You are on page 1of 21

Advances in Water Resources 30 (2007) 80–100

www.elsevier.com/locate/advwatres

A derived probability distribution approach to stormwater


quality modeling
Jieyun Chen *, Barry J. Adams
Department of Civil Engineering, University of Toronto, Toronto, Ont., Canada MIT 2R3

Received 2 January 2005; received in revised form 13 February 2006; accepted 20 February 2006
Available online 17 April 2006

Abstract

The closed-form analytical stormwater quality models are developed for simulating urban catchment pollutant buildup and washoff
processes. By integrating the rainfall–runoff transformation with pollutant buildup and washoff functions, stormwater quality measures,
such as the cumulative distribution functions (CDFs) of pollutant loads, the expected value of pollutant event mean concentrations
(EMCs) and the average annual pollutant load can be derived. This paper presents methodologies and major procedures for the devel-
opment of urban stormwater quality models based on derived probability distribution theory. In order to investigate the spatial variation
in model parameters and its impact on stormwater pollutant buildup and washoff processes as well as pollutant loads to receiving waters,
an extended form of the original rainfall–runoff transformation which is based on lumped runoff coefficient approach is proposed to dif-
ferentiate runoff generation mechanisms between the impervious and pervious areas of the catchment. In addition, as a contrast to the
aggregated pollutant buildup models formulated with a single lumped buildup parameter, the disaggregated form of the pollutant
buildup model is proposed by introducing a number of physically-based parameters associated with pollutant buildup and washoff pro-
cesses into the pollutant load models. The results from the case study indicate that analytical urban stormwater management model are
capable of providing results in good agreement with the field measurements, and can be employed as alternatives to continuous simu-
lation models in the evaluation of long-term stormwater quality measures.
Ó 2006 Elsevier Ltd. All rights reserved.

Keywords: Pollutant buildup and washoff; Stormwater quality control measures; Derived probability distribution; Analytical models; Continuous sim-
ulation models

1. Introduction models (including statistical and physically-based) and sto-


chastic models.
Different types of urban stormwater quality models are Statistical models are normally developed with well-
developed in the literature. These models with various established multivariate pattern recognition techniques
degrees of complexity are based on different modeling such as factor analysis, cluster analysis, correlation analy-
approaches. Huber [21,22] provided a review of these sis, etc. Other frequently used techniques for developing
stormwater quality models and noted that there were six stormwater quality models include linear, non-linear and
fully operational runoff quality models including the US stepwise regressions. Physically-based stormwater quality
EPA Stormwater Management Model (SWMM) [23], the models usually involve a number of simplified and inter-
most widely used continuous simulation model in North connected elements with some physical interpretation in
America. More generally, urban stormwater quality mod- attempting to describe the complex processes of pollutant
els fall into two broad categories, namely, deterministic generation and transport in urban catchments. Although
considered as an alternative to physically-based approach,
*
Corresponding author. Tel.: +1 416 299 3173. one apparent drawback of statistical modeling approach is
E-mail address: jy.chen@utoronto.ca (J. Chen). its inability to take into account the antecedent dry period

0309-1708/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advwatres.2006.02.006
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 81

which can be considered as one of the most important transformations and pollutant buildup function with was-
parameters in conventional physically-based models [20]. hoff model, stormwater quality measures such as the cumu-
In addition, the application of statistical models is some- lative distribution functions (CDFs) of pollutant loads, the
what restricted because of the onerous requirement of sup- expected value of pollutant event mean concentrations
porting data and the limited validation of results that (EMCs) and the average annual pollutant load to receiving
cannot be extrapolated or transferred [30]. The determina- waters are developed with derived probability distribution
tion of statistical relationships can be unduly influenced by theory. In order to evaluate the impacts of rainfall–runoff
high magnitude events with quite low frequencies of occur- transformation on pollutant loads, two different types of
rence [38]. In contrast to deterministic models that describe the rainfall–runoff transformations are proposed and inte-
independent variables and predict dependent variables with grated with the disaggregated form of the pollutant buildup
specific values of certainty, probabilistic models instead function in this study.
consider the possibility of occurrence of particular events
and determine the likelihood of their occurrence. For the 2. Analytical urban stormwater quality models
probabilistic models formulated with derived probability
distribution theory, major inputs to an urban drainage sys- 2.1. Derived probability distribution approach
tem similar to many other physical phenomena, can be typ-
ically considered as independent random variables with In contrast to other modeling approaches, the objective
specified known probability distributions. The probability of urban stormwater quality modeling with derived proba-
distributions of these independent random variables are bility distribution approach is essentially to transform the
then mathematically transformed into the probability dis- probability distributions of system inputs into the proba-
tributions of the dependent variables describing the system bility distributions of system outputs. This approach was
outputs. first outlined by Benjamin and Cornell [9] in the Civil Engi-
The derived probability distribution approach has been neering literature, where it was demonstrated that the
widely employed in flood frequency analysis [18,11,44]. probability distribution of a dependent random variable
Gottschalk et al. [17] showed that the probability distribu- might be derived from those related independent random
tion function of low flows could be derived by combining a variables using functional relationship between the depen-
regional rainfall model developed from long rainfall record dent and independent variables. For a given urban catch-
and a local model for runoff response. In addition to the ment, rainfall event volume (system input) is typically
modeling of stormwater quantity, the derived probability considered as one of the independent variables, and runoff
distribution approach has also been extended to include event volume, pollutant load at catchment outfall (typical
stormwater quality modeling as well [15,27,28,37,29]. system outputs) may be considered as dependent variables.
Among which Segarra-Garcia and Loganathan [37] derived Based on probability distribution theory, major procedures
a mathematical expression for the expected pollutant was- involved in the development of analytical stormwater qual-
hoff load by assuming that hydrologic variables such as ity models include analysis of probability distributions of
runoff event depth, duration and interevent time, were meteorological characteristics, conceptualization of the
exponentially distributed random variables. Loganathan rainfall–runoff transformation, formulation of pollutant
and Delleur [27] applied the derived distribution approach buildup and washoff models, development of stormwater
in evaluating the effect of urbanization on the frequency of quality measures such as the cumulative distribution func-
pollutant loads from storm sewer overflows. Akan [3] tion (CDF) of pollutant load, the expected value of pollu-
derived a mathematical frequency distribution for estima- tant event mean concentration (EMC), and the average
tion of stormwater runoff suspended solids load washed annual pollutant load to receiving waters.
off from the impervious surface of urban catchment. Li A starting point for the development of analytical
and Adams [26] developed analytical probabilistic models stormwater quality models with the derived probability dis-
with derived probability distribution theory to analyze tribution theory is to perform statistical analysis of long-
the runoff quantity/quality control performance of various term rainfall record for its characteristics (e.g., rainfall
combinations of storage/treatment systems. Xu et al. [45] event volume, duration and interevent time) in terms of
proposed a probabilistic stormwater quality model in the interevent time definition (IETD), which is defined as
which stormwater runoff quantity was estimated by the minimum interevent time period between two consecu-
STORM, and statistical estimation methods such as regres- tive pulses of rainfall. This basic concept of the IETD is
sion and maximum likelihood methods were employed for defined to discretize single storm event from a continuous
estimating the mean of censored concentrations and pollu- rainfall record. According to this definition, if the period
tant loads. between two pulses of rainfall is greater than the IETD,
This paper presents methodologies and major proce- the two rainfall pulses are categorized as belonging to dif-
dures for the development of urban stormwater quality ferent events; otherwise, the two rainfall pulses are catego-
model which consists of a number of functional compo- rized as belonging to the same event. Based on such a
nents such as rainfall–runoff transformation, pollutant definition, the rainfall event volume, duration and inter-
buildup and washoff, etc. By integrating rainfall–runoff event time can be extracted from historical rainfall record
82 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

and further analyzed with fitted probability distributions. where v is the rainfall event volume, vr is the runoff event
Thereafter, these probability distributions of the rainfall volume, / is a dimensionless runoff coefficient, and Sd is
characteristics are mathematically transformed by pollu- the depression storage. According to this formulation, the
tant washoff load model to create stormwater quality spatial variation of model parameters is not explicitly con-
measures. sidered. As a result, the analytical models developed for
runoff event volume are extremely compact and computa-
2.2. Schematic diagram of the analytical urban tionally efficient. However, the rainfall–runoff transforma-
stormwater quality model tion can be extended by dividing the entire catchment into
the impervious and pervious areas to consider heterogene-
The schematic diagram of the analytical stormwater ity of the catchment characteristics as follows:
quality model is shown in Fig. 1. As illustrated in the 8
diagram, the analytical model consists of a number of func- < 0;
> v 6 S di
tional components such as the rainfall–runoff transforma- vr ¼ hðv  S di Þ; S di < v 6 S dp
>
:
tion, pollutant buildup and washoff components, etc. ½h þ /dp ð1  hÞv  hS di  /dp S dp ð1  hÞ; v > S dp
Among which the key component is the rainfall–runoff ð2Þ
transformation, which characterizes runoff generation
mechanism in the catchment. While the three major com- where h is the fraction of the impervious area of the catch-
ponents are conceptualized with different methodological ment, Sdi is the impervious area depression storage, Sdp is
considerations, the pollutant washoff model is essentially the pervious area depression storage, and /dp is the pervi-
an integration of the rainfall–runoff transformation and ous area runoff coefficient. Based on the extended form of
pollutant buildup model. the rainfall–runoff transformation, the total runoff at the
The rainfall–runoff transformation component depicts catchment outlet is assumed to be a combination of the
runoff generation mechanism in terms of rainfall character- area-weighted runoff from the impervious and pervious
istics, based on which the probability density function portions of the catchment.
(PDF) of runoff event volume can be derived from the The pollutant buildup component can be formulated
PDFs of rainfall event volume and interevent time. More- with lumped parameters without dividing the entire catch-
over, based on the PDF of runoff event volume, the ment into the pervious and impervious areas and without
expected value of runoff event volume and the average considering different sources of pollutant deposition.
annual runoff volume can be derived as well. As one of According to such a formulation (e.g., exponential func-
the key elements, the rainfall–runoff transformation formu- tion), no physics is virtually involved in the lumped pollu-
lated in this component is based on different hydrologic tant buildup function depicted by a single parameter. To
considerations. As well known, the rainfall–runoff transfor- differentiate pollutant buildup processes in the impervious
mation may be formulated in a lumped form with two and pervious areas of the catchment in terms of different
lumped parameters for simplicity [42], in which the imper- sources of pollutant, and to enhance the physics of the
vious and pervious areas of the catchment are aggregated model as well, pollutant buildup model in the lumped form
together as follows: can be modified by incorporating more physically-based
 parameters.
0; v 6 Sd The component of pollutant washoff simulates the
vr ¼ ð1Þ
/ðv  S d Þ; v > S d removal of the accumulated pollutant on the catchment

Pollutant Buildup CDF of event


Atmospheric
Fallout pollutant load

Traffic Erosion
Impervious Pervious Pollutant Average annual
area area load model pollutant load
Litter Wind
effect

Average event
Pollutant PDF of runoff pollutant load
washoff event volume
PDFs of Rainfall
Characteristics Expected value
-Volume Rainfall-runoff of runoff
-Duration transformation event volume
-Interevent time Average
pollutant EMC

Fig. 1. Analytical urban stormwater management model.


