P (%)
B. SINGLE FAMILY RESIDENTIAL
1. GROUND COVER (%)
95%
1.06
0,15
0.0%
0.98
0,31
40%
2.06
0.17
loot
1.52
0.26
100%
95%
1.06
0.1S
0.0%
0.98
0.31
60%
2.06
0.17
100%
1.52
0.26
100%
98%
1.06
0.15
0.0%
1.22
O.38
75%
2.06
0.17
100%
1.52
0.26
100%
100%
1.06
0.15
0.0%
1.22
0.38
85%
2.52
0.21
100%
1.52
0.26
100%
100%
1.06
0.15
20%
1.22
0.38
92%
3.20
0.27
100%
1.52
O.26
100%
100%
1.88
0.26
50%
1.22
0.38
99%
3.20
0.27
100%
1.52
0.26
100%
100%
1.88
0.26
85%
1.83
0.58
99%
3.43
0.28
1OO%
1.52
0.26
100%
100%
1.88
0.26
90%
1.83
0.58
99%
3.43
0.28
100%
1.52
0.26
100%
tfftf
100%
i.ae
0.26
95%
1.83
0.58
99%
3.43
0.28
100%
1.52
0.26
100%
IA,£
98%
1.06
0.15
20%
0.98
0.31
70%
2.06
0.17
100%
1.52
0.26
100%
nL/v
95%
1.06
0.15
0.0%
0.98
0.31
30%
2.06
0.17
100%
1.52
0.26
100%
UfiU
95%
1.06
0.15
0.0
0.98
0.31
40%
2.06
0.17
100%
1.52
0.26
10O%
2. ACCUMULATION RATE: TOTAL K
(LBS/AC/DAY)
a. PERVIOUS FRACTION
b. IMPERVIOUS FRACTION
3. ACCUMULATION RATE: TOTAL P
(LBS/AC/DAY)
a. PERVIOUS FRACTION
b. IMPERVIOUS FRACTION
0.02
0.08
1.0035
0.01
0.02
0.08
0.0035
0.01
0.02
0.08
0.0035
0.01
0.02
0.08
0.0035
0.01
0.02
0.08
0.0035
0.01
0.02
0.08
0.0035
0.01
0.02
o.oa
0.0035
0.01
0.02
0.08
0.0035
0.01
0.02
O.O8
0.0035
0.01
0.02
o.oa
0.0035
0.01
0.02
0.08
0.0035
0.01
0.02
o.oe
O.O035
0.01
-------
MINIMUM TILLAGE CROPLAND
(OCCOQUAN 10)'TOTAL N
R*»0.92
SLOPE-0.93
2940676910
OBSERVED LOAD (lb»)
MEDIUM DENSITY SINGLE FAMILY RESIDENTIAL
(PEQUEA 4)- TOTAL N
R2 - 0.78
SLOPE • 1.37
5 IO IS 20 £8 3O 39
OBSERVED LOAD (lb»)
1.2
I 1.0
a 0.8
o
IU _ .
H 0.4
MINIMUM TILLAGE CROPLAND
(OCCOQUAN 10)-TOTAL P
0.2
tf>
00.
Re«0.89
SLOPE • 0.82
0.0 02 0.4 0.6 OB 1.0
OBSERVED LOAD (Ibs)
1.8
MEDIUM DENSITY SINGLE FAMILY RESIDENTIAL
(PEQUEA 4)' TOTAL P
R*»0.85
SLOPE » O.70
I 2 3 4 8 6 T 6
OBSERVED LOAD (lb«)
Figure 4. Sample Regressions of Simulated and Observed Loadings for Monitored Storms:
Occoquan #10 and Pequea #4
-------
Table 3
Comparison of Simulated and Observed Volume-Weighted Mean Concentrations
SITE (LAND USE)
PEQUEA 3 (H.T. CROP)
WARE 7 (H.T. CROP)
OCC. 2 (L.T. CROP)
OCC, 10 (L.T. CROP)
OCC. 9 (FOREST)
PEQUEA 2 (FOREST)
WARE 8 (FOREST)
OCC. 1 (PASTURE)
OCC. 5 (PASTURE)
PEQUEA 4 (RES ID.)
WARE 5 (RESID.)
OBS.
N
17
10
16
13
15
21
34
27
11
52
30
SIM.
(MG/L)
783
272
370
138
83
99
64
561
166
115
50
SEDIMENT
OBS.
(MG/L)
829
222
361
121
70
169
71
670
145
194
38
RATIO
0.94
1.23
1.02
1.14
1.19
0.58
0.90
0.84
1.14
0,56
1.32
SIM.
(MG/L)
4.53
0.70
1.96
0.47
0.13
0.1
0.05
0.94
0.43
0,24
0.12
TOTAL P
OBS.
(MG/L)
4.70
0.62
1.67
0.40
0.13
0.13
0.06
1.12
0.38
0.30
0.10
RATIO
0.96
1.13
1.17
1.18
1.0
0.74
0.83
0.84
1.13
0.80
1.23
SIM.
(MG/L)
18.7
1.6
6.8
4.6
1.1
3.8
0.37
5.3
2.5
1.8
0.95
TOTAL N
OBS.
(MG/L)
19.2
1.30
6.6
3.8
0.9
3.6
0.4
6.2
2.2
2.4
0.70
RATIO
0.98
1.20
1.03
1.22
1.22
1.05
0.93
0.86
1.15
0.75
1.36
Table 4
Nonparametric Goodness-of-Fit Statistics for Test Watershed Model
Calibration: Runoff Volumes (R.O.) and Total Phosphorus (TP),
Total Nitrogen (TN), and Sediment (SED) Loadings (Ibs)
SITE (LAND USE)
PEQUtA 3 (H.T. CROP)
WARE 7 (H.T. CROP)
OCC. 2 (L.T. CROP)
OCC. 10 (L.T. CROP)
OCC. 9 (FOREST)
PEQUEA 2 (FOREST)
WARE 8 (FOREST)
OCC. 1 (PASTURE)
OCC. 5 (PASTURE)
PEQUEA 4 (RESID.)
HARE 5 (RESID.)
RUNOFF
INTERVAL
(IN.)
sin.
STORMS
OBS.
STORMS
TWO-SIDED
K-S TEST WILCOXON RANK SUM
LEVEL OF SIGNIFICANCE
R.O. TP
> 0.025
(0.01-0.125
+ HURR.)
>0.015
>0.025
>0.03
<0.09
(0.075-0.9)
(0.025-0.5)
>0.01
> 0.025
>0.09
28
26
29
19
30
91
20
15
21
86
13
13
9
6
8
9
16
24
9
8
25
24
0.20
0.10
>0.20"
>o.ir
>0.20"
0.42
>0.20'
>0.10*
>0.20'
0.94
0.58
>0.20'
>0.20"
>0.20'
J>0.11'
"~ >0.20*
0.34
0.10
>0.10«
>0.20*
0.39
0.22
TN
>0.20§
>0.20*
0.10
>0.11'
>0.20*
0.26
0.10
>0.10*
0.20
0.44
0.29
SED
>0.20-
XK20"
>0.20"
>0.11'
N/A
N/A
N/A
N/A
N/A
N/A
N/A
TEST
LEVEL OF SIGNIFICANCE
R.O.
0.58
0.17
0.71
0.65
0.56
0.30
0.86
0.40
0.37
0.76
0.31
TP
0.96
0.23
0.91
0.97
0.45
0.61
0.31
0.95
0.54
0.70
0.64
TN
0.65
0.42
0.20
0.94
0.80
0.18
0.13
0.81
0.25
0.48
0.34
SED
0.83
0.84
0.39
0.98
N/A
N/A
N/A
N/A
N/A
N/A
N/A
"EXCEEDS MAXIMUM PROBABILITY VALUE CURRENTLY REPORTED IN STATISTICAL TABLES FOR SAMPLE SIZES
-------
T UPPER EXTREME
UPPER QUARTILE
MEDIAN
LOWER QUARTILE
•"• LOWER EXTREME
Figure 5. Configuration of a
Box and Whisker Plot
lines from the median line provides an indication of skewness of the
distribution. The mean value is sometimes plotted on the box and whisker
diagram to highlight departure from the median value and the non-normality
of the distribution. Box and whisker plots for the calibrated NFS loading
factors are shown in Figures 6, 1, and 8 for cropland, forest/pasture, and
urban land uses, respectively. To provide an indication of non-normality,
the location of the mean in each plot is indicated by an "o". Since sediment
is the driving force for simulated cropland loadings, sediment data is
presented for the cropland test watersheds in Figure 6.
As was the case with the simulated volume-weighted mean concentrations
reported in Table 3, the simulated box and whisker plots are based on all
storms which occurred during the test watershed monitoring period, including
those which were not monitored. The inclusion of all storms tended to
automatically skew the simulated distribution in the direction of minor
storms which typically were not monitored in the field due to the very small
runoff volumes. This skewness can be attributed to the fact that whereas
the mathematical model will calculate runoff volumes and loads from minor
storms, the test watershed monitoring studies relied upon runoff volume
thresholds associated with more significant rainfall events. Since similar
runoff volume distributions are required to ensure meaningful statistical
comparisons of storm loading datasets, minor storms were generally deleted
from the simulated dataset prior to the development of the box and whisker
plots shown in Figures 6-8. The establishment of the cutoff for minor
storms was based upon iterative analyses of box and whisker plots and
nonparametric statistics for the simulated and observed runoff volume
datasets. The storm runoff volume thresholds which resulted in box and
whisker plots and nonparametric statistics that indicated acceptable
agreement between the simulated and observed runoff volume distributions and
median values are summarized in the "runoff interval" column of Table 4. A
check of the total runoff volume and nonpoint pollution load produced by
115
-------
o.ov
OT
U 2.TS
HI
3 1.69
§
1.10
U.
1 030
IE
O.O
4
e
i
il
I
i
.OS OS OS OS
PEOUEA 3 WARE T OCC. 2 OCC W
(H.TCROP) (H.T. CROP) (LT. CROP) (LT. CROP)
ItO
UJ 1.3
= L2
§09
§
OS
U.
K
0.0
ll
0 9
]
-Li
0 8
I |
U
ii
ik
,
A!
OCC. 9 PEQ. 2 WARE B OCC I OCC. 5
(FOREST) (FOREST) (FOREST) (PASTURE) (MSTURE)
IOW,
G iso.
x
S 120.
-J
III
2
o 60.
Ill
CO
30.
an
r
(el |«
P
S) i
1
i,
1
.
A!
S OS
PEQUEA9 WARE 7 OCC. 2 OCC W
QitCROP) (H.T.CROP) (LT. CROP) (Lt CROP)
V.0U
5 0.28
V
to 020
O
J
** O.I8
« a 10
t-
O.O
I
i
IT
I
iU
I
1
1
OS OS OS 08 OS
OCC. 8 PEQ. 2 WARE a OCC. I OCC. 9
(FOREST) (FOREST) (FOREST) (PASTURE) (PASTURE)
ZJO
t.e
1
r
c
] r
1
i,
1
OS OS OS OS
PEOUEA 3 WARE T OCC. 2 OCC K>
(H.T.CROP) (H.T.CHOP) (LT. CROP) (LX CROP)
WJO
— aos
U
a
— aos
a.
-i oat
*~ Ml
0.0
[
il
I I
I
_3
1*
1
1
0.8
as
ai
no
'!
•
h
r
[
1 ,
I
i
J
1
^^ f
]
OCC. » PEQ. 2 WARE 8 OCC I OCC. 9
(FOREST) (FOREST) (FOREST) (PASTURE) (PASTURE)
Figure 7. Comparison of
Simulated (S) and
Observed (O) Box and
Whisker Plots for
Forest and Pasture
Watersheds
S OS OS OS
PEQUEA3 WARE r OCC 2 OCC W
(H.TCROP) (H.T.CROP) (LT. CROP) OJ. CROP)
Figure 6. Comparison of Simulated (S) and
Observed (O) Box and Whisker Plots
for Cropland Watersheds
-------
INCHES)
M KJ fri
b oi a
Ul
L
u.
lt
U.
I0"9
<
n
>
r
OS OS
PEQUEA4 WARES
(RESIOENTIAU (RESIDENTIAL)
08 08
PEQUEA4 WARE B
(RESIDENTIAL) (RESIDENTIAL)
v.vo
— O.OS
u
OT a04
m
"" OJOS
a.
^ 0.02
*~ OX)I
nn
• I
1 t
1
<
1
.
.
1
.
•
08 OS
PEQUEA4 WARE 5
(RESIDENTIAL) (RESIDENTIAL)
Figure 8. Comparisons of Simulated (S) and Observed (O) Box and
Whisker Plots for Single-Family Residential Watersheds
storm events which fall within the specified runoff interval indicates that
typically 90% or more of the total volume or load produced during the
monitoring period was generated by these storms. This check suggests that
the minor storms deleted from the simulated and observed datasets were
relatively insignificant in terms of seasonal or annual loads.
The runoff intervals shown in Table 4 were used as the basis for the box
and whisker plots of simulated and observed datasets in Figures 6-8.
Inspection of these box and whisker plots confirms the highly skewed
distributions of the runoff and nonpoint pollution loading datasets. The
similarities between the simulated and observed plots, in terms of
distribution and median value, also serves as graphical evidence of model
goodness-of-fit.
Nonparametric statistical tests (8) were performed for a quantitative
assessment of the goodness-of-fit for the datasets plotted in Figures 6-8.
Nonparametric statistics assume no shape for the population distribution and
therefore are valid for both normal and skewed distributions. Consequently,
nonparametric statistical techniques have e much higher power than normal
statistical techniques for analyses of datasets, such as the monitored storm
load dataset, which are characterized by small sample sizes and skewed
distributions. The results of two-sided Kolmogorov-Smirnov (K-S) and
Wilcoxon Rank Sum tests are summarized in Table 4. The K-S analysis is a
test for any significant deviation of the simulated distribution from the
observed distribution. The analysis involves checking the maximum
difference between simulated and observed distributions to determine if it
exceeds a critical value. Since it is a broad alternative test, the K-S
test has lower power for any specific alternative, such as a difference in
median values. The Wilcoxon Rank Sum analysis compensates for this
deficiency since it is designed to test for differences in median values,
under the assumption that the simulated and observed distributions may
117
-------
differ only with respect to this value. The Wilcoxon Rank Sum test assigns
ranks to the combined dataset of simulated and observed values and
calculates the sum of the data ranks for each dataset. If the simulated
median value differs significantly from the observed median value, the sum
of the simulated data ranks will be higher or lower than the sum of the
observed data ranks.
Based on a 0.05 probability cutoff for the 95% confidence interval, the
level of significance statistics in Table 4 indicate that the simulated
runoff volumes and nutrient loads do not vary significantly from those
monitored in each test watershed. The high significance levels of the
nonparametric tests summarized in Table 4 meet the primary objective of a
goodness-of-fit evaluation by indicating a low probability of accepting a
false model as true (Type II error).
Determination of Representative NPS Loading Factors. The purpose of the
test watershed studies was the development of nonpoint pollution loading
factors for application throughout the 64,000 sq mi drainage area of
Chesapeake Bay. Of the 11 modeled test watersheds, only the two urban sites
relied upon a single set of loading factors for the land use category.
Since the urban loading factors developed by a previous study (2) provided a
good representation of urban loadings in two different sections of the study
area, it was decided that these loading factors would be used for all
residential and commercial land uses in the .Chesapeake Bay Basin. The
transferability of urban nonpoint pollution loading factors is not
surprising because impervious cover is such an important contributor to
urban nonpoint pollution loadings (2) and an urban land use tends to exhibit
similar impervious cover patterns regardless of location.
Differences in calibrated loading factors at the test watersheds in each
rural-agricultural land use category can be attributed to variations in
management practices and in the significance of sediment loadings. For each
land use category, volume-weighted mean concentrations for modeled and
unmodeled (i.e., Patuxent and Chester rivers) test watersheds were compared
to ascertain long-term loading differences among the testing sites. For the
forest land use category, a review of long-term loading statistics for 3
modeled and 3 unmodeled sites indicates that Occoquan #9 is characterized by
mean concentrations which are similar to the mean concentrations at most
other forest sites. Therefore, the calibrated sediment potency factors and
pollutant accumulation rates for Occoquan #9 were selected as the most
representative forest loading factors for application throughout the
Chesapeake Bay drainage area.
The selection of representative pastureland loading factors was
influenced by the limitations of the land use database for the Chesapeake
Bay Basin which is based upon interpretations of LANDSAT satellite images
from the period 1977-1979. The LANDSAT data interpretations tend to
emphasize reasonably well-managed pasture (e.g., Occoquan 15) rather than
poorly-managed pastureland (e.g., Occoquan #1) since the latter is difficult
to distinguish from low-tillage cropland. Therefore, the calibrated loading
factors for Occoquan #5 were felt to be most appropriate for application to
the Chesapeake Bay Basin.
118
-------
For the cropland land use categories, variations in management practices
such as manure applications produced different monthly and average annual
potency factors for each watershed. Because sediment is modeled as the
driving force for nonpoint pollution loadings, comparisons of test
watersheds to identify representative loading factors were based upon
average annual sediment potency factors. Monitored pollutant loads were
regressed with monitored sediment loads for each site, and the slope of the
regression line was designated as an average annual sediment potency factor
(i.e., pollutant mass/sediment mass) which could be used to compare site
loading factors with factors for the high tillage or low tillage cropland
datasets. The monitored storm load datasets were then pooled by land use
category, and separate pollutant load vs. sediment load regressions were
Performed for the high-tillage cropland and low-tillage cropland datasets.
In this manner, pollutant loading factors for test watersheds which were not
suited to model calibration could be compared with factors for modeled
watersheds. Likewise, average annual sediment potency factors for
calibrated watersheds could be compared with average annual values for the
entire high tillage cropland or low tillage cropland datasets. The land
useisite" ratios of the regressed sediment potency factors were multiplied
by the calibrated average annual sediment potency factors for Pequea 13 and
Occoquan #10 to develop average annual sediment potency factors for
high-tillage cropland and low-tillage cropland, respectively. The average
annual potency factor was then distributed to monthly values based upon the
distributions calibrated for Pequea #3 and Occoquan #10.
The resultant NFS loading factors for rural-agricultural and urban land
use categories are summarized in Table 2. Based on test watershed model
calibration results, monthly ground cover (COWEC) for urban and pasture
land uses was set at 100% so that pervious area loadings are governed
entirely by the calibrated pollutant accumulation rate rather than soil
loss. For the other land use categories, ground cover was based upon the
calibrated values for the test watershed used to derive the representative
loading factors: Occoquan #9 for forestland, Pequea #3 for high-tillage
cropland, and Occoquan #10 for low tillage cropland. The ">re8tland and
cropland ground cover values shown in Table 2 were used to model the river
basins in the southern half of the Chesapeake Bay drainage area (e.g.,
Potomac and James river basins). For the Susquehanna River Basin, which
occupies the northern half of the Bay's drainage area, the ground cover and
corresponding sediment potency factors were shifted one month to represent
the shorter growing season and earlier crop harvest.
I. orae, to coe annual^onpoint poUution ioaain^f «. ^various
Sis i=a! o? t PUat .evince ana a „
overland flow slope. The NPS model was executed with hourly rainfall
records from the Virginia suburbs of Washington, D.C. to simulate annual
loadings for each land use based upon the NPS loading factors in Table 2 and
in Harfigan et al. (2). Annual loadings -re deve loped for a year of
Average wetnels-(1967) characterized by 40.6 in of "fal •
in
119
-------
area loads of total N and total P while forestland produces the lowest unit
area loads. Also of note are the higher loadings for wet year conditions.
For example, Table 5 shows significantly higher cropland loadings for wet
year conditions which can be attributed to percentage increases in soil loss
that are much greater than runoff increases, while urban land uses exhibit
increases proportional to runoff increases.
Verification of Test Watershed Loading Factors
A five-step procedure was followed to scale-up from the test watershed
models to the Chesapeake Bay river basin models. ,
First, the river basin models were subjected to an independent hydrology
calibration/verification study which is described elsewhere (6) .
Calibration/verification gages are shown in Figure 9. Hydrologic parameter
sets developed from the test watershed model calibration could not be
applied directly to the river basin models because the subsurface flow
component was often not detectable at the testing sites. Further, the
independent hydrology calibration for the river basin model permits an
accurate simulation of overland flow transport to the stream channel
system. An accurate representation of overland flow transport eliminates
the need for application of a "sediment delivery ratio" to simulated
Table 5
Simulated Annual Surface Washoff of Total N (as N) and Total P (as P)
For Average and Wet Years: Silt Loam Soils Typical of Piedmont Province
TOTAL N LOAD TOTAL P LOAD
LAND USE (Ibs/acre/yr) (Ibs/acre/yr)
AVG. YR. WET YR. AVG. YR. WET YR.
FOREST 0.6 0.8 0.08 0.12
PASTURE 2.6 3.0 0.45 0.52
SINGLE FAMILY RESIDENTIAL
(18% impervious) 6.0 6.8 0.86 0.97
COMMERCIAL (90% impervious) 10.7 12.2 1.29 1.46
LOW TILLAGE CROPLAND 6.3 33.2 0.52 2.74
HIGH TILLAGE CROPLAND 17.9 62.7 5.64 19.82
NOTE: BASED ON RAINFALL RECORDS FOR NORTHERN VIRGINIA GAGES
o AVG. YR. (1967) = 40.6 in.
o WET YR. (1975) = 54.1 in.
120
-------
LOCATION MAP
SUSQUEHANNA
RIVER
LEGEND
SU8-8ASIN
IVER 3ASIN
JAMES
RIVER
CHESAPEAKE
Y
Figure 9. Map of Chesapeake Bay Basin Showing
Sub-Basin/Channel Network for Basin
Model and Calibration/Verification
Gages (Susquehanna River at Cono-
wingo, MD (A), Potomac River near
Washington, DC (B), and James River
at Cartersville, VA (C))
121
-------
sediment and sediment-related loads. Based upon test watershed model
sensitivity studies, the river basin models relied upon a transport
coefficient (KSER) equal to 0.4 which provides a stable representation of
runoff transport of detached pollutants.
Second, an erodibility factor (KRER) was assigned to each sub-basin
based upon average soils characteristics. Since the NFS loading factors for
each land use category are the same in each river basin, the erodibility
factor is one of the most important parameters to represent locational
differences in cropland nonpoint pollution loadings. For example, high
tillage cropland in a sub-basin with highly erodible soils will produce
higher NFS loads of total N and total P than the same land use in a
sub-basin with less erodible soils, even though the sediment potency factors
are the same in both sub-basins.
Third, the NFS loading factors shown in Table 2 were assigned to the
land use categories in each sub-basin.
Fourth, a receiving water model with point source discharge files was
iteratively executed in a quasi-steady state mode for a typical low flow
condition (i.e., 25th percentile flow). Simulated baseflow concentrations
were compared to low flow monitoring data to establish an initial estimate
of baseflow/interflow concentrations.
Fifth, the sub-basin/receiving water models of each river basin were
calibrated for a two-year period (January 1974-December 1975) and verified
for a three-year period (January 1976-December 1978). The models were
iteratively executed for the two-year calibration period with the NFS
loading factors assigned in Step 3 to set instream process parameters and to
derive a final set of baseflow/interflow concentrations for each sub-basin.
The NFS loading factors were not adjusted during the calibration/
verification of the river basin models. Adjustments to instream process
parameters and subsurface flow concentrations were based upon comparisons
between simulated and observed water quality at two different levels: (1)
concentration time series (i.e., typically biweekly observations) for USGS
monitoring stations; and (2) nonpoint pollution loading records at USGS fall
line stations. A USGS fall line monitoring study (9) from January 1979
through April 1981, which focused on wet-weather loadings at the Susquehanna
River, Potomac River, and James River fall lines, produced an acceptable
database to verify the NFS loading factors in the river basin models. Since
sufficient NWS rainfall records were not available for the 1979-1981
monitoring period, regression equations (9) relating observed daily
streamflow and pollutant loads were used to synthesize a daily loading
database for the calibration/verification period. The period 1974-1978 was
selected for model calibration/verification because land use, wastewater
discharges, and flow-duration curves at the three fall line gages were
reasonably similar to conditions during the 1979-1981 monitoring study.
Comparisons of simulated loadings with flow-loading relationships from the
USGS fall line monitoring study were made on a daily, monthly, .and annual
basis. Daily loading comparisons provided a rigorous test of the nonpoint
pollution loading factors derived from the test watershed modeling studies,
while the monthly and annual loading comparisons were used to guide
122
-------
adjustments to receiving water model parameters and baseflow/interflow
concentrations. Following calibration, the receiving water model was
verified by operating it for the period January 1976-December 1978 with
constant instream process parameter sets and baseflow/interflow
concentrations. Since nonpoint pollution loading factors were not adjusted
during either the calibration or verification model runs, the test watershed
model calibration results were actually verified for a 5-yr period.
Comparisons of simulated fall line loads and synthesized loading records
based on the USGS fall line monitoring study are shown in Figures 10 and 11
for the Susquehanna and Potomac rivers, respectively. The good agreement
between simulated and synthesized nutrient loading records at the mouths of
the major river basins indicates that the testing site loading factors
Provide a reasonable representation of loadings from mixed land use river
basins.
Recommendations for Future Test Watershed Studies
Outlined below are recommendations for improved coordination between the
monitoring and modeling efforts to ensure maximum usefulness of the test
watershed database. The recommendations address problems with site
selection and certain elements in the monitoring work program which were
encountered during the modeling study described herein.
Site Selection. One problem which reduced the applications of
monitoring data from certain test watersheds was the selection of mixed land
use catchments with significantly different loading factors. Secondary land
uses which represent a relatively small percentage of the total catchment
area can distort monitoring characterizations of the primary land use if
runoff concentrations for the secondary land use are significantly higher.
While it may not always be possible to identify single land use watersheds
for monitoring studies, mixed land use catchments with a secondary land use
that is characterized by much higher NFS loading factors than the primary
land use should not be designated as test watersheds.
Two of the test watersheds in the Pequea Creek basin were located over
limestone formations that affected the quantity and probably the quality of
monitored baseflow during dry weather and storm periods. Since the purpose
of test watershed monitoring studies is to collect nonpoint pollution
loading data that are representative of larger basins, care should be taken
during site selection to ensure that underlying geology as well as land use
and upper soils characteristics are representative of the river basins in
which the data is to be applied.
Monitoring Work Program. As previously indicated, the test watershed
monitoring studies were designed and initiated in the absence of a specific
watershed modeling work program. The earlier start-up of the monitoring
study was intended to ensure sufficient time for modeling studies of the
monitoring data. However, if the monitoring and modeling work program had
been developed and implemented concurrently, it is likely that: some
different site selection decisions would have been made; collection of the
continuous rainfall and runoff records required for model calibration would
have received a higher priority in order to increase the amount of data
123
-------
SUSQUEHANNA RIVER (27,100 SQ. Ml.): MONTHLY TOTAL P
W00000 1/1/74 -.2/31/78
in
|j 2000000
g I.8O0000
UJ 1000000
«*
2 80O00O
R« -0.9Z
SLOPE • 1.03
800000 1000000 1000000 (000000 2,800000
OBSERVED LOAD UBS)
POTOMAC RIVER (11,560 SQ. M!.)' MONTHLY TOTAL P
I/1/74-12/31/78
100000 400000 600000 MO0OO I00O000 I^OOOOO
OBSERVED LOAD (LB3)
SUSQUEHANNA RIVER (27,100 SQ. Ml.)' MONTHLY TOTAL N
1/1/74 -1Z/3I/78
M.O
30.0
.
5 no
a
o 12.0
8
R*-0.92
SLOPE-0.92
0 «0 00 W.O Z40 30.0 34.0
OBSERVED LOAD (MILLION LBS)
POTOMAC RIVER (11,560 SQ. ML): MONTHLY TOTAL N
I/1/74-12/31/78
0 3000000 1000000 4000000 0000000 0000000
OBSERVED LOAD (LBS)
SUSQUEHANNA RIVER (27,100 SQ. Ml.): DAILY TOTAL P
I/I/74-IZ/3I/7B
TO
2"«o
•«»•
50
90
HJ
OBSERVED LOAD (lOOOO LBS)
Figure 10. Regression of Simulated
and Observed Loadings for
the Susquehanna River at
Conowingo, MD (Jan. 1,
1974-Dec. 31, 1978):
Monthly Total P, Monthly
Total N, and Daily Total P
POTOMAC RIVER (11,560 SQ. Ml.)' DAILY TOTAL P
I/I/74-K/3I/78
280000r
S0000 IOQ000 190000 200000 290000
OBSERVED LOAD (LBS)
Figure 11. Regressions of Simulated
and Observed Loadings for
the Potomac River near
Washington, D.C. (Jan. 1,
1974-Dec. 31, 1978):
Monthly Total P, Monthly
Total N and Daily Total P
-------
available for model calibration; data reduction requirements could have been
reduced considerably; and the results of model calibration studies would be
improved due to the expanded database. After the monitoring studies have
started, periodic interactions between the modeling and monitoring
investigators can facilitate any mid-course corrections necessary to enhance
the applications of the monitoring database. While it is often necessary to
initiate test watershed monitoring studies at the earliest possible date to
ensure the maximum amount of monitoring data and/or a sufficient amount of
time for data analysis, the advantages of better coordination between the
monitoring and modeling efforts from start to finish merits consideration.
