Sei sulla pagina 1di 8

34th IAHR World Congress - Balance and Uncertainty 26 June - 1 July 2011, Brisbane, Australia

33rd Hydrology & Water Resources Symposium


10th Hydraulics Conference

Probabilistic Hydrologic Simulation of Urbanized Catchments with


Sparse Data

J. Cantone1, Michelle Hollander2, and Arthur Schmidt2


1
CH2M-Hill
Chicago, IL 60606, USA
2
Department of Civil & Environmental Engineering
University of Illinois at Urbana-Champaign
Urbana, IL 61801, USA
E-mail: jcantone@ch2m.com

Abstract: Deterministic modeling of urban stormwater systems requires detailed data describing the
sewer network and areas contributing to each sewer. Records describing utility positions, when
available, are often incomplete and inaccurate with errors as high as 30%. Similarly, subcatchment
data are often unavailable, particularly at the scale of distributed models. Hence, urban hydrologic
modeling often requires either costly sewer mapping and field measurements, gross simplification, or
both. This paper uses the probabilistic Illinois Urban Hydrologic Model (IUHM), which predicts the
hydrologic response of urbanized catchments using the sewer layout and statistical properties of
sewers and subcatchments of each order. This allows random sampling, stratified by pipe order, to
produce the necessary model inputs, at a fraction of the effort and cost required to obtain data for
deterministic modeling. Example applications to catchments in the Chicago area are compared to
monitored flows for different levels of sampling.

Keywords: Hydrologic modeling, urban hydrology.

1. INTRODUCTION
The Illinois Urban Hydrologic Model (IUHM) is a probabilistically based approach for determining the
hydrologic response in highly urbanized catchments. IUHM was developed by Cantone and Schmidt
(2010a) at the University of Illinois at Urbana-Champaign to facilitate hydrologic and hydraulic
modeling of the plethora of combined sewer systems that contribute storm and sanitary flows to drop
shafts in MWRDGC’s Tunnel and Reservoir Plan (TARP) system. One of the primary reasons for
developing a new model, in preference to using an existing deterministic model like the widely
accepted and applied SWMM5 model, was the lack of input and calibration data needed to create a
more traditional all-pipe-all-subcatchment deterministic model of each combined sewer system. In
order to derive the input parameters required to run an IUHM model, it is not necessary to know the
hydraulic characteristics (e.g. diameter, length, slope etc.) of every conduit in the system or the
hydrologic characteristics (area, slope, imperviousness etc.) of every subcatchment contributing within
a given service area. The only information that is always required in an IUHM model is a plan view
describing the layout and connectivity of the sewer system being simulated. IUHM utilizes the structure
and layout of the sewer system, together with elements of the Strahler ordering scheme (Strahler,
1957), to breakdown the system into a number of conduit and overland states that define a discreet
number of possible flow paths through the system. The hydrologic and hydraulic characteristics in
each state are defined probabilistically using a mean and variance and may be determined using only
a sub-sample of the characteristics of all the conduits and subcatchments in the system.

2. ILLINOIS URBAN HYDROLOGIC MODEL (IUHM)


The IUHM is based on the hypothesis that the hydrologic response of a highly urbanized catchment is
inherently linked to the structure of the sewer system that conveys flow through the catchment to its
outlet. In the TARP service area, the sewer systems of interest are predominantly combined sewer
systems and the outlet is generally a conduit leading to a connecting structure that controls flow to the
interceptor and TARP sewer systems. A similar hypothesis to that utilized in IUHM was proven true for
natural watersheds by Rodriguez-Iturbe and Valdes (1979) and since that time has been successfully

ISBN 978-0-85825-868-6 1881 Engineers Australia


utilized in the development of the Geomorphologic Instantaneous Unit Hydrograph approach for
predicting the hydrologic response of natural watersheds. As explained by Cantone (2010) a number
of key differences between urban catchments and natural watersheds had to be addressed to IUHM
as a unique and robust model for predicting the hydrologic response of highly urbanized catchments.

