&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

-------