The majority of the hydrometeorologic data reduction required for model
calibration was performed by the modeling investigator. The monitoring
investigators were not required to reduce rainfall stripcharts and
reductions of flow stripcharts were generally restricted to the storms which
Produced water quality samples. Consequently, the modeling investigator was
required to perform most of the quality assurance checks on
hydrometeorologic data for the majority of the test watersheds. These
checks included assessments of rainfall-runoff relationships and comparisons
of runoff volumes recorded at the test watersheds in each river basin. Due
to the later start-up of the modeling study and delayed transmittal of
monitoring data to the modeling investigator, initial quality assurance
checks on the hydrometeorologic dataset were not completed until most test
watershed monitoring studies' had ended. As a result, onsite experiments to
resolve hydrometeorologic data problems could not be performed, and
mid-course corrections involving additional instrumentation, further
instrument calibration, or the selection of substitute testing sites could
not be considered. Further, an earlier quality assurance effort focusing on
model calibration needs probably would have flagged the significant^gaps in
the hydrometeorologic records required for model calibration in time to
Produce an expanded database. Therefore, it is recommended that extensive
Duality assurance checks be performed on the hydrometeorologic data very
early in the test watershed monitoring study so that problems and anomalies
can be identified in time for mid-course corrections.
For certain test watersheds, relatively long sampling periods (e.g.,
24-72 hrs) resulted in the inclusion of excessive baseflow volumes in the
flow-composite samples for monitored storm events. As a result, the
separation of baselflow volumes and loadings from the reported storm volumes
«a loadings was very difficult for these test watersheds, and model
calibration studies were significantly complicated. To ensurethe
Development of a reliable nonpoint pollution loading datasetby test
watershed monitoring studies, it is recommended that the sample collection/
retrieval schedule be designed to minimize baseflow contributions during
monitored storms.
c. - •*. ;„« et-uriips in four of the five river basins,
-^nsnrrr .s^'-sr. «»••--- ~s ?< -
arate rainaaae and lowmeer r
monitoring pe?io^.Lurof separate recorders generally resulted in
^synchronize^Rainfall and flow records due to ln^^^'JSStn
c"art speeds. Consequently, one of the more tim?-co"s™^ data reduction
tasks involved scanning the individual stripcharts to match rainfall and
125
-------
flow records for monitored storms. The use of a dual-pen recorder for the
raingage and the flowmeter would not only reduce data reduction requirements
for continuous simulation studies but would also facilitate the quality
assurance checks of hydrometeorlogic data recommended above.
Acknowledgements
The work described herein was funded through a Cooperative Agreement
(No. CR807816-01) with the U.S. Environmental Protection Agency's Chesapeake
Bay Program. James T. Smullen was the EPA Project Officer.
Several NVPDC engineer interns assisted with this project. Susan M.
Lees participated in the hydrologic and nonpoint pollution model calibration
tasks. Mark D. Taylor participated in the hydrologic model calibration
tasks. Cynthia D. Burch coordinated the reduction of the majority of the
hydrometeorologic data and participated in hydrologic model calibration
tasks. Mary Jo Rimkus participated in data reduction and hydrologic model
calibration tasks.
References
1. Southerland, E., et al., "A Modeling Study of Nonpoint Pollution
Loadings and Transport in the Chesapeake Bay Basin," Proceedings of
Thirteenth Annual Pittsburgh Conference on Modeling and Simulation,
School of Engineering, Univ. of Pittsburgh, Pittsburgh, PA, 1982.
2. Hartigan J.P., et al., "Calibration of Urban Nonpoint Pollution Loading
Models," Proceedings of ASCE Hydraulics Division Specialty Conference on
Verification of Mathematical and Physical Models in Hydraulic
Engineering, ASCE, New York, NY, August 1978, pp. 363-372.
3. Donigian, A.S. and Crawford, N.H., "Modeling Nonpoint Pollution from the
Land Surface," EPA-600/3-76—083, U.S. Environmental Protection Agency,
Environmental Research Laboratory, Athens, GA, July 1976.
4. USEPA Chesapeake Bay Program, "Monitoring Studies of Nonpoint Pollution
in Chesapeake Bay Test Watersheds: Final Completion Report," U.S.
Environmental Protection Agency, Annapolis, MD. (In Press).
5. Crawford, N.H and Linsley, R.K., "Digital Simulation in Hydrology:
Stanford Watershed Model IV," Dept. of Civil Engineering Technical
Report 39, Stanford University, Stanford, CA, 1966.
6. Cavacas, A., et al., "Hydrologic Modeling for Studies of Pollutant
Loadings and Transport in Large River Basins," Proceedings of Stormwater
and Water Quality Model Users Group Meeting; March 25-26, 1982, U.S.
Environmental Protection Agency, Environmental Research Laboratory,
Athens, GA, 1982.
7. Donigian, A.S. and Davis, H.H., "User's Manual for Agricultural Runoff
Management (ARM) Model," EPA-600/3-78-080, U.S. Environmental Protection
Agency, Environmental Research Laboratory, Athens, GA, August 1978.
126
-------
8- Hollander, M. and Wolfe, D.A., Nonparametric Statistical Methods. John
Wiley and Sons, New York, NY, 1973.
U.S. Geological Survey, "Water Quality of the Three Major Tributaries to
the Chesapeake Bay, January 1979-April 1981: Estimated Loads and
Examinations of Selected Water Quality Constituents," prepared for USEPA
Chesapeake Bay Program, November 1981.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
A9ency and no official endorsement should be inferred.
127
-------
HYDROMETEOROLOGICAL DATA AQUISITION: INNOVATIVE,
HIGH-RESOLUTION PROGRAMMABLE INSTRUMENTATION FOR
STORMWATER MANAGEMENT
BY
William James, Hector Haro, Mark A. Robinson,
Dale Henry and Reuven Kitai
Civil Engineering Department and
Electrical and Computer Engineering Department,
McMaster University, Hamilton, Ontario, Canada L8S MBS
ABSTRACT
The paper describes difficulties with existing instrumentation, a
new raingauge for sensing rainfall intensity at a high time resolution,
and associated new instrumentation. The data logger incorporates an
audio cassette unit for data retrieval and a single chip microcomputer
programmed to record appropriate time series data in compact form. The
data on the retrieved cassettes is input and stored on removable
hard-disc packs via a PDF 1123 microcomputer through a specially devised
decoder, also incorporating a single chip microcomputer. Special
interactive FORTRAN applications software on the 1123 processes the time
series data files and plots hyetographs on a desk-top plotter. Sample
input/output is provided. The entire system is designed to be
inexpensive, using cheap, mass-produced components. The system is being
extended to monitor water depth (discharges), temperature, conductivity
and pH. The system is designed for use in real-time control.
128
-------
INTRODUCTION
Our hydroraeteorological field program started in the summer of
1980. The data abstracted from these two field years enable calibration
of stormwater management models for the city of Hamilton. These models
are used in turn to determine average daily, monthly and annual amounts
of stormwater runoff and pollutant loadings entering Coote's paradise
and Hamilton Harbour, the receiving waters (Robinson and James, 1982).
The pollutants investigated include:
- Suspended Solids
- BOD 5
- Nitrogen
- Phosphate
- Coliforms
The models are also being used to investigate a wide range of design
alternatives and strategies for minimising pollutant overflows due to
stormwater from the city of Hamilton (or for example Kibler and Aron,
1980).
In the field program, rainfall intensity, stormwater quantity and
quality samples are collected from various field stations located
throughout the Hamilton-Wentworth region. Figure 1 presents an overview
of the data aquisition network, based on conventional instrumentation.
CONVENTIONAL FIELD INSTRUMENTS
For the benefit of readers not familiar with field work, the
purpose here is to review the general difficulties associated with
129
-------
Streamflow
Rainfall
Figure
Hydrome teorological Monitoring Network
for the City of Hamilton
130
-------
conventional instruments, and to elaborate on the new microcomputer-
based instruments.
For a general review of the subject, see Alleg (1977). Readers
familiar with conventional instruments should skip this part of the
Paper.
TIPPING BUCKET RAINGAUGES
Three types of tipping bucket raingauge (TBRG) were used in our
field work. The Atmospheric Environment Service (AES) provides a
standard TBRG which is used throughout Canada. It consists of a 10 inch
(25.4 cm) diameter brass funnel, an AES standard bridge and bucket
assembly and a heavy gauge brass and steel casing. Not all AES TBRGs
are calibrated to 0.2 mm per tip (0.00787 in. per tip). Older Imperial
gauges can be readily converted to metric standards by recalibrating
them using a Calibration Checking Device and a No. 2 nozzle and rainfall
simulator (Environment Canada, 1980).
Two other TBRGs were also used, manufactured in the United States
by the Belfort Instrument Company and Weathertronics Incorporated. Both
of these instruments were similar to the AES raingauge, except the
bridge and bucket assembly incorporates the use of a mercury switch
instead of a magnetic microswitch.
WATER LEVELS
Our pneumatic level sensors used air or nitrogen bubbles flowing at
a constant rate through a tube, to exit at an appropriate location (e.g.
weir). The components of a pneumatic level sensor site are a weir, an
air supply and tube, a bubble gauge, a recorder, a power supply and
131
-------
rating curves. Steven's float recorders, also used in our field work,
are simple and easy to use if properly installed and maintained. The
necessary components of this instrument are a stilling well, a recorder
in a waterproof recorder housing, a weir in a stable channel and a
rating curve. The recorder is powered by a clock which is driven by a
weight. The water level is sensed by a float inside the stilling well.
MAINTENANCE OF CONVENTIONAL INSTRUMENTS
With experience it was found advisable to carry out maintenance
twice a we$k. The following point by point maintenance schedule for the
pneumatic level sensors is presented to illustrate the effort involved:
1. Rewind spring operated clock to ensure accurate paper transport.
2. Check air or nitrogen pressure. If pressure is below 500 psi,
ipstall a new gas cylinder.
3. .Check bubbling rate to ensure that the apparatus is still
operating. One bubble emitted every 2 or 3 seconds is adequate.
^. Check battery to ensure adequate power supply.
5. On a monthly basis clean the upstream side of the weir to ensure
that the water level is unaffected by sedimentation.
Maintenance of chart recorders was as follows:
1. Indicate timing marks to ensure accurate timing.
2. Top off ink well and inspect ink nib.
3. Check chart paper to ensure sufficient quantity is available (new
rolls last two weeks).
y. Check battery to ensure an ample power source for the pen.
However, conventional spring-driven instruments failed to operate
for short periods at random intervals. The most common mechanical
failures experienced in our field work were:
1. Recorder ink running out before the next maintenance check. This
failure was caused by the lack of ink in the ink well, air bubbles
in the ink well (i.e. pen not primed) or pen releasing ink too
rapidly.
2. Recorder ink smearing causing distortions.
3. C^haft paper ripping and/or jamming.
132
-------
4. Electric failures (i.e. 110V and/or battery).
5. Erratic bubbler rate.
6. Bubble tube blocked.
7. Float sticking to side of stilling well.
DATA ABSTRACTION FOR CONVENTIONAL EQUIPMENT
The strip charts from the various rain recorders were removed after
each storm event. Timing marks were noted and timing errors manually
corrected. One minute rainfall intensities were calculated using
tipping capacities of 0.251 mm/tip or 0.200 mm/tip. A typical chart
strip recording and hyetographs is shown in Figure 2. The hyetographs
were plotted on a plotter using the software decribed below. It is
clear from the above descriptions that conventional instruments require
continuous maintenance and manual data massaging.
MICROCOMPUTER-BASED DATA
ACQUISITION SYSTEM
Modern microcomputers, on the other hand, provide an excellent tool
for stormwater management (James and Robinson, 1980).
A microcomputer is a digital integrated circuit (or group of
integrated circuits) containing all the functions required in a digital
processing system. A typical digital microcomputer consists of:
central processor unit (CPU), program memory, data memory and
input/output ports (I/O).
The program memory serves as a place to store instructions, the
coded pieces of information that direct the activities of the CPU. In
data memory the information processed by the CPU is stored. The input
ports are used by the CPU to receive information from an external device
133
-------
(.:
150 ,.
KG ._
130 ,.
120 ,.
110 _
ICO .
90 ._
80 ,,
50 .
40 .
30 .
20
10
0
16
r1!
TOTRL RfllNFRLL
CIRC RRQ
HTETOGRfiPH
1981/ 8/11
0.134E+02 MM
17
18 19
TIME IHDUR51
20
21
22
23
14
15
16
17
18
s
NE
sw
E
W
SE
NW
10
Figure
Strip Charts for the Rustrak and
30 Day Recorders
-------
such as: memory, paper tape reader, floppy disk, etc. at high rates of
speed and in large volumes. A computer also requires one or more output
ports that permit the CPU to communicate the results of its processing
to the outside world. The CPU controls the functions performed by all
the other components. The CPU must be able to fetch instructions from
memory, decode their contents, and execute them and output the results.
It must also be able to reference memory and I/O ports in the execution
of an instruction. Figure 3 presents a simplified overview of the
components in a microcomputer.
A microcomputer can receive digital information from rain gauges or
streamflow sensors and process the information into a useful format,
store this data on cassette tape and transmit the data to other remote
devices such as a central site computer.
A microcomputer-based system is more reliable, efficient and
accurate and also less expensive than a conventional data acquisition
system. Manual data manipulation is not necessary. Processed data is
stored permanently at each transducer on cassette tape recorders, and at
the central site where the processed data is dumped by the central
mini-computer onto magnetic tape. The processed data can be accessed by
an operator at the console for any desired time period. The processing
and transmitting procedures are so quick that real time control is
readily adaptable to this type of data acquisition system. At present
all programming in our existing raingauge microcomputers takes up only
370 milliseconds of the 60 second cycle time.
Mechanical failures are very unlikely. There are no moving parts
such as gears and switches to wear out and replace. Power input to the
system is minimal and a small power pack could supply the unit for a
135
-------
long period. Timing errors are eliminated since the internal timing
mechanism is extremely accurate. Synchronization of the entire data
acquisition network is easily obtained within an estimated acceptable
tolerable error of thirty seconds for the drainage control system.
Electronic devices require little maintenance. Remote devices can
be monitored by the operator at the central receiving site, thereby
reducing the number of field inspection trips. Faulty microcomputers
are easily replaced.
NEW MICROPROCESSOR-BASED RAINGAUGE
Conventional TBRGs and recorders are large, cumbersome and
expensive to purchase and maintain. Recorder parts and data recording
papers are expensive. We have accordingly attempted to produce a lower
cost, higher precision, reliable and automatic system for rainfall data
measurement, logging and presentation.
As shown in Figure 4, our raingauge consists basically of three
major components:
(A) The rain sensor collects precipitation and converts it into water
drops of almost constant size to be counted by the data logger.
(B) The data logger senses the drops and counts them for a programmable
time interval. The logger processes the time series and stores it
on standard audio cassette magnetic tape.
(C) The cassettes are removed, transported to our computational
laboratory, and read and interpreted by a decoder. The decoder
communicates the rainfall time series data to a PDF 1123 computer,
operating in a multi-user environment.
136
-------
r
CPU
FIGURE 3
RAINGAUGE
DATALOGGER
AC
POWER
CASSETTE
PLAYER
HARDCOPY
PLOTTER
FIGURE
137
-------
The microprocessor raingauge stations were not completed until late
1981. Therefore no complete storm event data were collected using these
instruments in 1981. Rigorous testing is now being completed at the
Canada Centre for Inland Waters (CCIW) in Burlington to ensure that the
microprocessor raingauge is functioning as expected. Ten gauges are now
being used in the field (1982) in Hamilton, and other are to be used
elsewhere (Ottawa, North West Territories, Precipitation sampling sites
in Ontario).
The raingauge is contained in two cylinders, numbers (1) and (2) in
Figure 5. Cylinder (1) contains the sensor while cylinder (2) is used
as the base. Both cylinders are easily assembled and the screws (3) are
used to clamp them together.
Inside cylinder (1) there is a plastic funnel (4) whose function
is to collect the rain. The rubber stopper (5) is pierced by a
capillary glass tube (6). The lower part of the funnel (7) supports the
sensor (8) which consists of two electrically conducting points.
The sensor is connected electrically to the connector (10) mounted
inside cylinder (1) and its sensor system may be separated mechanically
as well as electrically from cylinder (2).
At the top of the funnel inside the cylinder, there is a removable
coarse mesh (9) whose function is to trap leaves, etc. Under this mesh
at the point where the funnel shape changes, there is another
dome-shaped fine mesh (11) to prevent dust entering the glass tube.
The plastic surrounding (12) the top of cylinder (1) provides a
good aerodynamic shape, to reduce air turbulence above the gauge.
A clamp (13) secures the cable (14) to cylinder (2). Cable (14)
feeds the sensor to the electronic system.
138
-------
12
FIGURE 5
Mi crocompu ter
FIGURE
139
-------
The electronic system consists of five parts as shown in Figure 6.
The sensor interface senses the drops and converts them into
electrical pulses for the microcomputer. The microcomputer counts these
pulses and stores the count in memory for a time interval which is
predetermined when the microcomputer is programmed. At the end of this
time interval, the microcomputer turns on the cassette recorder through
the cassette interface and records the acquired information together
with the time the data was acquired.
A single chip microcomputer is used for temporary storage as well
as data processing.
The cassette interface controls the recorder power supply as well
as the signals to be recorded. The recorder is a mass-produced product
that has been modified slightly. Frequency shift keying is used because
of the limited frequency response of the recorder.
Power is normally provided by A.C. mains supply. A rechargable
battery backup provides continuity of supply in the event of main supply
failure.
The novel aspects of this device are as follows:
1. The raingauge is constructed from low-cost mass-produced plastic
components that are manufactured for other primary purposes.
2. A microcomputer in conjunction with peripheral electronic devices
is used to detect and count drops, together with precision time
measurement.
3. Digital rainfall rate and time are stored on audio magnetic tapes.
Limitations of the apparatus are as follows:
-------
For very high intensity rainfall, the flow changes from drops to a
continuous jet. The tests show that continuous jet flow commences only
above 25 millimeters of rain per minute, which is a very rare event.
The two meshes have to be wetted before the water can pass through
the funnel. Also, as the diameter of the glass tube is small to form
the drops, some water necessarily collects above the tube before surface
tension is overcome and drops pass the detectors.
The total initial water required to "wet" the funnel, from the dry
state, is of the order of 0.03 mm. of rain, roughly 15* of that required
to tip a standard tipping bucket type rain gauge. This is considered
negligible in most measurements.
Technical details for the raingauge are listed in Table 1 and for
the data logger in Table 2.
TABLE 1: RAINGAUGE DETAILS
2
Collecting surface 8 107 mm.
Drop size 31 mm.
3.81E-3 mm. of rain
Drop size variation (max.) +2.17 mm.
Maximum capacity 18*6 cm. /rain.
137 mra/hr of rain
Water retention from dry state 350 mm.
Size: Height 235 mm.
Width 235 mm.
-------
TABLE 2: DATA LOGGER DETAILS
Microcomputer
Storage media:
Magnetic cassette tape
Modulation
Carriers
Baud rate
Format
Data block
Error relation
Recorder capacity
Operating temperature
Power requirements
Steady state
peak (at last 4 sec. each 20 roins.)
Battery back-up
Intel 87*»8
Audio
Frequency shift keying
1.8 and 2.7 Khz.
300 bps.
50 bytes/block
11 bits/byte
10 bytes of synchronism
20 bytes of data
20 bytes of timing
Less than 1 error in 10
7 days of continuous rain
5 years of no rain
+10°C to -f50°C.
115 V.a.c.
75 ma. g 5 V.d.c.
225 ma. § 5 V.d.c.
17 hrs. 6 C size batteries
(Self Charging)
TABLE 3
system number 10010010050010080010110010160010210010270010310010360010380010420
01046001056001064001090001094001138001175001178001181001183001185001187001188001
18900119100119200119400119500119700119900120200120700121100121300121500121800122
00012270012330012400000990011250011580012340012400000720011050011320011410012400
00072001220001240000052001085001103001114001137001159001164001220001240000036001
STORM EVENT START MARCH,16,1982 12i02
FINISH MARCH,17,1932 10:00
1*42
-------
APPLICATIONS SOFTWARE FOR DATA PRESENTATION
A sample of the data stored on the data logger is presented in
Table 3. In order to present this information in a form readily
comprehended by hydrologists/engineers and also in a format for input to
SWMM, a considerable amount of program development was necessary.
The TRANSLATE and INTERPRETATION procedures developed for the
PDP1123 were written in a transportable subset of FORTRAN. The overall
flowcharts for the programs are presented in Figures 7-9 inclusive.
These programs interpret the information stored on the data logger
and prepare the plotting files for a Houston Instruments Plotter (Gausch
and Lomb, 1981). A typical hyetograph was presented in Figure 3.
FUTURE DEVELOPMENTS
REAL-TIME MICROCOMPUTER-BASED CONTROL SYSTEM
Using our existing calibrated continuous SWMM model of the drainage
network, simpler relations for the prediction of real time stormwater
flow (or water level) have been developed (Robinson and James, 1982).
Statistical analyses of the input and output from our calibrated model,
including multiple regressions, serial cross-correlations and auto-lag
correlations, provide reliable local prediction equations. It is our
intention to use these equations in existing field microcomputers on a
continuous basis.
The entire real time microcomputer control system envisaged would
include the following items (FigureT.
-------
FLOWCHART - GENERAL TRANSLATE PROCEDURE
(START)
Select disk
drive to be
uied
Specify
disk file
to be
translated
JL.
Translate
disk file
(INTERP)
k
Y
i
Generate
y-axie for
plot
(PI.OTY)
I
Generate
x-axis ,data
coordinates
(PLOT!)
1
Specify
port 1
plotter
connected to
X'PlotterX.
Steady /
^X
Pause until
N plotter is
ready
•A
V | M
Plot y-axii
x-axis and
hyetograph
./More >v^
\plot*>^
^***iS^
IN
( STOP J
FIGURE 7
-------
FLOWCHART - PROGRAM TNTEBP
f START j
_L
Initialiie program
variable! and
device numbers
_L
Prompt user for
data type, gage ID
units system, data
•ource, (tart tine
Prompt uier for
gage type (drop
counter or tipping
bucket)
Prompt user (or
inter-event period,
time-itep for
plotting hyetograph
I
1
Set leap
year flag
for «tart
y«jt r
FIGURE 8
145
-------
Create plotter
disk 'file of
intensities
Close disk file at
intensities to be
input to plot
routi ne 3
c
STOP
FIGURE 9
146
-------
DATA ACQUISITION
SYSTEMS
DATA ACQUISITION
SYSTEMS
DATA AQUISmON AND CONTROL
SYSTEMS
STREAM FLOW
CONTROLLED DIVERSION
STRUCTURES
RAINGAUGE
RAINGAUGE
RAINGAUGE
HARD COPY
DISPLAY
FIGURE 1Q ; Proposed Real Time Control System
-------
1. remote monitoring and telemetering stations each with microcomputer
and cassette recorders,
2. communication network (radio or existing leased telephone lines),
3. microcomputer controlled diversion structures (initially only one
is suggested),
t. central minicomputer with display, operator control console and
magnetic tape archive. Our existing PDF 1123 would be a suitable
machine for this.
Following the approach used elsewhere (EPA, 1980; McPherson and
Ammon, 1980), the microcomputer system will be programmed for the
following operational strategies:
Only these structures diverting combined sewage from catchments
lying in the storm track will be activated (Schtifter, 1981).
As much of the first flush as possible will be transported to the
sewage treatment plant (STP).
Discharges from the "dirtiest" catchment areas will have the
highest priority for transport to the STP.
The diversion gates close in a preset order; the least polluted
first, and to the sensitive outfall areas (if any) last.
When in-line storage to the STP becomes available, gates open in
the order of the more polluted runoff first, based on instantaneous
indicators.
Maximum inflow to the sanitary interceptor is increased to a
carefully computed increased proportion of dry weather flow, eg. M
times DWF. Manual override for all structures is provided.
Advantages of the proposed system are briefly listed below:
1. Inadvertent discharges to the natural receiving waters due to
erroneous storm forecasts are reduced.
2. Stormwater is selectively discharged so that the "dirtiest" flow is
diregted to the sewage treatment plant first.
3. Stormwater is anticipated so that maximum storage can be reserved
for highly polluted flows.
4. The central control site has an improved capability to divert some
or all of the Stormwater from the drainage network. At the present
time this is not possible. The central site will know the likely
pollutional loadings from the contributing areas, from the results
of our current modelling studies.
5. The improved system control could reduce the possibility of
basement flooding, through better monitoring and control.
6. Blockages of overflow structures are hard to detect at the present
time. The proposed system is able to detect them almost as soon as
they occur.
7. A good Stormwater data base is archived for future studies.
1148
-------
8. The microcomputer-based control system can be easily expanded. For
example, extra storage facilities can be added to the drainage
network using better estimates of cost-effectiveness from the
simulation programs.
9. Structural changes in the drainage system occur frequently;
re-evaluation of the prediction equations can easily be
incorporated in the microcomputers.
10. The storage available in most existing drainage networks is
minimal. The total amount of available storage in Hamilton,
converted to depth of water over the Central Business District
(CBD) amounts to only 1.2 mm. or about 4 mm. of rain on all
directly connected impervious areas of the CBD. Additional storage
would provide further control but would of course be expensive.
Additional storage can easily be incorporated in the system
programs.
11. The present continuous SWMM model for downtown industrial land uses
in the CBD has not yet been expressly calibrated for either
discrete or continuous storm events. New automatic sampling
equipment will facilitate calibration in the near future. New
calibration parameters can easily be incorporated into the
microprocessors.
ACKNOWLEDGEMENTS
The Computational Hydraulics Group at McMaster University, is a
small organization dedicated to researching and developing innovative
solutions for urban drainage and their environmental problems. The
group comprises nine members: W. James, Director, two full-time
research engineers (Mark Robinson and Dale Henry), one research
assistant (Carol Brown), a full-time text processor (Brenda Bon) and
five full-time graduate students (Hector Haro, Ron Scheckenberger, B.
Shivalingaiah, Alaa ElZawahary and Ali Unal). Two part-time under-
graduate members are Mark Stirrup and Peter Nimrarichter. Four summer
students and a visiting professor from Sweden usually join the group
each year.
-------
The Ministry of the Environment and the Hamilton-Wentworth Regional
Engineering Department jointly provided funds for investigations on the
urban drainage system in the city of Hamilton. Other studies have been
funded by the Natural Science and Engineering Research Council, and
Environment Canada, to develop computer software for stormwater
management, including pollutants washed off street surfaces.
The group possesses a Burroughs B1985 computer, with magnetic tape
and disc drives, line printer and six terminals, made available by
Burroughs Inc. The group also has a PDF 1123 with hard disc drives,
plotter, line printer, word processor and six terminals, obtained
through an NSERC grant.
REFERENCES
Environmental Protection Agency, "Urban .Stormwater Management
Technology: Case Histories", August, 1980, Report No.
EPA-600/8-80-035, 329 pp.
Kibler, D.F. and Aron, G., "Urban Runoff Management Strategies", Journal
of the Technical Councils, ASCE, Vol. 106, No. TC1, August 1980,
pp. 1-12.
Schtifter, Z., "A Kinematic Storm Model for an Urban Drainage Study",
M.Eng. Thesis, McMaster University, Hamilton, 1981, 152 pp.
James, W., and Robinson, M., "Potential Coordinated Multiprocessing
System for Field Data Acquisition and Real Time Control of Urban
Drainage in Hamilton", McMaster University, Hamilton, Ontario,
1980, 20 pp.
McPherson, M.B. and Ammon, D.C., "Integrated Control of Combined Sewer
Regulations Using Weather Radar", Municipal Environmental Research
Laboratory, Office of Research and Development, USA, EPA, R806702,
1980, 87 pp.
Alley, W.H., "Guide for Collection, Analysis and Use of Urban Stormwater
Data", ASCE New York, 1977, 115 pp.
150
-------
Environment Canada, "Field Test of Accuracy for Bridge and Bucket
Assemblies (Tipping Bucket Raingauge)", Atmospheric Environment
Services, Willowdale, Toronto, October, 1980, 10 pp.
Gausch and Lomb, "Houston Instrument, Hi-Plot, Operator's Instructions",
Gausch and Lomb, One Houston Square, Austin, Texas, 1981, 35 pp.
