Beruflich Dokumente
Kultur Dokumente
Title:
Authors:
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
ip
a
The Graduate Program in Environmental Science, State University of New York,
d cr
College of Environmental Science and Forestry (SUNY-ESF), 1 Forestry Drive,
te s
Syracuse, NY 13210-2778, Email: nsun@esf.edu
di nu
b
Department of Environmental Studies, SUNY-ESF, 1 Forestry Drive, Syracuse, NY
c
Department of Ecology and Evolutionary Biology, Cornell University, 103 Little
d
Department of Forest and Natural Resources Management, SUNY-ESF, 1 Forestry
Corresponding Author:
N ce
Key Words:
Validation
Abstract
This study examined how the level of catchment discretization influenced the
t
ip
minimum required number of subcatchments to retain the original sewer network
d cr
properties, and 2) the micro-scale model in which each subcatchment was defined for
te s
a unique soil and land use combination. For both scales, we calibrated the model
di nu
parameters and quantified the uncertainty of model outputs using the Generalized
sewershed, which was also delineated at both micro and macro scales, to test observed
C ted
versus simulated flows. The results indicated that the catchment disaggregation level
had a great impact on both parameterization and simulation results, and that the
ot p
N ce
majority of the parameters were sensitive to the modeling scales. Overall, the
sewersheds. Hence we argue that the calibrated parameters obtained, based upon the
macro delineation, would result in reduced confidence in simulated runoff for another
site unique in its characteristics, while the posterior parameters derived from the
1. Introduction
Urban sprawl over the last three decades has induced an expansion of impervious
where sanitary and storm sewers share a main sewer line. Instead of letting rain water
t
ip
run its pre-developed natural courses, the water flowing off city surfaces is connected
d cr
directly to the combined storm sewer system through pipes and gutters in most urban
te s
settings. Combined sewer overflows (CSOs) occur when stormwater runoff exceeds
di nu
infrastructure capacity. This normally is caused by large rainfall events. The CSO is
released into neighboring surface waters, and the resulting effluent pollutes receiving
ye a
op M
waters and poses a considerable threat to aquatic life, human health and groundwater
design effective and efficient stormwater control facilities and hydraulic structures
that can handle peak flows and hence reduce CSO events. In practice, the design
ot p
N ce
parameters of these facilities are generally derived from the actual hydrograph
however, are often not available for estimating these design parameters because of
sparsely installed flow meters that are used to measure stormwater runoff from urban
sewersheds. To overcome this problem, hydrologic models are often used to simulate
storm hydrographs and provide discharge estimates at ungaged sites. On the other side,
More than a quarter century ago, Huber et al. (1975) presented the possible
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
ip
Northwood section of Baltimore, Maryland. Zaghloul (1981) suggested using a finite
d cr
set of area- or length-weighted parameters to represent the spatially varying physical
te s
properties of continuous spaces. This is also known as the spatially lumped-parameter
di nu
approach to hydrological modeling. Since then, the majority of hydrological models
used in practice today are lumped parameter models (Butts et al., 2004). This method
ye a
op M
is widely used because it requires fewer input data and is more computationally
efficient than a model based on spatial units of a fine resolution. However, the lumped
C ted
hydrological modeling approach does not explicitly account for the heterogeneity of
surface properties (e.g., variation of land covers, soils, slopes and etc.) that influence
ot p
N ce
how water moves through the landscape. It is also very difficult to represent the
