Anda di halaman 1dari 10

WATER RESOURCES RESEARCH, VOL. 39, NO. 10, 1280, doi:10.

1029/2002WR001788, 2003

Comparative evaluation of urban storm water quality models


J. Vaze1 and Francis H. S. Chiew
Cooperative Research Centre for Catchment Hydrology, Department of Civil and Environmental Engineering,
University of Melbourne, Melbourne, Victoria, Australia
Received 21 October 2002; revised 27 May 2003; accepted 11 June 2003; published 9 October 2003.

[1] The estimation of urban storm water pollutant loads is required for the development

of mitigation and management strategies to minimize impacts to receiving environments.


Event pollutant loads are typically estimated using either regression equations or
process-based water quality models. The relative merit of using regression models
compared to process-based models is not clear. A modeling study is carried out here to
evaluate the comparative ability of the regression equations and process-based water
quality models to estimate event diffuse pollutant loads from impervious surfaces. The
results indicate that, once calibrated, both the regression equations and the process-based
model can estimate event pollutant loads satisfactorily. In fact, the loads estimated
using the regression equation as a function of rainfall intensity and runoff rate are better
than the loads estimated using the process-based model. Therefore, if only estimates
of event loads are required, regression models should be used because they are simpler and
INDEX TERMS: 0345 Atmospheric
require less data compared to process-based models.
Composition and Structure: Pollutionurban and regional (0305); 1871 Hydrology: Surface water quality;
4857 Oceanography: Biological and Chemical: Pollution; 4842 Oceanography: Biological and Chemical:
Modeling; KEYWORDS: diffuse pollution, event, regression equation, pollutant wash off, storm water, urban
Citation: Vaze, J., and F. H. S. Chiew, Comparative evaluation of urban storm water quality models, Water Resour. Res., 39(10),
1280, doi:10.1029/2002WR001788, 2003.

1. Introduction
[2] The estimation of urban storm water pollutant loads is
required to develop mitigation and management strategies
to minimize impacts to receiving environments. Long-term
estimates of pollutant loads are usually sufficient for management overviews and for management of receiving waters
with long response time. Estimates of storm wash off loads
and pollutant concentrations over storm events are required
to evaluate the impacts of elevated runoff and pollutant
concentrations and loads on macroinvertebrates and the
ecology of urban waterways; to investigate short-term
toxicity problems; to study the effectiveness of pollution
control measures during storms; and to design storage/
treatment facilities. Estimates of event pollutant loads are
also important because most of the annual loads are transported by the big events [Duncan, 1995]. Studies over
shorter timescales may also lead to a better understanding
of storm water runoff quality processes.
[3] Event pollutant loads are typically estimated using
either regression equations or process-based water quality models. The regression equations empirically relate the
event loads to the storm and catchment characteristics
[Jewell and Adrian, 1981, 1982; Driver and Lystrom,
1986; Driver and Tasker, 1988; Driver, 1990]. The process-based water quality models typically simulate the dry
weather accumulation of pollutants on the catchment sur1
Now at Department of Infrastructure, Planning and Natural Resources,
Centre for Natural Resources, Queanbeyan, New South Wales, Australia.

Copyright 2003 by the American Geophysical Union.


0043-1397/03/2002WR001788$09.00

SWC

face and the subsequent wash off over storms [Heaney et


al., 1976; Geiger and Dorsch, 1980; Johanson et al., 1984;
Hemain, 1986; Huber and Dickinson, 1988]. The relative
merit of using regression models compared to process-based
models is not clear. That is, it is not evident whether the
process-based models yield results that are worth the effort
required to make them operational.
[4] Data availability and reliability are the most critical
factors in the modeling of storm water pollution as, without
local data, none of the models can be tested or used
successfully. Detailed event data are required, where the
wash off is sampled at regular intervals throughout the
storm event and the rainfall and runoff are concurrently
measured. Thus the estimates derived from both approaches
depend heavily on the spatial and temporal quality and
quantity of the available data. As with nearly all-large scale
engineering problems, it is impossible to collect enough
data to fully characterize the pollutant loads that would
result from a full range of storm and catchment types; the
cost would simply be prohibitive.
[5] The main problem in modeling storm water pollution
loads is therefore the lack of event water quality data and
the large variability in the pollutant concentration data that
are available [Duncan, 1995]. The lack of data has led to a
diversity of opinions about the pollutant buildup-wash
off mechanisms. The apparent diversity of opinions for
explaining the wash off process of particulate pollutants
centers around four explanatory variables, rainfall intensity
[Sartor and Boyd, 1972; Yaziz et al., 1989], rainfall volume
[Pravoshinsky and Gatillo, 1969], runoff rate [Lager et al.,
1971; Ichikawa, 1981; Huber and Dickinson, 1988], and
runoff volume [Hartigan et al., 1978; Diniz, 1979; Barbe et
al., 1996], and two main processes, shear stress generated