J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 83

surface. The washoff of pollutants from the impervious sur- washoff models with various forms may be obtained by
face of the catchment is determined by a number of factors substituting different types of the rainfall–runoff transfor-
such as the types and conditions of the street surface, rain- mation and pollutant buildup model into the lumped was-
fall intensity and volume, particle size, street cleaning and hoff model.
traffic densities [16]. In contrast to the formulation of pol- Among different components as shown in Fig. 2, the
lutant accumulation process, in which different buildup rainfall–runoff transformation is the fundamental compo-
functions with various shapes may be employed, the forms nent, which not only determines the degree of the model
of washoff models or functions reported in literature are complexity, but also characterizes the forms of the pollu-
limited. The commonly used exponential washoff function tant load model. According to the rainfall–runoff transfor-
is employed to integrate with the buildup model and the mation, the PDF of runoff event volume can be employed
rainfall–runoff transformation. More detailed pollutant as a basis to derive the expected value of runoff event

250 900
EMC of TSS (mg/L)

EMC of TS (mg/L)
200 720
150 540
100 360
50 180
0 0
0 20 40 60 0 20 40 60
Rainfall Volume (mm) Rainfall Volume (mm)

1 1.5
EMC of TKN (mg/L)
EMC of TP (mg/L)

0.8 1.2
0.6 0.9
0.4 0.6
0.2 0.3
0 0
0 20 40 60 0 20 40 60
Rainfall Volume (mm) Rainfall Volume (mm)

50 10
EMC of BOD5 (mg/L)
EMC of COD (mg/L)

40 8
30 6
20 4
10 2
0 0
0 20 40 60 0 20 40 60
Rainfall Volume (mm) Rainfall Volume (mm)

0.15 10
EMC of Cu (mg/L)

EMC of Fe (mg/L)

0.12 8
0.09 6
0.06 4
0.03 2
0 0
0 20 40 60 0 20 40 60
Rainfall Volume (mm) Rainfall Volume (mm)

0.15
EMC of Zn (mg/L)

0.12
0.09
0.06
0.03
0
0 20 40 60
Rainfall Volume (mm)

Fig. 2. Scatter plot of the rainfall event volumes and the EMC of different pollutants.
84 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

volume, the CDF of event pollutant loads, the average cesses between the impervious and pervious areas of the
annual pollutant load as well as the average pollutant catchment, and the efficiency of source control for pollu-
EMCs. While rainfall–runoff models can be formulated tants cannot be explicitly evaluated since the buildup
with various degrees of complexity based on different meth- model is not specifically linked to major pollutant accumu-
odological considerations, the original rainfall–runoff lation and removal processes such as dust fall, street weep-
transformation was employed in the well-known STORM ing, etc. To reflect the reality, pollutant buildup process
model [42] despite its simplicity. As a comparison to may be better formulated to consider various sources of
another widely used NRCS Curve Number method, in pollutant deposition taking into account pollutant translo-
which the form of the rainfall–runoff model involved is cation from the impervious area to pervious area of the
not depicted as linear relationship as that employed in catchment as well [33].
STORM, it can be extremely difficult to integrate the The lumped form of buildup function in many previous
non-linear rainfall–runoff model with pollutant buildup studies was based on the assumption that the accumulated
and washoff models to derive closed-form analytical mod- pollutant on catchment surface was completely washed off
els for stormwater control measures. during the last runoff event (e.g., there is no residual
amount of pollutant available after the previous runoff
3. Pollutant buildup and washoff models washoff event). Alley and Smith [5] stressed that in deter-
mining pollutant accumulation rates, attention should be
Pollutant buildup and washoff is a continuous process given to the effects of the residual amount of pollutant
occurring over both dry and wet weather periods. Pollu- remaining after the previous period of storm runoff or
tants from various sources may enter stormwater runoff street sweeping. Actually, the amount of pollutant mass
via different pathways including atmospheric fallout, auto- accumulated on catchment surface may include two parts;
mobile emissions and corrosion, land surface erosion, one part is the pollutant mass buildup by dust and litter
pavement degradation, vegetation and leaf litter, etc. Dur- since the last rainfall event, and the other part is the resid-
ing the wet weather period, rainfall not only washes off pol- ual pollutant mass not washed off by the previous event.
lutants that have built up during dry weather period, but This assumption was adopted by a number of other studies
also deposits its own pollutants [24]. Although empirical in considering urban catchment pollutant buildup process
relationships may provide site-specific information on the [36,32,46,34]. According to the study by Osuch-Pajdzinska
buildup and washoff of pollutants, the simulation of pollu- and Zawilski [34], the change of pollutant accumulation
tant buildup and washoff processes in the catchment is usu- rate can be expressed as follows:
ally based on appropriately defined mathematical functions dM b
(e.g., exponential function). ¼ ð1  hÞmd b1 þ hmw b2 g  k b M b ð3Þ
db
3.1. Pollutant buildup model where Mb is the amount of pollutant per unit area on the
catchment surface, in kg/m2; h is the fraction of the imper-
Pollutant accumulation on an urban catchment may be vious area of the catchment, md is a constant rate of pollu-
represented by different types of buildup functions. Among tant deposition (dust fall) in kg/m2 h, mw is the quantity of
which include linear, power, exponential, and Michaelis– street sweepings accumulated on the streets, in kg/m2 h, g is
Menton functions as suggested by Ammon [6]. Huber the street sweeping effectiveness parameter, kb is a constant
and Dickinson [23] provided a general description of the pollutant removal rate, in h1, b is the time elapsed since
four types of buildup functions with plotted curves to show the last rainfall (i.e., interevent time) in h, while b1 de-
the differences in their shapes. While linear buildup func- scribes the conversion of the mass of particulate matter
tion is considered as a special case of the power function, into a parameter of a given type of pollutant, and b2 de-
and is employed in the earlier versions of SWMM, the lin- scribes the conversion of the mass of sweepings into a
ear buildup concept is not always adequate in depicting a parameter of a given type of pollutant. By denoting the
more realistic pollutant accumulation process in urban residual pollutant mass not washed off by the previous run-
catchments. The non-linear pollutant buildup was first ver- off event as M0, integrating Eq. (3) yields
ified from the field data collected by Sartor and Boyd [35], ð1  hÞmd b1 þ hmw b2 g
who were also among the first to suggest an exponential Mb ¼ ð1  ekb b Þ þ M 0 ekb b ð4Þ
kb
relationship between the amount of solids available on
the surface and the duration of the antecedent dry weather For this disaggregated form of pollutant buildup function,
period. Thereafter, this concept has been widely employed if the residual amount of pollutant (M0) is assumed to be
in many other studies [8,12,14,10,23,19,4,5]. zero, the second term in the equation can be ignored.
Despite the widespread application of exponential
buildup function in depicting pollutant buildup process, 3.2. Pollutant washoff model
the exponential buildup function is usually formulated in
the aggregated form with lumped parameters. As a result, Among the earliest studies on pollutant washoff process,
no distinction is made regarding the pollutant buildup pro- the parameter of pollutant washoff rate is usually consid-
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 85

ered strongly related to rainfall intensity. In urban storm- concentration is usually a preferred measure of stormwater
water quality modeling, the widely used exponential was- quality instead of instantaneous concentration since pollu-
hoff formulation suggested by Sartor and Boyd [35] states tant EMC is relatively less dependent on runoff volume.
that the pollutant washoff rate from the impervious surface The event mean concentration (flow-weighted mean)
in a given time interval is proportional to the runoff inten- should be a better measure of the pollution potential of
sity and the mass of pollutant remaining on the ground sur- urban stormwater runoff than an arithmetic mean, which
face. Akan [2] proposed a physically-based mathematical gives too much weight to low concentration samples usu-
model for the process of pollutant washoff, in which the ally associated with low flow rate sampling times [25].
pollutant washoff rate was assumed to be proportional to Charbeneau and Barrett [12] suggested that it was appro-
the bottom sheer stress of the overland flow and the distri- priate to use an event mean concentration in evaluating
bution density of the pollutant. In another study, Akan [3] the effects of stormwater runoff on receiving waters. The
suggested that the effect of the time pattern of rainfall study conducted by the US Nationwide Urban Runoff Pro-
intensity on pollutant washoff was negligible. Stormwater gram [41] found that there was no observable correlation
quality data collected from Champaign, IL, during the between pollutant event mean concentration and runoff
Nationwide Urban Runoff Program (NURP) concluded volume; also, the geographic location, the land use activi-
that rainfall characteristics such as total rainfall volume ties and population density did not significantly influence
and duration had the greatest influence on the constituent pollutant mean concentrations.
concentrations and washoff loads [40]. Since the runoff vol-
ume is recognized as a major driving force in the mobiliza- 4.1. Type 1 pollutant load model
tion, transport and deposition of pollutants in urban
pollution dynamics, pollutant washoff model may be better 4.1.1. Cumulative distribution function
described by runoff volume. Barbe et al. [7] employed a lin- Type 1 pollutant load model is formulated by substitut-
ear pollutant buildup function of the antecedent dry ing the disaggregated form of pollutant buildup model and
weather period that was linked with a pollutant washoff the original rainfall–runoff transformation into the washoff
model represented by a power function of the stormwater model, based on which the CDF (or PDF) of pollutant
runoff volume. Deletic et al. [14] found that an event aggre- event load, the expected value of pollutant EMC can be
gated load of suspended solids was strongly related to the derived. Mathematically, the Type 1 pollutant load model
event aggregated runoff volume. A number of other studies is expressed as
showed that the pollutant washoff might be better esti- 8
mated by using runoff volume rather than flow rate >
> 0; v 6 S d ; b > 0
<h i
[8,12,7,39,43]. l¼ M 0 ekb b þ ð1hÞmd bk1bþhmw b2 g ð1  ekb b Þ ð7Þ
In this study, it is assumed that the rate of pollutant >
>
:
washoff from the catchment surface is proportional to the ½1  ekw /ðvS d Þ ; v > S d ; b > 0
amount of pollutant buildup on the catchment surface
In the equation, event pollutant load, l, is a function of two
and is directly related to the volume of runoff. According
random variables; i.e., rainfall volume, v, and interevent
to this assumption, the mathematical expression is written
time, b. Other parameters in the equation, such as the pol-
as
lutant buildup rate, kb, washoff coefficient, kw, depression
dM b storage Sd, and dust fall, md, may be treated as spatially
¼ k wvr M b ð5Þ
db averaged constants. Physically, the phenomena of pollu-
where Mb is the amount of pollutant per unit area on tant washoff only take place when runoff events occur; in
catchment surface, in kg/m2, vr is the average runoff rate, other words, there will be no pollutant load if the rainfall
in mm/h, kw is the decay or washoff coefficient, in mm1. event volumes are not sufficient to generate runoff. Accord-
Performing integration yields ing to Eq. (7), runoff will not occur and there will be no
pollutant load if the rainfall volume is less than the depres-
M w ¼ M b ð1  ekwvr b Þ ¼ M b ð1  ekw vr Þ ð6Þ sion storage, Sd. Therefore, there is an impulse probability
of no pollutant washoff load when no runoff occurs.
An integrated form of the washoff model may be obtained
If the PDF and CDF of pollutant event load (l) are
by substituting the rainfall–runoff transformation (either
denoted as fL(b, v) and FL(b, v), respectively, where v and
Eq. (1) or (2)) into Eq. (6), while replacing Mb in the equa-
b are assumed statistically independent random variables,
tion by the disaggregated form of buildup function (Eq.
the CDF of washoff load, FL(b, v) or FL(l), is expressed as
(4)).
Z Z Z Z
F L ðlÞ ¼ F L ðb; vÞ ¼ fL ðb; vÞ dbdv ¼ wfewb efv dbdv
4. Development of stormwater quality control measures RL RL