(1991) argued that it may well be impossible to use the lumped value of spatially
(hydrological) modeling must take into account the lumping effect of subgrid scale
model, which ideally considers all spatial variability of the landscape by solving the
governing equation within each spatially-defined landscape unit (Baker, 1989; Chow
et al., 1988). It is important to note that any so called “lumped” model is rarely
entirely lumped, and so may be considered a partially “distributed” model. The model
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
scale, lumped or distributed, is relative to the size of the study area, scale of the input
t
ip
dataset, model complexity, and etc. The constraints of distributed-parameter models
d cr
result from the model complexity in terms of model structure and requirements for
te s
inputs. Consequently, the choice of modeling scale has always been a dilemma. Few
di nu
past studies examined the impact of modeling scale on distributed hydrologic model
performance. Mamillapalli et al. (1996) found that, using the Soil and Water
ye a
op M
Assessment Tool (SWAT), simulated streamflow predictions varied with the number
Watershed 6, New Hampshire into 208 cells led to an improvement in streamflow and
nutrient flux estimation simulated with Simple Nitrogen Cycle (SINIC) model (Hong
ot p
N ce
et al., 2006). Barco et al. (2008) applied the SWMM model coupled with GIS
procedures to simulate runoff for a large urban catchment in Southern California, and
Ac
SWMM runoff parameters. The results demonstrated that integrating GIS and a
watersheds. The results from Guo and Urbonas (2008), and Dankenbring and Mays
Management Model (SWMM) caused increasing predicted peak flows. Ghosh and
However, these studies were conducted in a deterministic fashion without taking the
structure and model parameterization can propagate to the model outputs. Thus it is
t
ip
desirable and necessary to integrate such uncertainties into the modeling processes in
d cr
order to quantify the overall uncertainty in model outputs. The Generalized
te s
Likelihood Uncertainty Estimation (GLUE) methodology, developed by Beven and
di nu
Binley (1992), is a widely adopted approach to evaluate model uncertainties
and Binley (1992), Lamb et al. (1998), Blazkova et al. (2002), Hong et al. (2005),
C ted
Choi and Beven (2007), Fang and Ball (2007), and Xiong and O’Connor (2008).
The research reported here focused on three questions: 1) how would model
ot p
N ce
outputs and parameter values respond to changing model scale? 2) Is there any scale
different discretization levels, i.e., micro and macro scale. For both catchment
delineations, we ran the SWMM model and quantified the uncertainties of model
predictions using the GLUE methodology. We had the measured hydrographs from
sewershed outlets available to us, which allowed us to calibrate the model parameters
and test parameter transferability through model validation. The ultimate goal of this
study was to investigate urban runoff modeling techniques and provide better
2. Methods
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
ip
SWMM is a physically based, spatially distributed model for simulating all
d cr
aspects of hydrologic and water quality cycles primarily within urban areas (Huber et
te s
al., 1988; Rossman, 2010). SWMM treats each catchment as a nonlinear reservoir and
di nu
employs the combined continuity equation (1) and Manning’s equation (2) on each
containing its own fraction of pervious and impervious surfaces whose runoff drains
(Rossman, 2010).
ot p
(1)
N ce
step, A denotes the area of the subcatchment, ie is the effective inflow, and Q is the
(2)
for U.S. metric units or 1.0 for SI units, n is the Manning’s roughness coefficient, DS
information systems (GIS) has been increasingly linked with SWMM for data pre-
parameters can be extracted from spatially defined GIS layers, e.g. catchment area
t
ip
and slope. The principal measured SWMM parameters are presented in Table 1.