5-1

SWC

5-2

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

Table 1. Land Use Breakdown Characteristics for the Three


Catchments Based on Aerial Photography
Blackburn Lake Sandy Creek Cressy Street
Total catchment area, ha
Total impervious fraction, %
Total pervious fraction, %
Total residential area, ha
Total industrial area, ha
Total open spaces, ha
Average catchment slope, %

200
67
33
94
106
0
3

227
42
58
149
0
40
4

107
44
56
107
0
0
2

by flow [Nakamura, 1984; Akan, 1987; Deletic et al., 1997]


and the energy input of rainfall [Price and Mance, 1978;
Coleman, 1993]. The problem with the four explanatory
variables is that they are inherently correlated with one
another, and it is difficult to discriminate accurately between
them [see Duncan, 1995; Chiew et al., 1997].
[6] This paper presents results from a modeling study
carried out to evaluate the comparative ability of the
regression equations and process-based water quality models to estimate event diffuse pollutant loads from impervious
surfaces. In the empirical modeling, simple regression
equations of storm-related variables are tested. In the
process-based modeling, a water quality model similar to
the commonly used U.S. Environmental Protection Agency
(EPA) Storm Water Management Model (SWMM) is coded
and used to simulate event hydrographs and loadographs.
Total suspended solids, total phosphorus, and total nitrogen
data from three residential catchments in Australia are used.

2. Methods
2.1. Catchment Description and Data Analysis
[7] Data from three Australian urban catchments are used
in this study. The catchments consist mainly of residential
land use and are located in Melbourne (Blackburn Lake
catchment) and Brisbane (Sandy Creek catchment and
Cressey Street catchment). All three catchments are relatively small with catchment areas of 200 ha (Blackburn
Lake), 227 ha (Sandy Creek), and 107 ha (Cressey Street).
Table 1 provides a summary of the land use breakdown and
proportions of imperviousness in the three catchments.
2.1.1. Blackburn Lake Catchment
[8] The Blackburn Lake catchment is located in the
eastern suburbs of Melbourne in the headwaters of the
Gardiners Creek catchment [RossRakesh et al., 1999].
The Blackburn Lake catchment has a temperate climate
with an average annual rainfall of 700 mm. The rainfall is
slightly higher in winter and early spring compared to the

rest of the year. The catchment is fully serviced by storm


water drains. Most of the drainage network consists of
underground pipe network with only about 10% above
ground, half of which is lined with rocks. The catchment
is relatively flat with average slopes of about 3%. The
catchment is completely developed, with predominantly
residential and industrial land uses. The catchment has very
small portions of lumped impervious and pervious areas,
which are formed by parks, schools, and shopping centers,
and practically no open space.
2.1.2. Sandy Creek Catchment
[9] The Sandy Creek catchment is located within the
suburb of Indooroopilly, approximately 6 km to the west of
Brisbanes central business district [City Design, 2000]. The
Sandy Creek catchment has a tropical climate with an average
annual rainfall of 1200 mm, practically all of which falls in
summer and early autumn. Sandy Creek discharges to the
Brisbane River and incorporates an extensive piped underground drainage network that flows to an open channel in its
lower reaches. The catchment is relatively steep in its upper
reaches, with gradients of between 5% and 10%. In the lower
to middle reaches, grades are moderate to flat. The catchment
is completely developed, with predominantly residential land
use. In addition, the catchment also supports considerable
portions of lumped impervious and pervious areas that are
formed by parks, schools and shopping centers.
2.1.3. Cressey Street Catchment
[10] The Cressey Street catchment is 107 ha in area. The
catchment is located in the suburbs of Wavell Heights and
Nundah, approximately 8 km to the north of Brisbanes
Central Business District [City Design, 2000]. The Cressey
Street catchment also has a tropical climate with an average
annual rainfall of 1200 mm, practically all of which falls in
summer and early autumn. The catchment topography is
relatively flat with average slopes of about 2%. Catchment
drainage is provided by a network of underground storm
water pipes which discharge to an open channel at the lower
end of the catchment. The catchment is completely urbanized with land use characteristics being dominated by
established residential development. Unlike the Sandy
Creek catchment, Cressey Street exhibits negligible proportion of lumped pervious and impervious areas, with the
majority of the catchment area supporting conventional
residential land use.
2.2. Data Collection and Analysis
[11] Data for the Blackburn Lake catchment are collected
as part of a study carried out by the Cooperative Research
Centre for Catchment Hydrology. Data for the Sandy Creek
and Cressey Street catchments are obtained from the Brisbane City Council.