ð8Þ
Stormwater quality measures include the CDF of pollu-
tant event load, the expected value of pollutant load and where the marginal PDFs of v (i.e., fefv), and b (i.e.,
pollutant EMC. In the literature, pollutant event mean wewb) are exponential functions, in which f equals inverse
86 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

of the average rainfall event volume, while w equals inverse From Eq. (7), the impulse probability of no washoff load is
of the average rainfall interevent time. equal to the probability that the rainfall event volume is
To simplify Eq. (8), the integral limits for FL(b, v) must not greater than the depression storage, Sd. Under such
be specified and the region of integration is defined by condition, the impulse probability of no washoff loads is gi-
Eq. (7). In particular, the region of integration should be ven by
bounded by the limits of the two independent random vari- Z Sd
ables (v and b). In order to determine the region for inte- F L ð0Þ ¼ Pr½V 6 S d  ¼ fefv dv ¼ 1  efS d ð15Þ
gration, solving Eq. (7) for b yields v¼0

1 Theoretically, the random variables b and v may take any


b¼
kb value including negative values for any positive value of
  l, and the CDF of pollutant load (Eq. (13)) should be appli-
½ð1  hÞmd b1 þ hmw b2 g½1  ekw /ðvS d Þ   k b l
 ln cable to the entire range of the random variables. However,
½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ½1  ekw /ðvS d Þ 
in practice, according to the physical interpretation, the
ð9Þ lower limits for b and v may be better set to zero, and
From the above expression, the limiting condition for the upper bounds for the two random variables may be
interevent time corresponding to an infinite rainfall event set to finite value instead of infinite. Such approximations
volume (v ! 1) is given by are made mainly for the convenience in conducting numer-
  ical evaluation of the CDF of pollutant load, and will
1 ð1  hÞmd b1 þ hmw b2 g  k b l
b ¼  ln ð10Þ not cause significant loss in accuracy considering that
kb ð1  hÞmd b1 þ hmw b2 g  k b M 0 the appropriate selection of limiting values actually accom-
Similarly, in order to determine the characteristic value of modates all the practical ranges of the two random
rainfall event volume as b approaches infinite, Eq. (7) may variables.
be also rearranged by expressing v as a function of l and b To facilitate numerical evaluation of the CDF of pollu-
as follows: tant event load, let Bm represent the maximum practical
1 value of the interevent time, and the corresponding rainfall
v ¼ Sd  event volume, V0, is obtained as
/k w
 
kbl 1
 ln 1  V 0 ¼ Sd 
ð1  ekb b Þ½ð1  hÞmd b1 þ hmw b2 g þ k b M 0 ekb b /k w
 
ð11Þ kbl
 ln 1 
ð1  ekb Bm Þ½ð1  hÞmd b1 þ hmw b2 g þ k b M 0 ekb Bm
The limiting condition for rainfall event volume corre-
ð16Þ
sponding to an infinite rainfall interevent time (b ! 1) is
obtained as Likewise, let Vm denote the maximum practical value of the
 
1 kbl rainfall event volume, then the corresponding rainfall dura-
v ¼ S d  ln 1  ð12Þ tion, B0, is obtained as
/k w ð1  hÞmd b1 þ hmw b2 g
Both the characteristic values of the rainfall event volume, 1
B0 ¼ 
v*, and interevent time, b*, are linear functions of pollutant kb
event load and can be represented by two straight lines.  
½ð1  hÞmd b1 þ hmw b2 g½1  ekw /ðV m S d Þ   k b l
According to the region of integration for FL(b, v), Eq.  ln
½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ½1  ekw /ðV m S d Þ 
(8) can be expressed more specifically as follows:
Z Z ð17Þ
F L ðlÞ ¼ F L ðb; vÞ ¼ wfewb efv db dv With these more specifically defined values for Bm, Vm,
RL
Z v¼v Z 1 B0 and V0, which actually form the region of integration
¼ wfewb efv db dv for the CDF of pollutant load, Eq. (13) can be rewritten
v¼0 b¼0 as
Z 1 Z gðl;vÞ
Z V 0 Z Bm
þ wfewb efv db dv ð13Þ
v¼v b¼0 F L ðlÞ ¼ F L ðb; vÞ ¼ wfewb efv db dv
v¼0 b¼0
where Z Vm Z gðl;vÞ
þ wfewb efv db dv ð18Þ
1 v¼V 0 b¼0
gðl; vÞ ¼ 
kb
  In the above equation, since the integral limits in the sec-
½ð1  hÞmd b1 þ hmw b2 g½1  ekw /ðvS d Þ   k b l ond term contain a complex function, g(l, v), analytical
 ln
½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ½1  ekw /ðvS d Þ  solution to the CDF of pollutant load cannot be explicitly
ð14Þ derived. Nevertheless, performing integration yields
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 87

Z Vm
surface after the previous washoff event, M0 can be as-
F L ðlÞ ¼ 1  fefv
V0 sumed to be zero, and the CDF of pollutant load becomes
 kw Z Vm
½ð1  hÞmd b1 þ hmw b2 g½1  ekw /ðvS d Þ   k b l b
F L ðlÞ ¼ 1  fefv
 dv
½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ½1  ekw /ðvS d Þ  V0
fV m wBm fV 0  kw
e e ð1  e Þ ð19Þ kbl b
 1 dv
Based on Eq. (19), given an event load, to obtain a specific ½ð1  hÞmd b1 þ hmw b2 g½1  ekw /ðV m S d Þ 
value for the CDF of pollutant event load, numerical eval-  efV m  ewBm ð1  efV 0 Þ ð25Þ
uation is necessary.
where V0 and B0 are also simplified as follows:
4.1.2. Expected value of pollutant event load and 1
event mean concentration V 0 ¼S d 
/k w
The general mathematical expression for the expected  
kbl
value of pollutant event load is given by  ln 1  k
ð26Þ
ð1  e b Bm Þ½ð1  hÞmd b1 þ hmw b2 g
Z V m Z Bm
1
EðlÞ ¼ l  fL ðb; vÞ db dv ð20Þ B0 ¼ 
v¼0 b¼0 kb
 
kbl
Substituting the joint PDF of v and b (a product of the  ln 1 
½ð1  hÞmd b1 þ hmw b2 g½1  ekw /ðV m S d Þ 
marginal PDFs of v and b) and event pollutant load (Eq.
(7)) into the above equation, performing integration yields ð27Þ
Z V m Z Bm  
and the expected value of pollutant load is simplified as
ð1  hÞmd b1 þ hmw b2 g  
EðlÞ ¼ M 0 ekb b þ 1  ekb b   
v¼0 b¼0 kb fekw /S d ð1  hÞmd b1 þ hmw b2 g
 l ¼ EðlÞ ¼ 1  ð28Þ
 1  ek w /ðvS d Þ wfewb efv dbdv f þ kw/ w þ kb
 
fV m fek w /S d ðfþk w /ÞV m Therefore, the expected value of event mean concentration
¼ 1e  ð1  e Þ
f þ kw/ (Eq. (24)) can also be simplified as
 
ðC m  M 0 Þw h ih i
 C m ð1  ewBm Þ  ð1  eðwþkb ÞBm Þ ð21Þ k w /S
w þ kb 1  fefþkw /d ð1hÞmwþk
d b1 þhmw b2 g
b
EðEMCÞ ¼ / fS d
ð29Þ
where C m ¼ ð1hÞmd bk1bþhmw b2 g. f
e
As Bm and Vm approach their limiting conditions Eqs. (22) and (24), and their simplified forms; i.e., Eqs. (28)
(Bm ! 1, Vm ! 1), Eq. (21) can be further simplified as and (29) are closed-form analytical models.
l ¼ EðlÞ
   4.2. Type 2 pollutant load model
fekw /S d ð1  hÞmd b1 þ hmw b2 g  wM 0
¼ 1 ð22Þ
f þ kw/ w þ kb 4.2.1. Cumulative distribution function
Based on the original rainfall–runoff transformation, the Following the similar procedures presented in the deri-
expected value of runoff event volume is derived as [1] vation of the Type 1 load model, a general mathematical
expression for the Type 2 load model is obtained by substi-
/ fS d tuting Eq. (4) into Eq. (6), that is
E½V r  ¼ e ð23Þ
f  
ð1  hÞmd b1 þ hmw b2 g
Therefore, the expected value of pollutant event mean con- l ¼ M 0 ekb b þ ð1  ekb b Þ ½1  ekw vr 
kb
centration per rainfall event can be expressed as
h ih i ð30Þ
k w /S
1  fefþkw /d ð1hÞmd b1wþk
þhmw b2 gwM 0
b By substituting the extended form of the rainfall–runoff
EðEMCÞ ¼ / fS d
ð24Þ
e transformation into the above equation, a more specific
f
expression for pollutant event load is obtained as
Note that the derivation of the expected value of pollutant 8
load and event mean concentration are based on the >
> 0; v 6 S di ; b > 0
>h
>
> k b b ð1hÞmd b1 þhmw b2 g k b b
i
assumption that there is always a residual or initial amount >
> M 0 e þ ð1  e Þ
>
< kb
of pollutant available on the catchment surface after the k w hðvS di Þ
previous runoff washoff or street sweeping event. However, l¼ ½1  e ; S di < v 6 S dp ; b > 0 ð31Þ
>
> h i
if the amount of pollutant accumulated on the catchment >
> ð1hÞm b þhm b g
>
> M 0 ekb b þ d 1 w 2
ð1  ekb b Þ
surface is completely washed off by the last runoff event, >
>
kb
:
there is no residual amount of pollutant on the catchment ½1  ekw /h ðvS dh Þ ; v > S dp ; b > 0
88 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

In order to determine the region of integration for the CDF To more specifically define the region of integration for the
of pollutant load, characteristic values for the rainfall event CDF of pollutant event load, let Bm represent the maxi-
volume and interevent time must be specified. For this pur- mum practical value of the interevent time, and Vm repre-
pose, Eq. (30) may be rearranged by expressing b as a func- sent the maximum practical value of the runoff event
tion of l and vr as follows: volume. Thus, the value of runoff event volume, Vr0, corre-
  sponding to Bm, and the value of interevent time, B0, cor-
1 ½ð1  hÞmd b1 þ hmw b2 gð1  ekw vr Þ  k b l responding to Vm, are obtained as follows:
b ¼  ln
kb ½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ð1  ekw vr Þ
1
ð32Þ V r0 ¼ 
kw
 
kb l
The limiting condition for the interevent time correspond-  ln 1 
ing to an infinite runoff volume (vr ! 1) is given by k b M 0 ek b Bm þ ð1  ekb Bm Þ½ð1  hÞmd b1 þ hmw b2 g
ð38Þ
 