d cr
Ideally, every SWMM parameter would be developed based on actual measurements
te s
of catchment characteristics. In practice, however, some parameters with a high
di nu
degree of spatial variability cannot be directly measured or extracted from GIS layers,
based on obtainable measurements, or established the ranges for the parameter values
C ted
are rarely well-defined due to model scale issues and the spatially and temporally
ot p
N ce
varying meteorological and geographical conditions under which the models were
built. In this study, based upon the actual sewer hydrographs, we aimed to refine the
Ac
storage (DS) for each surface cover, the coefficient for subcatchment width (KWidth),
and Manning's n for closed conduit (nConduit). Manning’s n, often called the surface
greater n value results in a lower runoff rate and prolongs the flow duration. DS is the
volume of water that can be held in natural depressions in the land surface (Horton,
1935). In other words, DS represents the depth of water that has to be exceeded on the
flow rates and earlier occurrence of runoff. Manning’s n and DS are difficult to
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
ip
context. Subcatchment width (W), conceptually, represents the width of the
d cr
downstream side of the idealized sloping rectangular subcatchment (Huber et al.,
te s
1988). A smaller W is equivalent to a longer flow path, leading to the attenuated peak
di nu
flow occurrence. Since real-world subcatchments are mostly irregular in shape with
drainage channels off the center, it can be very difficult to determine the actual
ye a
op M
subcatchment width. We therefore adopted the equation used in the InfoSWMM User
(3)
chose to use so the estimated flow width was less sensitive to the actual
subcatchment width (particularly important when the subcatchment has a very large
Ac
interior pipe walls, and determines the volume of water a pipe can convey (i.e., its
hydraulic capacity). The lower nConduit the smoother the wall, hence the greater is the
hydraulic capacity of the pipe. Although nConduit can be measured directly, it seldom
is, due to the fact that numerous studies have been done to establish a range of
Manning’s n values for various types of pipes. Nevertheless, the selection of the
In this study, the kinematic wave method (Lighthill and Whitham, 1955) was
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
chosen for the channel flow routing. The runoff from subcatchments was computed at
t
ip
a 5-minute time step, while the computational time used for flow routings was set to
d cr
30 seconds. Compared to the diffusive wave method, the kinematic wave method
te s
substitutes the simple stage-discharge relationship for the momentum equation by
di nu
assuming that the bed slope is approximately equal to the friction slope and stage-
pipes, are usually designed for partially full flow conditions, the kinematic wave
small urban basins (Overton and Meadows. 1976). It is also important to bear in mind
that the diffusion wave equation is more suitable than the kinematic wave method for
ot p
N ce
We developed two catchment delineations, one at the micro and one at the macro
scale, to test how much the model outputs are influenced by the aggregation level, or
scale, of input data. The two scales differ in the aggregation level of subcatchments
and the flow routing element settings. The routing element of each subcatchment is
comprised of two main components: outlet and internal subarea routing. The outlet
10
impervious and pervious subareas, internal subarea routing specifies the runoff
impervious subarea.
t
ip
2.2.1 The Micro-scale Delineation
d cr
The micro-scale model aims to reduce heterogeneity within each subcatchment of
te s
the sewershed in order to better replicate the heterogeneity across the sewershed and
di nu
thus reduce model uncertainty. Each subcatchment has a unique soil and land cover
within each subcatchment. Due to the homogeneity of each subcatchment, the micro
C ted
delineation can represent better the actual hydrologic behavior of different surface
covers. Slope variance within the subcatchment, however, could not be captured due
ot p
N ce
to the scale of the input digital elevation model (DEM), hence we used the average
overlaying the surface cover map with the sewer network map, and the sewer network
was preserved with the original level of detail. The majority of subcatchments
contain more than one surface cover type. The subarea routing direction within the
11
and the percent of runoff routed between subareas was assumed 100% for all
t
ip
Two urban sewersheds located in the City of Syracuse, NY were used in this
d cr
study, CSO 027 for calibration and CSO 044 for validation (Figure 1). CSO 027,
te s
46.76 hectares in size, is comprised of 68.4% impervious surfaces, 18% tree cover
di nu
and 13.6% lawns as viewed from aerial photography (Table 2). In CSO 044, 52.2% of
the total 17.75 hectares is covered by impervious surfaces, 21.3% is covered by trees,
ye a
op M
and the rest 26.5% is lawns. We defined impervious surfaces as all paved roads and
streets, sidewalks, rooftops, and parking lots, and assumed an identical hydrological
C ted
response from all impervious surfaces. Lawns and trees account for all pervious
surfaces in both sites. No adjustment is made for tree canopy over impervious
ot p
N ce
surfaces.