Table 2. Range of Storm Rainfall, Runoff and Pollutant Loads Used in the Study
Catchment

Data Set

Number of Events

Rainfall, mm

Runoff, mm

TSS, kg

TP, kg

TN, kg

Blackburn Lake

all events
small events
all events
small events
all events
small events

20
16
14
11
18
10

1 16
17
1 55
17
1 46
16

0.1 6.4
0.1 1.5
0.1 21
0.1 1.3
0.1 20
0.1 1

70 1700
70 1100
20 5200
20 1600
10 5600
10 200

0.1 1.8
0.1 1
0.1 13
0.1 1.6
0.06 7
0.06 0.5

0.7 15
0.7 6
0.8 45
0.8 14
0.4 50
0.4 2

Sandy Creek
Cressey Street

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

SWC

5-3

storm runoff, rainfall intensity, and runoff rate (stormrelated variables).


load atotal storm rainfallb

load atotal storm runoff b

load a

n
X

1  min rainfall intensityb

1  min runoff rateb

i1

load a

n
X
i1

Figure 1. Data details and procedure used to construct


pollutograph.
[12] Data from 20 storm events between February 1996
and August 1997 in Blackburn Lake catchment, 14 events
between October 1994 and January 1996 in Sandy Creek
catchment, and 18 events between January 1995 and May
1996 in Cressey Street catchment are used here. The
analysis is carried out separately using data for all events
and using data for small events only (see Table 2). The
analysis is carried out separately for small events to
identify pollutant wash off process from runoff only from
effective impervious areas. The small events are determined subjectively by examining the event rainfall-runoff
plot.
[13] One-minute runoff and rainfall data are used in the
modeling exercise. The 1-min runoff data are obtained
from continuous discharge measurements at the catchment
outlets. Rainfall data are recorded every 6 min in the
Blackburn Lake catchment and every 5 min in the Sandy
Creek and Cressey Street catchments, and the data are
disaggregated uniformly to provide the 1-min rainfall data.
Pollutant concentration data are obtained from laboratory
testing of storm water samples collected by automatic
sampling stations.
[14] The procedure for constructing the pollutograph
from the sampled pollutant concentrations at regular intervals and for calculating the total event pollutant loads is
shown in Figure 1. The event pollutant loads are calculated by summing the product of runoff and pollutant
concentration over 1-min intervals. The pollutant concentration is assumed to vary linearly between the recorded
concentrations (dotted line in Figure 1). The data are
included in the analysis only if at least 75% of the storm
runoff volume is sampled, assuming that a point water
quality measurement is representative of the pollutant
concentration over 5 min prior to and after the sample is
taken (shaded area in Figure 1).
2.3. Modeling
2.3.1. Empirical Modeling
[15] The use of five regression models (equations (1)
(5), where a, b, c, and d are model parameters) to estimate
event pollutant loads is investigated. The event loads are
estimated as power functions of total storm rainfall, total

load a

n
X

1  min rainfall intensityb

i1

n
X

1  min runoff rated

i1

2.3.2. Process-Based Modeling


[16] A process-based model similar to the runoff block
of the commonly used U.S. EPA Storm Water Management
Model (SWMM) [Huber and Dickinson, 1988] is used to
estimate event pollutant loads. The model is recoded to
facilitate automatic optimisation of model parameters and
used to simulate event hydrographs. As the main interest is
to simulate event hydrographs and pollutographs, the model
is run at a 1-min time step.
[17] To simulate event runoff, the model parameters are
calibrated separately for each event to obtain the best
possible estimate of runoff hydrographs to assess the model
simulation of pollutant wash off. Pollutant wash off is
simulated using the exponential decay function where the
surface pollutant load decays exponentially as a function of
the runoff rate,
Poff

dPp
Rc r n Pp
dt

where Poff is rate at which pollutant is washed off the


catchment at time t, Pp is the amount of pollutant on the

Figure 2. Commonly used pollutant buildup functions.