Robinson, M. and James, W., "Continuous SWMM Modelling of Hamilton
Summer Stormwater Including Certain Quality Indicators -
Preliminary Output Time Series Using Discrete-event Calibration for
Non-industrial Areas", published by CHI Publications (about 200
pages), March 1982.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
151
-------
The Separation of Boundary Layer
And Flow Turbulence Of Center-Feed Circular
Sedimentation Basins
by
T1eh-I1n Yin, M. ASCE
Project Manager/Engineer
Maryland-National Capital Park and Planning Commission
Upper Marlboro, Maryland
(I) Introduction
Sedimentation has been one of major treatments for
purification of waste and wastewater for many years. Settling
tanks are efficient In removal of suspended soil Ids at relatively
less cost. It 1s believed that they will continue to be one of the
primary unit operations in wastewater treatment process.
The efficiency of most wastewater treatment processes are
largely a function of hydraulic efficiencies of operations; and the
sedimentation 1s especially of the case. Several researchers have
Investigated the methods of determination of the hydraulic
efficiencies of sedimentation basins.
Basically the hydraulic efficiency of sedimentation basins are
evaluated through the analysis of the types of flow. The types of
flow may be considered plug flow and complete mixing flow, the
former being the ideal settling flow and never existing and the
latter having the assumption that any content, such as suspended
solids or a dye, will Immediately and completely disperse
throughout the volume of the whole tank at the beginning of the
influent. Both are extreme cases and the real basin falls in
between.
By plan view, sedimentation basins are generally either
rectangular or circular. The circular basins are further
classified as center-feed and peripheral-feed depending on the
location of inlet. Generally accepted through previous studies 1s
that the hydraulic performance of rectangular settling tanks are
better than that of circular tanks, and the peripheral-feed
circular settling tanks have better hydraulic efficiency than
center-feed circular settling tanks.
In previous research little emphasis has been focused on the
explanation into insight of the above fundamental phenomenon In
sedimentation basins. In this report It 1s Intended to analyze the
boundary layer flow separation In center-feed circular basins,
which Is believed to be the major reason for the difference of
hydraulic efficiency between rectangular and circular settling
basins.
152
-------
Primarily, this paper presents the analysis In an analytical
approach adopting some classical treatments of the problem, to
prove the occurrence of boundary layer flow separation 1n
center-feed circular settling basins.
(II) Fundamentals of Boundary Layer Flow Separation
There are certain phenomena In fluid flows, which cannot be
dealt solely by the frlctlonless flow theory. One example Is that
a submarine with constant cruising speed under water experiences
drag force due to viscous friction of the water. Another example
1s the experimental fact that the relative velocity between solid
surface and the flowing fluid 1s zero. Actually there Is a thin
layer Immediately adjacent to solid surface called boundary layer
where frlctlonless flow theory ceases to be completely valid.
Boundary Layer Approximations
It Is a basic assumption that In the study of boundary layer,
for flows with large Reynolds numbers the boundary layers are thin
compared with the characteristic d1m1nis1ons of the problem. Based
on this assumption we can have the following simplifying
approximations.
(1) Within the boundary layer the pressure 1s approximately
constant In the direction normal to the surface, and can
be calculated from frlctlonless flow theory.
(2) The flow within the boundary layer is basically parallel
and the shearing stress on a fluid element can be
approximated by TmjH{£!L) , where^
1s dynamic viscosity of the fluid, u 1s the tangential
component of the fluid velocity and y 1s the distance
normal from the body surface.
(3) The thickness of bounday layer is thin compared with the
body and with radius of curvature of the body surface.
The following boundary layer equations are based on these
assumptions.
Boundary Layer Equations
The following two equations are called Prandtl's bounday layer
equations, the derivation of which are primarily due to L. Prandtl.
153
-------
(2)
For the case of steady flow, Prandtl's boundary layer equations are
simplified to:
(3)
(4)
The derivation of Prandtl's boundary layer equations will not be
discussed here. In above equations, v is velocity component along
y; P is pressure; 4? is kinematic viscosity, and.p is fluid
density. '
Solution of Prandtl's Boundary Layer Equations
A number of solutions of boundary layer equations has been
developed in past years. Because of their mathematical
complication some approximate solutions have also been developed, a
famous one of which was attributed to Karman and Pohlhausen.
Basically this technique is used to prove the occurrence of
boundary layer separation in center-feed circular sedimentation
basins, as will be discussed later.
The problem of boundary layer is to find the solutions u (x,y)
and v (x,y) for equations (3), (4) for a given solid body in a flow
with large Reynolds number. The solutions must satisfy the
following conditions:
(1) On the body surface y=0, u=v=0
(2) At the outer edge of boundary layer u=U (x), the velocity
calculated from frictionless flow theory. P (x) and
U (x) are related by Bernoulli's equations P+l/2f> U* =
constant.
(3) At some upstream points in the boundary layer, the
velocity profile must satisfy some initial conditions,
such as X=0, U=o at stagnation point.
Boundary Layer Separation:
y u
u
154
-------
Along the streamwise direction, the pressure can either
increase or decrease depending on physical conditions, which will
make different consequences on the development of bounday layers
If the pressure decreases it is called favorable; if the pressure
increases, 1t will be called unfavorable. As the fluid element
within boundary layer experinces retarding viscous stress, the
favorable pressure gradient tends to accelerate the fluid element
in the boundary layer, thus helping it to overcome the viscousity.
In the case of unfavorable pressure, the increasing pressure will
add some retarding force to the fluid element in the boundary
layer. Near the bottom, the fluid elements have to overcome not
only the increasing pressure but also the wall shear stress which
is extremely large here compared with that in outer edge.
Therefore it may be a point downstream where the flow begins to be
forced in reverse direction near the bottom. This point Is called
separation point.
In mathematical expressions, the above argument can be as
follows: since the velocity profile 1s reversed at bottom part and
at the point of separation where the tangent 1s zero, there must be
a point of inflextion in the profile. In other words, the
separation can only occur when the flow is retarded.
Mathematically,
<&»•& >°
Or>
1s a necessary condition for the separation.
is the definition of and sufficient condition for the boundary
layer separation.
After the point of separation, the flow is reversed in
direction near the bottom and becomes highly turbulent due to
possible vertical eddies, Involving energy dissipation and making
the suspended solids difficult to settle. As to how to quantify
the turbulence and its effect on the sedimenation of suspended
solids, additional study involving experiments Is needed.
(Ill) Boundary Layer Separation and Flow Turbulence
In this section it 1s Intended to show that in center- feed
circular sedimentation tanks bounday layer separation occurs and it
makes the flow highly turbulent in the bottom area of the tank. We
try to obtain the conclusion through analytical approach based on
some mathematical developments about the boundary layer problem.
155
-------
This effort of trying to Investigate Insights of the efficiency of
circular sedimentation tanks has reached an explanation agreeable
with previous experimental results concerning hydraulic or
sedimentation efficiency of different kinds of settling basins.
Based on Van Kaman-Pohlhausen:
It 1s defined that
(8)
(9)
where, Q = memetum thickness of bounday layer; and V" =
kinematic vlsdslty.
At separation the velocity profile shape factor As-/£ or-|0.
A Is defined as
A— .£. Sjs S = boundary layer thickness (10)
-4* aX 9
If A =-J01s chosen, the corresponding value of K 1s -0.1369 as
derived by Hoi stein and Bohlen.
Therefore,
Or,
Let = (U)
(13)
Then equations (11) become \J z=0.l36q&
From mementum equation
=1-523 for A=-/Oalso by Holste1n and Bohlen'
i
Therefore, (5^= V .^2 ^^'^ ** '/ (14)
t> f V*-'X
If o >*\\ no separation (15)
// separation
156
-------
For the center- feed circular settling tank, the following
relationships hold as derived by Ch1u assuming point source of
influent, i i _ m
(16)
.
X2 (17)
where m is a constant.
Substitute equations (16) (17) (18) Into equation (14)
This result leads to conclude that the pressure Increase rate 1n
center-feed circular settling tanks is so large that separation 1s
far from being able to be avoided.
In real circular settling tanks, no point source of influent
exists. X should begin at Xo Instead of zero. Therefore,
(20)
U"(X) = 2 U0 -&3 (21)
Substituting equations (19) (20) (21) Into equation (14) also
gives as explained before.
— 2. •<• ^ //
Based on Curie and Skan's solution of Boundary layer equation,
y2/7p fQCf\i— f,Q4.X/O"g:=: |^ (22)
Where:
* = position of separation ^
CP • CP-PJ/f i f Uj) = I- -rk (23)
ii * I ^11
= max. main stream velocity.
* Function of pressure distribution
For the circular tank, substitute euquatlons (19) (20) (21) Into
equation (22) and get the following:
-2.
(24)
157
-------
Equation (24) becomes:
-f 354.4/5-4 x? =o (25)
As the solution of the equation (25), X has two values. One 1s
X=1.002 Xo and another one is X =4.369 Xo. The first value implies
the physical meaning that the flow separation occurs almost
Immediately after the influent point. The physical meaning of the
second value X=4.369 Xo Is perhaps more significant. It contains
the possibility that the boundary layer would go further turbulent
at around X=4.369 Xo due to continuous pressure increases after the
flow separation occurs at X=1.002 Xo. Alternatively, but
consistently with the foregoing interpretation, the physical
meaning can be conceived as follows: Since the flow faces a
continuous pressure increase along the outwardly radial direction,
the boundary layer is unstable almost at the beginning of the
influent, and the separation occurs at approximately X=4.369 Xo.
After this point, the boundary layer becomes turbulent due to the
flow separation and subsequent reversed flow and vertical eddies.
In other words, 1t can be concluded as follows: The
center-feed circular settling tank has such a large unfavorable
pressure gradient that the laminar boundary layer almost cannot
exist and the bottom flow is turbulent, especially after the flow
separation at X=4.369 Xo, all the way to the point of outlet. This
conclusion is parallel to the previous result that ^ = 2«11,
which Indicates that the pressure increase makes the bottom flow
much more turbulent than mere occurrence of boundary layer
separation. Obviously, experimental observation is needed to
quantify the turbulence under this condition.
Another approach can also lead to the similar result.
Also based on Kaman-Pohlhausen solution of boundary layer
equations, the velocity profile parameter >i — _£5"dU. determines
'*
the shape of velocity profile and occurrence of flow separation.
For the center- feed circular tank the boundary layer thickness at
(26)
'X
Substituting (UO , =-(L- and equation
X
(26) into definition equation of /\ gives
158
-------
/\ — — /£. is the value meaning occurrence of boundary layer
separation. This, also agreeable with the previous result, may be
interpreted as bottom flow being almost turbulent everywhere
in center-feed circular tanks. This can be further shown in a
family of curves derived by Kamen & Pohlhausen, involving M —
The above picture can be technically converted into the
following one with the conversion factor omitted.
Center-fee J
Circular
0 U
This converted y-u figure indicates that the boundary layer of
center-feed circular settling basins, with shape factor =-25, has
the phenomenon of reversed flow. The reversed flow at the bottom
of the basin would cause vertical eddies and, therefore, turbulence
of the flow. This turbulence would make it relatively difficult
for suspended particles to settle.
All the analytical results discussed in this paper support the
argument that center-feed circular settling basins have less
efficiency than rectangular basins, with the occurrence of boundary
layer flow separation and its subsequent reversed flow and eddies
believed as the main cause of the difference in the settling
efficiency.
159
-------
(IV). Conclusion
(1) In center-feed circular settling basins, the boundary
layer flow separation does occur and also the subsequent
reverse flow and vertical eddies at bottom of the settling
basins, due to the contlnous Increase of pressure.
Location of flow separation 1s analytically obtained.
(2) The flow separation, reversed flow and eddies which cause
turbulence of flow at basin bottom are believed to be the
fundamental reason for less settling efficiency of
center-feed settling basins compared with rectangular
ones.
(3) For practical purposes, it needs further study to quantify the
Impact of boundary layer flow separation on the turbulence of
bottom flow and on the settling efficiency.
(4) The reduction of settling efficiency due to boundary layer
flow separation in center-feed circular basins may be
considered not as Important as some other factors in
practical wastewater treatment operations and design.
However, the analysis of the Insight of the phenomenon and
the location of the separation point of bottom flow might
lead to Improvement of design in circular settling basins
or other related hydraulic structures.
160
-------
References:
1. Chlu, Y., "Boundary Layer Separation Concept", ASCE, Journal of
the Environmental Eng. Division, Dec., 1974.
2. SchUctlng, H.f "Boundary Layer Theory", 1968.
3. Kantian, Von; MUHkan, C.B., "On the Theory of Lamlnary
Boundary Layers Involving Separation", 1934, NACA Report.
4. Camp, T.R., "Sedimentation on the Design of Settling Tanks"
Transaction, ASCE Vol. Ill Paper No. 2285, 1946.
5. Goda, T., "A Study on the Mechanism of Transportation of
Suspended Sediment and Its Application to Increasing the
Efficiency of Sedimentation Basin", Kyoto University, Kyoto,
Japan, Vol. 15, No. 4, 1953.
6. Dague, R.R., "Hydraulics of Circular Settling Tanks by Model -
Prototype Comparison "M.S. Thesis, Iowa State University,
1960.
7. Teklppe, R. J. and Cleasby, J. L.t "Model Studies of a
Peripheral-Feed settling Tank", ASCE, Journal of Sanitary Eng.
Division, Feb., 1968.
8. Prandtl, L., "The Essentials of Fluid Dynamics" Hafner
Publishing Co., 1952.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
161
-------
DYNAMIC MODEL ADJUSTMENT
by
Dong Hoang
City of Portland, Oregon
MODEL DEVELOPMENT CONCEPT :
Generally all storm sewer computer raoiels are composed of two parts t the run-
off and the transport.
The first part is designed to computed flows versus time generated by rainfall
over watersheds which connect to the minor sewer system.The flow time curves
are called hydrographs.
The watersheds and the minor sewer system are represented by a series of nume-
rical values which represent the area of the watershed, the coefficients of
the Horton equation for infiltration,the slope, the diameter, the shape, the
length, the flow resistance of the pipe.,.
The hydrographs are then used as input flows to the major sewer system.The
hydr-ographs are input at pointtcalled inlets which are the connections bet-
ween minor and major sewer system,
The difference between the runoff and the transport models (or blocks) is in
the degree of sophistication between the two models.
The transport model uses the dynamic and continuity equations for gradually
varied unsteady flow to represent the behavior of.the flow in time and space.
Besides, with the idea of computing, for each time step, and for each node of
the sewer system, a surface area which is the sum of individual water surface
areas in each half length pipe connecting to the node, the transport model
allows the effect of back water to be considered. With those sophisticationsi
the flows are more reallistically represented in the major sewer system.
Before simulating the flow with the transport model, it is necessary to make
sure that the runoff is properly simulated»This process is called model cali-
bration.The model calibration is a serie of computer simulation runs which
attempt to match the simulated with the measured hydrograph,The discrepencies
betwwen the simulated and the measured hydrographs are believed,for simplicity,
to be influenced only by the percentage of impend.ousness,After a number of
tries, each with a different percentage of imperviousness, the adequacy of the
calibration is judged by visual comparisipri of the simulated and the measured
hydrographs in terms of total volume of runoff and the shape of the hydrograph.
This commonly used procedure of model calibration is empirical and subjective;
and due to the oversimplication of the process, flows distortions are always
apparent,Besides, it dosn't provide any quantifiable measure which allows the
computation of the model accuracy which is an impartial criterion to judge
the reliability of the simulation.
As we all know, the calibration is a much more complexed process, involving a
multitude of factors to be jointly considered, than the commonly used one,
162
-------
In fact the calibration can never be correctly realized if attempts are made
to adjust each influencing variables individually,
Theorically, a mathematical calibration model should take into account all the
influencing factors of all types which can be classified into two categories.
One is of statistical nature, the other regarding quantities which can be ex-
pressed by mathematical formulas*
To build a calibration model, in which all types of variables are involved
regardless of their nature, requires effort and resource which are economically
disproportionate to the field measurement accuracies* Not to mention the dif-
ficulty of handling it.
Logically, the calibration process aims at adjusting the a-pri-ori simulation
in which all the variables of statistical nature have been taking care of by
a careful model set-up which tries to properly and realistically distribute
the water over the subcatchments, the minor and major sewer system. See ap-
pendices.
The calibration model is then reduced to the one of reasonable size with
manageable number of variables to be handled.
MATHEMATICAL MODEL :
In general, no matter how careful the model is set-up, the flows simulated
never match the flows measured at the corresponding monitoring station. The
errors are caused by choosing wrong coefficients of pipeflow-resistance,
wrong coefficients of discharge of the diversion-devices, wrong constants
of Hortonfs equation for infiltration,
If QS is the simulated flows and QM is the measured flows of the corresponding
Pipe , then :
QM - Qg + ( qx + qf1 + qf2 + — + qD1 + qD2 + — + qE ) (2-1)
Where :
q,. : correction due to infiltration,
qf1 : correction due to choosing wrong pipe-flow-resistance
qf2 coefficients of the major sewer system used in the
transport block,
: correction due to choosing
163
-------
wrong discharge coefficients
of the diversion-devices (if any)|
extraneous water which is not formulated in the model,
Using the dynamic and continuity equations for gradually varied unsteady flow,
the rate of change of flows over time " r " can be established in the
following manner :
With reference to Fig-1 and by the energy principle the general dynamic
equation for gradually varied unsteady flow can be expressed in the following
form :
Fig-1
Now let !*y=H jOC=1 J Q=AV. And let use the continuity equation for unsteady
flow in open channels :
=: 0
Then we have :
164
-------
From Manning's Equation :
-f QM
Where
(2-3)
Substitute the value of S- derived from Manning's equation into equation (2-2)
and let^J^L equation (2-2 ) becomes :
Solving for 0. gives :
Q
t+vit
ft
Prom the above equation, the following dynamic equation, in finite form, used
to route the flows through the conduit system of the transport block is derived
. x
*
-* -ft
Where
Q : discharge,
165
-------
V
A
H
Sf
x
t
n
R
velocity,
cross-sectional area of the flow,
hydrolic head,
friction slope,
distance along conduit,
time
Manning's roughness coefficient,
hydrolic radius,
C-,0 : coefficients of entrance and exit lossese in the conduit.
The subscript i,j refer to upstream and downstream nodes respectively ;ra refers
to midpoint of the conduit. The barred symbol for R, V and A represent a weigh-
ted average along the conduit, eg,
The error drdueto errors committed by f and Q is :
elf =.df
-dQ.
If
then
(2'5)
m*m J /^p *
- Jf
(2-6)
(2-6b)
Equation (2-6) becomes t
dr ..
- T
For the time interval ^t, an error in the rate "r" induces an error inAQ»
In other words, during At, an error d(AQ) is committed in the conduit.
166
-------
a Jir At
f-T
+ <3 ) * dQ
Qt
+At
(2-8)
From equation (2-3)
we have
f s n
T -
(2-9)
Substitute the value of df/f defined by (2-9) into equation (2-8) gives :
J^
i-T
(2-10)
dQ of equation (2-10) is composed of
-------
For a point in time t, over a pervious area A, an error q,. is committed :
» AA
« BB
cc
(2-11)
(2-12)
(2-13)
then
The diverted flow over the diversion device Qr, is in error by a quantity qD.
This error is induced by choosing a wrong discharge coefficient C.Using the
weir formula,qD can be evaluated :
m CL
» c
(2-15)
Substitulng various valueB of 1^,1^ from equations (2-11),(2-12),(2-13),(2-15)
respectively In equation (2-10) we have :
n
-f BBdK2
CCdk3
(2-16)
168
-------
The factor (T/1-T) which is a function of the flov resistance coefficient,the
velocity, the hydraulic radius and the time interval is called dynamic coeff-
icient.For each time step, each conduit has different value of dynamic coeff-
icient.Equation (2-16) is the error committed in each conduit during the flow
routing with the transport block.
MODEL ADJUSTMENT BASED ON FLOW :
The basic formula used to form the observation equation is :
Q(measured)
@ time step t
of the last conduit
where the control
station is located
__..
Q( simulated)
@ time step t
of the last conduit
where the control
station is located
1
'
Sum of d(AQ)
@ time step t
of all the conduit from
the most upstream conduit
to the last conduit
connecting to the
control station*
!n. each tributary area, controlled by a control station the following
independent variables are to be determined :
1» d . correction to pipe-flow-resistance of the major sewer system,
2. d. 1 t correction to Morton's equation constant k^,
3» djg . correction to-Horton's equation constant kg,
^* ^3 • correction to Horton's equation constant k^,
5. oL. 9 correction to diversion-device discharge coefficient,
6, EQ t under a rainfall, EQ is a constant flow in cfs. For each conduit
and for each time step t, q£ is determined as follows :
, CM - OS
QM
AIMP : impervious area contributing to an inlet node,
SIAIMP t total impervious area of the tributary area,
QM : now measured at time step t of the last conduit connecting
169
-------
to the control station,
QS : flow simulated at time step t of the last conduit connecting
to the control station.
In total, if x is the number of pipe-flow-resistance coefficients of the
major sewer systiam, and y is the number of diversion-devices-discharge
coefficients, the number of independent variables are :(4+x+y)»
At each time step, the general form of 2Td(^lQ) is :
tn
mi,-
ma..-
(2-17)
Equation (2-17) is called observation equation for time step t,
m : total number of conduits,
ml, m2 : number of conduits having flow-resistances n1 , n2 respectively,
In general, if "v" is the residual flow, the following system of equation
is called system of observation equations ;
(2-18)
170
-------
AA, BB, GC are computed from (2-11), (2-12), (2-13) for each time step
over all the subcatchments contributing to each inlet-node.
MODEL ADJUSTMENT BASED ON DEPTH :
The sum of error in Q committed by different lines at time step t :
cross"sec tional ~ of the contro1
•= t*^t;
(2-19)
Vt is the velocity of the control conduit at time step t and dA is the
error of the cross-sectional area of the control conduit.
1
is also determined by :
Fig-2
(2-20)
dht 9 Wt are respectively the error of the flow depth and the width of the
control conduit at time step t . From equation (2-20) :
(2-21)
-------
Substituing value of dA. given by equation (2-19) into equation (2-21) giving:
dh =
t VtxWt
Divide equation (2-17) by Vt« Wt one obtains :
(2-22)
H
(2-23)
itv%^ivT=T7«LAWP/i QM /j
Equation (2-23) is called observation equation for time step t.
H ; Flow-depth measured @ time step t @ the control conduit,
H : Flow-depth simulated @ time step t @ the control conduit.
As for the case of flow adjustment a ^vstem of observation equations
can be derived
HMettt 8
The rest of the notations are explained in the previous section,
* See page 23 (INHX.FOR Program).
172
-------
THE LFST - SQUARE TECHNIUE :
System (2-18) is now presented in detail (similar presentation for system
2-24) :
1 - QS = a-n^
a6ldG1 4- ... + a^EQ + ^ = ^
(2-25)
- QSN = a1Ndn1 + a^dn,, + ... + a^ + a^dk,, -f a^ +
a6NdC1 + ... + a^EQ + VN = b^
Since the observation are not perfect, we must find the most probable systen
of values which, by the principle of least-squares, makes the sum of the
square of the residuals f 2J minimum
N
\
SSE :
SSE =V (b^ - &1idn1 - a2idn2 " ^i^l " ^4^2 " ^i^ ~ a6id°1 "
Since the variables are independent, the condition that SSE is a minimum is
that the partial differentials of function SSE with respect to each inde-
pendent variable must be zero. Realizing that condition, we generate the
following set of normal equations :
N_ „ JL N JL _N
dn
T~ T ~T 1 1
N
1 1 1
JL NL « N N J[
j ^^^"* ^**M» O ^^^"" I .31 X • ji
1 1 1 1 1
173
-------
N
N
N
N
1
N
dC.
N
1
N
N
N
S
3i
N
N
N : Number of observation equations. For simplicity purpose, we assume that
there are only 2 " dn " & 1 " dC " in the system above. Therefore the normal
equations are composed of 7 unknowns and 7 equations*
In terms of matrix, the normal equations can be written :
A X = G
where :
*i
dk2
EQ
N
N
JL
>•
MMH
1
-------
and
A » a'a
where a1 is the transpose of matrix a regarding the system of observation
equations (2-25).
If matrix A is nonsingular, we can write the solution for the normal
equations as '
X = A"1G
STANDARD ERROR :
Standard error of a single observation is computed by the following :
N-1
Standard error of the mean value can be computed by one of the following :
N : Number of observation equations,
v : Difference between corrected values and measured values,
VANCOUVER DRAINAGE BASIN :
The data blocks of this drainage basin are called DVANR.DAT for run-off
and DVANT.DAT for transport.
The model is simulated with an integration time of 30 seconds•
It is adjusted based on flow monitored by HYDRA STATIQN-18 of February 9,
1979 ( Hydrologic Data Retrieval & Alarm System ) Flows are recorded in
cubic feet per second and inches for every 5 seconds for flows and depth
respectively.
The calibration for this model is based on flow.
175
-------
In the simulation, base-flows are imput into nodes as net constant-flows
( QINST variable of the SWMM model ).
The number of independent variables to be solved for this model is 6 inclu-
ding 2 Manning's n ; 3 Norton's equation coefficientsK ; and One extraneous
water EQ. See Table-1.
VARIABLE
n1
n2
k1
k2
k3
EQ
ADJUSTED VARIABLES
ESTMATED VALUE
O.o15
0.013
0.400
0.800
0.00115
CORRECTION
+0.00033
-0.00131
-0.00595
-0.00360
-0.00005
+5.13746
CORRECTED VALUE
0.01533
0.01169
0.39405
0.79640
0.00110
5.13746
176
-------
TABLE-!
TIME
6h30
7.00
7.30
8.00
8.30
9.00
9.30
10.00
10.30
11.00
11.30
12.00
12.30
13.00
13.30
14.00
14,30
15.00
15.30
16.00
16.30
FLOW
MEASURED
QM
1.30
1.30
2.80
3.80
3.50
3.80
5.50
5.50
5.50
7.90
12.60
11.50
10.50
7.40
5.50
3.80
3.80
7.40
5.30
5.50
6.50
* See pages
** See pages
FLOW
SIMULATED
OS*
1.18
1.50
1.67
1.82
1.97
2.10
2.96
4.33
7.76
8.68
14.34
15.05
10.48
9.84
5.05
3.64
5.70
6.56
5.83
5.59
5.57
92-94
89-91
QM - QS
+0.12
-0.20
+ 1.13
+1.98
+ 1.53
+1.70
+2.54
+1.17
-2.26
-0.78
-1.74
-3.55
+0.02
-2.44
+0.45
+0.16
-1.90
+0.84
-0.33
-0.09
+0.93
FLOW
ADJUSTED
QA**
2.25
1.31
2.37
3.73
3.83
3.86
4.86
5.38
6.90
8.09
13.93
14.18
10.26
9.00
4.96
3.78
4.39
6.39
5.80
5.54
5*95
QM - QJ
-0.95
-0.01
+0.43
+0.07
-0.33
-0.06
+0.64
+0.12
-1.40
-0.19
-1.33
-2.68
+0.24
-1,60
+0.54
+0.02
-0.59
+1.01
-0.30
-0.04
+0.55
177
-------
TIME
17.00
17,30
18.00
18.30
19.00
FLOW
MEASURED
QM
4.20
3.80
2.40
2.40
1.90
6"Ln
FLOW
SIMULATED
OS
5.57
3.58
2.60
2.44
2.41
QM - QS
-1.37
+0.22
-0.20
-0.04
-0.51
53.3558
« *J
FLOW
ADJUSTED
QA
4.85
3.38
2.46
2.31
1,79
./.3251
QM - QJ
-0.65
+0.42
-0.06
+0.09
+0.11
26
*- ,, 17.8709
OArf-i'nc-Hflri =\/ = 0.82906
'Adjusted »l
26
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
178
-------
AN IMPROVED SURCHARGE COMPUTATION
IN EXTRAN
by John A. Aldrich and Larry A. Roesner
Camp Dresser & McKee
Annandale, Virginia
is EXTRAN?
EXTRAN is the Extended Transport Block of the Stormwater Management
Model, it is one of a select few models able to dynamically route gradually
varied flow through urban drainage systems and, in certain cases, natural
streams. This is accomplished by explicitly solving the full Navier-Stokes
equation for a wide variety of hydraulic conditions, including free surface
flow, pressure flow or surcharge, tidal and non-tidal backwater, and flow
reversals. EXTRAN represents a drainage system as a series of links (pipes
and. channels) and nodes (pipe junctions and flow diversion devices). The
"link-node" concept permits the user to model a wide range of system con-
figurations, including parallel pipes, branched and looped systems, and flow
diversion devices (orifices, weirs, and pumps).
EXTRAN was developed as the Transport Block of the San Francisco
Stormwater Model (1,2) by Water Resources Engineers (now part of Camp Dresser
* McKee, Inc). It was first included in EPA's SWMM package in 1974, pro-
viding detailed analysis of complex system configurations and special
hydraulic conditions which existing, simpler models in the SWMM package were
unable to simulate. At this time the model was renamed Extended Transport to
differentiate it from the University of Florida's Transport Block. Since
then, EXTRAN has been expanded and refined to more fully meet the needs of
SWMM users. Version III of SWMM, which has just been released, contains the
latest set of revisions to EXTRAN. The most important of these is a modifi-
cation to the surcharge computation routine.