1 ð1  hÞmd b1 þ hmw b2 g  k b l 1
b0 ¼  ln ð33Þ B0 ¼ 
kb ð1  hÞmd b1 þ hmw b2 g  k b M 0 kb
 
½ð1  hÞmd b1 þ hmw b2 gð1  ekw vm Þ  k b l
 ln ð39Þ
Similarly, to determine the characteristic value for runoff ½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ð1  ekw vm Þ
event volume, Eq. (30) may be rearranged by expressing
In the case of Vr0 6 h(Sdp  Sdi), let vr ¼ hðS dp  S di Þ, and
vr as a function of l and b as
define the function g(l, vr) as follows:
1 1
vr ¼  gðl; vr Þ ¼ 
kw kb
   
kb l ½ð1  hÞmd b1 þ hmw b2 gð1  ekw vr Þ  k b l
 ln 1 
k b M 0 ekb b þ ð1  ek b b Þ½ð1  hÞmd b1 þ hmw b2 g  ln
½ð1  hÞmd b1 þ hmw b2 g  k b M 0 ð1  ekw vr Þ
ð34Þ
ð40Þ
The limiting condition for the rainfall event volume corre- The cumulative distribution function of pollutant load for
sponding to an infinite rainfall interevent time (b ! 1) is V r0 6 vr is expressed as
given by Z vr0 Z Bm
  F L ðb; vr Þ ¼ F L ð0Þ þ fB;V r ðb; vr Þ db dvr
1 kbl
v0r ¼  ln 1  ð35Þ vr ¼0 b¼0
kw ð1  hÞmd b1 þ hmw b2 g Z vr Z gðl;vr Þ
þ fB;V r ðb; vr Þ db dvr
Actually, Eqs. (33) and (35) represent two asymptotic lines vr ¼V r0 b¼0
of the function embodied by Eq. (30). From Eq. (31), it can Z Vm Z gðl;vr Þ

be seen that when the rainfall event volume is less than the þ fB;V r ðb; vr Þ db dvr ð41Þ
vr ¼vr b¼0
impervious area depression storage, Sdi, no runoff will oc-
cur. Under such condition, the impulse probability of no Assuming that runoff event volume and rainfall interevent
pollutant washoff load is equal to the probability that a gi- time are independent random variables, given the PDF of
ven rainfall event volume does not exceed the impervious runoff event volume, the joint PDF of runoff event volume
area depression storage of the catchment. This impulse and rainfall interevent time is obtained as
probability is given by 8
> 1  efS di ; vr ¼ 0
Z S di >
>
< wb f fðvhr þS di Þ
we  he ; 0 < vr 6 vr ; b > 0
F L ð0Þ ¼ Pr½V 6 S di  ¼ fefv dv ¼ 1  efS di ð36Þ fB;V r ðb; vr Þ ¼

>
>
v¼0 >
: wb f f /vrh þS dh
we  /h e ; vr > vr ; b > 0
To obtain the cumulative distribution function of pollutant
load, instead of using the pollutant load model described ð42Þ
by the rainfall characteristics (Eq. (31)), it is desirable to Substituting Eq. (42) into Eq. (41) gives
utilize the PDF of runoff event volume under two condi- Z vr0 Z Bm
tions; i.e., Vr 6 h(Sdp  Sdi) and Vr > h(Sdp  Sdi) as fol- f
wewb  efð h þS di Þ dbdvr
vr
F L ðb; vr Þ ¼
lows [13]: vr ¼0 b¼0 h
8 Z vr Z gðl;vr Þ
f
wewb  efð h þS di Þ dbdvr
vr
>
> 1  efS di ; vr ¼ 0 þ
>
> h
>
< f fð h þS di Þ
vr
vr ¼V r0 b¼0


e ; 0 < vr 6 hðS dp  S di Þ Z V m Z gðl;vr Þ
fV r ðvr Þ ¼ h ð37Þ f f /vr þS dh
>
>
þ wewb  e h
dbdvr þ F L ð0Þ
>
> vr ¼vr /h
> vr
: f ef /h þS dh ; v > hðS  S Þ
b¼0

/h r dp di ð43Þ
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 89

Due to the presence of g(l, vr) in the integral limits, analyt- substituting the joint PDF of runoff event volume and rain-
ical solution to Eq. (43) cannot be derived. However, the fall interevent time into the above equation gives
CDF of pollutant load can be further simplified as follows: Z vr Z  Bm 
ð1  hÞmd b1 þ hmw b2 g
EðlÞ ¼ M 0 ek b b þ ð1  ekb b Þ
F L ðlÞ ¼ F L ðb; vr Þ vr ¼0 b¼0 kb


f f
 ½1  ekw vr   wewb  efð h þS di Þ dbdvr
vr
¼ 1  efS di þ efS di ð1  ewBm Þ 1  ehV r0
h

f
f  Z V m Z Bm  
f   v fV ð1  hÞmd b1 þ hmw b2 g
þ efS di ehV r0  ehvr þ efS dh e /h r  e /h m þ M 0e k b b
þ k b b
ð1  e Þ
Z vr vr ¼vr b¼0 kb
f f


 efS di ehvr  ewgðl;vr Þ dvr f f vr þS dh
h vr ¼V r0  ½1  ekw vr   wewb  e /h dbdvr ð48Þ
Z /h
f fS dh V m /f vr wgðl;vr Þ
 e e h e dvr ; V r0 6 hðS dp  S di Þ Performing integration yields
/h vr ¼vr
  
ð44Þ f

1  eðkw þhÞvr
f  f 
EðlÞ ¼ efS di 1  ehvr 
f þ hk w
In the case Vr0 > h(Sdp  Sdi) or V r0 P vr ,
the mathemati-   
cal expression for the CDF of pollutant load is given as f 
 v fekw vr fV fekw V m
þ efSdh e /h r 1   e /h m 1 
follows: f þ /h k w f þ /h k w

Z vr Z Bm ð1  hÞmd b1 þ hmw b2 g
f  ð1  ewBm Þ
wewb  efð h þS di Þ db dvr
vr
F L ðb; vr Þ ¼ kb
vr ¼0 b¼0 h 

½ð1  hÞmd b1 þ hmw b2 g  k b M 0 w
Z V r0 Z Bm  ð1  eðwþkb ÞBm Þ
wb f f /vrh þS dh ðw þ k b Þk b
þ we  e dbdvr
vr ¼vr b¼0 /h ð49Þ


Z V m Z gðl;vr Þ
wb f f /vrh þS dh Compared with the previous equation for estimating the
þ we  e dbdvr þ F L ð0Þ
vr ¼V r0 b¼0 /h expected value of pollutant event load (Eq. (22)) derived
ð45Þ from the original rainfall–runoff transformation, the above
equation shows increased degree of complexity. Neverthe-
the integral limits in Eq. (45) also contain a complex func- less, in contrast to the mathematical expression for the
tion g(l, vr), it is difficult to perform a complete integration CDF of pollutant event load, Eq. (49) is a closed-form ana-
of the equation. However, Eq. (45) can be further simpli- lytical model, and is more efficient in application once the
fied as formula is created in spreadsheet format. As Bm and Vm
F L ðlÞ ¼ F L ðb; vr Þ ¼ 1  efSdi þ ð1  ewBm Þ approach their limiting conditions (Bm ! 1, Vm ! 1),
h

f  i the equation is simplified as
f   v fV
 efS di 1  ehvr þ efS dh e /h r  e /h r0

f L ¼ EðlÞ
 V fV   
þ efS dh e /h r0  e /h m

f
1  e ð k w þhÞ v r
f  f 
fS di
Z Vm ¼ e 1  e hv r 
f fv f þ hk w
 efS dh e /h r  ewgðl;vr Þ dvr ; V r > hðS dp  S di Þ "
#)
/h V r0 f
f 
 v f  k w þ vr
ð46Þ þefS dh e /h r  e /h

f þ /h k w
 
As with the Type 1 pollutant load model, there is no expli- ð1  hÞmd b1 þ hmw b2 g  wM 0
cit analytical solution for the CDF of the pollutant load  ð50Þ
w þ kb
based on the Type 2 load model. In both cases, the CDFs
of pollutant load must be evaluated numerically. To derive the expected value of pollutant EMC, it is neces-
sary to know the average runoff event volume. The ex-
4.2.2. Expected values of pollutant event load and pected value of runoff event volume based on the
event mean concentration extended form of the rainfall–runoff transformation is ex-
In terms of the Type 2 pollutant load model, the math- presses as
ematical expression for the expected value of washoff load Z ðS dp S di Þh
per rainfall event can be expressed as follows: EðV r Þ ¼ 0  Pr½V r ¼ 0 þ vr  fV r ðvr Þ dvr
Z vr Z Bm Z 1
vr ¼0

EðlÞ ¼ 0  F L ð0Þ þ l  fB;V r ðb; vr Þ db dvr þ vr  fV r ðvr Þ dvr ð51Þ


vr ¼0 b¼0 vr ¼ðS dp S di Þh
Z Vm Z Bm
þ l  fB;V r ðb; vr Þ db dvr ð47Þ substituting Eq. (37) into the above equation and perform-
vr ¼vr b¼0 ing integration yields
90 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

h /dp fS dp [31], the catchment is composed of primarily residential


Eðvr Þ ¼ efS di þ ð1  hÞ e ð52Þ
f f land use with the majority of the site made up of table-
lands. The vegetation cover through almost all the site is
Therefore, the expected value of pollutant EMC per rain- characterized by uniform, well-manicured lawn of horticul-
fall event is obtained as tural sod grasses. With a moderately high clay content, the

(
)
h
i f S dh þ/r
v h i h i
fS di hfvr f ðk w þhfÞvr fek w vr ð1hÞmd b1 þhmw b2 gwM 0
e 1e  fþhk w
1e þe h
1  fþ/h kw wþk b
EðEMCÞ ¼ /dp
ð53Þ
h fS di
f
e þ ð1  hÞ f
efS dp

Eqs. (50) and (53) may be also simplified by assuming type of soils is generally classified as imperfectly drained
that the accumulated pollutant mass on the catchment sur- with slopes typically ranging from 0% to 3%. The percent-
face is completely washed off by the previous runoff event, age of the imperviousness of the catchment is estimated to
and the estimation of the pollutant mass accumulated on be 31%. The mean annual rainfall volume (exclude snow) is
catchment surface only needs to take into account the approximately 665 mm.
amount of pollutant deposited by dust and litter since the There is no existing rainfall gauge in the catchment. As a
last runoff event. Under such circumstance, the expected result, long-term rainfall data is taken from the nearby
value of pollutant load (Eq. (50)) is simplified as meteorological station (Toronto Bloor Street) and the rain-
L ¼ EðlÞ fall record available from 1960 to 1994 is used to perform
(  long-term continuous simulation in this study. From the
f

1  e ð k w þhÞ v r
f 
f 
¼ efS di 1  ehvr  long-term rainfall record, the mean monthly rainfall vol-
f þ hk w ume during the summer/fall period (May–November) is
"
#) estimated to be 68 mm in depth. The evaporation rates
f
f 
 v f  k w þ vr
used to perform continuous simulation are based on pan
þ efS dh e /h r  e /h

f þ /h k w evaporation rates obtained from the Richmond Hill, part


  of the Greater Toronto Area. The maximum mean
ð1  hÞmd b1 þ hmw b2 g0
 ð54Þ monthly evaporation rate normally occurs in July with a
w þ kb
value of approximately 140 mm/month. For the analytical
and the expected value of pollutant EMC (Eq. (53)) is sim- stormwater quality models, the characteristics of rainfall
plified as (e.g., rainfall event volume, duration and interevent time)