CSO 027 and CSO 044 were both delineated on micro and macro scales. The
Ac
delineation contained 292 subcatchments (Figure 2). The macro-scale CSO 044
contained 19 subcatchments and the micro-scale CSO 044 was partitioned into 478
homogeneous subcatchments (Figure 3). The GIS layers used to prepare SWMM
input parameters included the Digital Elevation Model (DEM), the soil map, the
sewershed surface cover map that we digitized from the sewershed orthoimages, and
12
the sewer network map (all described in Table 3). We employed the SCS Curve
Number Method (USDA SCS, 1986) for infiltration estimations. The hydrologic soil
group (HSG) map derived from the 30-m soil map (NRCS, 2010) was overlaid with
the surface cover map to determine the curve number for each surface cover
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
according to the table “Runoff Curve Numbers for Urban Area” (USDA SCS, 1986).
t
ip
This method was selected because it is conceptually simple, has less intensive
d cr
parameterization requirements, yet provides reliable estimates of infiltration rates
te s
(Bales and Betson, 1982; Xiong and Melching, 2005). Additionally, the curve number
di nu
method can easily incorporate changing land uses and covers in hydrologic modeling.
The 5-min rainfall data, daily temperature and monthly evaporation data from a
ye a
op M
nearby weather station provided the model meteorological inputs. The selected storm
events used in model calibration and validation are presented in Table 4. The
C ted
2009 at the two sewersheds, CSO 027 and CSO 044, were provided by Onondaga
ot p
N ce
County consultants, CH2MHill. The data were used for model calibration and
represents the sanitary flow contribution only. To do this we adjusted the sanitary
hourly) of human water use. The adjustment factors were derived from the collected
5-minute sanitary inflow data during dry weather conditions. A full description of the
13
of uncertainties for inputs and outputs. GLUE uses a likelihood function to measure
the goodness-of-fit between simulated and observed data and produces credible limits
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
for model input parameters and simulated outputs for a given uncertainty level
t
ip
(Beven, 2006). A full mathematical description can be found in Beven and Binley
d cr
(1992), Lamb et al. (1998), Hong et al. (2005). The steps for the calibration and
te s
validation procedures are briefly described as follows.
di nu
A uniform prior distribution was assumed for each parameter over a sufficiently
wide range of parameter values (Table 5). We ran SWMM for the micro/macro CSO
ye a
op M
027 delineation using 25,000 sample parameter sets drawn from the prior distribution
using a Monte Carlo technique. From this the prior output distributions were
C ted
of the 25,000 simulations against total flows measured from each of the selected rain
ot p
N ce
events using the Gaussian distribution function. The selection of rainfall events was
based on the observed continuous hydrographs. The lag time between the selected events
Ac
had to be long enough so that the events could be considered independent from one another
(Table 4). In each instance the runoff from a previous event had returned to base flow,
meaning that the contribution of the direct runoff generated from the previous storm events to
the hydrograph was negligible when the following storm event started. The likelihood
function was then updated using Bayes’ theorem (Box and Tiao, 1992) when
14
(4)
where D denotes observed data (DQ for observed total flow volume and DP for
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
observed peak flow rate); θ denotes model parameter sets; θQ is simulated total flow
t
ip
volume and θP is simulated peak flow rate; SQ and SP is the standard deviation of
d cr
observed total flows and peak flow rates, respectively; i is numbered rainfall event
te s
and n is the total number of events. Note that the underlying assumption of applying
di nu
the Gaussian distribution function (Equation 4) is that the total flow and peak flow
data are independent of each other. The Gaussian likelihood function was used here in
ye a
op M
order to approximate the true form of posterior probability distributions under the
resampling (SIR) algorithm (Rubin, 1987, 1988) to generate the posterior probability
ot p
N ce
distributions for each inferred parameter of interest and the uncertainty bounds for
and flow uncertainty bounds for the micro- and macro- CSO 027 to test how
parameter values and model outputs would respond to the changing model scale.