179
-------
Shortcomings of Previous EXTRAN Surcharge Computations
The original WRE Transport Model treated a system surcharge as If the
excess water rose Into a surge chamber located at the junction node. In many
hydraulic situations, however, this resulted in minor, and 1n a few cases,
major misrepresentations of the system flows. To remedy this, a new approach
was taken 1n more recent versions of the model. This approach 1s based on
fact that the continuity equation for a surcharged node j at time t 1s:
i Q,(t) = 0 (1)
If 9Q/8H- Is computed for each link connected to node j, the continuity
equation can be rewritten as
J
and solved for a head correction,
J
1n the half and full time steps. It has been found, however, that the solu-
tion 1s more stable when only half of the head correction 1s applied 1n the
half time step. Also, only 0.3 and 0.6 of the correction are applied In the
half and full steps, respectively, at upstream terminal nodes to prevent
oscillations In head at these points.
Surcharges computed in the above manner were found to give accurate
results under most surcharge conditions. When several consecutive nodes are
1n surcharge simultaneously, though, predicted heads and flows can be signi-
ficantly underestimated, especially when no lateral Inflows exist.
The University of Ottawa has been studying this problem (3). They
tested a system consisting of a long series of pipes with a restricted out-
fall. Figure 1 shows a comparison of the surcharge heads obtained from
EXTRAN and a similar model, the HVM Dorsch model. The Dorsch model is
180
-------
8
8
8-
8
o
CM
§
?«°
£
^8
III P
Q ^.
8
00
8
*'
OTIO - JUN 13 TRGI
DORSCH-HVM
WRE-TRANSPORT
^^^^^
=^
o :**.• »j
0.00
40.00
120.00 160.00
TIME(min-)
200.00 240.00 280.00
Figure 1 - Comparison of HVM Dorsch Model and
EXTRAN Without Surcharge Iterations
181
-------
designed to handle the same hydraulic conditions as EXTRAN, but by using an
implicit solution technique rather than EXTRAN's explicit solution. The
implicit character of the Dorsch model makes it a reliably accurate predic-
tion tool, therefore its results are considered a good basis of comparison.
This accuracy is gained, however, through a much larger expenditure of com-
puter time than is required by EXTRAN.
The findings of the University of Ottawa indicated that the flexibi-
lity and, consequently, the ultimate usefulness of EXTRAN was limited, even
though only the most severe surcharge conditions cause errors of this magni-
tude. It thus was important to understand where the physical realities of
surcharged flow, represented mathematically by equations 1-3, and the model
differed.
Analysis of The Explicit Surcharge Solution Technique
The most basic difference between a physical situation and a model of
this situation is, of course, the mathematical approximations required to
solve the fundamental flow equations, and it was here that the deficiency of
the surcharge calculation was found. The explicit solution technique used by
EXTRAN is advantageous for computational efficiency because the unknowns in
the present time step are computed solely from previous values in time of
that quantity. Storage requirements are thus kept small and complicated
arithmetic such as matrix inversion is unnecessary. Unfortunately, the
simplification and efficiency acquired from an explicit solution does not
permit any spacial interrlationships, thus the computation of the head
correction, AHj, is independent of other head corrections for adjacent nodes.
Another cause of error in computing surcharge heads rests in the fact
that computed values are approximations. Therefore, the continuity equation
at a surcharged node is approximated as:
I Qj(t) - .j (4)
182
-------
where £j = small nodal flow differential. At each surcharged node, this com-
putational error is an insignificant fraction of the flow passing through the
junction. Thus if only a few isolated nodes surcharge, the net flow pattern
in the system will not be drastically affected and EXTRAN's results will be
reliable.
When several adjacent nodes surcharge, there is net flow within the
entire surcharge area, i.e.
^entire surcharge section ^ ' (5)
Computationally, however, a small flow error exists at each surcharged node.
Since an explicit solution technique is used, no relationship exists between
these nodes and individual nodal errors accumulate, giving:
E ^entire surcharge section * ' " Zej
(6)
This cumulative error can be a large fraction of the flow in the surcharge
section under certain conditions and can significantly alter the net flow
pattern, even though total system continuity is preserved. The problem Is
especially apparent for a long line (many links) where there is little or no
lateral inflow. Therefore a refinement to the explicit solution technique is
needed to compute surcharge heads and flows.
.Semi-implicit Surcharge Iteration Loop
In modifying the surcharge routine, a balance was sought in which the
area in surcharge could be considered as an entity while the explicit
character of the general solution technique would be retained. To do this,
the full-step computations for heads and flows in areas undergoing surcharge
were placed in an Iteration loop. The Iterations continue until the sum of
inflows to and outflows from all surcharged nodes in equation 6 Is suf-
ficiently small, i.e.
183
-------
to 0 (7)
This technique achieves a relationship between the surcharged nodes, giving
the surcharge computation a semi-Implicit character.
To demonstrate the effectiveness of the semi-Implicit Iteration loop,
a closer look at the computation method employed 1n EXTRAN is required.
Figure 2 shows how this technique, the Modified Euler method, is used to com-
pute discharge. Steps 1 and 2 show that the half step flow and, con-
sequently, head are based solely on system properties at time t which are
assumed to be true. Since values at time t cannot be altered, an Iteration
of the half-step computation would be useless. Therefore, the Iteration Is
restricted to the full-step computations, steps 3 and 4.
Once an estimate of the flow is made in the half-step, Q(t + At/2),
the average of the flows in all links connected to surcharged nodes 1s com-
puted. A fraction of this average flow, input by the user as SURTOL, is then
used as a check on the convergence of the Iteration. This test, in essence,
simply checks the validity of equation 7 above. Once an Initial estimate of
full step flows and heads 1s made, the sum of the flow differentials at nodes
under surcharge 1s compared with the test differential computed above. If
the computed flow differential 1s greater than the test value, the full step
computations are repeated for surcharged nodes and their connecting links.
This continues until either the computed flow differentials are sufficiently
small or, to prevent a possible infinite loop, until a user-input maximum
number of Iterations 1s exceeded.
The surcharge Iteration loop was found to dramatically improve the
accuracy of EXTRAN under severe surcharge conditions. Figure 3 demonstrates
this improvement by superimposing the results from the surcharge Iteration
loop version of EXTRAN onto Figure 1. EXTRAN is now found to closely
184
-------
COMPUTED
VALUE
TIME
T) Compute (|§.) from properties of system at time t
=
) Project Q(t+f ) as Q(t+ ) = Q(t) + (
7) a. Compute system properties at
LAt
b. Form
At
from properties of system at time
^ Project Q(t+At) as Q(t+At) = 0(t) + (^) At At
t+
T"
Figure 2 - Modified Euler Solution Method For
Discharge Based on Half-Step,
Full-Step Projection
185
-------
8
00
CM
0.00
OTIO - JUN 13TRGI
o DORSCH-HVM
A WRE-TRANSPORT
0 EXTRAN (Revised)
At = 10 sec
ITMAX = 30
CPU time = 1:15.11
Error Cont. = 1.82%
SURTOL = 5%
4/23/81
40.00
80.00
120.00 160.00
TIME(min.)
200.00
240.00
280.00
Figure 3 - Comparison of HVM Dorsch Model and
EXTRAN With and Without Surcharge
Iterations.
186
-------
approximate the Dorsch results while using less computer time. In addition,
the use of smaller values of SURTOL further increased the accuracy of EXTRAN,
but also increased the computation time required.
.User Control of the Iteration Loop
There is, of course, a certain amount of experience required to effi-
ciently and accurately utilize the iteration loop. The two variables which
control the loop, the fraction SURTOL of the average surcharge flow and
the maximum number of iterations ITMAX, act both individually and jointly.
It is clear that a small SURTOL along with a large ITMAX will yield the
highest degree of accuracy. This is gained, however, at the expense of com-
puter time. The user, therefore, has a tradeoff between accuracy and effi-
ciency.
To assist in the optimization of SURTOL and ITMAX, EXTRAN's inter-
mediate printout shows the sum of the surcharge flow differentials and the
iterations required in each print cycle where iterations occur. It may be
advisible, then, to design a small, short problem similar to a portion of a
system prone to surcharge in order to adjust these two variables. It is also
advisible in any case to initially set SURTOL = 0.05 and ITMAX = 30, values
which have been shown to give good results.
Another troublesome situation which is likely to arise in large
complex systems concerns the case where two or more separate areas of the
system are in surcharge at the same time. In this situation, the net flow in
each surcharge area could easily differ by a significant amount. EXTRAN,
however, computes the cutoff for the surcharge iteration as a fraction of the
average flow through all surcharged sections. This means that equation 7 may
be satisfied in one surcharged section but not in another if the flows are
relatively different in each section.
187
-------
Several ways exist to compensate for this problem. The sum flow dif-
ferential, £ej, may remain relatively large only 1n surcharge areas with flow
substantially less than the gross average surcharge flow. Therefore, errors
In these sections may not Impact the flows 1n the system as a whole simply by
the fact that these flows are small. If, on the other hand, reliable values
of flow are desired in these sections of the system, the value chosen for
SURTOL can be reduced to lower the cutoff value for the Iteration. This
would give more accurate results overall, but at a cost of computer time.
Finally, EXTRAN could be revised to check the conveyance of each surcharge
area Individually. It was determined, though, that the effort Involved and
loss of computational efficiency rendered by these changes would hardly be
warranted by greater automation of the surcharge computation. In addition,
the authors believe that the user should always check the intermediate prin-
tout to determine the accuracy of the surcharge procedure rather than blindly
accept EXTRAN's solution of complex flow patterns.
Interpretation of the intermediate results for partially surcharged
systems is straightforward. As noted above, the actual sum flow differential
and number of Iterations required 1s printed at each print cycle where itera-
tions occur. Also, nodes 1n surcharge are designated with an asterisk. If
nodes in unconnected areas are found In surcharge at the same time, a quick
calculation can determine the sum flow differential In each section in the
same way that EXTRAN computes It for all surcharged nodes. Figure 4
demonstrates how this is done for one large area In surcharge. First, the
links connecting to each surcharged node need to be found 1n the Internal
connectivity summary within the Input data echo. The flow differential at
each node, £j, 1s simply
Cj = I Qj (8)
where Inflows to the node are positive arid outflows are negative. These
flows are shown in the Intermediate printout using the sign convention
established by the user. Once the differentials are found, they can be
summed over each area in surcharge and compared with the average flow 1n this
-------
:reu .,, mr
JUNCTIONS / OECTHS
tin/ 16,3*« I/
«/ lf.r,. n,
'ONOUITS / FIPUS
1"3/ S.S.Z4 If]/
!'.»/ I"*.!!"! 1C1/
1 HP* ' 71.10 KIN FLOU
1*.71« ?/ 17. *5«
!•>.«•.« in/ 11.7*.
ar.Sf rjz/ 02.93
H9.11 I10/ !>?.«•
OIFFFRtNTUL IN Jl
S/ !?.»
-------
section. If the user finds this value to be too large, say greater than 5 or
10 percent of the averge flow, he can reduce SURTOL and re-run EXTRAN. In
this way, an acceptable degree of accuracy can be gained. It is clear,
however, that this determination should be made if possible before a major
run of EXTRAN is attempted to avoid unnecessary computer expense.
Future Research
EXTRAN has been in use for several years and in most cases has been
found to give reliable results. It has also been continually revised, both by
the authors and by the many users of the model. Most verification of the
model's accuracy, however, has been done by comparing the results with simi-
lar models, as was done against the HVM Dorsch model by the University of
Ottawa, or by running simple cases which can be checked with hand calcula-
tions.
Camp Dresser and McKee, Inc. is presently preparing to further verify
EXTRAN using a large data base collected on a major metropolitan sewer
system. The author's believe that this data is diverse enough to fully uti-
lize all features of EXTRAN and demonstrate its accuracy as well as point out
any remaining shortcomings and sources of error. The results of this study,
which should be of interest to all EXTRAN users, will be reported at a future
date.
References
1. Shubinski, R. P., and L. A. Roesner. Linked Process Routing Models,
paper presented at the Symposium on Models for Urban Hydrology,
American Geophysical Union Meeting, Washington, D.C., 1973.
2. Kibler, D. F., J. R. Mouser, and L. A. Roesner. San Francisco
Stormwater Model, User's Manual and Program Documentation, prepared
for the Division of Sanitary Engineering, City and County of San
Francisco, Water Resources Engineers, Walnut Creek, California, 1975.
3. Personal Communication with Atef M. Kassem, Research Associate,
University of Ottawa, Ottawa, Ontario, Canada, November 27, 1980.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
190
-------
PREPARING A DESIGN STORM
By
Stephen A. McKelvie, P. Eng,
Project Manager
Gore & Storrie Limited
Toronto, Ontario.
Introduction
The application of all individual event oriented models analyzing an
urban drainage problem requires that the user establish various
characteristics to accurately describe both the rainfall and the watershed.
The decisions regarding the rainfall used are generally the most important
as this input has the greatest effect on the runoff from the watershed under
consideration. Thus it is important that the user of SWMM or any of the
other hydrologic models fully understand the basis for the rainfall selected
and entered into the model. It is hoped that the comments presented herein
will aid SWMM users in understanding the rainfall input of the watershed
model.
jjse of Design Storms
The concept of a design storm dates back to the development of the
Rational Method. The design storm has maintained its popularity until
relatively recent times when the widespread use of computer models have made
the use of continuous modelling or historical storms a real possibility and
in many cases desirable.
In spite of this, in most practical applications in Ontario design
storms have continued to be used. This is largely due to the encouragement
of provincial and municipal governments who in many cases specify the
rainfall to be used in urban drainage analyses. At the SWMM Users Group
Meeting (1) held in Montreal in 1979 it was concluded by those present that,
while the use of "synthetic" design storms presents many advantages, their
selection is an important part of the modelling process and should be
analyzed on a case-by-case basis.
191
-------
In Ontario the most commonly used synthetic design storms are the
"Chicago" design storms, the SCS Type II distribution and the direct use of
Intensity-Duration-Frequency (I.D.F.) curve. Some authorities specifically
provide the design storms, others suggest various design storms based on the
particular analysis.
In order to understand the "synthetic" design storm, one must
understand how it is generated.
Construction of the Synthetic Design Storm
Construction is an appropriate word to use in conjunction with
synthetic design storms as it is often built by using various blocks of
actual storms. The origins of most synthetic design storms or in the case
of the Rational Method, the design storm is the Intensity-Duration-Frequency
(I.D.F.) curve.
The I.D.F. curves are based on statistical relationships of the
return period of the average intensity of rainfall for a specific duration.
For example, the highest annual rainfalls in a 10-minute period for the
length of available records would be statistically analyzed with the result
being the 1/2 year design average intensity, the 1/5 year design average
intensity and so on. This procedure is repeated for various durations and
the results are plotted in a manner shown on Figure 1, (2). As can be
envisioned from this procedure, one design rainfall event is built of blocks
of rainfall from several actual rainfall events, not from one rainfall.
Thus, the design storm is a storm that never really occurred. This
fact is not always recognized, but is important to understand.
The decision to use a design storm, continuous rainfall or
historical rains will have to be considered carefuly by those involved in
the analysis. A detailed discussion of the merits of the various type of
rainfalls that could be used is beyond the scope of this presentation.
Some of the perceived benefits of design storms are as follows:
192
-------
1248
y-IOOYR STORM (i* ~
117.98
50 YR. STORM (1=7
5O 60 70
TIME (MINUTES)
110 120
FIGURE 1 - RAINFALL INTENSITY-DURATION-FREQUENCY CURVES
193
-------
- a jurisdiction may want to use a design storm to maintain a
consistency of design standards.
- in most cases, use of a design storm yields conservative results
(i.e. high peak discharge rates and volumes) when compared to
actual storms.
- the use of design storms is relatively inexpensive and not very time
consuming.
On the other hand, there are a number of concerns that must be
realized when a modeller and other people involved with stormwater runoff
use design storms.
- a basic misinterpretation is often made that the flows generated
from various design storms of a given frequency have the same
probability of occuring as the design rainfall itself. One must
realize that there are many other factors that influence stormwater
runoff.
- methods to construct a design storm may be misinterpreted
- design storms usually do not cover all the possibilities
(i.e. early peak rainfall, late peak rainfall, snowmelt combined
with rainfall etc.)
- usually design storms do not generate the levels of runoff
(i.e. long duration, relatively high flows) that can cause problems
in detention/retention facilities.
All of these concerns must be realized before one proceeds with the use of a
design storm.
-------
.Preparing Intensity-Duration-Frequency Curves
In Canada, the Atmospheric Environmental Service of the Department
of the Environment operates the weather stations from coast to coast. These
weather stations can provide the necessary information to formulate
Intensity-Duration-Frequency curves. In most urban areas of Canada a copy
of the continuous strip chart of the rain gauge record is available from the
Atmospheric Environment Service. This will provide sufficient detail of
historical rainfall events for further processing. By referring to annual
summaries, one can quickly "zero-in" on the significant rainfall events of
that year and thus, an entire year of continuous strip chart records will
not have to be analyzed.
t
In many urban areas there are several weather stations that may be
used as the data base for the design storms. The three most important
factors in the comparison of several stations are:
(1) length of period of record
(2) distance from the study area
(3) climatic similarity to the study area
The influence of these factors must be carefully considered when
reviewing the records at several stations close to the study area.
For the statistical analysis of the rainfall data up to a return
period of 10 years a partial duration series is usually used. In a partial
duration series the extreme values are analyzed without regard for the year
of occurence. Thus, for example, if one was analyzing the maximum rainfall
during 30 minutes for a period of 12 years, one would select the 12 largest
30 minute rainfall events for analysis rather than the largest 30 minute
rainfall for each of the 12 years.
The partial duration series can be converted to an annual series by
the emperical factors (4) shown in Table 1 as follows;
195
-------
TABLE 1
FACTORS TO CONVERT PARTIAL-DURATION
SERIES TO ANNUAL DURATION SERIES
Return Period Conversion
(years) Factor
2 0.88
5 0.96
10 0.99
, >10 1.00
In deriving intensity-duration-frequency relationships, rainfall
intensity values for each selected duration are considered independently
from other durations. The initial step beinq a separate ranking of rainfall
intensities for each selected duration in descending order of size. A
mathematical fit is made to the array of intensities for each selected
duration using the Gumbel analysis or the Log-Pearson Type III analysis.
The Gumbel Method is generally accepted for extreme value analysis of
rainfall events and is used by the Atmospheric Environment Service in
Canada.
The Gumbel Extreme Value Type I Distribution is as follows (6);
p = 1 - e-e^ - (1)
where p = probability of being equaled or exceeded
e = base of napierian logarithms
y = reduced variate (function of probability)
This distribution may be written as
X =7 + K (S) - (2)
where X = mean value of annual maxima
S = standard deviation of annual maxima
K = frequency factor
X = rainfall value
196
-------
vo
-4
TABLE 2
VALUES OF K FOR EXTREME-VALUE (TYPE-I) DISTRIBUTION
RETURN
PERIOD
YEARS
1.58
2,00
2.33
5 ,
10
20
25
50
75
100
200
400
1000
RECORD LENGTH. YEARS
PROBABILITY
0.63
0.50
0.43
0.20
0.10
0.05
0.04
0.02
0.013
0.01
0.005
0.0025
0.001
15
1.703
2.410
2.632
3.321
3.721
4.005
6.265
20
-0.492
-0.147
0.052
0.919
1.625
2.302
2.517
3.179
3.563
3.636
4.49
5.15
6.006
25
1.575
2.235
2.444
3.088
3.463
3.729
5.847
30
-0.482
-0.152
0.038
0.866
1.541
2.188
2.393
3.026
3.393
3.653
4.28
4.91
5.727
40
-0.476
-0.155
0.031
0.838
1.495
2.196
2.326
2.943
3.301
3.554
4.16
4.78
5.476
50
-0.473
-0.156
-0.026
0.820
1.466
2.086
2.283
2.889
3.241
3.491
4.08
4.56
5.478
60
1.446
2.059
2.253
2.852
3.200
3.446
70
1.430
2.038
2.230
2.824
3.169
3.413
5.359
75
1.423
2.079
2.220
2.812
3.155
3.400
100
-0.464
-0.160
0.016
0.779
1.401
1.998
2.187
2.770
3.109
3.349
5.261
200
-0.459
-0.162
0.010
0.755
1.36
1.94
2.70
3.27
3.83
4.40
CO
-0.450
-0.164
0.001
0.719
1.30
1.87
2.59
3.14
3.68
4.23
-------
The frequency factor, K, varies with the return period and record
length. The values of K, for the extreme value Type I distribution are
given on Table 2, (5, 6).
The foregoing analysis is usually used to determine the extreme
values of rainfall for the rainfall events with a return frequency of 2, 5,
10, 25 and 100 years, although one or two other storms may be required. For
erosion control projects the return period of concern may be the 2 month
rainfall event.
The rainfall durations that are usually computed are 5, 10, 15, 20,
30, 60, 120, 180, 360, 720 and 1440 minutes. This provides sufficient data
to use the Rational Method, the SCS procedures or the many rainfal1-runoff
models in use.
The information resulting from this analysis is commonly displayed
in the form of the Intensity-Duration-Frequency Curves as shown on Figure 1.
The information in this form is suitable for use in the Rational Formula
which is still frequently used in small watersheds. It is important to
realize that the frequency curves link occurrences that are not necessarily
from the same storm. They do not represent a sequence of intensities during
a single storm, but only the average intensity of rainfall expected for the
specified duration.
Due to the effect of the length of rainfall records it is important
that the intensity-duration-frequency data and consequently all other data
derived from it, be reviewed and revised ff necessary, at periodic intervals
of about five years. Of course this depends upon the length of rainfall
records used initially.
In order to use the information provided in the Intensity-Duration-
Frequency curves in computer aided calculations it is usually necessary to
determine the equation of the rainfall curves. The most common forms of the
equations used to describe these curves are as follows:
198
-------
7.0-
6.0-
5,0-
™
3
0
c
c
4.0
^
5>
z
ui
z
^ 3.0
*
z
5
2.0
1.0-
^_T^
•v . • . - .. . --.
I ' i i i i
0 10 20 30 40 50
• -••;:
: ;
1
.
'
;
;
.
:;
TIME (hour)
W— v____
. _ ,:.,.:. ,., ,„;
1 1 I .O
- 152.4
-127.0
—
0
c
E
-101.6 1
t—
«
Z
1—
z
-76.2 J
4
u.
Z
Z
•50.8
-25.4
1 • ' '" ""''' ' • ' — i — ' " ' U
2 3
Figure 2 1 710 YEAR DESIGN STORM - CHICAGO DISTRIBUTION
-------
- (3)
+ c
i =
(t + c)b
- (4)
where i = average intensity
t = duration of rainfall
a, b, c = constants
The suitability of these equations will depend upon the shape of the
Intensity-Duration-Frequency curves.
The constants in these equations are obtained for each return period
by fitting the Gumbel intensity-duration data to the general form of
Equations 3 and 4 and using regression equations of the following forms;
For Equation 3;
Log i = Log a - Log (tb + c) - (5)
For Equation 4;
Log i = Log a - b Log (t + c) .
For most design storms in Ontario, Equation 4 provides the best fit.
One of the most frequently used types of rainfall distribution used
for SWMM analysis in Ontario is the "Chicago type" hyetograph. A typical
hyetograph is shown on Figure 2, (3).
For Intensity-Duration-Frequehcy curves similar to Equation 3 the
following distribution is used, (7);
Before the Peak
i = a
- (7)
1 2
200
-------
~-4 6 •
0 10 20 90 40 90
INTENSITY-DURATION FREQUENCY CURVE
* duration of rainfall.
»c > tint* of concentration
2-YEAR CURVE
DESIGN STORM HYETOGRAPH
10 t TO , 50 t «0
TIME
VOLUME® « VOLUME©
From equation of in ten t it y-
duration curv« of farm
t> • 7
e • o.o»r
From ttatlit(ea) enelytli of
dittribulian of rainfall bcfor*
Odd oftar »»ok itorm imtntilni.
• t>m* mtoturad front lh» peak time
10 the left
« lnnt mtowrtd from tM prok tint
•o tht right.
DERIVATION OF 2 YEAR
DESIGN STORM FROM
INTENSITY-DURATION
FREQUENCY CURVE
FIGURE 3
201
-------
After the Peak
i = a
( 1 -
+ c
- (8)
where tb = time before peak
ta = time after peak
r = advancement of storm pattern
The measurement of the storm advancement, r, is the elapsed time
•from the beginning of the design storm to its peak, divided by the total
duration of the design storm.
For Intensity-Duration-Frequency curves similar to Equation 4 the
following distribution is used, (8);
Before the peak
i = a
tb + c
,7/ _
- (9)
n 1+b
After the Peak
i = a
(1 - b)[_tj
- (10)
ta\ + c
T^FJ
1+b
where tb = time before peak
ta = time after peak
r = advancement of storm pattern
These "Chicago-type" storms attempt to distribute rain such that for
any time interval the average intensity is equal to that of the
intensity-duration-frequency curves. This is shown on Figure 3, (11).
202
-------
The location of the peak rainfall intensity rates, r, for the design
storm is based on observed storm characteristics. Table 3 presents valutas
of r used in various cities in Canada.
TABLE 3
City Country £
Kitchener Canada 0.40
Burlington Canada 0.46
Oakville Canada 1/2-0.345, 1/5-0.366, 1/10-0.488
Richmond Hill Canada 0.35
Winnipeg Canada 0.31
East York Canada 0.35
Nepeari Canada 0.41
Mississauga Canada 0.30
SCS Type II U.S.A. 0.50
It is not possible to draw any significant conclusions from this
comparison other than the typical range of the storm advancement, r, is from
0.30 to 0.50. The storm advancement should be investigated with respect to
local experience.
The selected duration of the design storm should be based on the
size of the watershed and more specifically on the watershed response. The
duration of the design storm should be longer than the time of travel from
headwater to outlet of the largest watershed. To most urban municipalities,
the duration is 3 - 4 hours is sufficient. In the analysis of
pre-developrnent or rural conditions, the duration may have to be
significantly longer.
In many design storms based on the Chicago type design storm the
peak rainfall intensity is significantly greater than in the I.D.F.curve on
which it is based. This may be one of the causes of the reputation that the
Chicago type storms are "peaky". In some cases the volume of rainfall
during the duration of the storm may be different than that obtained by
using the intensity-duration-frequency curve.
203
-------
When the design storm hyetograph is prepared it should be compared
to the I-D-F curve on which it was based. The differencies, if any, should
be identified and understood by all parties. This becomes important when
comparing flows estimated by the direct use of the I.D.F. curves with those
estimated by use of the hyetograph. The most important considerations when
comparing the I.D.F. curves with the hyetograph is the peak rainfall
intensity and the average rainfall intensity and hence the rainfall volume.
As mentioned previously, in Ontario the Intensity-Duration-
Frequency curves are most often of the general form described by Equation 4.
Thus, Equations 9 and 10 are used to describe the hyetograph. However, for
modelling purposes, the hyetograph curve will usually have to be
discretized.
In urban areas hyetographs are usually discretized into time steps
of 5 - 15 minutes. The size of the time step has a significant effect on
the peak rainfall intensity. This will affect the generated peak flows in
highly impervious areas. The selected time step for the discretized
hyetograph should not be less than 5 minutes in most cases as the
intensity-duration-frequency curves which are the basis for the hyetograph
usually do not provide rainfall intensity data for duration less than 5
minutes.
The method recommended by the Ontario Ministry of the Environment to
discretize the hyetograph curves (9) as identified by Equations 9 and 10 is
as follows;
1. Select the time step, At
2. Compute the peak rainfall intensity using the following equation
1P = a - (11)
M (At + cjl>
3. Distribute the time interval selected (At) around the peak as r At
before the peak and (1 - r) At after the peak.
4. Compute additional points before and after the peak by integrating
the design curve and calculating the intensity value by equating the
volumes for each time increment of At.
-------
13-
R-
SI
i
1C
a
oo
ro
O
VI
S £
-H =8
•
n*4 3
B *
B f?
oo S-
i
OO c
3 = 5
CD
g
"O
|h . 93-58 (0'24B*b+9)
886
93-58(0-211*0-r-9)
l-8B6
Ttivit Troii
Stirt ef Star
0.0
5.00
10.00
IS. 00
20.00
25,00
10.00
IS. 00
• 0.00
• s.oo
SO. 00
ss.oo
SO. 00
cs.oo
70.00
75.00
•0.00
IS. 00
90.00
9S.OO
100.00
IDS. 00
110.00
11S.OO
120.00
us. oo
130.00
US. 00
MO. 00
MS. 00
ISO. 00
155.00
1(0.00
its. oo
170.00
17S.OO
110.00
165.00
HO. 00
19S.OO
200.00
205.00
210.00
215.00
220.00
DMcrttlltd
Iflt*Mf I*
_[ln./Hr.)
0.0
0.12
O.U
0.14
0.1S
0.16
0.17
O.U
0.20
0.22
0.2S
0.21
O.JJ
0.11
0.41
B.SI
0.76
l.OS
1.X
S.77
9.03
1.11
2.00
\.tt
0.10
0.8!