SWC

5-4

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

Figure 3. Comparison of estimated and recorded event pollutant loads in cross-verification exercise for
small events only.

catchment at time t, Rc is wash off coefficient, r is runoff


rate, and n is an exponent.
[18] For each catchment, the same starting surface pollutant load is used for the simulation of all the events (see
Figure 2, catchment load). This is because studies have
shown that the accumulated load on the catchment surface is
often not the limiting factor for pollutant wash off [Vaze and

Chiew, 2002, 2003b; Chiew et al., 1997; Malmquist, 1978].


The surface pollutant generally accumulates quickly and
reaches an equilibrium load over several dry days, and a
storm event typically removes only a small proportion of the
surface pollutant load. There are therefore three parameters in
the pollutant wash off model: the starting pollutant load (Pp),
wash off coefficient (Rc), and the wash off exponent (n).

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

SWC

5-5

Figure 4. Comparison of estimated and recorded event pollutant loads in split sampling exercise for
small events only.

2.3.3. Model Testing


[19] Two types of model calibration and testing are
carried out. In the first method, a split sampling exercise
is carried out where the model is calibrated against the first

half of event pollutant load data. The optimized parameter


values are then used to estimate the event loads for the
remaining storms, and the estimated and recorded loads are
compared to assess the model. In the second method, a

SWC

5-6

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

Figure 5. Coefficient of efficiency (see equation (8)) indicating the correlations between the estimated
and recorded event pollutant loads.

cross-verification exercise is carried out where each event is


left out in turn, and the model is calibrated against the other
data. The load for the event that was left out is then
estimated using the optimized parameter values and compared with the recorded load. Both methods thus assess the
loads estimated by the models against event pollutant load
data that are not used in the model calibration.
[20] In both methods, the parameters are optimized
to minimize the sum of squares of errors between the
estimated and recorded loads,
n
X
i1

ESTi  RECi 2

where n is the number of storm events and ESTi and


RECi are the estimated and recorded total event loads,
respectively.

3. Presentation of Results
[21] The plots in Figures 3 and 4 compare the event
pollutant loads estimated by the five regression equations
and the process-based model with the recorded loads for the
cross-verification and split sampling exercises, respectively,
for the analysis of only the small events. Simulations were
also carried out using all the data (small and big events), but
the results are not shown here because it is difficult to
differentiate between different model outputs. Instead, two

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

SWC

5-7

Figure 6. Comparison of total pollutant loads estimated by the five regression equations and the
process-based model with the total recorded loads.

overall measures are used to summarise the model results.


Figure 5 shows the coefficient of efficiency, E, for the crossverification and split sampling exercises, for modeling using
all data and modeling using only data from the small events,
for the three catchments. The coefficient of efficiency is
defined as
n 
P

E i1

n
2 P
RECi  REC  ESTi  RECi 2
i1
n 
P

RECi  REC

2

i1

where RECi and ESTi are the recorded and estimated loads,
respectively, and REC is the mean value of all the recorded

loads. The coefficient expresses the proportion of variance


of the recorded loads that is explained by the model [Nash
and Sutcliffe, 1970; Chiew and McMahon, 1993] and
provides a measure of the closeness of the points in the plots
in Figures 3 and 4 to the 1:1 line. It is a similar measure to
the objective function used here (see equation (7)), and they
both place more weight on the adequate simulation of the
higher loads.
[22] Figure 6 compares the total estimated and recorded
loads presented as (the total estimated load minus the total
recorded load from all events modeled) divided by the total
recorded load. The scales in the plots indicate values within
50% of the total loads, and the shaded areas indicate that

SWC

5-8

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

the estimated total loads are within 25% of the total


recorded loads.