of interest, we applied the first-order sensitivity method (Saltelli, 2002) at both model
scales. This method quantifies the interactions between input parameters and
identifies the relative influence of each input parameter on the model output
15
uncertainty (Saltelli et al., 2000). The fundamental equation used to compute the
fractional contribution (IX) of each model input parameter θX to the total variance of
(5)
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
The sum of IX closer to one indicates less interaction among the parameters. A higher
ip
IX value represents higher sensitivity, indicating greater influence on the model output
d cr
from changes in an individual parameter (Saltelli, 2004).
te s
Finally, we conducted two same-scale and two cross-scale model validation runs
di nu
to identify whether finer scale catchment parameterization produces better model
ye a
performance when transported to another catchment. The model validation was based
op M
on the assumption that DS and Manning’s n for impervious surfaces, tree and lawns,
C ted
KWidth and nConduit were the same from one catchment to the other. Two same-scale
validations include: the posterior parameters derived from the micro CSO 027 were
ot p
applied on the micro delineations of the second watershed CSO 044 to simulate the
N ce
flows, and the parameters from the macro CSO 027 applied on the macro CSO 044.
Ac
Conversely, parameters from the macro-scale CSO 027 applied on the micro CSO 044
and parameters derived from the micro-scale CSO 027 applied on the macro CSO 044
comprise the cross-scale model validations. The simulation results were validated
against the measurements taken at the CSO 044 outlet under storm events of varying
magnitude.
16
We obtained over 1300 posterior parameter sets with unique values for both
delineations using the SIR approach. This number was considered as a sufficient
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
ip
(Bates et al, 2000; Lamb et al. 1998; Xiong and O’Connor, 2008). The posterior
d cr
parameter distributions generated from the macro CSO 027 delineation were
te s
compared with those from the micro delineation in Figure 4. Overall, a greater degree
di nu
of updating from prior to posterior parameter distributions was observed in the micro
from the uniform prior to the posterior for DSIS, nIS, DSTree and nTree in the micro
delineation compared to the macro one, indicating that observations of flow data
C ted
better constrained the likely values of these parameters at the micro delineation. Since
the same likelihood measure and posterior sampling approach conditioned on same
ot p
N ce
parameters for CSO 027 at micro and macro scales, we argue that the difference
Ac
between estimated posterior parameters from the micro and macro scale models is due
primarily to the modeling scale. The posterior distribution for KWidth and nConduit
showed slight differences between the micro and macro delineation, indicating that
the scale change did not have considerable impact on these two parameters. This may
be because the original properties of the pipes were retained on both scales without
spatial aggregation and the subcatchment width was dependent on the corresponding
17
subcatchment area that was measured based on the GIS map. In addition, DSLawn and
nLawn at both scales showed little update from the prior distribution.
The first-order sensitivity analysis suggested that: 1) at both micro and macro
scales, DSIS was the most important parameter, followed by nIS in determining the
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
total flow (Figure 5). Relatively speaking, DSIS and nIS were more important in
t
ip
determining the total flow at micro scale than at macro scale; 2) the peak flow showed
d cr
greatest sensitivity to nConduit followed by DSIS at both scales; 3) the total flow showed
te s
very little first-order sensitivity (Ix < 4%) to DSTree and nTree at both scales, while
di nu
DSTree and nTree each accounted for a significant percentage (Ix > 9%) of the total
variation in the peak flow in the micro model. However, the peak flow of the macro
ye a
op M
model showed less response to these two parameters; and 4) the peak flow and total
flow at both scales exhibited little sensitivity to DSLawn or nLawn, suggesting that the
C ted
variability in these lawn parameters had almost no effect on the variability in total and
peak flow predictions. For the micro-scale model, the sum of IX of all parameters was
ot p
N ce