0.55
0.41
0.39
0.34
0.30
0.27
O.Z4
0.22
0.20
O.U
0.17
O.U
0.1S
0.14
0.14
O.U
0.12
O.lt
0.0
10 ZO 30 40 50 60 70 BO 90 100
Time in Minutes
10
120
130
140
' a
160
ITO
180
190
-------
The general integral form of the hyetograph curve before the peak is
given by Equation 12, (9).
atb
c +
- (12)
The general integral form of the hyetograph curve after the peak is
given by Equation 13, (9).
at=
c + ta
- (13)
If this procedure is used similarity between the intensity-duration-
frequency curve and the design storm hyetograph is maintained. This is
important as the intensity-duration-frequency curve is based on recorded
rainfall information. An example of this is shown as Figure 4, (10).
Design storm hyetographs can easily be converted into mass rainfall
curves that are required for some hydrologic models such as HYMO.
As mentioned previously the SCS Type II distribution is used
frequently in Ontario. The SCS Type I/I distribution takes the total
rainfall over a given period of time, usually 6, 12 or 24 hours and
distributes the rainfall throughout that period by a mass curve shown on
Figure 5, (12). Used in this manner, the distribution of rainfall intensity
within the duration of the rainstorm is not related to historical data at a
particular location.
The SCS Type II distribution is usually expressed in hourly and
half-hourly intervals and is most applicable in this form to larger rural
206
-------
100
to
o
—I
90
100
PERCENT DURATION
FIGURE 5 - NORMALIZED SCS TYPE H RAINFALL DISTRIBUTION
-------
watersheds. This distribution can be used in pipe system analysis if the
time interval is smaller. In the analysis of urban drainage systems, the
rainfall period is usually not required to be longer than 12 hours. Long
duration storms may be of interest to test the performances of
detention/retention facilities.
For the 1/10 year SCS Type II distribution of the rainfall at the
Toronto Bloor Street .Station the observations shown on Table 4 are
interesting. As can be seen in Table 4 the 12 hour storm duration is the
best fit to the I.D.F. curve with respect to peak and average rainfall
intensities in this case. This may vary for other locations.
TABLE 4
Comparison of I.D.F. Curves/SCS Type II Rainfall Distribution
Toronto Bloor Street Station 1/10 Year Rainfall
Storm
Duration
24 hour
12 hour
6 hour
3 hour
Ave.
Intensity
mm/hr.
3.10
5.68
10.30
17.30
Total
Rainfall
mm
74.4
68.2
61.8
51.9
Peak
SCS
12 min.
mm/hr.
83.7
114.2
119.9
127.9
Peak
I.D.F.
12 min.
rmi/hr.
113.7
113.7
113.7
113.7
SCS
Peak/Ave.
27.0
20.1
11.6
7.4
I.D.F.
Peak/Ave.
36.7
20.0
11.0
6.6
Intensity = 875 nm/hr.
(t + 3)0-75
When using design storms, the modeller and others involved in
stormwater management should fully understand the procedures used to develop
the design storm and the limitations inherent of the design storm concept.
It is hoped that this commentary will aid in this matter.
200
-------
REFERENCES
1. Summary - Seminar on the Design Storm Concept - "Proceedings Stormwater
Management Model (SUMM) Users Group Meeting - May~24 - 25, 1979"^GTST
Environmental Protection Agency - EPA 600/9-79-026.
2. Gore & Storrie Limited - "Storm Water Management for Nepean, Merivale
Area" Research Report No^89, Canada-Ontario Agreement on Great
Lakes Water Quality.
3. Andrew Brodie Associates Inc. "Town of Oakville Storm Drainage
Policies and Criteria" 1979.
4. V.T.Chow "Handbook of Applied Hydrology" McGraw Hill Book Company.
5. Viessman, Knapp, Lewis, & Harbaugh - "Introduction to
Hydrology" Second Edition Harper & Row Publishers Inc.
6. Linsley, Kohler, & Paulhus - "Hydrology for Engineers" Second
Edition, McGraw Hill Inc.
7. C.J. Keifer & H.H. Chu - "Synthetic Storm Pattern for Drainage
Design" Proceedings ASCE, August 1957.
8. M. Bandyopadhyay - "Synthetic Storm Pattern and Run-off for Gauhati.
India" Journal of the Hydraulics Diversion, ASCE HY5, May 1972.
9. Ontario Ministry of the Environment and Municipal Engineer's
Association - "Training Manual - Sewer and Watermain Design
Course" September 1981.
10. M.M. Dillion Limited - "City of Burlington Storm Drainage
Manual" 1977.
11. "Second Canadian Stormwater Management Model Workshop" October 19 -
21, 1976.~~
12. K.R. Cooley - "Erosivity Values for Individual Design Storms"
Journal of the Irrigation and Drainage Division, ASCE, IR2 June, 1980.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
209
-------
A PREDICTIVE MODEL FOR HIGHWAY RUNOFF POLLUTANT CONCENTRATIONS AND LOADINGS
by
Brian W. Mar, Richard R. Horner *• ^
Introduction
Highway runoff is not recognized in most literature as a separate con-
stituent in nonpoint source pollution (Browne and Grizzard, 1979; Browne,
1981). Its threat to water resources has traditionally been aggregated with
general urban runoff and characterized by mass/unit area/unit time loadings,
as well as concentrations. Another approach has been to hypothesize a pollut-
ant deposition model for the periods preceding storms and a washoff model for
contaminant removal in runoff from individual storms (Sartor and Boyd, 1972;
Sylvester and DeWalle, 1972; Meinholz et_ a^., 1978; Kobriger et^ aj_., 1981).
There was reason to believe that these approaches do not adequately represent
the important operative factors in determining the character of highway run-
off for impact assessment in the climatic conditions prevalent in the Pacific
Northwest. Accordingly, the Washington State Department of Transportation
funded a five-year research effort to improve the understanding of these fac-
tors and model highway runoff pollutant loadings for use in impact analysis.
One of the first issues faced in the research program was whether to
base monitoring on discrete samples collected throughout storms or composites
representing entire storms. A system was developed to economically collect
composite samples from a storm, and it was decided to sacrifice the better
characterization of the pollutographs of a relatively few storms through dis-
crete samples for storm composite data for many events. We have used the
system to sample approximately 500 storms at nine locations in Washington
State.
Experimental Design
Figure 1 illustrates the composite sampling system schematically. Its
major elements are a calibrated flow splitter and composite sample collector.
Flow splitters were sized to capture a set proportion of the design storm for
each site, typically about 1-2 percent.
The data collected at each site included continuous, automatic traffic
counts and, for each storm, precipitation and runoff volumes. Samples were
analyzed for total suspended solids (TSS), three metals (lead, zinc, copper),
nutrients (total phosphorus, total Kjeldahl nitrogen, nitrate-plus-nitrite-
nitrogen), and general measures of organic constituents (chemical oxygen
demand, total organic carbon). Measured concentrations and flow volumes were
used to estimate pollutant loadings. Tests demonstrated that comparable
*• •* Environmental Engineering and Science Program; Department of Civil Engi-
neering, University of Washington, Seattle, Washington 98195.
210
-------
Approximate Scale 1" = 2.5'
Hi ghway
Channel to
Curb Flow Splitter
Channel to
Tank
Hi ghway
Shoulder
Terrain
Composite
Sample
Storage
Tank
Figure 1: Layout of the Composite Sampling System on a Curbed Highway
-------
loading estimates resulted from using samples from the composite tank and
composites made from discrete samples collected simultaneously be an automatic
sampler (Clark et^ ad., 1981).
Pollutant Transport Mechanisms
Figure 2 illustrates the pollutant deposition mechanisms considered to
be operating, including contributions from the surroundings, traffic deposi-
tion, pavement wear, maintenance operations, and spills. Data collected sug-
gested that vehicles traveling during storms were a very important source of
pollutants in the extended wet periods of Western Washington. Vehicles appar-
ently pick up and retain contaminants, which are then spray-washed from their
undercarriages while driving on wet roads.
Mechanisms tending to remove pollutants from highways, diagramed in Fig-
ure 3, include hydrologic and vehicular scrubbing, maintenance, and natural
and traffic-generated winds. The eruption of Mt. St. Helens midway in the
project provided an opportunity to directly observe the latter mechanism,
and we consider it to be of major importance in pollutant removal. In the
Pacific Northwest transport of highway pollutants appears to be more a func-
tion of kinetic energy provided by moving vehicles than by the low-intensity
rainfall.
Pollutant Loading Model
As the data base developed, we investigated the associations among pol-
lutant loadings and a number of site and storm characteristics, including
volume, duration, and intensity of precipitation, antecedent dry period,
total traffic, and vehicles traveling during storm periods. The analysis
exhibiting the most consistent pattern for the various sampling sites and con-
taminants monitored was cumulative pollutant mass per unit highway length ver-
sus cumulative vehicles during storms (YDS), pictured in Figure 4 for TSS at
pne station. The relationship assumed a "stair-step" form, the steps being
associated with the occurrences of winter sanding or, on a few occasions, vol-
canic eruptions. The fall and spring periods were characterized by linear
relationships. Viewed for all sites (Figure 5), the slopes of the lines dif-
fered among sites and between winter and other seasons at each site.
Observing these differences, it was natural to hypothesize that site run-
off coefficients (RC) should have a major influence on the cumulative pollut-
ant mass loading entering the runoff. When this variable was introduced, TSS
runoff rates at the various stations grouped as shown in Figure 6. The ele-
vated rates at arid Eastern Washington locations result from deposition of
the loose soils on roadways by relatively high and continuous winds. The
relationships illustrated can be expressed by a model in which TSS loading is
proportional to the product of VDS and runoff coefficient:
TSS Loading = (K)(VDS)(RC)
212
-------
Wind Deposition and Atmospheric Fallout
1*0
Accidents
and
Spills
Maintenance
Sanding
Operations
Traffic
Preceding
and During
the Storm
Pavement
Destruction
Figure 2: Total Suspended Solids and Pollutant Deposition Mechanisms
-------
Meterological Conditions
Wind and Rainfall
Traffic
Generated
Winds
Vehicular Scrubbing
of the Pavement
During and Preceding
the Storm
Maintenance
Sweeping and
Vacuuming
Operations
Rainfall Intensity
and
Solids Transport
Along the Curb
Figure 3: Total Suspended Solids and Pollutant Removal Mechanisms
-------
8000
6000
_a
3
U
GO
TJ
O
GO
•o
O)
-o
°- 4000
en
(0
+•>
2000
winter period
sanding
Sand1ng
sanding
spring period
fall-spring period
Y = 6.6X-28
r = 0.98
200 400 600 800 1000 1200
Cumulative Vehicles During the Storm (1000 vehicles)
Figure 4:
Cumulative Traff1c Volumes During the Stonn
-------
N>
•o
c
-------
CO
Q
O
o
O
0)
I—I
•H
e
£>
u
^
J3
i-H
^—/
4)
O
c
CO
CO
20
15 •
10 •
5 •
A Eastern Washington
O Western Washington
A Pasco
A Spokane
Pullman
O Vancouver
1-5 (w/grit)
Snoqualmie Pass
SR-520
e i-s*
1-5 (w/o grit)
0 -1 -2 .3 .4 .5 .6 .7 .8 .9 1.0
Average Runoff Coefficient
Figure 6: TSS Runoff Rate versus Average Runoff Coefficient for All Sites,
The constant of proportionality (K), which is the TSS runoff rate at a runoff
coefficient of 1, may be established as follows, since TSS loading is directly
proportional to runoff flow rate:
„ K(RC=n)
(RC=1) ~ n
The mean constant (+_ one standard error) for Western Washington locations is
6.4 ^0.8 lb/curb-mi/1000 YDS. It was found by observing runoff following
large, intense storms which thoroughly cleaned highway surfaces that the con-
stant fell to approximately 3 lb/curb-mi/1000 YDS under those conditions,
representing the direct contribution of vehicles alone and excluding import
from adjacent land uses. The reduction in the loading factor is short-lived,
and one or more dry days restores enough solids to return it to approximately
the mean value. K for Eastern Washington locations was estimated on the basis
217
-------
of considerably less data than available for Western Washington to be
26 lb/curb-mi/1000 YDS.
Other pollutants generally observed a relationship to cumulative VDS sim-
ilar to that of TSS. Illustrated as an example (Figure 7) is the chemical
oxygen demand plot for one sampling site. Again, a linear relationship dur-
ing the spring and fall was evident, broken by steps coincident with sanding.
The similarities of form among the plots suggested that loadings of the vari-
ous contaminants could be estimated as proportions of TSS loadings:
Specific Pollutant Loading = (P)(TSS Loading)
The coefficients of proportionality (P) are analogous to potency factors
employed in SWMM, STORM and other models. The coefficients derived from the
data (Table 1) may be taken as constants at any Washington State location for
some pollutants or as linear functions of traffic for other contaminants. We
are continuing to refine this aspect of the model, with a particular interest
in expressing loadings of the soluble, more biologically available forms of
the metals.
The model was developed on the basis of cumulative measures and thus is
applicable to assessing total loadings over a time span encompassing a number
of storms (monthly or annually). We have also adopted an approach to charac-
terizing individual storm loadings, as discussed below.
Table 1: Expressions of Specific Pollutant Ratios Recommended for Use with
Washington State Highway Runoff Model.
Pollutant Expression
R2 Specifications
vss
COD
TOG
Pb
P
P
P
P
VSS
COD = '
TOC = *
OK = 2
2
4
8
x 10"5 + (8.55 x 10 "8)*(ADT)
For
For
For
I .987 For
all
all
all
all
sites
sites
sites
sites
Zn
Cu
TKN
except Pasco
PZn = 4>48 x
'37 x 10~8)*(ADT) -820 For
except Spokane
P = 7.05 x 10~5 + (2.89 x 10"9)*(ADT) .888 For all sites
Cu
p = 2 x
TKN
~3
For Western
Washington sites
P = 5.36 x 10"3 + (3.06 x 10"9)*(ADT) .995 For Eastern
Washington sites
NO3 + N02
TP
N PMn ..„ M = 2 x 10"3
N03 + N02 - N
PTp = 2 x 10"3
218
For all sites
For all sites
-------
£ 4000
-Q
— 3000
TJ
E
QJ
O
C
|> 2000
(U
-------
Analysis of Individual Event Runoff Concentrations and Loadings
To this point this paper has dwelled on pollutant loadings and said noth-
ing about concentrations, which generally are of more direct biological sig-
nificance. Loadings are primarily useful to evaluate the relative stresses
on aquatic biota in two given situations and to assess the accumulation of
contaminants in sinks such as sediments and standing bodies of water of long
residence times. Toxic responses in dynamic systems are more a function of
concentrations, however. The plot in Figure 8, often termed a pollutograph,
illustrates a typical storm runoff concentration pattern with time, with
greatly elevated concentrations in the first fraction of an hour declining
rapidly to fairly stable, low levels. Of course, this pattern is not always
observed; sometimes intense bursts of precipitation midway in a storm result
in concentrations far higher than the "first-flush".
Considering this variability in discrete sample concentrations and the
lack of knowledge to assess the consequences of brief, high level exposures,
we decided to base our analysis on the concentrations in composite samples,
which are the event mean values. This approach admittedly neglects the maxi-
mums; nevertheless, it should represent the approximate conditions affecting
receiving water biota for all but very short periods. We reinforced this pro-
cedure with bioassays enabling direct observation of impacts (although presen-
tation of the results of these experiments is beyond the scope of this paper).
The composite samples from our various monitoring stations exhibited wide
concentration ranges. It appeared that many variables would greatly compli-
cate any deterministic or stochastic attempt to express concentrations in
terms of those variables. Inspired by the technique being applied to the
Nationwide Urban Runoff Program data (U.S. Environmental Protection Agency,
1981), our solution to the problem of expressing individual storm loadings
and concentrations was to analyze cumulative distributions to determine the
probability of exceeding specific values with given storm and site conditions.
After we plotted the individual site distributions, it became clear that
aggregation of the data into Eastern and Western Washington groupings would
again be warranted. As shown in Figure 9 for TSS concentrations, plotting
the aggregated cumulative distributions on logarithmic axes against concentra-
tions (or loadings) produced straight lines useful in plotting probability
distributions. Using TSS concentration in Western Washington highway runoff
as an example (Figure 10), these distributions demonstrate that the data are
essentially log-normal. In addition to showing the probability of surpassing
any given concentration in any storm in untreated runoff, the graphs include
lines representing various levels of pollutant reduction through treatment,
dilution by receiving waters, or a combination of the two. Where available,
criteria were shown to provide a basis for judgment.
220
-------
NJ
700
600-
500-
400-
300
200.
100.
Drain 1
Drain 2
0:00 0:30
1:00 1:30 2:00 2:30 3:00 3:30 4:00
Time Since Beginning of Storm (Hr)
1 1 I
4:30 5:00 5:30
Figure 8: Total Suspended Solids Concentration versus Time, SR-520, 2/25/78.
-------
1000
•100
00
CO
CO
O
oo
u_
o
CD
10
EASTERN WASHINGTON
WESTERN WASHINGTON
0 200 400 600 800 1000
TSS CONCENTRATION (MG/L)
Figure 9: Cumulative Distributions of TSS Concentrations.
Conclusions
Our model thus consists of a component for predicting total pollutant
loadings over an extended time period, plus a series of charts with which
individual event impacts may be assessed probabilistically. These elements
have been assembled in a guidebook for evaluating aquatic impacts due to high-
way operations and maintenance. We believe that the specific research find-
ings and the proposed impact assessment procedures apply throughout the
Pacific Northwest and that the techniques used to monitor storms and analyze
the data are more generally applicable.
222
-------
NJ
NJ
V/J
1000
i 100 -
CO
CO
WATER QUALITY
CRITERION
10
99,99 99 95 70 20 5 1 ,01
PROBABILITY (%) THAT TSS IN ANY STORM > CONCENTRATION
Figure 10: TSS Concentration-Probability Distributions for Western Washington
Storms.
-------
REFERENCES
Browne, F.X. and T.J. Grizzard, "Non-point Sources", J. Water Poll. Control
Fed., 51., No. 6, June 1979, pp. 1428 - 1444.
Browne, F.X., "Non-point Sources", J. Water Poll. Control Fed., 55, No. 6,
June 1981, pp. 901 - 908.
Clark, D.L., R. Aspluiid, J. Ferguson, and B.W. Mar, "Composite Sampling of
Highway Runoff", J. Environ. Eng. Div. ASCE, 107, No. EE5, October 1981,
pp. 1067 - 1081.
Kobriger, N.P., T.L. Meinholz, M.K. Gupta, and R.W. Agnew, "Constituents of
Highway Runoff, Vol. Ill: Predictive Procedure for Determining Pollut-
ant Characteristics in Highway Runoff", FHWA/RD-81/044, Envirex, Inc.,
Milwaukee, WA, 1981.
Meinholz, T.L., N.P. Kobriger, M.K. Gupta, and R.W. Agnew, "Predictive Pro-
cedure for Determining Pollutant Characteristics in Highway Runoff,
Vol. Ill, Final Report", Report to U.S. Department of Transportation,
Federal Highway Administration by Envirex, Inc., Milwaukee, WI, 1978.
Sartor, J.D. and G.B. Boyd, "Water Pollution Aspects of Street Surface Con-
taminants", EPA-R2-72-081. U.S. Environmental Protection Agency, Wash-
ington, D.C., 1972.
Sylvester, R.O. and F.B. DeWalle, "Character and Significance of Highway Run-
off Waters: A Preliminary Appraisal", Report to Washington State High-
way Commission by Department of Civil Engineering, University of Wash-
ington, Seattle, WA, 1972.
U.S. Environmental Protection Agency, "Preliminary Results of the Nationwide
Urban Runoff Program", Vol. I (Draft), USEPA Water Planning Division,
Washington, D.C., 1981.
-------
Highway Runoff Water Quality Reports
Report No. 1. Homer, R.R. and E.B. Welch, "Effects of Velocity and Nutri-
ent Alterations on Stream Primary Producers and Associated Organisms,"
November 1978.
Velocity and nutrient studies at 12 sites in Western Washington
streams indicated that 50 cm/sec is the critical average current
velocity where the productive base of the food web is impacted.
Swiftly flowing streams rich in nutrients should not be slowed to
this value, and slowly flowing streams should not be altered to have
velocities greater than this value.
Report No. 2. Homer, R.R., S.J. Burges, J.F. Ferguson, B.W. Mar, and E.B.
Welch, "Highway Runoff Monitoring: The Initial Year," January 1979.
This report covers the initial 15 months of effort to review the lit-
erature, select a prototype site, compare the performance of several
automatic sampling devices, and install a prototype sampling site on
1-5 north of Seattle.
Report No. 3. Clark, D.L. and B.W. Mar, "Composite Sampling of Highway
Runoff: Year 2," January 1980.
A composite sampling device was developed that can be installed at
less than ten percent of the cost of automatic sampling systems cur-
rently used in Federal highway runoff studies. This device was oper-
ated for one year, along-side an automatic sampler at the 1-5 site, to
demonstrate that the two systems provide statistically identical storm
composites.
Report No. 4. Vause, K.H., J.F. Ferguson, and B.W. Mar, "Water Quality
Impacts Associated with Leachates from Highway Woodwaste Embankments,"
September 1980.
Laboratory and field studies of a woodchip fill on SR-302 demonstrated
that the ultimate amounts of COD, TOC and BOD per ton of woodchips can
be defined and that this material is leached exponentially by water.
After a year the majority of the pollutant has been removed, suggest-
ing that pre-treating of the woodchips prior to use in the fill can
reduce the pollutant release from a fill. Thus, chips should be pro-
tected from rainfall and groundwater intrusion to avoid the release
of leachate. Release of leachate onto tidal lands can cause beach
discoloration, and an underground deep outfall may be required.
Report No. 5. Aye, R.C., "Criteria and Requirements for Statewide Highway
Runoff Monitoring Sites," July 1979.
225
-------
Criteria for selecting statewide monitoring sites for highway runoff
were established to provide representative combinations of climate,
traffic, highway, land use, geographic and topographic characteristics.
Using these criteria, a minimum of six sites were recommended for use
in this research.
Report No. 6. Asplund, R., J.F. Ferguson, and B.W. Mar, "Characterization
of Highway Runoff in Washington State," December 1980.
A total of 241 storm events were sampled at ten sites during the first
full year of statewide monitoring of highway runoff. Analyses of
these data indicate that more than half of the observed solids in this
runoff is traced to sanding operations. The total solids loading at
each site was correlated with traffic during the storm. The ratio of
other pollutants to solids was linear when there was sufficient traf-
fic-generated pollutants to saturate the available solids.
Report No. 7. Mar, B.W., J.F. Ferguson, and E.B. Welch, "Year 3 - Runoff
Water Quality, August 1979 - August 1980," January 1981.
This report summarizes findings detailed in Report Nos. 4 and 6 plus
the work of Zawlocki on trace organics in highway runoff. Several
hundred compounds tentatively identified by GC-MS were grouped into
nine categories, which were not mutually exclusive. Major components
of these categories were petroleum products used by vehicles and incom-
pletely combusted hydrocarbons. The concentrations of these trace
organics groups were low compared to criteria proposed for protection
of aquatic life.
Report No. 8. Eagen, P.O., "Views of Risk and Highway Transportation of
Hazardous Materials - A Case Study in Gasoline," November 1981.
While gasoline represents one-third of all hazardous materials trans-
ported in the country by trucks, the risk associated with gas trans-
portation, as viewed by the private sector, is small. Public percep-
tions of risk are much greater due to lack of knowledge on probabili-
ties and consequences of spills. Methods to improve knowledge avail-
able to the public on gasoline spills and methods to improve estimates
of environmental damages from gasoline spills is presented. General-
ization of methodologies to hazardous materials in general are dis-
cussed.
Report No. 9. Zawlocki, K.R., J.F. Ferguson, and B.W. Mar, "A Survey of
Trace Organics in Highway Runoff in Seattle, Washington," November 1981.
Trace organics were surveyed using gas chromotography coupled to mass
spectrometry for highway runoff samples from two Seattle sites. The
characterization of the organics exhibited concentrations of aliphatic,
aromatic and complex oxygenated compounds. Vehicles, including exhaust
emissions, were concluded to be the source of many of the organics.
226
-------
10 Wang, T.S., D.E. Spyridakis, R.R. Homer, and B.W. Mar,
^position and Control of Heavy Metals in Highway Runoff," Jan-
uary 1982.
Mass balances conducted on soils adjacent to highways indicated low
mobility of metals deposited on well-vegetated surfaces. Grass drain-
age channels were shown to effectively capture and retain metals (e.g.
a 60 m channel removed 80 percent of the original Pb concentration) .
Mud or paved channels, however, demonstrated little or no ability to
remove metals from runoff. Metal release studies suggested that acid
precipitation could release metals bound in the soil, especially where
low buffering capacity exists.
Report No. 11. Portele, G.J., B.W. Mar, R.R. Horner, and E.B Welch
"Effects of Seattle Area Highway Stormwater Runoff On Aquatic Biota," Janu-
ary 1982.
The impacts of stormwater runoff from Washington State freeways on
aquatic ecosystems were investigated through a series of bioassays
utilizing algae, zooplankton and fish. Algae and zooplankton were
adversely affected by the soluble fraction of the runoff, while sus-
pended solids caused high mortalities of rainbow trout fry. In addi-
tion BOD. values similar to those reported in the stormwater litera-
ture 'were measured; however, there were indications that results were
influenced by toxicity to microbial populations.
Report No. 12. Chui, T.W., B.W. Mar, and R.R. Horner, "Highway Runoff in
Washington State: Model Validation and Statistical Analysis," November
1981.
Results of the second year of full-time operation of nine monitoring
sites in the State of Washington produced 260 observations of highway
storm runoff A predictive model was developed based on the data from
two years of observation for total suspended solid loads. A high cor-
relation was demonstrated between total suspended solids and COD, met-
als and nutrients. The major factor controlling pollution loads from
highways in Washington State is the number of vehicles passing during
each storm, not those preceding storms.
Renort No 13 Mar, B.W., J.F. Ferguson, D.E. Spyridakis, E.B. Welch, and
R?R? Sorner, "Year 4 - Runoff Water Quality, August 1980 - August 1981."
This report summarizes findings presented in Report Nos. 10 - 12.
Included are the results of studies aimed at improving and extending
Asplund's solids loading model, increasing data on the ratios of van-
ous pollutants to TSS in the runoff, investigating the fate of heavy
metals in drainage systems, and conducting bioassays on sensitive
organisms exposed to highway runoff,
14 Horner, R.R. and B.W. Mar, "Guide for Water Quality Impact
of Highway Operations and Maintenance." (Draft issued Fall,
1981).
227
-------
Procedures particularly applicable to Washington State have been devel-
oped to assist the highway designer in evaluating and minimizing the
impacts of highway runoff on receiving waters. The guide provides com-
putation procedures to estimate pollutant concentrations and annual
loadings for three levels of analysis which depend on the watershed,
the discharge system and traffic. It further provides means to judge
the potential impacts of the runoff on receiving waters.
Report No. 15. Horner, R.R. and E.B. Welch, "Impacts of Channel Reconstruc-
tion in the Pilchuck River." (To be issued Winter, 1982.)
Report No. 16. Report on dissertation project during Year 5. (To be issued
Summer, 1982.)
Report No. 17. Final report. (To be issued Summer, 1982.)
Please send me copies of Report No. 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12,
13, 14
Name ^
Address
Phone (
Return to: B.W. Mar
Department of Civil Engineering
University of Washington
307 More Hall, FX-10
Seattle, WA 98195
There is a charge of $2.00 per report to cover copying and postage.
e,nclpse a check to the University of Washington.
Please
The, work described in this paper was not flanded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
228
-------
CHIMNEY HILL OFF-SITE DRAINAGE STUDY
Jerome M. Normann, P.E.1 and
Edward R. Estes III, EIT2
March 26, 1982
Introduction
The City of Virginia Beach, Virginia, is one of the
most rapidly developing cities in the United States. It is
located on the Atlantic Coast in the Tidewater area of
Virginia (Figure 1). The topography of the Tidewater area
is very flat and lowlying, with a high groundwater level.
The shoreline areas of Virginia Beach and the
Chesapeake Bay, are already extensively developed, and resi-
dential growth is now occurring in the large interior land
areas of the city. These inland areas are drained by a
system of long drainage canals with large cross sections,
which discharge into tidal estuaries. Systems of feeder
canals, in turn, drain the subdivisions and convey the water
to the main drainage canals.
Many of these feeder canals are very wide, retain water
continuously, and have aesthetic appeal as well as serving a
storm drainage function.
Chimney Hill is one of the city's many subdivisions.
Chimney Hill and the adjacent developments compromise about
800 acres, which are drained by seven feeder canals in
series, with a total length of about 13,000 feet, intercon-
nected by culverts under roadways and draining into Canal
No. 2, one of the city's main drainage canals (Figure 2).