IUHM is based on a probabilistic approach that uses the structure of the network and the Strahler
ordering scheme to determine the probability, Pw, of a drop of water taking one of a finite number of
possible flow paths through the system. Any given path w is made up of a series of states, assumed to
be either overland (xoi) or conduit (xci). Each state, in each order, has associated with it a length,
celerity and hydrodynamic dispersion coefficient that are used to derive the travel time through the
state A finite number of paths, 2Ω, are possible for a catchment of order, Ω. If w denotes a specified
path then the probability of a raindrop following that path is the probability of starting out in the initial
overland state times the probabilities of making successive transitions to conduits of higher order
along the path. This probability is combined with other characteristics of each path, such as length,
celerity and a hydrodynamic dispersion coefficient to determine the network impulse response function
for the catchment.

Conceptually the IUHM model assumes that rainfall initially falls on an overland flow region. This
overland region is split into an impervious and pervious portion where water is assumed to flow from
pervious to impervious regions. After accounting for depression storage (a spatially uniform initial
loss) and infiltration losses (using the Green and Ampt (1911) equation) the direct runoff from each
path is determined and assumed to feed an artificial gutter. The artificial gutter is used to drive flow to
an inlet/catch basin that controls flow into each conduit state. An ith-order conduit receives external
flow from the inlets for the ith-order overland regions and also inflow from lower order conduits.

2.1. Parameter Determination


IUHM assumes that each conduit in the system is fed by a corresponding overland flow region. Ideally,
delineation of the subcatchment/overland region contributing to each conduit is based on the inlets
surrounding the upstream junction of the conduit of interest. For each inlet contributing to the system a
subcatchment should be delineated to describe the area of flow contributing to the inlet. To do this for
every inlet contributing to an urban sewer system would be cumbersome and would counteract the
advantages that IUHM has over deterministic models. A significant advantage of IUHM is that mean
and variance of the various subcatchment input parameters for each order do not necessarily need to
be based on the full data set. As such it is possible to derive the overland region input parameters
using a sub-set of the full data-set. Note here that this sub-sampling is only used for generation of
input parameters relating to subcatchments in the basin. The entire network is still used to generate
transition and path probabilities and conduit region parameters, which are fundamental to the model.
Four different methods for delineating subcatchments and deriving the corresponding IUHM
parameters for each overland state were evaluated:
1. Manual delineation of subcatchments for every inlet in the sewer system.
2. Manual delineation for a sub-sample of randomly selected overland regions.
3. No delineation – IUHM parameters are based on neighboring or geometrically similar (same
order) catchments.
4. Automatic delineation using Thiessen Polygon method.

In natural watersheds a number of authors (Horton, 1945, Strahler, 1950, Yen and Lee, 1997) have
explored the use of a series of stream order laws in determining watershed characteristics. It was
hypothesized by Cantone (2010) that a similar set of ratios could be developed for urban sewer
systems and used to help determine IUHM parameters in basins that had limited or missing input data.

IUHM models were developed for five service areas in the Calumet TARP system conduit information
for every conduit in the system and by delineating subcatchments for every inlet in the system. It is
possible to use these IUHM parameters to derive a set of ratios for urban sewer systems similar to
those postulated by Horton (1945) and Yen and Lee (1997). These ratios, which are presented in
Table 1 are :
L
Conduit Length Ratio: RL = i for i=2, 3…Ω; (1)
Li−1

1882
Di
Diameter Ratio: RD = for i=2, 3…Ω; (2)
Di−1
Sc
Conduit Slope Ratio: Rsc = i for i=2, 3…Ω; (3)
Sci−1

and are based on the geometric mean of the ratios for each order within a catchment. It can be
identified from Table 1 that there is similarity in the ratios calculated across the catchments. A
consideration of only five service areas is not exhaustive but it does indicate that there may be
potential to use these ratios to generate IUHM parameters in situations where limited input data are
available. Using these ratios and knowing the mean conduit length, diameter, and slope in Ωth-order
conduits would allow the average conduit length, diameter, and slope in other ith-order conduits to be
determined.