0.92 for the total flow and 0.88 for the peak flow. For the macro-scale model, the sum
of IX was 0.82 for the total flow and 0.89 for the peak flow. The IX value of 0.82 for
Ac
the total flow suggested that approximately 18% of the variability in the total flow
data was left unexplained by considering the first-order effects of the model input
parameters. With techniques described in Ratto et al. (2001), we also examined the
flow and peak flow for the micro and macro scale. The results (not shown here) did
not show a significant level of interaction between any parameters. Overall, the
18
parameter sensitivity results were consistent with the findings shown in Figure 4,
where the posterior distribution of the parameters with higher first-order sensitivity
t
ip
observations for 6 representative events of different magnitudes, for CSO 027 were
d cr
estimated from the prior parameters (labeled “Prior”), and the posterior parameters for
te s
the macro-scale (labeled “Macro”), and the micro-scale delineation (labeled “Micro”)
di nu
(Figure 6). The “Micro” and “Macro” flow uncertainty bounds were substantially
refined compared to the “Prior” uncertainty bounds (Figure 6A-E), indicating that the
ye a
op M
combined peak flow and total flow were informative in reducing model output
uncertainty at both scales. Comparing the “Micro” and “Macro” uncertainty bounds,
C ted
the “Micro” bounds were better constrained towards the observations while the wider
did not show any noticeable influence on the timing of peak flow occurrence. The
results suggested that, given the same observed data, the macro-scale model was less
Ac
informative in updating the “Prior” bounds to the posterior uncertainty bounds than
the macro-scale model. The only exception is shown in Figure 6F, in which both
“Micro” and “Macro” flow bounds fail to envelope the observations of the small rain
event (< 1.94 mm/hour), implying that the SWMM model has limitations in
19
based upon the micro- and macro-scale CSO 027 were first applied to the micro- and
macro-scale CSO 044 to generate the flow rates, respectively (i.e., same-scale
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
validation). Figure 7 shows the flow uncertainty bounds at α = 90%. Overall, the
t
ip
simulation uncertainty bounds estimated from the posterior parameters of the macro
d cr
CSO 027 delineation (labeled “Macro”) were wider than those computed from the
te s
posterior parameters of the micro CSO 027 delineation. The results indicated that the
di nu
employment of the posterior parameters calibrated from the macro-scale model led to
acceptable simulation results with the observations falling within the uncertainty
ye a
op M
bounds. However, the posterior parameters derived from the micro-scale model
modeling scale. We then applied the posterior parameters derived from the micro- and
macro-scale CSO 027 on the macro- and micro-scale CSO 044 (i.e., cross-scale
ot p
N ce
validation). We found that all uncertainty bounds generated from the posterior
parameters calibrated based upon the macro-scale CSO 027 led to a much greater
Ac
degree of uncertainty represented by wider flow bounds, while the flow uncertainty
bounds estimated from the posterior parameters of the micro CSO 027 narrowed
down the uncertainties of flow predictions at the macro-scale CSO 044 (Figure 8). In
order to determine why the posterior parameters derived from the micro-scale CSO
027 outperformed the ones from the macro-scale CSO 027, we computed the 25% and
75% credible interval and the mode value (as an alternative to the optimal parameter
20
value) for each posterior parameter distribution derived from micro and macro CSO
027 (Table 6). It indicated that the credible interval of DSIS and nTree generated from
micro-scale CSO 027 was much smaller than those from macro-scale CSO 027. As
shown in the sensitivity analysis (Figure 5), DSIS showed greatest influence on the
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
simulated peak flow rates. In this case, it may be the wider range of DSIS value
t
ip
derived from the macro CSO 027 that contributed largely to the greater uncertainty in
d cr
flow predictions. Hence we argue that the calibrated parameters obtained based upon
te s
the macro delineation might not be adequate to be applied to simulate another
di nu
sewershed delineated at a different scale, while the posterior parameters derived from
different catchment aggregation level to a much better degree and thus provide more
reliable predictions.