Several routing calculations were performed previously
by hand to determine flow conditions in the feeder canals
during the design storm. However, due to the flat longi-
tudinal slopes of the canals, it was suspected that flow
might move upstream in some canals during the early part of
the runoff event, invalidating the results of the simple
routing calculations which proceeded from the upstream to
the downstream end of the canal system.
In order to evaluate flow conditions in the system of
large feeder canals, it was decided to attempt to apply the
SWMM Version II model to the problem on an experimental
basis. The Extended Transport module of the program was
I/.Director of Water Resources, MMM Design Group, Norfolk,
Virginia
2/.Civil Hydraulic Engineer, MMM Design Group, Norfolk,
Virginia.
229
-------
NORTHAMPTON
COUNTY
POQUOSON
C hesapeake
Bay
Hampton Roadi
eol \Bridae Tunnel
Hampton
Roods
s^ *
PROJECT
CHESAPEAKE
FIGURE I
LOCATION MAP
230
-------
NODE NUMBER
run CULVERTS (EVEN NUMBERS)
QBE CANALS (000 NUMBERS)
0T£: COWOUlTHSja AN OVERFLOW CHAIMCL.
MEPKESCNTIHC OVEMTOWIM OF ROADWtT
MOOC (jj$)lS AT WEIK, OMLY USED WITH Tw. 2*
50' WIDE «IR
FIGURE 2
SUBCATCHMENT AREAS.
CONDUITS, CANlALS AND NODES
CHIMNEY HILL OFF-SITE
DRAINAGE STUDY
-------
selected due to the flat slopes and high tidal tailwater
conditions in the canals.
This paper describes the application of the SWMM model
to the rather complex drainage system, and the results of
that study.
The Watershed
The watershed drained by the system of canals is shown
in Figure 2. The development consists of commercial areas,
multi-family housing, and single-family housing. Since the
purpose of the study was to test the capabilities of the
EXTRAN module of the SWMM model, no attempt was made to
perform a detailed analysis of the watershed runoff.
The 800 acre drainage basin was subdivided into 26
subcatchment areas as shown in Figure 2. Subcatchment areas
and the composition of the development in each subcatchment
area are shown in Table 1.
TABLE 1
Subcatchment Areas and Types of Development
Sub-
catchment
Nos.
1,2
3,4,5,6
7,8,9,10
11,12,13,14
15,16,17,18
19,20,21,22
23,24,25,26
Total
Area,
Acres
149.0
125.1
129.1
67.1
116.9
72.6
133.9
Canal
Acres
13.9
6.3
5.6
4.4
7.4
7.0
9.0
Commercial
Acres
21.0
9.6
30.6
42.7
0.0
15.6
42.2
Multi
Family
Acres
108.6
41.2
0.0
0.0
20.2
0.0
15.7
Single
Family
Acres
5.5
92.9
68.0
20.0
89.3
50.0
67.0
The following parameters were chosen for the subcatch-
ment areas .
Percent Imperviousness: A weighted percentage of
imperviousness was calculated for each subcatchment based on
the percentages of subcatchment area associated with each
type of development. The following values were assumed;
232
-------
based on observed development characteristics in the
subcatchments:
Percent
Type of Development Impervious
Single-Family 50
Multi-Family 70
Commercial 85
Water Surfaces 100
Slope: A generalized slope of 0.003 ft./ft. was used
for all subcatchments. These values could be adjusted based
on more detailed topographic information.
Resistance Factors: A Manning n value of 0.013 was
selected for impervious areas, and a value of 0.100 was
chosen for grassy, pervious areas.
Surface Storage: The SWMM program default values,
0.062 inch of storage for impervious areas and 0.184 inch of
storage for pervious areas, were used.
Infiltration Rates: Based on soil conditions in
Virginia Beach (6) and discussions with the local office of
the Soil Conservation Services (5) the following
infiltration rates were assumed for the Horton Equation used
by the SWMM model:
Maximum infiltration rate : 2.0 in/hr.
Minimum infiltration rate : 0.4 in/hr.
Decay rate (default value): 0.00115 I/sec.
Rainfall
The synthetic rainfall hyetograph was developed based
on the 50 year recurrence interval rainfall values given in
Technical Paper No 40 (1). While more up-to-date rainfall
information is available from the National Weather Service
(2) and state sources, it was decided to use the values from
Technical Paper No. 40 to provide for comparisons with the
previous runoff studies which had also utilized that publi-
cation.
A small computer program entitled "Rain", obtained from
the U S Army Corps of Engineers, was used to derive the
synthetic rainfall hyetograph, which is tabulated in Table 2
and shown in Figure 3. The resultant hyetograph is some-
what conservative, because it is synthesized to contain all
storms shorter than 24 hours in duration with a 50 year
recurrence interval. For example, the peak 30 minute period
233
-------
has a total rainfall of 2.89 inches or an average intensity
of 5.78 inches per hour over the 30 minute period. Given
total rainfall in inches for various durations, the "Rain"
program develops a 24 hour rainfall distribution for the
requested time increment. The peak rainfall intensity is
located at the center of the hyetograph, and decreasing
intensities alternate to either side of the peak.
10
e
8
£7
X
,6
0
Sc
RAINFALL INT
v> w *
1— *•
-
_/
J-J
°0 2
RflIN
CITY
RECU
DURA
INTEF
h
^_
FIGURE 3
FALL HYETOGRflPH FOR
OF VIRGINIA BEACH. VA.
RRENCE INTERVAL = SO YRS
riON = 6 HOURS
WAL = 10 MIN.
-« — t,
3456
TIME, HOURS
-------
Table 2
Rainfall Distribution for Virginia Beach, Virginia
based on Technical paper 40 (1)
Time
Hour Minute
0 10
20
30
40
50
1 00
10
20
30
40
50
2 00
10
20
30
40
50
3 00
10
20
30
40
50
4 00
10
20
30
40
50
5 00
10
20
30
40
50
60
Rainfall
Inches
0.05
.05
.05
.07
.07
0.08
.11
.13
.13
.20
.28
0.74
1.65
.50
.22
.14
.13
.12
.08
.07
.07
.06
.05
0.05
.05
.05
.05
.05
.05
0.05
.03
.03
.03
.03
.03
.03
Inches/Hour
0.30
0.30
0.30
0.42
0.42
0.48
0.66
0.78
0.78
1.20
1.68
4.44
9.90
3.00
1.32
.84
.78
0.72
.48
.42
.42
.36
.30
0.30
.30
.30
.30
.30
.30
0.30
.18
.18
.18
.18
.18
.18
235
-------
A six hour segment was selected from the 24 hour hyeto-
graph for this simulation, with about two hours of antece-
dent precipitation prior to the peak rainfall. A ten minute
time increment was used since the response time for the
watershed was estimated to be much longer than ten minutes;
on the order of one to two hours.
Any other recurrence interval rainfall, such as the 100
year rainfall, could easily be inserted in the model and the
resultant runoff calculated.
Evaporation Rate
Due to the short time duration of 6 hours used in the
program run, losses due to evaporation are negligible in the
analysis. The program default value of 0.1 inch/day was
used.
Tidal Conditions
Canal Number 2, into which the canal system under con-
sideration discharges, has complex tidal characteristics.
The canal connects to two separate tidal estuaries;
Lynnhaven Inlet on the north and West Neck Creek, leading to
Currituck Sound, on the south. The canal has been studied
extensively by the Corps of Engineers for the Canal No. 2
Project, and it would be possible to define a tidal cycle at
the exit of the Chimney Hill Canal system. This may be done
in future studies; however, for this preliminary analysis,
two constant tailwater elevations were chosen. One eleva-
tion, +2.0 ft. MSL, represents the mean spring high tide.
The other elevation, +4.75 ft, MSL, represents a tide about
halfway between mean high tide and the 100 year tide (+7.8
ft. MSL) . (7)
The EXTRAN tidal tailwater elevation option in Version
II of the SWMM model did not operate properly due to a cod-
ing error. It is hoped that this option in Version III will
operate properly, and that this program capability can be
used in future simulations.
Selection of Simulation Model
The SWMM model with the Extended Transport module was
chosen for evaluation because of its capability to analyze
both surcharged flow conditions and tailwater effects. Most
other stormwater routing programs, such as the Corps of
Engineers HEC-1 program, begin at the most upstream channel
or pipe and progress downstream, neglecting back pressure
effects due to tailwater.
236
-------
These programs are inadequate in lowlying tidal areas
where backflow can occur and there are nearly always tail-
water effects present.
While the SWMM model is normally applied to prismatic
drainage conduits rather than to the irregular canals in-
volved in this study, it was decided to attempt to utilize
the model because of its potential benefits in improved
analysis of the complex hydraulics involved.
Application of the SWMM Model
In this rather unusual application of the SWMM Version
II model, several different approaches were formulated and
attempted.
One of the first approaches considered involved the
representation of the system as a series of very large sto-
rage nodes (the canals) linked by conduits (the culverts).
This concept was quickly abandoned because adequate repre-
sentations of the geometry of the canals (storage versus
volume relationships) were not possible using vertical
walled nodes. Problems with system connections were also
encountered in attempting to apply the storage nodes in the
above manner.
A more conventional approach was then attempted, repre-
senting the canal system as a series of links and nodes,
with hypothetical nodes at either end of each canal and
culvert. The volumes of the nodes were negligible when
compared with the large volumes of the canals in this
system.
Both canals and culverts were represented as links in
the computer model. Multiple barrels and non-circular cul-
verts were represented as equivalent rectangular conduits on
the basis of cross sectional areas and roughness, main-
taining the actual heights of the structures due to flow
depth considerations. The dimensions of the culverts are
given in Table 3.
Canals were first represented as trapezoidal channels
with 2 (horizontal) to 1 (vertical) side slopes, based on
the original canal designs. This was later found to be in
error, and the bottom widths and side slopes were adjusted
to match surveyed stage versus volume information as closely
as possible. The canal lengths used were actual centerline
lengths. By dividing the canal volume by the canal length,
cross sectional areas were determined at two water depths of
interest. Then, simultaneous equations were formulated and
solved for the base width and side slopes as follows:
237
-------
A = Bd
Where
and
Zd2
A is
d is
B is
Z is
the cross sectional area of the flow prism
the corresponding depth of flow
the computed bottom width
the cotangent of the side slope.
Table 3
Descriptions of
Culverts
Struc
ture Link
No. No.
No.
Barrels
& Shape
Mat
erial
Size
B D
Length
Elev.
Inlet
Invert
Elev.
Outlet
Invert
Slope
1
2
3 *
4
6
7
214
212
210
208
206
204
202
3-Rect
4-Circ
2-Rect
3-Circ
3-Circ
2-Pipe
Arch
3-Circ
Cone
Cone
Cone
Cone
Cone
Corr
Metal
Cone
9'x4'
6'
9'x6'
6'
6'
10'-1
V *7 • •»_ T *
A / -L
6'
100'
132'
116'
120'
194'
110'
120'
1.30'
-3.50'
-0.92'
-0.79'
-0.79'
-0.79'
-0.79'
1.10'
-3.50'
-1.10'
-0.79'
-0.79'
-0.79'
-0.79'
0.0020
0.0000
0.0016
0.0000
0.0000
0.0000
0.0000
* Structure No. 3 has a 30 ft. wide rectangular weir
upstream of the structure with a crest elevation
of 4.21 feet MSL.. This weir maintains canal levels
upstream of Structure No. 3.
Therefore, the canals are represented in the model as
prismatic, trapezoidal channels which produce the proper
stage versus volume relationships, but which may not physi-
cally resemble the actual canals. Table 4 shows the dimen-
sions of the hypothetical canals used in this model.
Table 4
Descriptions o_f Canals used in Model
Link No.
201
203
205
207
209
211
213
Length/ft.
3130
1460
1400
800
1350
1630
2170
Bottom Width,ft.*
91.0
123.0
110.0
166.0
200.0
202.0
250.0
Side Slopes
5.0 :1
7.2 :1
8.4 :1
9.4 :1
10.5 :1
12.6 :1
3.5 :1
* At a bed elevation of +4.75 ft MSL,( the upper
tailwater elevation)
The subcatchments, shown in Figure 2, were divided so
that for each canal, one half of its contributing drainage
area is directed at each end on the canal. This approxi-
mates the true flow condition which is a combination of
238
-------
of spatially varied flow and point discharges into the
canals. No attempt was made to model the details of the
drainage subsystems (pipes, small channels, gutters, etc.)
in each subcatchment area.
As was expected, no particular problems were
encountered in utilizing the runoff module to simulate the
surface runoff from the drainage subcatchments, although
some fine tuning was performed. Use of the EXTRAN module to
model the flows in the canals and culverts was much more
difficult. To avoid instability in the relatively short
culverts, 10 second time steps were selected for use in the
EXTRAN model.
Since the slopes of the canals and culverts were very
flat in some reaches, small incremental "Z" values were
input; ie., node bottom elevations were staggered slightly
to avoid longitudinal conduit slopes of zero.
One of the first major problems encountered was that
the Version II EXTRAN module had no method of establishing
starting water surface elevations in the canals. Thus,
starting with actual invert elevations (Table 3) adjusted
to avoid zero slopes, most of the runoff was expended in
filling the canals to their proper starting water surface
elevations.
This problem led to several attempts at solutions,
using trial and error procedures.
First, it was attempted to input a high initial
rainfall in the Runoff module to fill the canals and thus
reach a stable starting condition prior to inputting the
design storm. This ploy worked partially, but the
watersheds became so saturated that very low losses occured
when the design hyetograph was finally applied. Thus,
almost all of the rainfall ran off of the subcatchments,
and very high peak runoff rates were observed.
Then, it was attempted to compensate by raising the in-
vert elevations of the conduits (links), while leaving the
nodal inverts at their actual levels. This did not satisfy
the initial storage in the canals, and the rainfall was
still expended in filling the canals.
It was noticed that the program apparently regards the
nodal invert elevations as the starting water surface eleva-
tions.
Therefore, on the next trial, it was decided to raise
the nodal invert elevations to the desired starting water
surface elevations, and to use zero Z values. This seemed to
239
-------
controlled by a weir where two different starting water
surface elevations were neccessary. Note that there is a
weir upstream of node 110 which maintains water surface
elevations at +4.21 upstream, while downstream levels could
be at a lower level, such as + 2.0 ft. MSL. If the upstream
and downstream nodes, with large differences in invert ele-
vations, were connected by a steeply sloping conduit, huge
flow rates through the steep conduits would occur early in
the run, even before actual runoff had started.
However, for the higher starting tailwater elevations,
when the weir at node 110 was submerged, the ploy of raising
nodal invert elevations seemed to work quite well. It was
neccessary to adjust the canal cross sections so as to neg-
lect the portions of the cross sections submerged at the
beginning of the run, but this was of little consequence
since friction losses in the large canal sections were al-
most negligible in either case.
Culvert friction losses were also minor in most cases,
at least at low flows. Thus, raising culvert inverts tended
to increase friction losses slightly, but not to a degree
which was of significance. Also, at the peak of the runoff
event, water levels are such that the culverts are submerged
in either their true physical profile or with their inverts
artificially raised to control starting water surface eleva-
tions .
In these early successful program runs, reverse or
ups.tream flow was indeed observed in some of the canals.
The negative flows appeared to be true flows and not merely
products of instability in the model.
However, at this point in the study, it was discovered
that our initial canal geometry which had assumed 2 H to 1 V
side slopes was incorrect, and the canal geometry was ad-
justed to match the measured stage versus volume character-
istics. This correction in geometry eliminated negative
flows of consequence, somewhat to our disappointment.
The model run with a +4-75 ft. MSL starting water sur-
face was performing very well by this time; canal sections
had been corrected, and stage and flow output seemed cor-
rect. Hand calculations were performed to spot check cul-
vert and canal flow results, which seemed correct. Some
flooding of Riverbend Road at conduit 212 was observed, so
it was attempted to insert wide weirs at nodes 112 and 114
to simulate flooding across Riverbend Road and, just in
case, across Lynnhaven Parkway. It was attempted to simu-
late the roadways using in-line broad crested weirs. This
had been successfully accomplished in an earlier study (3),
but, in that case, the flow was discharged from the system
-------
and the weirs were only used as devices to track the total
overflow. In this study, it was attempted to reinsert the
overflow from node 112 at node 113 downstream, and this did
not work, probably due to incorrect connectivity in the
model. Therefore, a 275 ft. wide, low rectangular conduit
with inverts matching the road elevations was inserted be-
tween nodes 112 and 113 to simulate flow across Riverbend
Road. The broad crested weir was left at node 114, since
the flow leaves the system at that location in any case. As
it turned out, there was no flooding at node 114 for the 50
year design discharge.
For the lower system tailwater (El. +2.0 MSL), high
flows occured between nodes 110 (invert elevations 4.21) and
node 111 (invert elevation 2.0) at the beginning of the run.
In order to preserve the characteristics of culvert 210
under Holland Road, an additional node (No 116) was added
upstream of node 110, and a conduit with the same width as
the weir (30 ft.) was added to simulate the upstream opera-
tion. (A weir at conduit 110 had been tried previously, but
had been unsuccessful. Apparently, in-line weirs are not
acceptable in the EXTRAN module. This information would be
helpful if presented in the User's manual for the program.)
The 30 ft. wide rectangular conduit was set at a rather
flat slope (.003 ft. per ft.) by raising the downstream end
in node 110 in order to prevent high initial flows, How-
ever, the conduit first chosen was only 70 feet long, and
use of this hydraulically short pipe resulted in extreme
flow instability downstream of culvert 208. Therefore, the
30 ft. wide conduit was lengthened to 150 ft. and the slope
was flattened to 0.001 ft. per ft., which eliminated a great
deal of the model instability and provided good estimates of
flow and water surface elevations in the system. A shorter
time increment may be necessary to completely eliminate
instability for low tailwater runs.
Results of the Model Applications.
Some representative calculations have been chosen to
illustrate the results of the model runs. Figure 4 depicts
a typical runoff hydrograph from the subcatchments into
Canal No. 201, the most upstream canal in the system. This
hydrograph was developed by summing the inflow hydrographs
from nodes 101 and 102. The outflow hydrograph through
culvert 202 is also shown in Figure 4, illustrating the
routing effect of the wide, long canals. Note that the peak
flow reduction is due to live storage above elevation +4.75
ft. MSL only.
-------
TYPICflL INFLOH HYDROGRflPH
TO UPPER CflNflL flND
ROUTING EFFECT THROUGH
CULVERT
TfllLWflTER = +4.75 FT. MSL
TOTflL INFLOW
TO CflNflL 201
r—TOTflL OUTFLOW
\ THROUGH CULVERT 202
2 3 4
TIME, HOURS
Table 5 shows the peak inflow rates into each canal,
also derived by adding the inflows at each end of the
canals.
Table 5 Peak Inflows to Canals
Canal
No.
201
203
205
207
209
211
213
Downstream
Structure
202
204
206
208
210
212
214
Upstream
Node
~T0T
103
105
107
109
111
113
Downstream
Node
102
104
106
108
110
112
114
Peak
Inflowycfs
897.
624.
600.
278.
578.
398.
738.
-------
These peak inflow rates were similar to, but higher
than the inflow rates derived by using the methods of SCS
TR-55 (4) which were utilized in the previous hand calcula-
tions .
Figure 5 is presented to illustrate the negative flows
obtained in early runs of the model prior to correction of
the canal geometry. The inflow hydrograph and location are
the same as in Figure 4; however, note that at about hour
2.4, a negative flow of about -60 cubic feet per second
occurred, even though the inflow hydrograph was rising ra-
pidly. This was a function of canal storage in the up-
stream and downstream canals, resulting in differential
water surface elevations producing upstream flow.
EXAMPLE OF REVERSE FLOW
TAILWATER = +4.75 FT. MSL
PRIOR TO ADJUSTMENT OF
CflNflL GEOMETRY
TOTAL INFLOH
TO CANAL 201
TOTAL OUTFLOW
THROUGH CANAL 202
TIME, HOURS
-------
Use of the wide retangular conduit between nodes 112
and 113 to simulate flow across Riverbend Road is illus-
trated in Figure 6. Flow under the road is depicted by flow
in conduit 212, flow over the road is carried by conduit 215
which is the rectangular conduit 275 feet wide and 1 foot
high, and the total flow is represented by the solid curve.
The water surface stage upstream of the road is also shown
at the top of the figure.
8
7
§6
0
u
LU
CO
LU
O.
B
"-
, .
CD
<-> 3
Q
a
X
0
u! i
UPSTRE
flM STflG
(NODE 112)
FT . ,
i ' ^' — •
nsL.
J
^*
FLOH UNDER
ROflD, CFS
CONDUIT 212 -i
A
l\
\
\ '
^
\ /
*f
-/ f
V
; TC
"* — — -X*
Nj
\
TflL FLOI
\
\
\
i flT \
I ROflD, CFS ^\
1
1
V •
1 >
i
i
i
•J '
i :
N
^ \*
\
\
k -
\
\
V
\
/ ' FLOH flCROSS N
/ ' RQflD. CFS ^
i CONDUIT 215— -^ *
t
i
i
i
FIGURI
E 6
\
^k
** 0<
^
\
X
FLOODING flT RIVERBEND ROflD
TfllLWflTER EL. = +4.75 FT. MSL
CURB EL. =.+6.75 FT. MSL
a
6
CO
t-
4m
UJ
LU
,
U
K
1
0123456
TIME. HOURS
Results of the study are summarized in Figure 1, which
shows profiles of peak water surface elevations in all of
the canals, as well as peak flow rates, for the two tail-
water elevations of interest. Note the friction losses in
the culverts, and the fact that water surface profiles in
the canals are nearly horizontal, indicating very low fric-
tion losses.
-------
Jr-
un
J»
J
a-
c
I*
I
\
1*
1*1
i»
•*•
/tt M.OO
v®
PEAK MA
TAILWAT
PEAK FL
TAIL*
VMM
ER SURf
R • 2 !
WS THRL
ATER •
HTR .
,(
ACE EL6
0
STRUCT
* 7S
-too —
IOUMD SW
VATIONS
IRES
\ _o-A
&
~* o noo 2000 ^
LE6END
WATER ELEV. BEFORE STORy (4.73)
PEAK WATER ELEV. DURING STORM
WATER ELEV. BEFORE STORy (ZOO)
PEAK WATER ELEV. DURING STORM
-
•
£
' '
*•
*
i
1
;
:-
-^
i
no
.44
i IT.W • 4.75
EL
•-
CLM;
».44i — '.EL
-^-r
:
•
^
:
'
'
4000
• 49
to
r
—
E
EL
^X
•-
8?
- .
-
k
»
tL •« C
CL.TTfl E
BftfA
^ '
/EL. tl.77
A,'
\
*r
--
i;
'
«
c
LI 30
L.T.OB
went CKSI
a*.
D
—
EL
£LT
°""**'\
@
-
-
!
^
i
!
' |
:
s
'-
-
. ;
9000 «000 7000 •000
DISTANCE IN FEET FROM UPSTREAM END
n.oa
^
tso
• so
1*
-\
\
^^
§
^ —
•^®
^^
^^
EL.-
71
J7
^
I
"v.
i
!e*t
ISU:
2
=
a
s
:
L4Z
ft
U479
D
c-
•000 KXOOO EI.OOO IZ.OOO 14UUU
FISURE T
PRORLE OF DRAINAGE SYSTEM
WITH PEAK WATER SURFACE
ELEVATIONS AND FLOW RATES
CHIMNEY HILL OFF-SITE
DRAINAGE STUDY
-------
For the lower tailwater elevations (-1-2.0 ft. MSL) ,
supercritical flow exists in conduit 214, indicating that
the culvert is in so-called "inlet control." This accounts
for the steep water surface profile at that location.
Conclusions
The SWMM Version II model with EXTRAN option has been
successfully applied to the simulation of a series of large
storage canals, linked by culverts, and discharging into a
tidal estuary.
The results from the SWMM Version II model seem reason-
able; however, due to shortcomings in the model, it would be
beneficial to apply the Version III model to the data. For
example, the tidal tailwater option in Version II EXTRAN did
not operate properly, and it was not possible to input
starting water surface elevations.
A capability which would be very useful in the EXTRAN
model would be the ability to simulate in-line weirs, as
well as diversion weirs. In this study, weirs were
simulated using short conduits, but instability in the model
was a problem.
Overall, SWMM Version II worked quite well in simulat-
ing the canal system, and the results confirmed the results
of the hand calculations performed earlier.
Acknowledgements
The authors wish to express their thanks to the City of
Virginia Beach Department of Public Works especially to Mr.
Donald R. Trueblood,P.E., City Engineer, and Mr. Keith
Slicer,P.E., for their assistance and support during this
study.
Appendix References
1."Rainfall Frequency Atlas of the United States," Technical
Paper No. 40, U. S. Department of Commerce, Weather Bureau,
Washington D.C., May 1961.
2."Five-to 60-Minute Precipitation Frequency for the Eastern
and Central United States," NOAA Technical Memorandum NWS
HYDRO-35, National Oceanic and Atmospheric Administration,
Silver Spring, Maryland, June 1977.
3."Modeling of Storm Drainage in a Coastal City," Normann,
Jerome M., Proceeding of the National Symposium on Urban
Stormwater Management in Coastal Areas, Hydraulics Division,
ASCE, Blacksburg, Virginia, June 20, 1980.
-------
4."Urban Hydrology for Small Watersheds," Technical Release
No. 55, Engineering division, Soil Conservation- Service,
U.S. Dept of Agriculture, Washington, D. C., January 1975.
5.Personal communications with Louis Cullipher, District
Conservationist, U. S. Department of Agriulture, Soil
Conservation Service, Virginia Beach, Virginia, May 1980.
6."Soil Survey, Norfolk County Virginia," Series 1953, No.
5, U. S. Department of Agriculture, Soil Conservation
Service, May 1959.
7.Tidal studies conducted by Norfolk District, Corps of
Engineers, in June 1969, file H-31-10-47, and May 1974,
Files H-31-10-80(8), J-31-10-80(9), Referenced in Storm
Drainage Seminar presented at Old Dominion University,
Norfolk, Virginia, October 8, 1975.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
2^7
-------
DESK TOP METHODOLOGY FOR
NONPOINT SOURCE LOAD EVALUATION
by
Arun K. Deb, Ph.D., P.E.
Project Director
Environmental Systems Department
Roy F. Weston, Inc.
INTRODUCTION
Nonpoint source pollutants are materials which degrade surface
water and groundwater quality and which originate in a diffuse
manner from the land surface. Much attention has been given by
water quality managers and planners to the problems associated
with this form of pollutant loading. Impacts of nonpoint source
loading on surface water quality have been studied for limited
large urban areas under Section 208 of Public Law 92-500. In
such areas, nonpoint pollutant load controls have been analyzed
by computer models in developing most cost-effective areawide
water qualilty management. With the completion of a fairly
large number of 208 studies, a large amount of nonpoint source
data have been developed in characterizing pollutant development
in various land use areas. Importance of evaluation of nonpoint
source loading from urban and agricultural areas other than 208
study areas and its impact in water quality has been recognized.
The use of computerized calibrated models in nonpoint source
loading analysis is expensive and may not be cost-effective
in many small water quality management studies. An alternative
to computerized approaches for the; analysis of the nonpoint
pollutant loadings is desktop anlaysis techniques. A desktop
model for nonpoint source pollution assessment is any hand-
calculation technique or procedure that can be used to determine
the stormwater runoff pollutant loading from a study area. This
type of approach involves an estimation of pollution loadings
sufficiently comprehensive to provide an environmentally and
economically satisfactory basis for realistic decisions con-
cerning nonpoint source pollution impacts.
248
-------
OBJECTIVES
This paper outlines a simplistic methodology for nonpoint source
loading evaluation for areas where rate coefficients and load-
ings can be estimated. The methodology has limited application
and can be used for:
o First level of analysis;
o Areas with similar characteristics where rate
coefficients and loadings have been developed;
o Determining the need for a detailed computer
modeling.
METHODOLOGY
The methodology used in this paper was developed for island
wide urban nonpoint source pollution analysis for Puerto Rico.
A detailed calibrated computer modeling method was used to
develop nonpoint source loadings for five typical urban areas
of Puerto Rico. Using the rate coefficients and pollutant
loading rates for various land use groups obtained by field
sampling and computer modeling, a desktop methodology was
developed for the evaluation of urban nonpoint source loadings
for other similar cities on the island.
The simplified methodology developed in this paper contains the
following basic elements:
1. Rainfall data analysis and development of a
design storm event.
2. Characterization of the drainage basin.
3. Run-off quality and flow characterization.
Essentially, the methodology relates phenomena in the drainage
area to their associated effect on stormwater loads. The rate
coefficients which depend on the cause and effect relationships
of storm events and land use characteristics are determined
previously for a similar urban area.
The step-by-step methodology developed in this paper for evalua-
tion of nonpoint source loads incorporates the following steps:
1. Historical rainfall data analysis and selection
of a design storm.
2. Determination of the land use pattern of the
study area.
249
-------
3. Determination of the percent of imperviousness
of the various land use types.
4. Estimation of the composite run-off coefficient
for the study area.
5. Estimation of the stormwater run-off rates.
6. Calculation of pollutant accumulation on each
land use at the start of the storm.