(
)
h
i f S dh þ/r
v h k w v
i h i
efS di 1  e hfvr f
 fþhk w
1  e ð k w þhf Þ vr
þe h fe
1  fþ/ h kw
r ð1hÞmd b1 þhmw b2 g0
wþk b
EðEMCÞ ¼ /dp
ð55Þ
h fS di
f
e þ ð1  hÞ f
efS dp

After obtaining the expected value of pollutant event based on a 6-h interevent time definition is given in Table 1
load, the average annual pollutant load from the urban [1].
catchment outfall may be estimated as a product of the While there is no long-term measurement of runoff quan-
average annual number of rainfall event and the expected tity and quality available at the site. However, during 1992
value of pollutant event load. and 1994 Summer/Autumn period, field measurements were
conducted in the study area for a set of 17 rainfall events
5. Model application: case study [31]. Among these rainfall events, field measurements of
runoff event volume are available for only 11 rainfall events.
5.1. Study area and data condition Table 2 gives the measured rainfall event volumes and cor-
responding measured runoff volumes for the 11 rainfall
In this study, the test catchment with an area of 16.1 ha events. The runoff coefficients given in the table are esti-
is located within the Upper East Don subwatershed, part of mated as a quotient of the runoff volume divided by the rain-
the Don River watershed in the city of Toronto. The study fall volume. As indicated in Table 2, for the 11 events, the
area selected from one of the two subcatchments within the rainfall event volumes are in a range of 4–51 mm, and the
tributary to the Upper East Don River is drained to the corresponding runoff event volumes are in a range of
Heritage Estates pond. According to the previous study 0.71–11.64 mm. The maximum runoff event volume (i.e.,
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 91

Table 1
Rainfall statistics and PDFs of rainfall characteristics of Toronto Bloor Street station (IETD = 6 h)
Rainfall statistics Parameters of PDF Exponential PDF
Rainfall volume, v (mm) v ¼ 7:03 f ¼ 1=v ¼ 0:142 fv(v) = f exp(fv)
Duration, t (h) t ¼ 5:81 k ¼ 1=t ¼ 0:172 fT(t) = k exp(kt)
Interevent time, b (h) b ¼ 94:5 w ¼ 1=b ¼ 0:0106 fB(b) = w exp(wb)
Average annual number of rainfall events, h = 65.8
Average annual rainfall volume, v = 665 (mm)

Table 2
Simulated runoff volumes and estimated runoff coefficients based on field measurements of rainfall and runoff volumes
Event no. Date Rainfall Observed runoff Simulated runoff Runoff
volume (mm) volume (mm) volume (mm) coefficient
1 27/08/92 18.6 3.52 3.68 0.19
2 16/10/92 17 9.94 7.01 0.58
3 19/06/93 51 11.64 9.40 0.23
4 29/07/93 8.2 1.12 2.07 0.14
5 02/09/93 19 4.60 6.75 0.24
6 08/09/93 8 0.92 0.20 0.11
7 16/10/93 33.6 9.94 10.5 0.30
8 20/10/93 10 3.31 4.04 0.33
9 06/05/94 7.4 1.60 1.38 0.22
10 22/07/94 4 0.71 0.50 0.18
11 26/07/94 4.4 0.69 1.56 0.16

11.64 mm) is generated from the rainfall event with the max- 5.2. Calibration of continuous simulation model
imum rainfall event volume (51 mm), whereas the two low-
est runoff event volumes are generated from the two rainfall 5.2.1. Calibration of SWMM runoff quantity parameters
events with the lowest rainfall event volumes. However, the One of the major purposes of performing continuous
relationship between the rainfall event volumes and the run- simulation with SWMM is to generate long-term detailed
off event volumes are not definitive. runoff hydrograph, based on which runoff event volumes
Field measurements of runoff quality data (event mean can be extracted for those rainfall events whose field mea-
concentrations) are available. These field data for a total surements of runoff volumes are missing while pollutant
of 17 rainfall events are measured for nine major pollutants EMCs are available. In calibration of SWMM, the most
(i.e., TSS, TS, TKN, TP, COD, BOD5, Cu, Fe and Zn). sensitive parameters involved in the RUNOFF module
From the 17 measured pollutant EMCs, the 11 events with are the size of the catchment area, slope, the impervious
corresponding measurements of runoff volumes are and pervious area depression storages, Manning ‘‘n’’ value,
selected and plotted in Fig. 2 for different pollutants. These the initial and ultimate infiltration capacities.
figures show the ranges and variations of pollutant EMCs With long-term rainfall and initial parameter values as
against rainfall event volumes. The TS EMCs are highly inputs, continuous simulation with SWMM is performed,
variable which are in a range of 130–838 mg/L with a mean and runoff event volumes are extracted from long-term
value of 372 mg/L, and the EMCs of TSS in a range of 3.9– simulation results for those specific rainfall events and
214 mg/L with a mean value of 43.4 mg/L are also highly compared with the field measurements. The model calibra-
variable from event to event. The highest EMC of TKN tion is an iterative process, in which the parameter values
is 1.3 mg/L and the lowest is 0.05 mg/L. For pollutant of RUNOFF module are constantly adjusted until the
TP, the EMCs are generally below 0.5 mg/L, the EMC of deviation or standard error between the simulated and
COD generally varies from 1.0 to 47.1 mg/L with a mean observed runoff is minimized or reduced to a satisfactory
value of 26.3 mg/L. For BOD5, there seems no substantial level. As a result, the impervious and pervious Manning’s
variation in the observed EMCs as indicated in Fig. 2. As n is calibrated to be 0.015 and 0.25, respectively, with a
for heavy metals, the EMCs vary from 0.004 to catchment width of 550 m and an average slope of 0.016;
0.114 mg/L for Cu, from 0.02 to 5.53 mg/L for Fe and the impervious and pervious area depression storage are
from 0.007 to 0.13 mg/L for Zn. In addition, for the 11 set to 1.5 mm and 4.5 mm, respectively; the initial and ulti-
rainfall events listed in Table 2, the pollutant event loads mate infiltration capacity are adjusted to be 36.8 and
at the catchment outfall were also monitored for the nine 0.58 mm/h, respectively, the infiltration decay coefficient
pollutants [31]. And these field measurements of pollutant is calibrated to be 4.14 h1. With the calibrated parameters,
loads are used to calibrate continuous simulation model the long-term SWMM simulated runoff volumes for the 11
for runoff quality. rainfall events are shown in Table 2, and these results are
92 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

Observed Runoff (mm)


15 8

Observed Load (g/m2)


12
6
9
6 4

3 2
0
0 3 6 9 12 15 0
Predicted Runoff (mm) 0 2 4 6 8
Estimated Load (g/m2)
Fig. 3. Calibration of SWMM with observed runoff event volumes.
Fig. 4a. Calibration of SWMM runoff quality parameters with observed
event TS loads.
plotted in Fig. 3. The R-squared value of the regression
(R2) is estimated to be 0.94.
As indicated in Table 2 and Fig. 3, continuous SWMM 0.4

Observed Load (g/m2)


simulation is able to generate good estimates to the
observed runoff volumes. However, it must be pointed out 0.3
that the calibration of SWMM RUNOFF block is limited
0.2
to available field data. It may be desirable to recalibrate
SWMM parameters when more field data become available. 0.1

5.2.2. Calibration of SWMM stormwater quality parameters 0


The calibration of stormwater quality parameters as the 0 0.1 0.2 0.3 0.4
second step following the calibration of runoff quantity Estimated Load (g/m2)
parameters is intended to determine the appropriate param- Fig. 4b. Calibration of SWMM runoff quality parameters with observed
eter values for the pollutant buildup and washoff functions. event TSS loads.
In SWMM, based on the exponential buildup function, the
two major parameters are the maximum amount of pollu-
tant per unit area on catchment surface (QFACT1) and pol- 2
Observed Load (mg/m2)

lutant buildup rate (QFACT1). In the exponential washoff


function, two major parameters involved are WASHPO 1.5
(exponent) and RCOEF (washoff coefficient).
Similar to the calibration of runoff quantity parameters, 1
with field measurements of pollutant event loads and initial
parameter values for pollutant buildup and washoff, long- 0.5
term continuous simulation is conducted, pollutant event
0
loads for the specified rainfall events are extracted from 0 0.5 1 1.5 2
the continuous simulation results (pollutograph) and com- Estimated Load (mg/m2)
pared with the observed data. The parameter values of pol-
lutant buildup and washoff in SWMM RUNOFF block are Fig. 4c. Calibration of SWMM runoff quality parameters with observed
event TP loads.
constantly adjusted until the deviation or standard error
between the simulated and the observed pollutant loads is
minimized or reduced to a satisfactory level. The major cal- 4a–4f. In these figures, the R-squared value of the regres-
ibrated parameter values for SWMM are shown in Table 3. sion (R2) is estimated and given in Table 3. In terms of
The comparisons of the observed and long-term the R2 value, the lowest is the correlation between the
SWMM simulated pollutant loads for the six pollutants observed and simulated TSS load with a value of 0.86,
(TS, TSS, TP, COD, BOD5 and Fe) are illustrated in Figs. while the highest is the correlation between the observed

Table 3
Major calibrated runoff quality parameters for the six pollutants
TSS TS COD BOD5 Fe TP
QFACT1 (kg/ha) 105 1110 200 0.95 1.7 0.25
QFACT2 (day1) 0.0245 0.024 0.0257 0.60 0.67 0.54
WASHPO 0.96 0.95 0.55 1.25 0.75 0.90
RCOEF 0.0112 0.0087 0.0032 0.032 0.006 0.0075
R2 0.86 0.93 0.91 0.88 0.92 0.87
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 93

0.5 catchment into the impervious and pervious areas. Consid-


Observed Load (g/m2) ering that the depression storage is seldom measured
0.4
directly from field data, a reasonable estimate of the lumped
0.3 depression storage may be obtained by taking area-weighted
0.2 depression storages of the pervious and impervious areas of
the catchment. For example, according to the fraction of the
0.1 impervious area of the catchment (estimated to be 0.31), the
0 area-weighted depression storage should have a value of
0 0.1 0.2 0.3 0.4 0.5 3.57 mm, noting that the calibrated pervious and impervi-
Estimated Load (g/m2) ous area depression storages in SWMM are 4.5 and
Fig. 4d. Calibration of SWMM runoff quality parameters with observed 1.5 mm, respectively.
event COD loads. As for the lumped runoff coefficient, without observed
field data, the determination of the parameter value for /
may be based on trial-and-error. That is by assuming an ini-
60 tial value to /, the average annual runoff volume is estimated
Observed Load (mg/m2)