C ted
4. Conclusions
findings we concluded that the manner in which continuous geographic spaces were
Ac
disaggregated into discrete spatial units had a significant impact on both the model
parameterization and simulation results. The results showed that the parameter values
were quite sensitive to the modeling scales. Furthermore, both model calibration and
validation results showed that the posterior parameter sets calibrated based upon the
compared to those from the macro delineation. Accordingly, we argue that the
21
provide acceptable simulations for this specific site at the matching scale. However, it
might not be considered reliable to be used for simulating another site unique in its
framework at micro scale takes care of spatial connectivity between the homogeneous
t
ip
subcatchments via flow routing component, the posterior parameter distributions
d cr
derived from the micro delineation might better accommodate the variation in
te s
catchment characteristics, and provide higher confidence level in terms of parameter
di nu
transferability for modeling other sites. These findings are important for future
which case model parameters cannot be calibrated for better model fit or validated to
C ted
test its acceptance. In this instance, we recommend using the posterior parameter
distributions calibrated based upon a micro-scale model of a gaged site to model other
ot p
N ce
ungaged sites delineated on the same fine scale, although such parameters derived
from small-scale model may lose physical significance at larger scales. Furthermore,
Ac
the micro-scale model can easily incorporate changes in surface covers at a small
scale. This is significant for assessment of the potential runoff control benefits given
implementation, e.g., planting more trees and replacing impervious roofs with green
22
net utility of decentralized GI facilities at the parcel scale while the lumped models
Acknowledgements
0948952 for an Urban Long Term Research Area Exploratory project (ULTRA-EX)
t
ip
that supported and inspired this research. We also want to thank the Onondaga County
d cr
Water and Environment Program and CH2MHILL, Syracuse for the provision of
te s
monitored flow data.
di nu
ye a
op M
C ted
ot p
N ce
Ac
23
References
111-133.
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Bales, J., Betson, R. (1982). The curve number as a hydrologic index, 371-386, ed. V.
t
P. Singh. Rainfall-Runoff Relationships. Water Resources Publications,
ip
Highlands Ranch, CO.
d cr
te s
Barco, J., Wong, K.M., Stenstrom, M. K. (2008). Automatic calibration of the
di nu
U.S.EPA SWMM model for a large urban catchment. J. Hydr. Engrg., 134(4),
466-474.
ye a
op M
320(1-2), 18-36.
Beven, K. and Binley, A. (1992). The future of distributed models: model calibration
24
Box, G.E.P. and Tiao, G.C. (1992). Bayesian inference in statistical analysis. Wiley
Online Library.
Butts, M.B., Payne, J.T., Kristensen, M., Madsen, H. (2004). An evaluation of the
t
ip
Choi, H. T., Beven, K. (2007). Multi-period and multi-criteria model conditioning to
d cr
reduce prediction uncertainty in an application of TOPMODEL within the
te s
GLUE framework. J. Hydrol, 332(3-4), 316-336.
di nu
Chow, V. T., Maidment, D. R., Mays, L. W. (1988). Applied hydrology. McGraw-
hydrograph procedure: a case study in the East Toll Gate creek watershed.
C ted
Fang, T., Ball, J. E., 2007. Evaluation of spatially variable control parameters in a
ot p
N ce
Gironás, J., Roesner, L. A., Davis, J., Rossman, L. A. (2009). Storm Water
25
Guo, J.C.Y. and Urbonas, B. (2008). Consistency between CUHP and rational
Hong, B., Strawderman, R.L., Swaney, D.P., Weinstein, D.A. (2005). Bayesian
t
ip
estimation of input parameters of a nitrogen cycle model applied to a
d cr
forested reference watershed, Hubbard Brook watershed six. Water Resour.
te s
Res., 41(3), W03007.
di nu
Hong B, Swaney, D.P., Weinstein, D.A. (2006). Simulating spatial nitrogen dynamics
Huber, W. C., Heaney, J. P., Medina, M. A., Peltz, W. A., Sheikh, H., Smith, G. F.