4. General Discussion
[23] In the following discussions, the five regression
models (equations (1) (5)) will be referred to as the total
rainfall, total runoff, rainfall intensity, runoff rate, and
combined rainfall and runoff equations, respectively. The
results for total suspended solids (TSS), total phosphorus
(TP), and total nitrogen (TN) are generally similar. (TP is
mainly a particulate pollutant, but there is a significant
dissolved component in TN [Vaze and Chiew, 2003a]), and
unless otherwise stated, their results are discussed together.)
[24] The simulations in the cross-verification testing are
better than those in the split sampling testing, because the
models are calibrated against data from more storm events
in the cross-verification method. The E values are higher
and the total estimated, and recorded loads are in closer
agreement in the cross-verification testing compared to the
split sampling testing (see Figures 5 and 6). In the crossverification testing all five regression models estimated the
total pollutant loads remarkably well. Except for TSS in
Sandy Creek, the total estimated loads are always within
25% of the total recorded loads, and when only small events
are considered, the total estimated loads are within 15% of
the total recorded loads.
[25] In the split sampling exercise, the total pollutant
loads estimated by the five equations for the small events
only are generally better than those estimated when considering all the events (see Figure 6). This is mainly because of
the lower variability in the small event data subset; in the
small events, wash off occurs mainly from the impervious
surfaces, while in the big events, there are various proportions of wash off from impervious and pervious surfaces.
[26] The process-based model estimated the total wash
off loads reasonably accurately for all the three catchments.
Except for TSS in the split sampling exercise for Sandy
Creek, the E values are generally greater than zero, and the
total estimated loads are within 50% of the total recorded
loads (see Figure 6).

5. Discussion of Models and Processes


[27] The split sampling and cross-verification results lead
to similar conclusions in the interpretation of the relative
performance of the five regression equations and the process-based model. As the cross-verification exercise uses
more data to calibrate the model, the following discussion
concentrates only on the cross-verification simulations.
5.1. Rainfall Intensity or Runoff Rate?
[28] The alternative methods for modeling pollutant wash
off center around two main processes: the loosening of
surface pollutants created by the energy of falling raindrops
and the shear stress generated by flow.
[29] The cross-verification modeling results indicate that
the total loads estimated by the rainfall intensity and runoff
rate equations for all the three catchments are always within
25% of the recorded loads (see Figure 6). The results are
significantly better when only the small events are considered, with the estimated loads being within 10% of the
recorded loads.

[30] The E values in the cross-verification exercise are


generally greater than 0.5 for the Blackburn and Sandy
Creek catchments (except for the estimation of TSS in
Sandy Creek) and are greater than zero for the Cressey
Street catchment (except for rainfall intensity equation for
TN), indicating that once calibrated, the two equations can
estimate event pollutant loads reasonably well.
[31] There are two main limitations in the data used here
to reflect rainfall energy for detaching surface pollutants and
runoff energy associated with overland flow. First, the
runoff is measured at the catchment outlet and not in the
catchment itself where the wash off processes are occurring.
Nevertheless, as the catchments are relatively small, the
runoff hydrograph in the catchment may not be too different
from the hydrograph at the catchment outlet. Second, as the
time of concentrations in the catchments are relatively
small, rainfall data over periods shorter than the 5 or
6 min used here may be required to capture the surface
pollutant detachment process driven by rainfall energy. The
quality of the pollutant loads estimated by the rainfall
intensity and runoff rate equations are similar here, but it
is likely that in bigger catchments, the rainfall intensity
equation would perform better because the time of concentration is longer and because the runoff hydrograph at the
catchment outlet is a lot different than the hydrograph in the
catchment itself.
5.2. Rate or Volume?
[32] The total pollutant loads estimated by the total
rainfall and rainfall intensity equations in the cross-verification exercise are within 15% of the total recorded loads
(see Figure 6). The E values in Figure 5 indicate that the
rainfall intensity equation consistently performs better than
the total rainfall equation (except for TSS for all events).
Except for TSS in Sandy Creek and TN in Cressey Street,
the E values are above zero, suggesting that once calibrated,
the rainfall intensity equation can be used to provide a
reasonable estimate of individual event pollutant loads.
[33] The total pollutant loads estimated by the total runoff
and runoff rate equations in the cross-verification exercise
are also within 15% of the total recorded loads (except for
the loads estimated by the total runoff equation for Sandy
Creek for all events; see Figure 6). The E values indicate
that the runoff rate equation performs similarly or better
than the total runoff equation. This is particularly evident in
the Cressey Street simulations, where the E values for the
runoff rate equation are positive, while the E values for the
total runoff equation are negative (see Figure 5). The results
therefore indicate that the rainfall energy for detaching
surface pollutants and the shear stress generated by the
flow rate are bigger factors governing pollutant wash off
from impervious surfaces than the total storm rainfall or
runoff.
5.3. Rainfall Intensity and Runoff Rate?
[34] The most striking observation is the remarkable
performance of the equation combining the rainfall intensity
and runoff rate (equation (5)). The total pollutant loads
estimated by this equation in the cross-verification exercise
are practically the same as the total recorded loads (see
Figure 6). The total loads estimated by this equation are also
generally within 20% of the total recorded loads in the split
sampling exercise, while in some cases, the other equations

VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

cannot estimate the total loads to within 50% of the


recorded loads. The E values comparing the pollutant loads
estimated by this combined equation with the recorded
loads are also generally above 0.5, compared to very low
or negative values in the simulations using the other four
equations (see Figure 5). The simulations here therefore
indicate that once calibrated, the combined rainfall intensity
and runoff rate equation can estimate event pollutant loads
satisfactorily.
[35] It is possible that the combined rainfall intensity and
runoff rate equation performed best because it has two
additional parameters. In some calibrations, the four parameters in the equation are optimized using only data from
several events (e.g., six events in the split sampling testing
for small events in Sandy Creek and Cressey Street).
However, it is unlikely that the accurate estimates given
by equation (5) are due to an artificial predictive skill
because the split sampling and cross-verification test the
model performances on independent data that are not used
in the calibration.
5.4. Process-Based Model
[36] For all the three catchments, comparison of the
simulated and recorded hydrographs indicates that the
simulated flow hydrographs for all the events matched
the recorded hydrographs adequately (not shown here),
although there is considerable variability in the optimized
parameter values for the different events.
[37] For all the three catchments, comparison of the
simulated and recorded loadographs indicates that the
model-simulated loadographs for all the events are not as
good as the simulated hydrographs. The main reason for this
is because unlike the runoff hydrographs, the same set of
parameter values are used to simulate all the storm events in
a catchment. The parameters are also optimized against the
total wash off loads and not against the loadographs.
[38] In any case, the intention here is to assess the
models ability in estimating the total wash off loads.
Figures 5 and 6 indicate that the process-based model
almost always performs better than the total rainfall equation (1) and total runoff equation (2) and similarly or
slightly better than then rainfall intensity equation (3) and
runoff rate equation (4). However, like the other four
empirical equations, the simulations from the process-based
model are considerably poorer than those from the combined rainfall intensity and runoff rate equation (5).

6. Conclusions
[39] This paper assesses the use of simple regression
equations and a process-based model to estimate event
pollutant loads. Five types of regression equations are used
where the event load is estimated as a power function of
total storm rainfall, total storm runoff, rainfall intensity,
runoff rate, or a combination of rainfall intensity and runoff
rate. A process-based water quality model similar to the
Storm Water Management Model (SWMM) is also used.
The models are used to estimate TSS, TP, and TN loads for
three residential catchments in Brisbane and Melbourne.
[40] The results indicate that the combined rainfall intensity and runoff rate equation performed the best, and once
calibrated, the equation can estimate event pollutant loads
satisfactorily. The rainfall intensity equation generally per-

SWC

5-9

forms better than the total rainfall equation, and the runoff
rate equation generally performs better than the total runoff
equation, indicating that the rainfall energy for detaching
surface pollutants and the shear stress generated by the flow
rate are bigger factors governing pollutant wash off from
impervious surfaces than the total storm rainfall or runoff.
[41] The process-based model performs better than the
total rainfall equation and total runoff equation and performs similarly or slightly better than then rainfall intensity
equation and runoff rate equation but performs considerably
poorer compared to the combined rainfall intensity and
runoff rate equation. The results therefore suggest that
where only event pollutant load estimates are required,
regression equations (particularly the combined rainfall
intensity and runoff rate equation) should be used because
they are simpler and require less data. Nevertheless, unlike
the regression models, which only estimate the event
pollutant load, the process-based model also estimates
pollutant concentration and load over the storm event.