50 and compared with the SWMM estimate obtained from


40
long-term continuous simulation. The parameter value of
/ in the original rainfall–runoff transformation is adjusted
30 until the estimate of runoff volume from the analytical
20 model is close to the SWMM estimate. Alternatively, an esti-
10
mate for / may be obtained from the mean runoff coefficient
value (0.25) estimated from the observed rainfall event vol-
0 umes and corresponding runoff event volumes. However,
0 10 20 30 40 50 60
considering the rainfall loss in the depression storage, a rea-
Estimated Load (mg/m2)
sonable estimate for / may take a value that should be
Fig. 4e. Calibration of SWMM runoff quality parameters with observed greater than 0.25. For example, / may take a value of 0.35
event BOD5 loads. if the depression storage is assumed to be 3.57 mm. Fig. 5
shows the correlation between the observed rainfall (runoff)
event volumes and estimated runoff coefficients.
The calibration of the extended form of the rainfall–run-
15
Observed Load (mg/m2)

off transformation employed in the Type 2 load model


12 needs to determine parameter values for the pervious area
9 runoff coefficient, /dp, the impervious area depression stor-
age, Sdi, and the pervious area depression storage, Sdp. The
6
parameters Sdi and Sdp in the extended form of the rain-
3 fall–runoff transformation may adopt the calibrated values;
0 i.e., Sdi and Sdp have values of 1.5 and 4.5 mm, respec-
0 3 6 9 12 15 tively. The parameter value of the pervious area runoff
Estimated Load (mg/m2) coefficient may be determined in a procedure that is similar
to the procedure used to determine the parameter value for
Fig. 4f. Calibration of SWMM runoff quality parameters with observed
event Fe loads. /. In this study, it is found that when /dp is set to 0.30
(which should be generally lower than the parameter value

and simulated TS load with a value of 0.93. From these


estimated R2 values that are in a range of 0.86–0.93, it 60
appears that SWMM is able to provide reasonable esti- Rainfall Event Volume
50
mates of pollutant event load for various pollutants.
Rainfall (Runoff)

Runoff Event Volume


Volume (mm)

40
5.3. Calibration of analytical stormwater models 30
20
5.3.1. Evaluation of rainfall–runoff transformation
parameters 10
For the original rainfall–runoff transformation employed 0
0 0.2 0.4 0.6 0.8
in the Type 1 pollutant load model, it is necessary to deter-
Runoff Coefficient
mine the two parameters; i.e., runoff coefficient, /, and
depression storage, Sd. The two lumped parameters repre- Fig. 5. Correlations between the observed rainfall (runoff) event volumes
sent spatially averaged values without dividing the entire and runoff coefficients.
94 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

for /), the average annual runoff volume estimated from amount of pollutant after the last runoff or street sweeping
the extended form of the rainfall–runoff transformation is event, M0; and the constant pollutant removal rate, kb.
167 mm, close to SWMM estimate (169 mm) with a relative Retaining the two parameters of M0 and kb, the disaggre-
difference of 1.4%. gated form of buildup function involves five additional
To further evaluate the reliability of the parameter value parameters; namely, a constant rate of pollutant deposition
for /dp, the fraction of the impervious area of the catch- (dust fall), md; the quantity of street sweepings accumu-
ment is assumed to change in a range of 0.1–0.9. It is found lated on the streets, mw; the street sweeping effectiveness
that even with a fixed value of 0.30 for /dp, the average parameter, g; the factor converting particulate matter into
annual runoff volumes estimated from the extended form parameters of a given type of pollutant, b1; and the factor
of the rainfall–runoff transformation are in good agree- converting the mass of street sweepings into parameters of
ment with continuous SWMM simulation results. The esti- a given type of pollutant, b2.
mated runoff volumes from the extended form of the In the calibrations of the Types 1 and 2 pollutant load
rainfall- runoff transformation are shown in Table 4. The models, the objective is to minimize the deviation or stan-
estimated average annual runoff volumes from analytical dard error between the estimated and the observed cumu-
models and continuous simulation are plotted in Fig. 6. lative distribution functions (CDFs) of pollutant event
The comparable results in the table (or the plotted figure) loads. For the Type 1 pollutant load model, the mathemat-
show that the determined parameter value for the pervious ical expression of the CDFs of pollutant event load are
area runoff coefficient is a reasonable estimate for the numerically evaluated for different pollutants with sampled
extended form of the rainfall–runoff transformation. pollutant event loads, and compared with the empirical
CDFs of pollutant event loads obtained from the field data.
5.3.2. Calibration of pollutant load models Since the mathematical expression developed for estimat-
The major distinctions between the aggregated and ing the CDF of pollutant event loads is not an analytical
disaggregated forms of the pollutant load models are solution, to obtain the specific CDF values for the pollu-
reflected in formulating pollutant buildup function. The tant, numerical evaluation is necessary. The observed
aggregated form of the buildup function uses three param- CDFs of pollutant load for different pollutants are
eters to depict the pollutant buildup process; namely, the obtained from the 17 measured events and their empirical
maximum amount of pollutant buildup, Mm; the residual plotting positions are estimated by Weibull formula, which
means that the observed event washoff load series must be
sorted and ranked in a ascending order of the magnitude
Table 4 before the non-exceedance probabilities can be calculated.
Verification of the pervious area runoff coefficient for the extended form of The calibrated values of pollutant buildup and washoff
the rainfall–runoff transformation
functions for SWMM can be used as the initial trial values
h Annual runoff volume (mm) Relative difference (%) in the calibration of analytical models. For the Type 1 load
Type I model SWMM model, the major calibrated parameters are given in Table
0.2 134 128 4.4 5.
0.3 164 165 0.8 The calibration of the Type 2 load model follows the
0.4 195 202 4.1 similar procedures to those described in the calibration of
0.5 224 238 5.9
the Type 1 load model. Since the Type 2 load model
0.6 253 273 6.9
0.7 284 307 7.4 employs pollutant buildup and washoff functions in the
0.8 315 340 7.6 same forms as those employed in the Type 1 load model,
0.9 344 373 7.7 and the major difference between the two types of models
is that the Type 2 load model is based on the extended form
of the rainfall–runoff transformation. The calibrated
stormwater quality parameters for the disaggregated form
400
of the Type 2 load model are given in Table 6. In Types
Average Annual Runoff

1 and 2 pollutant load models, the appropriate parameter


300
Volume (mm)

values for dust fall (md), the amount of street sweepings


200
accumulated on the streets (mw) and the street sweeping
effectiveness parameter (g) may be determined according
Type I Model to field measurements or based on empirical data. In this
100
SWMM study, md is set to 0.4 g/m2 day; and mw is set to 2.5 kg/
0 m2 year. In the calibration process, the initially assigned
0 0.2 0.4 0.6 0.8 1 values for these parameters can be adjusted along with
Fraction of Imperviousness other buildup and washoff parameters in order to minimize
Fig. 6. Comparison of average annual runoff volumes from the extended the standard deviation between the empirical CDF curve of
form of the rainfall–runoff transformation and continuous SWMM pollutant load and the numerically evaluated CDF curve
simulation. from pollutant load models. Figs. 7a–7f show the compar-
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 95

Table 5
Calibrated parameters for the Type 1 pollutant load model
Runoff quality constituents Buildup parameter Washoff parameter
2 1
M0 (g/m ) b1 b2 kb (h )
TSS 1.5 0.29 0.54 0.0047 0.0143
TS 5.5 0.48 0.75 0.00102 0.065
TKN 0.0035 0.0093 0.0107 0.0213 0.0192
TP 0.002 0.00335 0.0064 0.0235 0.0131
COD 0.85 0.345 0.49 0.0041 0.013
BOD5 0.01 0.0135 0.027 0.0297 0.068
Cu 0.00025 0.000121 0.000142 0.0012 0.025
Fe 0.025 0.01 0.026 0.023 0.0184
Zn 0.003 0.00024 0.001 0.0108 0.0141

Table 6
Calibrated parameters for the Type 2 pollutant load model
Runoff quality constituent Buildup parameter Washoff parameter
2 1
M0 (g/m ) b1 b2 kb (h )
TSS 0.095 0.45 0.64 0.0102 0.0085
TS 2.0 0.48 0.67 0.00101 0.058
TKN 0.001 0.0051 0.008 0.0205 0.0135
TP 0.001 0.00184 0.0034 0.0225 0.0115
COD 0.195 0.0605 0.13 0.00107 0.018
BOD5 0.005 0.011 0.014 0.0248 0.057
Cu 0.0001 0.000117 0.000125 0.0017 0.012
Fe 0.02 0.0123 0.029 0.028 0.0084
Zn 0.001 0.00057 0.0007 0.0118 0.0101

1 1
Cumulative Probability

Cumulative Probability

0.8 0.8

0.6 0.6
Type I Model Type I Model
0.4 Observed 0.4 Observed
Type II Model Type II Model
0.2 0.2

0 0
0 1 2 3 4 5 6 0 0.1 0.2 0.3 0.4 0.5
TS load per event (g/m2) COD load per event ( g/m2)

Fig. 7a. Comparison of the observed CDF of TS load and that derived Fig. 7c. Comparison of the observed CDF of COD load and that derived
from the Types 1 and 2 pollutant load models. from the Types 1 and 2 pollutant load models.

1 1
Cumulative Probability
Cumulative Probability

0.8 0.8

0.6 0.6
Type I Model
0.4 0.4 Observed
Type I Model
Type II Model
Observed
0.2 0.2
Type II Model
0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0 10 20 30 40 50 60
TSS load per event (g/m2) BOD load per event ( mg/m2)

Fig. 7b. Comparison of the observed CDF of TSS load and that derived Fig. 7d. Comparison of the observed CDF of BOD5 load and that derived
from the Types 1 and 2 pollutant load models. from the Types 1 and 2 pollutant load models.
96 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

1 for rainfall events with relatively low washoff loads. The


Cumulative Probability

reason may be partially explained by the different forms


0.8
of the rainfall–runoff transformations employed in the
0.6 Types 1 and 2 load models.
Type I Model
0.4
The original rainfall–runoff transformation aggregates
Observed
the impervious and pervious areas of the catchment
0.2 Type II Model
together, and employs two lumped parameters (runoff coef-
0 ficient and depression storage) in depicting the runoff gen-
0 3 6 9 12 15 eration mechanism for the entire catchment, whereas the
Fe load per event (mg/m2) extended form of the rainfall–runoff transformation
Fig. 7e. Comparison of the observed CDF of Fe load and that derived
employs the pervious and impervious area depression sto-
from the Types 1 and 2 pollutant load models. rages and impervious area runoff coefficient in depicting
the separated runoff generation mechanisms in the two
areas of the catchment. The parameter value of the lumped
depression storage usually takes an area weighted value
Cumulative Probability