(1975). Storm Water Management Model User's Manual, Version II. EPA-
ot p
N ce
Huber, W.C., Dickinson, R.E., Barnwell Jr, T.O. (1988). Storm water management
Ac
26
Lamb, R., Beven, K., Myrabų, S. (1998). Use of spatially distributed water table
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
t
ip
Resour., 22(4), 305-317.
d cr
Leavesley, G.H. and Stannard, L.G. (1990). Application of remotely sensed data in a
te s
distributed-parameter watershed model. Proc., Workshop on Applications of
di nu
Remote Sensing in Hydrol, 47–68.
Mamillapalli, S., Srinivasan, R., Arnold, J.G., Engel, B.A. (1996). Effect of spatial
Moore, I.D., and Gallant, J.C. (1991). Overview of hydrologic and water quality
MWH Soft Inc. (2005), InfoSWMM User Manual. MWH Soft Inc., Pasadena,
California.
27
t
ip
Ratto, M., Tarantola, S., Saltelli, A. (2001). Sensitivity analysis in model calibration:
d cr
GSA-GLUE approach. Comput. Phys. Commun., 136(3), 212-224.
te s
Rossman, L.A. (2010). Storm water management model user's manual, version 5.0.
di nu
National Risk Management Research Laboratory, Office of Research and
Rubin, D.B. (1988). Using the SIR algorithm to simulate posterior distributions.
Ac
28
Tsihrintzis, V. A., and Hamid, R. (1998). Runoff quality prediction from small urban
t
ip
hydrology for small watersheds. Technical release 55, 2nd ed., NTIS PB87-
d cr
101580. USDA SCS, Springfield, Virginia.
te s
Xiong, L., O’Connor, K. M. (2008). An empirical method to improve the prediction
di nu
limits of the GLUE methodology in rainfall–runoff modeling. Journal of
10(1), 39-49.
Zaghloul, N.A. (1981). SWMM model and level of discretization. Journal of the
ot p
N ce
29
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Copyright 2012 by the American Society of Civil Engineers
J. Hydrol. Eng.
Figure 4
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Copyright 2012 by the American Society of Civil Engineers
J. Hydrol. Eng.
Figure 5
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Journal of Hydrologic Engineering. Submitted March 28, 2012; accepted December 21, 2012;
posted ahead of print December 26, 2012. doi:10.1061/(ASCE)HE.1943-5584.0000777
Downloaded from ascelibrary.org by UNIVERSIDADE DE BRASILIA on 05/31/13. Copyright ASCE. For personal use only; all rights reserved.
Accepted Manuscript
Not Copyedited
Accepted Manuscript
Not Copyedited
Accepted Manuscript
Not Copyedited
The Sewer The map includes georeferenced sewerlines, Onondaga County consultants,
Network Map junctions and sewershed outlets in a vector format. CH2M Hill, Syracuse
Accepted Manuscript
Not Copyedited
Accepted Manuscript
Not Copyedited
Manning's
n 0.02~0.8 for pervious surfaces
nTree Manning's n for trees Uniform (0, 1)
(Huber et al., 1988; Huber
and Dickinson, 1992)
nLawn Manning's n for lawn Uniform (0, 0.5)
Depth of depression 0.012~0.1 (Huber et al., 1988;
DSIS storage of impervious Uniform (0, 0.5) Huber and Dickinson, 1992;
surfaces Tsihrintzis and Hamid, 1998)
Depression
Storage Depth of depression 0.1~0.2 for pervious surfaces
DSTree Uniform (0, 0.8)
(inch) storage of trees (Huber et al., 1988; Huber
Depth of depression and Dickinson, 1992;
DSLawn Uniform (0, 0.5) Tsihrintzis and Hamid, 1998)
storage of lawns
Fraction of the square
0.2~0.5 (MWH Soft Inc.,
KWidth root of subcatchment Uniform (0, 1)
2005)
width
0.011~0.017 (Huber et al.,
Manning's roughness
nConduit Uniform (0, 0.1) 1988; Huber and Dickinson,
coefficient for conduits
1992)
Accepted Manuscript
Not Copyedited
Table 6. The 25% and 75% credible interval and the mode value of the posterior parameter distribution
Accepted Manuscript
Not Copyedited