References
Akan, A. O., Pollutant washoff by overland flow, J. Environ. Eng., 113,
811 823, 1987.
Barbe, D. E., J. F. Cruise, and X. Mo, Modelling the buildup and washoff of
pollutants on urban watersheds, Water Resour. Bull., 32(3), 511 519,
1996.
Chiew, F. H. S., and T. A. McMahon, Assessing the adequacy of catchment
streamflow yield estimates, Aust. J. Soil Res., 31, 665 680, 1993.
Chiew, F. H. S., H. P. Duncan, and W. Smith, Modeling pollutant buildup
and washoff: Keep it simple, paper presented at 24th International
and Water Resources Symposium, N. Z. Hydrol. Soc., Auckland, New
Zealand, Nov. 1997.
City Design, Stormwater quality monitoring program, report, Brisbane City
Counc. Waterway Programs, Brisbane, Australia, 2000.
Coleman, T. J., A comparison of the modelling of suspended solids using
SWMM3 quality prediction algorithms with a model based on sediment
transport theory, in Proceedings of the Sixth International Conference on
Urban Storm Drainage, Niagara Falls, Ontario, Canada, vol. 1, edited
by J. Marsalek and H. C. Torno, pp. 790 795, 1990.
Deletic, A., C. Maksimovic, and M. Ivetic, Modelling of storm wash-off of
suspended solids from impervious areas, J. Hydraul. Res., 35, 99 118,
1997.
Diniz, E. V., Water quality prediction for urban runoff, an alternative
approach, in Proceedings Storm Water Management Model (SWMM)
Users Group Meeting, Misc. Res. Ser. EPA-600/9-79-026, pp. 112
134, Environ. Prot. Agency, Washington, D. C., 1979.
Driver, N. E., Summary of nationwide analysis of storm-runoff quality and
quantity in urban watersheds, in Proceedings of the Fifth International
Conference on Urban Storm Drainage, Osaka, Japan, vol. 1, edited by
Y. Iwasa and T. Sueishi, pp. 333 338, 1990.
Driver, N. E., and D. J. Lystrom, Estimation of urban storm runoff loads,
urban runoff qualityImpact and quality enhancement technology, in
Proceedings of an Engineering Foundation Conference, New Hampshire,
edited by B. Urbonas and L. A. Roesner, pp. 222 232, Am. Soc. of Civ.
Eng., Reston, Va., 1986.
Driver, N. E., and G. D. Tasker, Techniques for estimation of storm-runoff
loads, volumes and selected constituent concentrations in urban watersheds in the United States, U.S. Geol. Surv. Open File Rep., 88 191,
Denver, Colo., 1988.
Duncan, H. P., A review of urban stormwater quality processes, Rep. 95/9,
Coop. Res. Cent. for Catchment Hydrol., Melbourne, N. S. W., Australia,
1995.
Geiger, W. F., and H. R. Dorsch, Quantity-Quality Simulation (QQS): A
detailed continuous planning model for urban runoff control, in Model
Description, Testing and Application, vol. I, Rep. EPA-600/2-80-011,
U.S. Environ. Prot. Agency, Cincinnati, Ohio, 1980.
Hartigan, J. P., A. M. Lumb, J. T. Smullen, T. J. Grizzard, and P. A.
Alcivar, Calibration of urban nonpoint pollution loading models, in Verification of Mathematical and Physical Models in Hydraulic Engineering, Proceedings of the 26th Annual Hydraulics Division Specialty
Conference, pp. 363 372, Hydraul. Div., Am. Soc. of Civ. Eng., Reston,
Va., 1978.