1
0.8 that is greater than the impervious area runoff depression
storage and less than the pervious area depression storage.
0.6
Type I Model As a result, for those rainfall events with relatively low
0.4 Observed rainfall volumes, runoff may be not generated from the
Type II Model
0.2 entire catchment based on the original rainfall–runoff
0 transformation (if the rainfall cannot satisfy the lumped
0 1 2 3 4 5 6 depression storage), but runoff may still occur from the
TP load per event (mg/m2) impervious area for the same rainfall events due to the rel-
Fig. 7f. Comparison of the observed CDF of TP load and that derived
atively low impervious area depression storage. Therefore,
from the Types 1 and 2 pollutant load models. no runoff events are more likely to occur based on the ori-
ginal rainfall–runoff transformation than the extended
form of the transformation. For such small rainfall events,
isons of the observed and the numerically evaluated CDFs it may be reasonable to assume that only the impervious
of pollutant loads from Types 1 and 2 load models for TS, area of the catchment contributes to runoff and generates
TSS, COD, BOD5, Fe and TP, respectively. washoff loads. In other words, the impulse probability esti-
Fig. 7a illustrates a comparison of the empirical CDF of mated from the Type 1 load model can be higher than the
TS load and the corresponding CDF curves derived from Type 2 load model. As a result, the Type 1 model appears
the Types 1 and 2 load models. The figure shows that both insensitive to the small rainfall events.
the Types 1 and 2 load models are able to provide compa- As a summary, considering that the total number of
rable estimates to the observed results regardless of the dif- parameters to be calibrated is 9 for the Type 1 load model
ference in the formulation of the rainfall–runoff and 11 for the Type 2 Model, there seems no big difference
transformation. Generally, the CDF curve derived from compared with a total of 12 parameters which must be cal-
the Type 2 load model is in good shape with the CDF curve ibrated for SWMM. However, the effort required for cali-
based on the observed results. The estimates from the Type bration of analytical models and SWMM are significantly
1 load model appear to be larger than the observed results different. Continuous simulation with SWMM usually
as pollutant load increases. Except for the plotting posi- employs long-term rainfall records at a short time step as
tions for the observed TSS load in Fig. 7b, which seem to input to generate long-term outputs (in this study, a total
be difficult to be fitted by a smooth line, for other pollu- of 35-year long-term hourly rainfall records are used to
tants, Figs. 7c–7f show that Types 1 and 2 models are able generate hourly runoff hydrograph and pollutograph).
to produce estimates that fit the observed results reason- The long-term continuous simulation outputs are then ana-
ably well. Nevertheless, the Type 1 load model appears to lyzed for its characteristics. It is tedious and time-consum-
be more likely to underestimate the pollutant load than ing process to extract and summarize the detailed runoff
the Type 2 load model for those rainfall events with rela- statistics, such as runoff event volume and pollutant event
tively low washoff loads. The impulse probabilities pre- loads. Furthermore, the calibration of continuous model
dicted by the Type 1 load model are higher than the is actually an iterative process in which parameter values
observed results in Figs. 7b–7d and 7f. From these figures, are constantly adjusted in each simulation run, and the sta-
it appears that the Type 2 load model produces the CDFs tistics of system outputs must be computed each time. To
of pollutant load in better agreement with the observed find an ‘‘optimized’’ parameter set, the calibration of con-
CDFs than the Type 1 load model. The Type 1 load model tinuous simulation model usually requires a significant
is more likely to produce a higher impulse probability than number of simulation runs. The computational burden
the observed. There are apparent deviations between the and the time requirements for system analysis may be pro-
observed CDFs and the one derived from the Type 1 model hibitive for continuous simulation with short time steps. In
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 97

contrast, the Types 1 and 2 models for long-term storm- Table 8


water quality control measures are closed-form analytical Comparison of the observed and Type 2 load model predicted average
event mean concentrations (EMCs)
models, which are compact in mathematical expression,
efficient in practical applications, and can be a cost-effective Runoff Average EMCs (mg/L) Relative
quality difference (%)
approach for screen level analysis at planning stage in constituents
Type 2 load Observed
which a large number of alternatives need to be evaluated. model
Analytical models can be extremely efficient if the models TSS 42.16 43.41 2.88
are created in a spreadsheet format. However, the advanta- TS 381.20 372.40 2.36
TKN 0.549 0.59 6.77
ges of performing continuous simulation with SWMM are TP 0.185 0.19 1.25
also numerous. Based on a wide range of meteorological COD 28.13 26.30 6.96
conditions and physical characteristics of the catchment, BOD5 3.12 2.85 9.38
continuous simulation can produce detailed hydrograph Cu 0.018 0.016 12.12
or pollutograph at a desirable time step with consideration Fe 0.963 1.03 6.53
Zn 0.047 0.044 6.23
of spatial and temporal variations of model parameters.
Continuous simulation models like SWMM have been
widely recognized as a reliable tool in engineering planning
and design. from the model are lower than the continuous SWMM
simulation results.
5.3.3. Evaluation of the average EMCs and annual For the Type 2 load model, the estimated average EMCs
pollutant loads and the corresponding relative differences for different pol-
With the calibrated pollutant buildup and washoff lutants are shown in Table 8. Compared to the estimates
parameters, a comparison of the observed average pollu- from the Type 1 load model given in Table 7, the estimated
tant event mean concentrations (EMCs) and the Type 1 average EMCs from the Type 2 load model for different
load model predicted results is given in Table 7. The table pollutants are encouraging. For example, the estimated
also shows the estimated average annual pollutant loads average TSS EMC from the Type 2 load model is
for different pollutants based on the Type 1 load model 42.16 mg/L, with a relative difference of 2.88%, while
and the results obtained from continuous simulation. Since the estimated average TSS EMC from the Type 1 load
there is no field measurement of the average annual pollu- model is 54.42 mg/L, with a relative difference of 25.36%.
tant loads for different pollutants, to verify analytical The estimated TS EMC from the Type 1 load model is
model results, long-term continuous simulation is per- 335.6 mg/L with a relative difference of 9.9%, it is
formed to estimate the average annual pollutant loads for 381.2 mg/L with a relative difference of 2.36% from the
different pollutants. These continuous SWMM simulation Type 2 load model. The Type 1 load model overestimates
results are compared to analytical model results in Table the average EMCs of COD and BOD5, the Type 2 load
7. The comparison indicates that while the Type 1 load model estimates (i.e., 28.13 mg/L with a relative difference
model is able to produce reasonable estimates to the of 6.96 for COD, and 3.12 mg/L with a relative difference
observed in terms of the CDF of pollutant load, the esti- of 9.38% for BOD5) are closer to the observed results.
mated average pollutant EMCs for different pollutants For other pollutants such as heavy metals, similar conclu-
(except for TS) are generally higher than the observed sions can be drawn in terms of the estimated average EMC.
results, and the estimated average annual pollutant loads With all the relative differences in Table 8 are less than 10%
(except for Cu), the Type 2 load model outperforms the
Type 1 load model.
Table 7 The performance of the Type 2 load model is also veri-
Comparison of the Type 1 load model predicted average EMCs and fied by comparing the estimated average annual pollutant
annual pollutant loads with the observed and SWMM simulation results loads with continuous SWMM simulation results as shown
Runoff Average Average in Table 9. In terms of the estimated average annual pollu-
quality pollutant annual pollutant tant loads and corresponding relative differences in the
pollutants EMCs (mg/L) load (g/m2) table, it is found that the Type 2 load model is capable
Type 1 Observed Type 1 SWMM of providing comparable results to those obtained from
model model continuous simulation. For example, the estimated average
TSS 54.42 43.41 3.98 5.54 annual TSS load from the Type 2 load model is 5.39 g/m2
TS 335.6 372.40 32.16 46.5 (continuous simulation result is 5.54 g/m2) with a relative
TKN 0.84 0.68 0.045 0.075
difference of 2.7%, while the estimated average annual
TP 0.32 0.19 0.013 0.027
COD 51.35 36.30 2.86 3.96 TSS load from the Type 1 load model is 3.98 g/m2, with
BOD5 5.19 2.85 0.27 0.38 a relative difference of 28.7%. The estimated average
Cu 0.035 0.016 0.0018 0.0023 annual TS load from the Type 2 load model is 48.76 g/
Fe 1.72 1.43 0.094 0.142 m2 with a relative difference of 4.9%, it is 32.16 g/m2 with
Zn 0.068 0.044 0.0038 0.0059
a relative difference of 30.8% from the Type 1 load model.
98 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

Table 9 Table 11
Comparison of the Type 2 load model predicted average annual pollutant Average annual pollutant loads obtained from long-term continuous
loads with continuous SWMM simulation results SWMM simulation
Runoff Average annual Relative h Long-term average annual pollutant load (g/m2)
quality pollutant load (g/m2) difference (%)
TSS TS COD BOD5 Fe TP
constituents
Type 2 load SWMM 0.1 3.07 25.7 2.57 0.19 0.09 0.015
model
0.2 4.28 35.8 3.30 0.28 0.12 0.021
TSS 5.39 5.54 2.7 0.3 5.43 45.6 3.90 0.37 0.15 0.026
TS 48.76 46.5 4.9 0.4 6.50 54.7 4.42 0.46 0.17 0.031
TKN 0.070 0.075 6.3 0.5 7.49 63.2 4.92 0.55 0.20 0.037
TP 0.024 0.027 12.5 0.6 8.39 71.0 5.36 0.63 0.22 0.041
COD 3.60 3.96 9.1 0.7 9.25 78.4 5.77 0.70 0.24 0.046
BOD5 0.40 0.38 4.9 0.8 10.04 85.5 6.16 0.78 0.26 0.050
Cu 0.0024 0.0023 5.3 0.9 10.70 91.4 6.50 0.84 0.28 0.054
Fe 0.123 0.142 13.3
Zn 0.0060 0.0059 1.6

example, when the fraction of the imperviousness of the


catchment is assumed to be 0.1, the Type 2 model predicted
For pollutants COD and BOD5, the estimated average
annual TSS load is 3.18 g/m2, and SWMM simulated result
annual loads are 3.6 g/m2 (with a relative difference of
is 3.07 g/m2. For the same pollutant, if the fraction of the
9.1%) and 0.4 g/m2 (with a relative difference of 4.9%),
imperviousness is increased to 0.9, the average annual
respectively, while the Type 1 load model estimates are
TSS load simulated from SWMM is 10.7 g/m2, whereas
2.86 and 0.27 g/m2 for COD and BOD5, respectively. For
the Type 2 load model estimate is 11.39 g/m2 with a relative
heavy metals, the relative differences of the estimated aver-
difference of 6.4%. For heavy metals, such as Fe, if the frac-
age annual loads from the Type 1 load model are in a range
tion of the imperviousness is relatively low, say 0.1, the
of 21.7% to 35.6%, the relative differences of the esti-
Type 2 load model estimated average annual load is
mates from the Type 2 load model are in a range of 1.6%
0.08 g/m2, close to the SWMM simulated result (0.09 g/
to 13.3%.
m2). If the impervious area of the catchment is significantly
To further verify the model performance, the Type 2
increased, say 0.9, the SWMM simulated average annual
load model is employed to evaluate the impact of land
Fe load is 0.28 g/m2, whereas the Type 2 load model esti-
use on long-term pollutant loads for different pollutants.
mated annual load for Fe is 0.29 g/m2 with a relative differ-
The average annual pollutant loads are estimated from
ence of 3.6%. For other pollutants, good agreements are
the Type 2 load model by assuming that the fraction of
also observed between the estimates from the Type 2 load
the imperviousness of the catchment varies from 0.1 to
model and continuous SWMM simulated results. With
0.9. Since there is no long-term measurement for annual
the capability of providing close estimates to continuous
pollutant load for different pollutants, the average annual
simulation results, the Type 2 load model can be used as
pollutant loads from the Type 2 load model are compared
an alternative in the evaluation of long-term pollutant
to continuous SWMM simulation results. Table 10 pre-
loads.
sents the Type 2 load model estimated average annual
loads for different pollutants. The results obtained from
SWMM continuous simulation are given in Table 11. 6. Summary and conclusions
From the two tables, it can be seen that the average
annual loads estimated from the Type 2 load model are This paper presents methodologies for the development
close to the continuous SWMM simulation results. For of analytical stormwater quality models based on the
derived probability distribution approach. By integrating
rainfall–runoff transformation with pollutant buildup and
Table 10 washoff models, long-term stormwater quality measures;
The Type 2 load model predicted average annual pollutant loads for
e.g., the average pollutant event mean concentration, and
different pollutants
annual pollutant load to receiving waters are developed
h Long-term average annual pollutant load (g/m2)
in terms of catchment meteorological characteristics. As
TSS TS COD BOD5 Fe TP demonstrated in this study, analytical models for estimat-
0.1 3.18 30.1 2.14 0.25 0.08 0.014 ing long-term stormwater quality control measures can be
0.2 4.24 39.2 2.84 0.32 0.11 0.019 developed with various degrees of complexity with the flex-
0.3 5.29 47.9 3.53 0.39 0.13 0.023
ibility in choosing different types of the rainfall–runoff
0.4 6.33 56.1 4.21 0.46 0.16 0.028
0.5 7.36 64.0 4.87 0.52 0.19 0.032 transformations and in the formulation of different forms
0.6 8.38 71.4 5.52 0.58 0.21 0.037 of pollutant washoff load models. As a result, the selection
0.7 9.39 78.4 6.16 0.64 0.24 0.041 of the appropriate analytical models for practical applica-
0.8 10.40 85.2 6.79 0.70 0.26 0.045 tions can be a tradeoff between model complexity and the
0.9 11.39 91.6 7.40 0.75 0.29 0.049
desired levels of performance. While the proposed Types
J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100 99