7. Estimation of the stormwater run-off loads.
Step 1; Selection of a Design Storm Event
The important parameters for design storm characterization are
the duration of a storm event, the dry period between storm
events, the maximum hourly intensity, and the total rainfall.
The maximum hourly rainfall and the total rainfall parameters
are defined by statistical recurrence intervals.
Historical rainfall data can be used to develop design storms
for each urban study area. These storms can then be used for
the development of nonpoint source pollution loads. The
specific technique used consists of the following steps:
o Select the maximum hourly intensity for the
desired design storm in a particular urban
study area.
o Select the average duration for design storms
in the urban study area.
o If the average duration is three or four hours,
assume that the maximum hourly intensity occurs
during the second hour. If the average dura-
tion is two hours, assume that the maximum
hourly intensity occurs during the second hour.
o Select the total design storm volume for the
particular design storm-in the urban study area.
o For an urban area, where storms have an average
duration of three or four hours, determine the
difference between the total rainfall for the
design storm and the maximum hourly rainfall,
and divide this difference equally between the
remaining hours of the design storm. For an
urban area where storms have an average duration
of two hours, determine the difference between
the total rainfall for the design storm and the
250
-------
maximum hourly rainfall, and assume that this
difference occurs during the first hour of the
storm.
The rainfall amounts for the different hours of the design
storm, along with the dry period before th storm would occur,
defines the design storm event. The design storm used in
illustrating the methodology is given in Table 1.
TABLE 1
Sample Study Area - Design Storm Characteristics
Total Rainfall - 0.9 inches
Storm Duration - 3 hours
Antecedent Dry Period - 10 days
Rainfall by Hour
o First - 0.25 Inch
o Second - 0.40 Inch
o Third - 0.25 Inch
Step 2; Estimation of the Land Use Pattern of the Study Area
The key element in making reliable storm load estimates with
the desktop methodology is the determination of study area
characteristics which are used in relationships that define
pollutant concentrations for particular amounts of rainfall
which will leave the area as runoff.
Initially, the overall study area, i.e., that area for which
nonpoint source loadings are desired, must be defined. The
urban study area will typically be defined on the basis of
political boundaries. However, in order to evaluate all
factors that affect nonpoint source runoff and resulting water
quality impacts, it may be necessary to look beyond the politi-
cal boundaries of the study area.
The type and quantity of nonpoint source loads depend on land
use. In order to estimate these loads, it is first necessary
to determine the distribution of the land use types in the
urban study area, and the areal extent of each land use type.
The procedure to use in determining the areal extent of land
use types is one of obtaining, from the best possible source,
up-to-date land use maps. If no maps are found to be available,
251
-------
site investigations should be used to define the general land
use characteristics of the study area.
The sample study area for this example is assumed to have the
land use characteristics listed in Table 2.
TABLE 2
Sample Study Area - Land Use Characteristics
Area Percent of
Land Use Type In Acres Total Area (LUP)
Single Family Residential 150 7
Multiple Family Residential 1000 50
Commercial 100 5
Industrial 50 3
Open and Park 700 35
TOTAL 2000 100
Step 3; Estimation of the Percent Imperviousness of the Various
Land Use Types
Imperviousness portions of the urban land use type contributes
maximum to the nonpoint source pollution loads. Therefore,
it is essential that the percent of impervious cover for each
land use type should be estimated for the study area. Table
3 lists average percent imperviousness values used for the
example.
TABLE 3
Sample Study Area - Percent of Imperviousness
Values for Particular Land Use Types
Land Use Type Percent of Imperviousness Cover
(PIC)
Single Family Residential 52
Multiple Family Residential 63
Commercial 70
Industrial 90
Open or Park 15
252
-------
Step 4; Determination of the Composite Runoff Coefficient
for the Study Area
The volumetric runoff coefficient measures the fraction of the
storm volume that reaches the receiving water body as runoff.
It is assumed that the average runoff coefficient for imper-
vious sections of an urban drainage area is 0.92, and the
value for pervious sections is 0.18. The major portion of
urban nonpoint source runoff which impacts a receiving water
body originates from the impervious sections of the drainage
basin. This is due to the fact that runoff from pervious
sections of a predominantly urban drainage basin is attenuated,
and possibly even permanently lost to infiltration.
The composite Runoff Coefficient (CRC) for the entire urban
study is determined by the following relationship:
CRC = [(IRC X PIC/100) '+ (1 - PIC/100) PRC] LUP
where:
CRC = Composite runoff coefficient for impervious
areas in each land use type.
LUP = The percentage of the total drainage area
in each land use.
IRC = Impervious runoff coefficient.
PRC = Pervious runoff coefficient.
Table 4 illustrates the calculations for determining the CRC
for the sample study area.
Step 5: Determination of the Stormwater Runoff Rate
The rate at which rainfaill is assumed to run off the drainage
basin is defined by the amount of rainfall for the storm event,
its intensity, and the Composite Runoff Coefficient (CRC). An
additional factor that must be considered when analyzing urban
nonpoint source runoff is depression storage. This parameter
defines the volume of water that is retained on the surface
in small depressions and does not become surface runoff. The
value of depression storage that was used in the sample area
was 0.02 inches.
253
-------
TABLE 4
Sample Study Area - Composite Runoff Coefficient
(CRC) Calculations
ISJ
vn
Percent
Land Use Type of Area
Percent
Imperviousness
Pervious
Area
Land Use CRC Calculation
Total
Impervious
Area
Single Family
Residential
Multiple Family
Residential
Commercial
Industrial
Open or Park
7
50
5
3
35
52
63
70
90
15
0.18 X 0.48
= 0.08
0.18 X 0.37
= 0.07
0.18 X 0.10
= 0.05
0.18 X 0.10
= 0.02
0.18 X 0.85
= 0.15
0.92 X 0.52
= 0.48
0.92 X 0.63
= 0.58
0.02 X 0.70
= 0.64
0.92 X 0.90
= 0.83
0.92 X 0.15
= 0.14
0.56
0.65
0.69
0.85
0.29
CRC for imperviousness area = (0.48 X 0.07) + (0.58 X 0.50) + (0.64 X 0.05)
+ (0.83 X 0.03) + (0.14 X 0.35)
= 0.43
= (0.56 X 0.07) + (0.65 X 0.50) (0.69 X 0.05)
+ (0.85 X 0.03) + (0.29 X 0.35)
= 0.53
CRC for the total area
-------
The relationship for determining total stormwater runoff volume
for a study area is:
Total Stormwater Runoff = (Total Rainfall - Depression
Storage) X (Composite Runoff
Coefficient)
For this example, the values in this relationship are:
Total Stormwater Runoff = (0.9 inches - 0.02 inches)
X 0.53
Total Stormwater Runoff = 0.46 inches
This same runoff relationship can be used to determine the
volume of stormwater runoff on an hourly basis. Where hourly
Total Stormwater Runoff is desired, hourly Total Rainfall must
be provided.
For the design storm being used in this example, Table 1 lists
the total rainfall for each hour of the event. The total run-
off for each hour is given in Table 5.
TABLE 5
Hourly Runoff
Rainfall
Inches
Runoff
Impervious
Inches
Total
Inches
Hour 1
Hour 2
Hour 3
0.25
0.40
0.25
0.25 X 0.43
= 0.11
0.40 X 0.43
= 0.16
0.25 X 0.43
= 0.11
(0.25-0.02) X
0.53
= 0.12
0.40 X 0.53
= 0.21
0.25 X 0.53
= 0.13
255
-------
Step 6; Calculation of Pollutant Accumulation on Each Land Use
at the Start of the Storm
The washoff of pollutants from an urban drainage area depends
on the amount of pollutants that are built up on the area at
the strat of the storm event. Table 6 lists the specific pollu-
tant accumulation values (in terms of pounds per acre per day)
as obtained from detailed studies of similar areas for the
land use types used in this example. The pollutant accumulation
given in Table 6 can be used to determine the weighted average
Pollutant Accumulation Rate (PAR) for the sample study area.
The relationship used to determine the weighted average PAR
(in pounds/acre/day) is:
Weighted Average PAR = I (PAR x LUP)
where:
PAR = The pollutant accumulation rate, in
pounds/acre/day, for each land use
(from Table 6)
LUP = The percent of the total drainage
area for each land use.
Table 6 also illustrates the calculations required to determine
the Weighted Average PAR. The example is presented for bio-
chemical oxygen demand (BOD).
TABLE 6
Sample Study Area - Weighted Average PAR Calculations
BOD Pollutant Accumu-
lation Rate
Land Use Type Percent of Total Area (pounds/acre/day)
Single Family
Residential 7 0.22
Multiple Family
Residential 50 0.40
Commercial 5 0.48
Industrial 3 0.50
Open or Park 35 0.20
Weighted Average PAR = (0.22 X 0.07) + (0.40 X 0.50) +
(0.48 X 0.05) + (0.50 X 0.03) +
(0.20 X 0.35)
Weighted Average PAR = 0.32 pounds/acre/day
256
-------
Step 7; Determination of Stormwater Runoff Loads
Stormwater runoff pollutant loads are a function of the mass of
pollutants on an acre, the runoff intensity, and an exponential
washoff coefficient. The relationship which incorporates these
factors in order to determine pollutant loads is an empirical
one based on studies of surface buildup and washoff of pollu-
tants, and is expresed as:1
Mp = Pp (1 - e-KRl)
where:
Mp = the mass of a pollutant that is washed from
a surface during a given (hourly) time period.
Pp = Accumulated pollutant remaining on the
surface at the beginning of the time step.
K = Washoff decay coefficient, assumed to be
equal to 2.0 for Puerto Rico urban areas.
R.J. = Stormwater runoff rate in inches per
hour from the impervious section of the
surface
The value for Pp at the beginning of the rainfall event depends
on the Weighted Average Pollutant Accumulation Rate, Table 6,
the Antecedent Dry Period (Table 6), and the total acreage of
the study area. For this example, the total pollutant on the
surface at the start of the rainfall event is:
Pp = (Weighted Average PAR) X (Antecedent Dry Period)
X Acres in Study Area
Pp = 0.32 pounds/acre/day X 10 days X 2000 acres
Pp = 6400 pounds
Using the empirical pollutant washoff relationship, the total
amount of pollutants washed off the total study area during
each hour of the design storm is:
Hour 1
Pp = 6400 pounds
Rj = 0.11 inches/hour
K = 2.0
^Ann .. -2.0 X O.llv
Mp = 6400 (1 - e )
Mp = 1264 pounds
257
-------
Hour 2
The Pp value at the start of hour 2 is equal
to the Pp value at the start of the storm minus
the Mp value for the first hour, and so on
throughout the storm event.
Pp =
Tf —
Mp =
Mp =
6400 pounds - 1264 pounds
0.16 inches/hour
2.0
= 5136 pounds
5136 (1 - e
1406 pounds
-2.0 X 0.16
Hour 3
Pp =
K
Mp =
Mp =
5136 pounds - 1406 pounds = 3730 pounds
0.11 inches/hour
2.0
3730 (1 - e
737 pounds
02.0 X 0.11
The total pollutant load for this design storm event is there-
fore the sum of the hourly pollutant loads, or 3407 pounds,
Pollutant Concentration
Hour 1 Pollutant Concentration = Load
Runoff Volume
1264
0,12 X 2000 X 0.0272 X 8,33
23,2 mg/L
Hour 2 Pollutant Concentration = 1406
0,21 X 2000 X 0.0272 X 8,33
= 14.8 mg/L
Hour 3 Pollutant Concentration = 737
0.13 X 2000 X 0.027 X 8,33
= 12,5 mg/L
Average Concentration of Nonpoint Source Pollutant during the
storm event =
Total Load
Total Runoff
= 1264 t 1406 + 737
0.46 X 2000 X 0,0272 X 8.33
258
=16,3 mg/L
-------
CONCLUSIONS
The desktop methodology discussed, in this paper is simple and
can be used with minimum available information. However,
limitations for using this methodology should be observed and
can only be used as first level analysis and extrapolation of
analysis to areas where good values of nonpoint source loading
rates and rate coefficients are established.
REFERENCE
1. "Storage, Treatment, Overflow, Runoff Model *- STORM,
User's Manual", The Hydrologic Engineering Center,
The Army Corps of Engineers, July 1976.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
259
-------
CONTINUOUS SIMULATION OF INSTREAM
FECAL COLIFORM BACTERIA
A.C.Rowney (1) L.A.Roesner (2)
ABSTRACT
The simulation of fecal coliform bacteria in an urban stream is
discussed in light of a recent study of the Rideau River at
Ottawa, Canada. As part of this study, it became necessary to
verify the use of a one-dimensional advection/dispersion river
model incorporating first order decay in the simulation of fecal
coliform bacteria. An extensive time-series data base was
gathered for calibration and verification of the model QUAL-II.
This data base was supplemented with runoff data simulated using
STORM. It was found that the selected models provided a good
representation'of major processes in the test reach selected,
certainly adequate for planning purposes. Some preliminary
statistics representing the goodness of fit of simulated to
observed instream concentrations are presented. In addition,
modifications made to QUAL-II to permit simulation of time series
input conditions are briefly discussed.
INTRODUCTION
The Rideau River at Ottawa has been the subject of an on-going
investigation designed to protect and enhance water quality for
several years. This investigation is motivated by a number of
general concerns, one of which is that the bathing beaches
located on the river and within the city of Ottawa have been
closed by the medical officer due to unacceptable indicator
bacteria levels.
A major component of the investigations has therefore been to
isolate sources of bacterial contamination so that proper
management techniques to reduce bacterial levels to acceptable
limits may be undertaken. Past studies, discussed during the 1981
SWMM users group meeting in Niagra Falls, have shown that a major
contribution to dry-weather sources is likely to be carryover of
elevated instream concentrations from one storm event to the
next. It is evident, however, that significant non-point sources
of dry weather bacterial pollution also exist. Prior to any
(1) Proctor and Redfern Ltd
(2) Camp Dresser and McKee Inc.
260
-------
attempt to institute controls on storm-water related pollution,
it is necessary to investigate the likely magnitude of these dry
weather sources.
A careful examination of all likely sources of fecal coliform
contamination is therefore required, so that the nature and
likely relative importance of sources may be estimated. A part of
that investigation involves modelling the instream transport of
bacteria concentrations, and it is the calibration and
verification of the QUAL-II model in this context which is
discussed in this paper.
The work discussed herein was carried out by a joint venture
between Proctor and Redfern Ltd and Gore and Storrie Ltd in
association with Camp Dresser and McKee Inc. on behalf of the
Rideau River Stormwater Management Study.
2.0 SELECTION OF TEST REACH
The selected test reach for calibration and verification is
depicted in Figure 1. As shown, the test reach is a fairly
straight section of river, bounded at the top and bottom by
bridges. The section was selected for several reasons. It is
known to have a fairly regular cross-section throughout its
length, which facilitates simulation of hydraulics. In addition,
the area directly tributary to this reach is minimal, which
reduces the size of un-monitored stormwater sources and
consequently simplifies simulation. The bridges above and below
the test reach provided convenient markers for locating
monitoring stations, and facilities for 24-hour monitoring and
sample storage could be located nearby conveniently.
Unfortunately, the site had some drawbacks, chief of which was
its length, some 1.6 km. This short distance means that during
high flow periods, changes in bacteria concentration due to
die-off are small, and therefore are difficult to assess.
However, increasing the length of the test reach would increase
the number and size of non-quantified sources, so a balance had
to be struck. The test reach shown was the result.
3.0 AVAILABLE DATA BASE
3. 1 Hydraulic Data
The Rideau River at Ottawa is characterised by a series of deep
and shallow sections. As part of a navigational waterway, the
river is highly regulated, and flows within the study section are
controlled by backwater effects from the locks and dams at the
outlet. For this reason, the option for calculation of hydraulics
261
-------
using the normal depth trapezoid contained by QUAL-II was
abandoned in favour of inputs derived using the backwater HEC-2
model. HEC-2 was set up and run for a range of flows which
encompassed the range measured during collection of the bacteria
data in the test reach. Examination of the flow and elevation
data simulated and observed in the test reach showed that changes
in velocity in this section can be well represented by a linear
function, and that depths are relatively constant over the range
of flows which is of interest. Therefore, QUAL-II was coded to
use a constant depth and linear velocity-flow relation.
River flow data during the period of monitoring was taken from a
Water Survey of Canada gauge located about 5 km. upstream of
the test reach. A seperate gauge was established at the outlet of
the Sawmill Cr., since flows from this tributary can govern flows
in the test reach during storm events. Lateral inflows not
accounted for by the two gauges were estimated using runoff
volumes calculated by a calibrated STORM model.
3.2 Fecal Coliform Bacteria
Bacteria concentration data were gathered at four instream points
and on one tributary. The chosen sites, depicted in Figure 1, are
located evenly down the length of the test reach, and on the
Sa wm i 11 C r .
Samples were taken hourly on a 24-hour basis for a period of just
over one month, beginning on the sixth of July, 1981. Samples
were taken at the mid-stream point at a depth of one meter, and
were stored in cooler chests in ice for a maximum of six hours
prior to shipping to the laboratory for analysis. A membrane
filter technique was used in estimating fecal coliform
concentrations in the samples so collected.
The measured sources therefore accounted for all runnoff related
bacteria except for a small local tributary area. This area was
simulated using the STORM model, and results were input to
QUAL-II along with the measured values.
Bacteria die-off rate parameters wfere taken from findings of a
study of the University of Ottawa, which measured bacteria
die-off in plastic bags suspended in the river from floats. An
average measured T90 of 43 hours was calculated in these
experiments, and a range of 24-48 hours was tested in this study.
4.0 MODIFICATIONS TO QUAL-II
QUAL-II is currently available in a version which only accepts
inputs of steady state or initial condition data. Applied in the
steady state mode, it is useful for evaluation of continuous
discharges during periods of known constant river flow, or for
similar applications. The model can be extended to dynamic uses
262
-------
BRIGHTON
PARK
STORM SEWER
COMBINED SEWER OVERFLOW
TEST REACH SAMPLE
RIDEAU RIVER STORMWATER
MANAGEMENT STUDY
TEST REACH FOR
BACTERIA CALIBRATION
263
-------
If initial conditions resulting from an event are known, since
these initial conditions can be specified in the model, and the
subsequent recovery of the river can be 'viewed'.
However, it is not possible to simulate time varying inputs of
concentration or flow with the standard version of the model.
While this was not deemed a necessary condition for completion of
modeling in the overall study, this capability was required for
calibration to time series data during this specific part of the
study. Rather than employ a model already using dynamic inputs,
it was decided for reasons of consistancy and economy to modify
QUAL-II to accept this type of data.
In practise, these changes proved to be quite simple to
implement. A number of possible modifications were tested, and a
final version of QUAL which allows dynamic input of concentration
and flow data at headwater, point source, and uniform lateral
load points was devised. The modified MAIN flowchart for QUAL-II
is shown in Figure 2.
The new version of QUAL is designed to read data at arbitrary
time intervals from disc files, and will accept point source,
headwater , and distributed loads that vary independently.
Outputs from the model are printed to disc or hard copy devices
at user defined time increments in a format which facilitates
subsequent computer plotting. In addition, the time series
outputs have been made slightly less unweildy by allowing the
user to specify any number of individual elements for display.
To allow a rapid and economical sorting of data by users, a LINK
program was also created. This program is a simple utility
routine which reads data from coded field observations and
interpolates flow and concentration data at any specified time .
increment and in any combination desired by the user. The primary
benefit to this approach is that it allows the user to rapidly
evaluate the significance of various combinations of individual
sources, and in addition makes it possible to smooth data
somewhat by averageing several observations into one value
representative of each time increment.
It is interesting to note that the' modified QUAL-II reflected
hydraulic conditions just as well when headwater flows were held
fixed as when they were allowed toy vary as observed. This
implies that an average flow, velocity, and depth adequately
represent hydraulic conditions in this system, which is
consistent with observations made in prior studies. However, it
was found that the resulting simulation did not accurately
represent in-stream concentrations. The model would conserve
mass, but without the proper input volumes to calculate
dilutions, concentration simulation was not possible. The dynamic
hydraulic inputs used in this calibration were therefore provided
261+
-------
ff
4
, h
2
x
u
i
$
£
3
S
3
HI
at
a
<
i
j
. ,
^
S»
5
S
5
^
B
a »
w ui
>
^
U £
Az
2
w
3
*
I
2g
o
J
£
c
^
a
K
a
«
|
-C>
rW
* t
t
3 r
* b
s-
h
« D
- p
9
10 D
,, p
11
12 B
13 ^
14 t
,. p
15
!S t
i^
17
18 D
19
r
21 t
23 Ci
24 *!
25 D
9B
28
27 CI
R
28 c
J
^
INDATA
IuvnoAii
TBIUAY
TKIMAT
ICONSVT
ITBUBAvmjBttC
rtUHrar rUHr«Kf
Ioflne
ouua
IAI (SACQ
MbvMca
IBAAe
rV^a
INHSfi
MA99
n\j£st
MOSS
QPAPQf*
nCACKV
DCS
iCOLlS
RADIOS
WRPT2
pi OALIfl
MtOOFT^
WRPT3
1 i^l r*UAMI
—^^^V\ WclMHu
i^^^^^J ugftTgv I'ucn^ro
^^^^^^ riCAl CJW FI&M 1 &n
i "" "i 1
I
I
I
^ D Y N 1 N
1
o
v
T
I
HHYDRAU
T R I M AT
FIGURE 2. MODIFIED QUAL-II STRUCTURE
265
-------
to allow proper computation of diluted concentrations, rather
than for the impact they make on advection parameters.
5.0 MODEL APPLICATION
5.1 Calibration
A calibration period consisting of the last 200 hours of
observations was selected for calibration. This period includes
about one quarter of the available time series observations, and
covered periods representative of both dry and storm flow
conditions.
Calibration was a fairly simple process, since flows input to
the model were fixed. Parameters available for calibration were
therefore advection, dispersion, and die-off.
It was observed that the advection of pollutants, as measured by
timing of peak events moving through the system, was fairly well
represented as input, and this factor was not altered. The
initial flow/velocity relation calculated by HEC-2 was maintained
throughout the calibration/verification series.
Dispersion in the model as applied in this study was limited to
the numerical dispersion induced by the particular element size
and time step incorporated in the model. An associated study
showed the impact of dispersion on the model results is not
significant, and this factor was not altered during calibration.
Thus, the major parameter left for calibration was the bacteria
die-off rate. Since previous die-off studies by the University of
Ottawa had quantified this value, the range of die-off rates
tested was known. In this case, T90 times averaged 43 hours with
a range of approximately 20 to 60 hours.
There was, however, one condition not accounted for in the above
set of input and rate parameters, and that is the rate of input
of continuous sources of bacterial pollution. These sources which
included the combined effect of such things as illegal
connections, animal populations, and other diffuse sources, were
not quantifiable and therefore had to be input to the model by
inference. The model was adjusted so that events causing peaks of
concentration (i.e. storms) were well represented, and then the
rate of constant distributed side loads was adjusted to increase
dry weather levels to observed values.
This proved to be an effective and rapid means of obtaining a
good calibration to observed conditions, although as noted below
the size'of the implied constant source 1s a direct function of
the assumed rate and form of bacterial decay.
266
-------
5.2 Veri fication
Once the calibration was completed, the model was verified by
completing a run over the entire 800 hour data set with all
calibration parameters held constant.
6.0 DISCUSSION OF RESULTS
In general, the model was found to reproduce observations
remarkably well, considering the highly variable nature of fecal
coliform bacteria. Figure 3 shows part of the calibration data
set. It is evident that major peaks are reproduced quite
closely, and that base levels do range about the simulated
values. Figure 4, representing part of the verification period,
presents a similar picture, although somewhat more scattered.
Subjectively, one may observe what appears to be a diurnal
variation of observed concentration about the simulated level
during some dry periods, but the mechanism for this effect is not
known for certain.
Comparisons were made of observed and simulated conditions in
several ways, as discussed below:
Mean errors were calculated to provide some indication of how
wel1simulations were centered within the range of observed
bacterial concentrations.
TABLE 1. MEAN DIFFERENCES BETWEEN SIMULATED AND
OBSERVED F.C. CONCENTRATIONS
location ^points mean error mean log error
(no/dl) log(no/dl)
TR-2 760 -8.9 -0.0646
TR-1 778 15.6 -0.0112
TR-5 353 -52.6 -0.0627
Average concentrations observed were about 300 no/dl at all three
stations. Although the plotted curve does not duplicate the
observed values exactly, it is (on average) well centered in the
range of observed values.
RMS errors were calculated to provide an index of spread about
the simulated values.
267
-------
STATION
TR 1
Aug.
STATION
TR 2
TIME (days)
FIGURE 3. CALIBRATION PERIOD
+ Observed
- Computed
268
-------
s
July 6
ro
a
u
<~.
u
-7
-w
,olh
FIGURE 4. VERIFICATION - STATION TR1
PARTIAL RESULTS
!
H,
TIME (days)
+ Observed
Computed
-------
TABLE 2. RMS DIFFERENCES BETWEEN SIMULATED AND
OBSERVED F.C. CONCENTRATIONS
location #points RMS error
1og(no/dl)
TR-2 760 0.2342
TR-1 778 0.1962
TR-5 353 0.1687
The average concentration observed during the above simulation
period was about 2.5 log(no/dl).
Frequency of departures was calculated. While the RMS shown
above is a common measure of fit, it does not provide much
information about the likelihood of any given error unless the
distribution relating simulations and observations is known. To
provide a measure of this factor, the absolute differences
between 1og(observed) and 1og(simulated) results were collected
and grouped in classes of width .02. From this data, % less than
curves were determined as shown in Figure 5. It is evident from
these curves that the three stations are represented
approximately equally well, and that for practical purposes all
observations may be considered to be within .2 to .3 logs of
the simulation. Again, for.the variability of the phenomenon,
this is quite a good approximation.
Changes in concentration during peak periods were estimated.
This was by necessity a subjective assessment, but gives some
feel for the 'goodness1 of peak event representation. As shown
in Table 3, observed changes are on average 6% higher than
simulated.
Absolute peaks were compared, again subjectively, to provide
some indication of how well peak concentrations are predicted, as
shown in Table 4.
In summary, several points are clear from the above results.
o While the natural variability of fecal coliform bacteria
in the environment makes it impossible to exactly
duplicate observed values, simple model techniques are
capable of providing a reasonable representation of the
instream processes affecting the persistence of fecal
coliform bacteria.
o QUAL-II provides a quick, economical, and effective model
for use in this type of analysis.
270
-------
100 -
90
SMYTH
CENTENNIAL
• BRIGHTON
0.0
0.1
0.2
0.3
0.4
0.5
0.6
O.7
0.8
0.9
1 r—
1.0 1.1
1
1.2
TEST REACH ANALYSIS
DEPARTURE OF OBSERVATIONS FROM
SIMULATED RESULTS
FIGURE 5.
271
-------
TABLE
COMPARISON OF SIMULATED AND
OBSERVED CHANGES FROM BASE LEVEL
FECAL COLIFORM CONCENTRATIONS
LOCATION
TR-1
TR-2
TR-5
EVENT CHANGE (log(no/dl)) RATIO
OBSERVED SIMULATED (0/S)
1 1. 05 .66 1. 59
2 .75 .33 2.27
3 .89 .95 .94
4 .50 .42 1.19
5 1.45 1. 40 1. 04
1 .70 1.02 .69
2 .55 .50 1.10
3 .75 1.25 .60
4 .35 .78 .45
5 1.15 1.61 .71
4 .40 .35 1.14
5 1.30 1.38 .94
Average= 1.06
TABLE
COMPARISON OF SIMULATED AND
OBSERVED FECAL COLIFORM
PEAK CONCENTRAIONS
LOCATION
TR-1
TR-2
TR-5
EVENT
1
2
3
4
5
1
2
3
4
5
4
5
PEAK (log(no/dl))
OBSERVED SIMULATED
3
3
3
2
3,
2,
3,
55
20
00
72
3. 70
3.25
3.15
00
80
65
2. 75
3. 65
3.22
2
3,
2
66
21
88
3.84
3,
2,
3,
3
3
54
80
42
12
98
2.8
3.8
Average'
Average as %•
DIFFERENCE
(0-S)
.33
.54
-.21
-. 16
-. 14
-.29
.35
-.42
-.32
-.33
-.05
-.15
-.07
17% high
272
-------
o In the Rideau river, a significant continuous source of
fecal coliform bacteria contamination exists.
o Quantification of such a continuous source is not in this
case possible by direct measurement. It is, however,
possible to estimate the size of the source by inference
from modelling results. Care must be taken in such a
calculation since the estimated rate of bacteria die-off
has a great impact on the implied load, and the meachanism
of instream die-off is only approximated by a simple first
order decay assumption.