SWC

5 - 10 VAZE AND CHIEW: EVALUATION OF URBAN STORM WATER QUALITY MODELS

Heaney, J. P., W. C. Huber, and S. J. Nix, Storm water management model:


Level IPreliminary screening procedures, Rep. EPA-600/2-76-275,
U.S. Environ. Prot. Agency, Cincinnati, Ohio, 1976.
Hemain, J. C., Statistically based modelling of urban runoff quality: State
of the art, in Urban Runoff Pollution, NATO ASI Ser., Ser. G, vol. 10,
edited by H. C. Torno et al., pp. 277 304, Springer-Verlag, New York,
1986.
Huber, W. C., and R. E. Dickinson, Storm water management model, in
Users Manual Version 4, Rep. EPA/600/3-88/001a, U.S. Environ. Prot.
Agency, Athens, Ga., 1988.
Ichikawa, A., Estimation of nonpoint source loading, in Urban Stormwater
Quality, Management, and Planning, Proceedings of the Second International Conference on Urban Storm Drainage, edited by B. C. Yen, pp.
585 586, Water Resour. Publ., Highlands Ranch, Colo., 1981.
Jewell, T. K., and D. D. Adrian, Development of improved stormwater
quality models, J. Environ. Eng. Div. Am. Soc. Civ. Eng., 107, 957
974, 1981.
Jewell, T. K., and D. D. Adrian, Statistical analysis to derive improved
stormwater quality models, J. Water Pollut. Control Fed., 54, 489
499, 1982.
Johanson, R. C., J. C. Imhoff, J. L. Kittle, and S. Donigan, Hydrological
Simulation Program-Fortran (HSPF): User manual for release 8.0, Rep.
EPA/600/3-84/066, U.S. Environ. Prot. Agency, Athens, Ga., 1984.
Lager, J. A., R. P. Shubinski, and L. W. Russell, Development of a simulation model for stormwater management, J. Water Pollut. Control Fed.,
43, 2424 2435, 1971.
Malmquist, P., Atmospheric fallout and street cleaningEffects of urban
stormwater and snow, Prog. Water Technol., 10, 495 505, 1978.
Nakamura, E., Factors affecting stormwater quality decay coefficient, in
Proceedings of the Third International Conference on Urban Storm
Drainage, vol. 3, edited by P. Balmer, P. Malmquist, and A. Sjoberg,
pp. 979 988, Goteborg, Sweden, 1984.
Nash, J. E., and J. V. Sutcliffe, River forecasting through conceptual models, 1. A discussion of principles, J. Hydrol., 10, 282 290, 1970.

Pravoshinsky, N. A., and P. D. Gatillo, Determination of the pollutional


effect of surface runoff, in Advances in Water Pollution Research, Proceedings of the Fourth International Conference, Prague, edited by S. H.
Jenkins, pp. 187 203, Pergamon, New York, 1969.
Price, R. K., and G. Mance, A suspended solids model for storm water
runoff, urban storm drainage, in Proceedings of the International Conference Held at the University of Southampton, edited by P. R. Helliwell,
pp. 546 555, Pentech, London, 1978.
RossRakesh, S., C. Gippel, F. Chiew, and P. Breen, Blackburn Lake discharge and water quality monitoring program: Data summary and interpretation, Tech. Rep. 99/13, Coop. Res. Cent. for Catchmernt Hydrol., 98
pp., Melbourne, N. S. W., Australia, 1999.
Sartor, J. D., and G. B. Boyd, Water pollution aspects of street surface
contaminants, Rep. EPA-R2-72-081, U.S. Environ. Prot. Agency,
Washington, D. C., 1972.
Vaze, J., and F. H. S. Chiew, Experimental study of pollutant accumulation
on an urban road surface, Urban Water, 4, 379 389, 2002.
Vaze, J., and F. H. S. Chiew, Nutrient loads associated with different sediment sizes in urban stormwater and surface pollutants, J. Environ. Eng,
in press, 2003a.
Vaze, J., and F. H. S. Chiew, Study of pollutant washoff from small
impervious experimental plots, Water Resour. Res., 39(6), 1160,
doi:10.1029/2002WR001786, 2003b.
Yaziz, M. I., H. Gunting, N. Sapari, and A. W. Ghazali, Variations in rainwater quality from roof catchments, Water Res., 23, 761 765, 1989.




F. H. S. Chiew, Cooperative Research Centre for Catchment Hydrology,


Department of Civil and Environmental Engineering, University of
Melbourne, Melbourne, Victoria 3010, Australia.
J. Vaze, Department of Infrastructure, Planning and Natural Resources,
Centre for Natural Resources, P.O. Box 189, Queanbeyan, N. S. W. 2620,
Australia. (jai.vaze@dipnr.nsw.gov.au)

Anda mungkin juga menyukai