1 and 2 load models are able to provide comparable results runoff pollution. Berlin: Springer-Verlag; 1986. p. 277–304 [Pub-
to the field measurements, the Type 2 load model generally lished in Cooperation with NATO Scientific Affairs Division].
[21] Huber WC. Modeling urban runoff quality: state-of-the art. In:
outperforms the Type 1 model in estimation of the average Urbonas B, Roesner LA, editors. Urban runoff quality – impact and
pollutant EMCs and annual pollutant loads. By comparing quality enhancement technology. Proceedings of an engineering
analytical model results with continuous simulation, it is foundation conference, Henniker, New Hampshire. New Yor-
concluded that analytical stormwater quality models can k: ASCE; 1986. p. 34–48.
be employed as effective tools or alternatives to continuous [22] Huber WC. Deterministic modeling of urban runoff quality. In:
Torno HC, Marsalek J, Desbordes M, editors. Urban runoff
simulation models in the evaluation of long-term storm- pollution. Berlin: Springer-Verlag; 1986. p. 167–242 [Published in
water quality control measures. Cooperation with NATO Scientific Affairs Division].
[23] Huber WC, Dickinson RE. Stormwater management model Version
4, User’s Manual, EPA-600/3-88-001a. Athens (GA): US EPA;
References 1988.
[24] James W, Thompson MK. Contaminants from four new pervious and
[1] Adams BJ, Papa F. Urban stormwater management planning with impervious pavements in parking lot. In: James W, editor. Advances
analytical probabilistic models. New York: John Wiley; 2000. in modeling the management of stormwater impacts, vol. 5. CHI
[2] Akan AO. Pollutant washoff by overland flow. J Environ Engrg Publications; 1997. p. 207–22.
ASCE 1987;113(4):811–23. [25] Jewell TK, Adrian DD. Improved techniques for modeling storm-
[3] Akan AO. Derived frequency distribution for storm runoff pollution. water quality. In: Yen BC, editors. Proceedings of the second
J Environ Engrg ASCE 1988;114(6):1344–51. international conference on urban storm drainage, vol. II, June 15–
[4] Alley WM. Estimation of impervious-area washoff parameters. Water 19. Urbana (IL); 1981. p. 156–65.
Resour Res 1981;17(4):1161–6. [26] Li JY, Adams BJ. Probabilistic models for analysis of urban
[5] Alley WM, Smith PE. Estimation of accumulation parameters for runoff control systems. J Environ Engrg ASCE 2000;126(3):
urban runoff quality modeling. Water Resour Res 217–24.
1981;17(6):1657–64. [27] Loganathan GV, Delleur JW. Effects of urbanization on frequen-
[6] Ammon DC. Urban stormwater pollutant buildup and washoff cies of overflows and pollutant loadings from storm sewer
relationships. Master of Engineering thesis. Gainesville (FL): Depart- overflows: a derived distribution approach. Water Resour Res
ment of Environmental Engineering Sciences, University of Florida; 1984;20(7):857–65.
1979. [28] Loganathan GV, Delleur JW, Segarra RI. Planning detention storage
[7] Barbe DE, Cruise JF, Mo X. Modeling the buildup and washoff of for stormwater management. J Water Resour Plng Mgmt ASCE
pollutants on urban watersheds. Water Resour Bull AWRA 1985;111(4):382–98.
1996;32(3):511–9. [29] Loganathan GV, Watkins EW, Kibler DF. Sizing stormwater
[8] Behera PK. Urban stormwater quality control analysis. PhD disser- detention basins for pollutant removal. J Environ Engrg ASCE
tation. Toronto (Ont.): Department of Civil Engineering, University 1994;120(6):1380–99.
of Toronto; 2001. [30] Marsalek J. Report on NATO workshop on urban runoff quality. In:
[9] Benjamin JR, Cornell CA. Probability, statistics, and decision for Urbonas B, Roesner LA, editors. Urban runoff quality – impact and
civil engineers. New York: McGraw-Hill Book Company; 1970. quality enhancement technology. Proceedings of an engineering
[10] Bertrand-Krajewski JL, Scrivener O, Briat P. Sewer sediment foundation conference, Henniker, New Hampshire. New Yor-
production and transport modeling: a literature review. J Hydrol k: ASCE; 1986. p. 15–28.
Res 1993;31(4):435–60. [31] Ministry of Ontario Environment and Energy (MOEE) and the
[11] Cadavid L, Obeysekera JTB, Shen HW. Flood frequency derivation Metropolitan Toronto and Region Conservation Authority
from kinematic wave. J Hydrol Engrg ASCE 1991;117(4):489–510. (MTRCA). Performance assessment of an off-line stormwater man-
[12] Charbeneau RJ, Barrett ME. Evaluation of methods for estimating agement pond. Richmond Hill (Ont.): Heritage Estates Stormwater
stormwater pollutant loadings. Water Environ Res Management Pond; 1996.
1998;70(7):1295–302. [32] Novotny V, Sung HM, Bannerman R, Baum B. Estimating nonpoint
[13] Chen JY. Integrated urban stormwater management modeling. PhD pollution from small urban watersheds. J Water Poll Contr Fed
dissertation. Toronto (Ont.): Department of Civil Engineering, 1985;57:339–48.
University of Toronto; 2003. [33] Novotny V. Modeling of sewer flow quality. In: Cao C, Yen BC,
[14] Deletic AB, Maksimovic CT, Ivetic M. Modeling of storm washoff of Benedini M, editors. Urban storm drainage, proceedings, US–Italy
suspended solids from impervious surfaces. J Hydrol Res bilateral seminar. Littleton (CO): Water Resources Publications;
1997;35(1):99–117. 1993. p. 275–306.
[15] DiToro DM. Probability model of stream quality due to runoff. J [34] Osuch-Pajdzinska E, Zawilski M. Model of storm sewer discharge, I:
Environ Engrg ASCE 1984;110(3):607–28. description. J Environ Engrg ASCE 1998;124(7):593–8.
[16] Ellis JB. Pollutional aspects of urban runoff. In: Torno HC, Marsalek [35] Sartor JD, Boyd G. Water pollution aspects of street surface
J, Desbordes M, editors. Urban runoff pollution. Berlin: Springer- contaminants. Report EPA-R2-72-081. Washington (DC): US EPA;
Verlag; 1986. p. 1–38 [Published in Cooperation with NATO 1972.
Scientific Affairs Division]. [36] Sartor JD, Boyd G, Agardy FJ. Water pollution aspects of street
[17] Gottschalk L, Tallaksen L, Perzyna G. Derivation of low flow surface contaminants. J Water Poll Contr Fed 1974;46(3):
distribution functions using recession curves. J Hydrol 458–67.
1997;194:239–62. [37] Segarra-Garcia R, Loganathan VG. Stormwater detention storage
[18] Gottschalk L, Weingartner R. Distribution of peak flow derived from design under random pollutant loadings. J Water Resour Plng Mgmt
a distribution of rainfall volume and runoff coefficient, and a unit ASCE 1992;118(5):475–90.
hydrograph. J Hydrol 1998;208:148–62. [38] Servat E. TSS, BOD5, and COD accumulation and transport over
[19] Grottker M. Runoff quality from a street with medium traffic loading. urban catchment surfaces: a modeling approach. In: Torno HC,
J Sci Total Environ 1987;59:457–66. Marsalek J, Desbordes M, editors. Urban runoff pollution. Ber-
[20] Hemain JC. Statistically based modeling of urban runoff quality: state lin: Springer-Verlag; 1986. p. 325–42 [Published in Cooperation with
of the art. In: Torno HC, Marsalek J, Desbordes M, editors. Urban NATO Scientific Affairs Division].
100 J. Chen, B.J. Adams / Advances in Water Resources 30 (2007) 80–100

[39] Singh VP, Krstanovic F. A stochastic model for water quality [43] Wegener KR, Malone RF. A methodology for determining the
constituents. In: Proceedings – 6th congress of IAHR-APD, Kyoto, predictive accuracy of mathematical models from a limited database.
Japan; 1988. In: Camenroth WK, Skogerboe GV, Flug M, editors. Analysis of
[40] Terstriep ML, Noel DC, Bender GM. Sources of urban pollutions – ecological systems: state-of-the-art in ecological modeling. 1982. p.
do we know enough? In: Urbonas B, Roesner LA, editors. Urban 75–80.
runoff quality – impact and quality enhancement technology. [44] Wood EF, Hebson CS. On hydrologic similarity 1, derivation of the
Proceedings of an engineering foundation conference. New Yor- dimensionless flood frequency curve. Water Resour Res
k: ASCE; 1986. p. 107–21. 1986;22(11):1549–54.
[41] US Environmental Protection Agency (EPA). Final Report of the [45] Xu P, Tsanis IK, Marsalek J. PMSQ-A probabilistic model for
nationwide urban runoff program, vol. I. Washington (DC): US EPA; calculating toxic pollutant loads in urban stormwater runoff. In:
1983. Marsalek J, Torno HC, editors. Proceedings of the 6th international
[42] US Army Corps of Engineers. Storage, treatment, overflow, runoff conference on urban storm drainage, Niagra Falls, Ont., Canada, vol.
model (STORM). Gen. Computer Program 723-S8-L7520. Davis 1. 1993. p. 921–6.
(CA): the Hydrologic Engineering Center, US Army Corps of [46] Zhang HP, Yamada K. Estimation for urban runoff quality modeling.
Engineers; 1977. Wat Sci Tech 1996;34(3–4):49–54.

You might also like