Currently, work is underway to resolve some of these
uncertainties. For example, improved information on bacterial
die-off is being generated from larger scale and longer duration
measurements. Subsequent stages of analysis will use the models
developed in this study to provide information on bacteria
sources sufficient to permit planning of measures to protect the
river from further degradation, and if possible to improve the
current situation.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
273
-------
ATMOSPHERIC POLLUTION IN RELATION TO STORM WATER
QUALITY MODELLING; LITERATURE REVIEW FOR AN
INDUSTRIAL CITY
by
Shivalingaiah, B.
and
William James
Civil Engineering Department
McMaster University
Hamilton, Ontario, Canada
ABSTRACT
The atmosphere is one of the largest sources of pollu-
tants in surface runoff in an industrial city. The location
of the major industrial areas, wind direction and velocity
and source concentrations are important parameters in the
prediction of the atmospheric fallout component of surface
loadings. Isopleths of atmospheric fallout are correlated
with these parameters and superimposed on the discretized
catchment. The total accumulated dry weather loading is
input to the water quality sections of SWMM-RUNOFF. The
water quality algorithms in SWMM III are reviewed and im-
provements suggested, eg. inclusion of atmospheric fallout
and scavenging, street sweeping time series data, separation
of pollutant source areas, variable time step hydrology and
moving storm analysis.
274
-------
INTRODUCTION
Atmospheric pollutants are washed out by precipitation
or fallout as sediments, or are deposited by chemical
processes. The atmosphere is one of the identified sources
of nutrients and solids to runoff. The interrelation
between atmospheric pollution, rain and storm water contami-
nation has not been fully established. An attempt is made
in this paper to review the literature for an industrial
city, Hamilton, on atmospheric pollution in relation to
storm water quality modelling. A discussion of the SWMM III
runoff water quality algorithms and possible areas for im-
provement concludes the paper.
GEOGRAPHICAL FEATURES OF HAMILTON
Hamilton is highly industrialized and is located on the
south western shore of Lake Ontario. The iron and steel in-
dustry is the major activity in the city of 306,640 (1980)
occupying most of the southern shore of Hamilton Harbour.
The downtown area lies to the southwest. The Niagara Es-
carpment almost surrounds Hamilton and the southern arm di-
vides the city into upper and lower sectors having an aver-
age height difference of 325 feet (figure 1). The escarp-
ment is cut by a number of deep valleys, the most important
of which is the southwest-northeast aligned Dundas Valley.
275
-------
LOCATION OF HAMILTON
Figure 1: Geographical features of Hamilton
276
-------
The major steel plants are located in the northeast of Ham-
ilton, with associated industries including machinery,
electrical and chemical manufacturing. The business dis-
trict is made up of a core of multi-storey commercial build-
ings of limited areal extent surrounded by a lower level of
mixed commercial and residential properties. The prevailing
wind is dominantly west and southwest. Though the frequency
of easterly wind (30-40 percent) is less than westerly wind,
easterly wind transports industrial pollutants west across
the city. It is especially unpleasant when occasional anti-
cyclonic conditions result in light easterly winds which
have been cooled over the cold lake surface. Then, very
stable lower air layers are created due to the temperature
inversion which locks polluted air over most of the lower
city.
SOURCES OF ATMOSPHERIC POLLUTANTS
In an urban situation industries, automobiles, house
heating and resuspension of solids due to wind and vehicle
movements contribute significant quantities of pollutants to
the atmosphere. Winds on unpaved areas and unprotected in-
dustrial waste products also adds to the problem. The aver-
age values of pollutants produced due to combustion are
given in Table 1 (Joe 0. Ledbetter, 1972).
277
-------
COMBUSTION
TABLE 1
POLLUTANTS
Particu- Oxides of
lates Nitrogen Organic Hydrocarbons Aldehydes
Natural Gas
Ib/million
cu.ft.
of gas 15
Fuel oil
lbs/1000 gal
of oil 8
Coal
Ib/million
BTU 5
Autos lb/
1000 gal/fuel 12
Diesel lb/
1000 gal/
fuel 110
390
140
1.9
113
222
3.2
1,0
0.6
0.4
200
0.0013
4
136
10
REMOVAL PROCESSES
Pollutants are removed from the atmosphere by dry and
wet processes. In the case of dry process (absence of rain)
particles greater than 20 urn. size settle under gravita-
tional force] and particles less than 20 urn. size are tran-
sported close to the surface by means of turbulence. Wet
processes refer to removal of pollutants by precipitation.
This is further divided into washout and rainout processes.
Precipitation particles collecting dry particles by inertial
collection or/ in the case of smaller aerosol particles, by
278
-------
diffusion and phoretic forces, is known as washout (scaveng-
ing). Rainout refers to particles which are nucleated for
water condensation and thereby grow into cloud drops and in
turn grew into precipitation particles.
Scavenging of gases by water drops due to molecular
diffusion occurs in accordance with the vapour pressure and
solubility of the free and collected gases (Junge, 1963;
Chamberlain, 1960; and Griffiths, 1963). The maximum wash-
out possible at a particular distance, assuming the rain
starts as soon as the release of pollutants, is related to
source strength, washout coefficient and average wind velo-
city (Culkowski, 1963; Guthrie and Nichols, 1964).
Among the above processes, wet processes contribute re-
latively more pollutants to runoff water. During dry days,
dust accumulates (dry process) on the surface and washs off
during precipitation. Wet processes scavenge pollutants
during rain. It is evidently essential to account for both
dry and wet processes in modelling surface water quality.
CONTRIBUTION OF ATMOSPHERIC POLLUTANTS TO PRECIPITATION
Increased urbanization and industrialization auguments
pollutants in rain water. Suspended solids, BOD, COD, TOC,
nitrate and phosphates are often commonly measured in rain
279
-------
and runoff water. The average values of COD and TOC in rain
water are 20.7-322 rag/lit and 2.8-18 ing/lit respectively
(C.W. Randall et al 1978; Per-Arne Malmquist 1978).
Nitrate and phosphate in rain water has been found to exceed
the level recommended for controlling eutrophication (Vladi-
mir Novotny 1981, Randall et al, 1978, Rutherford 1967, John
A. Frizzola, 1975; Albert Goettle, 1978; Dean Stuart,
1975, Chan and Kantz 1981). The concentration of pollutants
in the atmosphere and in rain water may be site specific in
an highly industrialized city. Maximum levels of pollutants
have been observed to occur during the early stages of rain,
similar to the first flush effect observed in runoff water
(Randall et al, 1978). The ground surface acts as a net
pollutant sink rather than a source of water pollution (Ran-
dall 1981). Exceptions to this condition are tilled land
and highly impervious business areas.
AIR POLLUTION AND RAINWATER STUDIES IN THE CITY OF HAMILTON
The first report of an air pollution survey was pub-
)
lished inyDecember 1956 (Ontario Research Foundation, 1956).
Since then a number of studies have been conducted on dif-
ferent aspects such as atmospheric fallout, total suspended
solids, soil index, sulpher dioxide, oxides of nitrogen
(Stewart, 1968; Mathehson, 1969; Rouse et al, 1970;
Heidron, 1978; MOE reports 1977-80; Barton, 1981). The
280
-------
dust concentration has been increasing even though a decre-
asing tendency has been observed for other pollutants since
1970. Effective controls for point source emissions have
been enforced since 1970 (MOE 1980 Hamilton air quality re-
port) . This shows that other pollution sources (industrial
and non-industrial) such as blowoff from unpaved areas, ex-
cavation, construction, demolition, road traffic (Syd Barton
1981) , uncontrolled stockpiles and other non stack industri-
al emissions are adding to the problem. Yearly average
dustfall observations showed that a portion (15 sq. km.) of
the lower city and the beach strip near the industrial area
was encompassed by the 9.0 gram/sq.m - 30 days isopleth.
This is twice the Ontario Ministry of the Environment objec-
tive. Another 57 sq. km. was encompassed by 4.5
gram/sq.m.-30 days contour. Remaining areas of the city re-
ceive less than 4.5 gram/sq.m.-30 days.
An average concentration of 3.5 ug/cu.m. of nitrate
and 3-24 ug/cu.m of organic carbon (5-14% of TSP) (Barton
1981) was observed in suspended particulates over Hamilton.
Maximum atmospheric pollution index incidents (32 to 50
level) were observed during a lake breeze regime with moder-
ately consistent winds from the east-northeast at speeds
less than 6 miles/hr (Heidorn, 1978).
Not many studies appear to have been conducted on rain
281
-------
water quality over Hamilton, except for the work of Dean
Stuart (1975). Relative to surface waters, precipitation is
normally low in conductivity and pH, with elevated heavy me-
tals (10-1000 ug/lit) and nutrients (50-100 mg/litP;
400-2000 mg/litN) concentration. The above mentioned re-
sults are mean values obtained at a single station. Samples
collected at a number of stations give a better understand-
ing of wet process contributions.
FACTORS AFFECTING SPATIAL DISTRIBUTION OF POLLUTANTS
The dispersion of pollutants in the atmosphere is the
result of three dominant mechanisms: (a) the general mean
air motion that transports the pollutants downwind, (b) the
turbulent velocity fluctuations that disperse the pollutants
in all directions and (c) mass diffusion due to concentra-
tion gradients. In addition, the general aerodynamic char-
acteristics such as size, shape and weight, affect the rate
at which the nongaseous pollutant particles settle to the
ground or are buoyed upwards.
The wind patterns near the shore of lakes, oceans, and
bays are complicated because of differences in the rate of
war-ming between the land and water. The expansion of the
rising warmer air over the land causes a general air move-
ment horizontally from the water to the land (sea or lake
282
-------
breeze). At night the land surface cools at a faster rate
by radiation than does the water. The air over the land
gradually becomes cooler and more dense than the air over
the water. Hence, a general local horizontal air movement
occurs from the land to the water (land breeze). Most large
industrial cities in Canada are located near large bodies of
water and hence wind patterns over cities are complex. In
addition to this, the presence of escarpments, mountains,
large hills or prominant drainage valleys near the city
further complicates the wind pattern.
Atmosphere stability is related to pollutant dispersal.
Under stable conditions, pollutants do not exhibit much
vertical mixing or motion. This is of importance in esti-
mating the quantity of air pollutants at a given location,
SWMM III RUNOFF QUALITY ALGORITHMS
Pollutants are assumed to build up on an impervious
area during the dry days preceding a storm and then washoff
into the drains during a storm. Land use types for pollu-
tant build up rate are the same as those in SWMM II but many
more relations (power, linear, exponential or
Michhaelis-Menton) are included in SWMM III for dust and
dirt accumulation. A maximum limit for accumulation of dust
and dirt and for specific pollutants is also included in
283
-------
SWMM III (1982).
Rain causes washoff of dust and dirt from impervious
areas for each time step, (POFF), proportional to runoff rate
to a power, (WASHPO).
- POFF(t) = d/dt (PSHED) = -RCOEFX*(R**WASHPO)*PSHED
where POFF = constituent load washed off at time, t,
PSHED = quantity of constituent available for washoff
at time 't', mg.
RCOEFX = washhoff coefficient = RCOEF/3600,
(in/hr)**-WASHPO *(I/sec), and
RX = runoff rate, in/hr.
RCOEF - coefficient, includes units conversion
WASHPO = exponent
A rating curve method is included as an alternative to
the use of a buildup - washoff formulation. Loads
(mass/time) may be generated proportional to flow power
POFFV RCOEF*(WFLOW**WASHPO)
where:
WFLOW = subcatchment runoff, cfs,
(or cu.m/sec)
RCOEF = coefficient, includes units
conversion,
-------
WASHPO = Exponent
parameters RCOEF and WASHPO are entered for each
particular constituent.
Soil erosion from the pervious area is calculated using
the universal soil loss equation.
L = R.K.LS.C.P.
where L = average annual soil loss per unit area
R = rainfall factor
K = soil erodibility factory
LS = the slope length gradient ratio
C = the cropping management factor or cover
index factor,
P = the erosion control practice factor
R = E. RAINIT
where E = total rainfall energy for time period
of summation, 100 ft.-ton/ac.
= (9.16 + 3.31 log(RNINHR(J)))*RNINHR(J)*DELT
RNINHR = rainfall intensity at time interval J, in/
hr and
DELT = time interval, hr,
RAINIT = maximum average 30-minute rainfall
intensity for the storm (single event) or
the period of simulation (continuous) in/hr.
Provision has been made to add erosion to other
285
-------
constituents.
The contribution of pollutants from precipitation is
allowed in SWMM by permitting a constant concentration of
constituents as an input to the model for a complete catch-
ment.
AREAS FOR IMPROVEMENT
Dust and dirt buildup ideas used in the SWMM do not
consider the physics of generation of pollutants from
sources such as street pavements, vehicles, atmospheric fal-
lout, scavenging, vegetation, land surfaces, litter, spills,
antiskid compounds and chemicals, construction and drainage
networks. The model lumps all these sources together in es-
timating the buildup rates for different land uses. It may
be simple to separate some major sources like atmospheric
fallout, scavenging, etc. Atmospheric fallout, and the con-
tribution from scavenging, is a local phenomena which mainly
depends on the type of city, wind velocity and direction,
location of industrial area or major industries and topogra-
phy of the catchment. Comparison of build-up rates given in
SWMM for various land uses with observed atmospheric fallout
in Hamilton showed that atmospheric fallout contributes a
major portion of dust and dirt (Table 2). The average
monthly variations in fallout of 1977-80 is given in Table
286
-------
TABLE 2
COMPARISON OF CONSTITUENT BUILD UP RATES OF SWUM WITH
ATMOSPHERIC FALLOUT OF HAMILTON % Contri
Atmos. fall- buttons
SWUM dd Rates out of Hamilt. from
Mo. Land Use DDlb/100 ft-dryday DDlb/acre-day DDlb/acre-day atmos.
Single Family
Residence
Multi-Family
Residence
Commercial
Industrial
Undeveloped/park
Open/Instutional
O. 7
2. 3
3. 3
4.6
1.5
1.54
5.06
9.9
5.98
2.40
1. 344
2.688
2.688
2.688
1.344
86»
54%
26\
41*
54%
TABLE 3
Average Monthly Variation of Duitfall, qram/m2-30 dayi (1977-1980).
Station
Jan
Fab
March
April
May
June
July
Aug
Sept
Oet
Dnc
29001
29006
29008
29009
29010
29011
29012
29017
29019
29025
29026
29030
29031
29036
29037
29044
29046
29067
5.48
5.53
26.83
S.OO
9.50
9.58
9.20
7.75
2.68
9.88
4.55
3.55
5.80
5.73
15.73
14.08
3.58
5.03
6.28
6.3
17.5
5.9
13.4
15.53
7.53
7.95
3.08
7.75
5.15
4.53
7.9
9.1
14.9
12.35
3.03
5.2
12.65
12.08
14.75
8.4
24.68
22.4
14.47
15.93
6.58
14.28
6.70
7.40
11.25
14.98
26.83
14.78
4.93
9.68
10.58
11.23
11.23
7.95
20. 3
22.35
12.05
11.96
4.83
13.28
6.28
6.25
10.03
14.85
26.57
18.08
5.23
7.63
9.8
8.03
9.78
6.53
18.15
19.73
14.48
10.3
6.27
14.13
6.53
8.18
8.08
12.65
18.68
10.88
14.5
8.17
6. SB
9.83
8.37
6.65
12.77
16.33
9.93
8.93
5.35
11.60
5.95
8.75
9.35
13.88
17.50
11.35
9.25
6.00
6.0
6.18
12.40
5.10
13.90
9.03
7.20
8.80
7.88
9.30
4.70
6.63
5.03
10.90
17.80
10.58
9.03
5.35
5.73
7.68
11.93
5.1
16.25
12.10
7.88
8 ..50
3.55
9.53
4.53
6.25
6.00
9.48
21.53
9.53
3.53
4.48
6. 75
5.93
10.45
6.03
15.57
16.23
13.28
9.00
5.80
11.20
4.83
6.00
7.83
10.45
19.95
14.95
4.03
7.08
5.35
5.93
13. 38
4.70
16.05
12.48
7.15
14.00
3.80
8.18
4.75
4.85
5.80
9.75
22.0
8.98
2.83
4.4
5.4
5. 1
13. 3
4. 53
19.93
13.95
8. IS
8.63
3.40
8. 48
4.30
4.40
6.08
8.35
24.20
9.83
3.88
3.60
e, . (• »
5.75
1 9 . G 'I
5.n^
17. 33
13.C,
12. 1
11 .:n
3.53
7 .nn
•1 ."I
3.7r'
fi . ? •>
22. 2(1
1 4 . 7 '.
11.10
3 ,<10
6. mi
287
-------
3. The variation is mainly due to meteorological parameters
and location relative to the industrial area. Location of
the industrial area/ dust sampling stations and subcatch-
ments for runoff calculations are shown in figure 2.
Monthly and yearly average wind roses are shown in Figure 3.
The yearly average isopleths of atmospheric fallout are
shown in Figure 4 (MOE report 1980). Attempts are now
underway to correlate the dust concentrations at various
stations with wind direction and velocity. General conclu-
sions will be reported in the near future.
Scavenging of pollutants by rain depends on the concen-
tration and source of pollutants, atmospheric stability, in-
tensity of rainfall, wind direction and storm characteris-
tics. Therefore atmospheric fallout and scavenging should
be modelled separately. These processes should not be con-
sidered build up rates as they are independent of land use.
Provision has to be made in the model to generate concentra-
tion isopleths for dry periods to obtain a loading rate due
to atmospheric fallout and scavenging process for every in-
)
dividual-subcatchment, based on source location, atmospheric
stability and meteorological condition. Constituents based
on land use are added to the above loading rate to obtain
total buildup rate for every subcatchment area. This is
better than relating build up rates to land use. The au-
thors will attempt to suitably modify SWMM III during summer
288
-------
'
Figure 2: Location of Industrial Area
289
-------
October
Novembar
December
40%
9-10 11-20 21
MILES PER HR.
B 10 18 20
— ii i ]
PERCENT FREQUENCY
Figure 3: Yearly and monthly average wind rose, Hamilton
(RBG, 1978-80)
290
-------
Figure 4: Yearly average isopleths of atmospheric fallout
291
-------
1982.
The quality of stormwater depends on the concentrations
of dust and dirt accumulated on the land surface at the
onset of the storm. SWMM assumes a constant cleaning inter-
val for each different land use. There is no control over
antecedent conditions, whether or not the street has been
swept or not before the storm. Therefore, inclusion of
street sweeping data as a time series better simulates the
actual dust and dirt accumulation at the beginning of the
storm. Support for this modification to SWMM III has been
secured for 1982.
SWMM washoff algorithm uses a power exponent of runoff
rate and a washoff coefficient. Provision is made to input
different values for each constituent in estimating washoff
rate. It does not allow for the effect of slope in washoff.
In the case of steep roofs, washoff of dust and dirt is much
faster and responsive to small rain intensities. Therefore
separation of steep sloped roofs in the runoff water quality
analysis might yield better results. Not all constituents
found on roads and gutters may be expected on roofs or im-
pervious areas, and vice-versa. A modification to SWMM III
is currently underway.
SWMM uses a constant time step to integrate over the
292
-------
period for quantity and quality prediction. The smaller the
time step, the greater would be the accuracy in prediction,
and higher the computation cost. Introducing a variable
time step, allows the user to run the program at smaller
time steps during storms and longer time steps in the dry
periods for continuous modelling. This modification will be
carried out at the same time as the street sweeping input
modification.
Thunderstorms are not static, and the intensity of rain
varies in each subcatchment (Shtifter and w. James, 1981).
Inclusion of storm intensity predictions on a given sub-
catchment using the storm direction and cell width, age of
the storm cell, wind direction and velocity yields better
input to the surface runoff calculations. Our SWMM block
entitled THOR computes spatially and time averaged hyeto-
graphs for kinematic storms.
All of these changes have been or are being implemented
in the SWMM package used by the Computational Hydraulics
Group at McMaster University as part of our on-going stu-
dies. Results obtained for Hamilton have so far shown en-
couraging improvements (Robinson and James, 1982); (James
and Shtifter, 1981).
293
-------
REFERENCES
Air Quality Report 1978, Hamilton city, MOE, Ontario.
Air Quality Report 1979, Hamilton city, MOE, Ontario.
Air Quality Report 1980, Hamilton city, MOE, Ontario.
Barton, S. et al (1981), "An Assessment of Street Dust and
Other Sources of Airborne Particulate Matter in Hamilton,
Ontario", Technology Transfer Seminar, Toronto, November 24,
1981. Sponsored by MOE 11. 11 pp.
Chamberlain, A.C., (1960), "Aspects of the Deposition of Ra-
dioactive and Other Gases and Particles", Inter, J. Air
Pollution 3 (1-3).
Culkowski, W.M., "Deposition and Washout Computations Based
on the Generalized Gaussian Plume Model", USAEC Report
ORO-599, Weather Bureau, Oak Ridge, Tenn. (1963).
Stuart, Dean, "Ontario Precipitation Chemistry and Heavy
Metal Speciation", Ph.D. Thesis, McMaster University, 1975.
pp.1-125.
Griffiths, V., "The Removal of Iodine from the Atmosphere by
Sprays", British Report AHSB(S) R-45 (1963).
Heidorn, K.C., "Air Pollution Incidents and Wind Variability
in Southern Ontario", J. Atmospheric Environment Vol. 12,
pp. 2251-2257, 1978.
Geottle, A. "Atmospheric Contaminants, Fallout and Their
Effects on Stormwater Quality", Prog. Wat. Tech., Vol.
10, Nos. 5/6, pp. 455-467, 1978.
Frizzola, John A., et al, "Contaminants in rainwater and
their Relation to Water Quality", Part I, Water and Sewage
Works, August, 1975. pp. 72-75.
Frizzola, John A., et al, "Contaminants in rainwater and
their Relation to Water Quality", Part II, Water and Sewage
Works, September 1975. pp. 94-95.
James, W., and Shtifter, Z., "Implications of Storm Dynamics
on Design Storm Inputs", proceedings of the Conference on
Water Quality and Storm Water Management Modelling, Niagara
Falls, Ontario USEPA, October, 1981, pp. 55-78.
Ledbetter, Joe 0., "Air Pollution Part A: Analysis", Marcel
Dekker, Inc., New York, 1972.
-------
Junge, C,E., "Air Chemistry and Radioactivity", Academic
Press Inc., New York, 1963.
Matheson, D.H., et al, "Air Pollution Survey for Hamilton,
Ontario", Atmosphere Environment, Pergamon Press 1969. vol.
3, pp. 11-23. printed in Great Britain.
Ontario Research Foundation Report, "Air Pollution in Hamil-
ton City", Department of Chemistry, December 11, 1957, np
1-50.
Malmquist, Per-Arne, (1978), "Atmospheric Fallout and Street
Cleaning - Effects on Urban Storm Water and Snow", Prog.
Wat. Tech. 1978, vol. 10, nos. 5/6 pp. 495-505.
Randall, C.W., "The Impact of Atmospheric Contaminants on
Stormwater Quality in an Urban Area", Prog. Wat. Tech
1978 Vol. 101, no. 5/6, pp. 417-431.
Randall, C.W., et al, "Comparison of Pollutant Mass Loads in
Precipitation and Runoff in Urban Areas", Second Interna-
tional Conference on Urban Storm Drainage, Urbana, Illinois,
U.S.A., June 14-19, 1981.
Robinson, M., arid James, W., "Continuous SWMM Modelling of
Hamilton Summer Stormwater Including Certain Quality Indica-
tors - Preliminary Output Time Series Using Discrete-event
Calibration for Non-industrial Areas", published by CHI Pub-
lications, March 1982.
Rouse, W.R., and McCutcheon, John G., "The Effect of the Re-
gional Wind on Air Pollution in Hamilton, Ontario", Canadian
Geographer, XIV, 4, 1970 pp. 271-285.
Rutherford, G.K. "A Preliminary Study of the Composition of
Precipitation in S.E. Ontario", Canadian Journal of Earth
Sciences, vol. 4, 1967, pp. 1151-1160.
Stewart I.M., et al, "Methods of Relating High Volume
Sampler Particulate Loadings to Wind Direction", Atmospheric
Environment, Pergamon Press, 1968, Vol. 2, pp. 181-185.
Novotny, Vladimir, "Acidity of Urban Precipitation and its
Buffering during Overland Flow", second International
Conference on Urban Storm Drainage, Urbana, Illinois, U.S.A.
June 1981.
Huber, Wayne C., et al, "Storm Water Managment Model User's
Manual Version III, November 1981.** pp.
The work described in this paper was not funded by the U.S. Environmental
Protection Agency. The contents do not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
295
-------
SWMM CONFERENCE - MARCH 25-26, 1982
LIST OF ATTENDEES
Roger K. Wells - HMM Associates, Inc., Raleigh, NC
Curtis H. Dalton - Maryland DHMH, OEP
Stephen L. Luckman - Maryland DHMH, OEP
Joan Lefler - U.S. EPA, Washington, DC
Bastien Jean-N - St. Laurent, Quebec, Irmwada
John Roberts - MacLaren Plansearch, Toronto
Steve McKelvie - Gore & Storrie Ltd., Toronto
Bob Walker - Beak Consultants Ltd., Toronto
Gary Woodruff - Tulsa City-County Health Dept.
Joseph L. Norton - Ford Thornton Norton & Assoc., Vicksburg
Fred Morris - S. Florida Water Management District
Robb Startzman - S. Florida Water Management District
Claire Welty - U.S. EPA
Warren Viessman, Jr. - Congressional Research Service
Charles Delos - U.S. EPA
John Weeks - Ketron, Inc., Arlington, VA
Bob Rallison - U.S. Soil Conservation Service, Wash., DC
Gerald Dougherty - Purdum & Jeschke, Balto., MD
Michal D. Norn's - Purdum & Jeschke, Balto., MD
Raymond Wright - University of Rhode Island
Wayne Huber - University of Florida
Paul MacLeod - Giffels Associates, Ltd.
Richard Horner - University of Washington
Frank 0. Marrazza - GKY Assoc^, Springfield, VA
John Barile - GKY Assoc., Springfield, VA
Jack Hartigan - N. Virginia Planning District Comm.
Tom Quasebarth - N. Virginia Planning District Comm.
Betsy Southerland - N. Virginia Planning District Comm.
Dave Gubarry - Woodward-Clyde Consultants
Ellen Petticrew - Canada Centre for Inland Waters
Mark Robinson - McMaster University, Canada
Ron Scheckenberger - McMaster University, Canada
296
-------
SMvalingaiah. B - McMaster University, Canada
? - McMaster University, Canada
Michael Akerbergs - Woodward-Clyde Consultants
Raymond Dever - Montgomery County DEP, MD
Thomas Nesbitt - EPA HQ
Samuel R. Martin - Regional Planning Council, Balto., MD
Alan Lamb - U.S. Geological Survey
Grace Weik - Dames & Moore
Robert Pasley - USNA Soil Conservation Service
Padma Datta - EPA - OPP
A. Charles Rowney - Proctor & Redfern, Ottawa
Marcel!a McTaggart - Alcoa
Phil Rosten - AWARE Corp.
Lai it Sinha - EPA, IL
Charles D. Woo - FHWA
Don Hoang - Portland
Ray Whittemore - NCASI
James McKeown - NCASI
Michael J. Hudson - IEP, Inc.
Malcolm A. MacGregor - IEP, Inc.
John Aldrich - COM, Annandale, VA
David Schafer - COM
H.S. Loijens - Cananda
Peter G. Robertson - MD OEP
Bernard Ross - USF
W. Janes - McMaster University
Ivan Chou - ESE, Inc.
Stephen E. Wall - Greeley & Hansen
Edward R. Ester, III - MMM Design Group
James E. Scholl - CHgM Hill
Roger L. Long - General Software Corp.
Jan-Tai Kuo - General Software Corp.
Karl Hemmerich - City of Toronto
Mike Kangas - Dal ton-Dai ton-Newport, Cleveland
Bob Cole - Dal ton-Dai ton-Newport, Cleveland
297
-------
Ginger Klingelhoefer - Anne Arundel Co., MD
Wan Wong - Ontario Environment
Don Urban - USDA-SES
Charles R. Terrell - USDA-Soil Conservation Service
Bruce Bird - AACC
Jonathan Young - Brown & Caldwell
Larry A. Roesner - Camp Dresser & McKee, Annandale, VA
Prekimi V. Tawari - Engineering & Economics Research, Inc.
Byron Lord - FHWA R&D
John Segna - US EPA
Deborah McCall - Roy F. Weston, Inc.
Marlene Conaway - Anne Arundel Co., MD
Tom Schaffer - MW Council of Gov't.
Jerry Klafter-Snyder - Roy F. Weston, Inc.
Link Haghighaf - Howard County DPW, MD
Arun K. Deb - Roy F. Weston, Inc.
Bob Ambrose - U.S. EPA, Athens, GA
Alan Cavacas - Northern Virginia Planning Dist. Comm.
Tieh Yin - MNCPPC, Upper Marlboro, MD
John A. Friedman - Northern Virginia Planning Dist. Comm.
Donald Groff - Western County State College
Harry Torno - U.S. EPA, SAB
Tom Barnwell - U.S. EPA
Larry Roesner - COM, Inc., Alexandria, VA
298
a U.S. GOVERNMt NT PRINTING OFFICE: 1982-559-092/0444
------- |