Table 1 Ratios for Calumet TARP service areas

CDS- CDS- CDS- CDS- CDS-


Ratio MEAN STDEV RANGE
51 36L 20 17 33
RL 1.12 1.23 1.03 1.27 1.32 1.20 0.12 1.03 to 1.32
RD 1.58 1.90 1.51 1.52 1.48 1.60 0.17 1.48 to 1.90
RSc 0.75 0.49 1.47 0.98 0.20 0.78 0.48 0.20 to 1.47

Thiessen (Voronoi) polygons (Han and Bray, 2006) define individual areas of influence around each of
a set of points. Thiessen polygons are polygons whose boundaries define the area that is closest to
each point relative to all other points. They are mathematically defined by the perpendicular bisectors
of the lines between all points. In hydrology, Thiessen polygons are commonly used to compute the
areal average of rainfall that falls within a catchment from a series of nearby rain gages. In this
approach, it is hypothesized that Thiessen polygons can be used to represent the subcatchment area
associated with each junction in a combined or storm sewer system. In this way it is inherently
assumed that the inlets contributing in junctions within a sewer network are systematically placed
based on the location of other inlets within the system. Given that IUHM is based on structure and
layout of the conduits, it was recognized that the simplest approach would be to draw polygons based
on the location of all junctions with the sewer system being analyzed. In this way one subcatchment
would be drawn corresponding to each conduit in the system. Once the polygons are drawn, IUHM
subcatchment parameters can be determined in the same way as if the subcatchements were
manually digitized.

Fundamental to IUHM is the assignment of a pipe order to every conduit in the sewer system under
investigation. Once an order has been assigned to each conduit, the conduit region properties can be
determined from the available conduit attributes. The geometric network created should include any
characteristics of the sewer system available in the local sewer atlas. Such characteristics could
include conduit diameter, conduit length, conduit material, upstream and downstream invert elevation
and inlet locations. This information, if available, can be used directly to calculate the mean and
variance for the required conduit region inputs.

Based on an assessment of the local sewer atlases made available for Chicago and the 29 additional
municipalities in the Chicago Metropolitan area service area, conduit layout and inlet/manhole
locations were available for 93% of the municipalities, and pipe lengths could be determined from the
maps. Conduit diameters were available for at least some of the pipes for 86 percent of the
municipalities. However, upstream and downstream conduit invert information was missing for almost
all (75%) of the municipalities outside the City of Chicago and for those with invert information, it is
generally limited to a subset of the pipes. Of the seven municipalities with sewer invert data, one had
invert elevation data for approximately 80% of the pipes and the rest (including the City of Chicago)
had invert elevation data for between 30% and 60% of the pipes. One of the advantages of the IUHM
is that is not reliant on having the input data for every conduit in the system. A fraction of the full
deterministic dataset can be used to derive IUHM input parameters, such as the mean and variance of
conduit slope or diameter, without significantly affecting the predicted direct runoff hydrograph. This
provides the user with a number of options when input data are missing. In situations where there are
some data about conduit slope but not for every conduit, the mean and variance of the conduit slope

1883
in each order could simply be determined using the available information. In situations where no data
are available we propose to derive the conduit slope based on pipe diameter under typical design
assumptions.

Combined sewer systems were designed based on set of commonly accepted principles. For
example, the Concrete Pipe Design Manual (2002) published by the American Concrete Pipe
Association prescribes a simple methodology for designing storm sewers that revolves around the
following assumptions:
• Pipe size is selected based on Manning’s formula for the pipe flowing full:
• Design flow is calculated based on the rational method.
• A minimum velocity of 0.91 m/s (3 ft/s) is recommended for storm sewers
Utilizing these assumptions and Manning’s formula, it is possible to derive a relationship between
conduit diameter (D) and slope (S) that may be used to predict conduit slopes when it is unknown but
the conduit diameter is available. If we assume the pipe full flow velocity (V) to be the minimum
specified by the local design criteria and assume a Manning’s coefficient (n) for the pipe material then
we can derive an explicit relationship between slope (%) and diameter (m) (Kn is 1.0 for S.I. units and
1.49 for ft-lb units):
.
100 (4)
Equation 4 could be used to derive the slope for each conduit in the system based on the diameter of
the conduit. The sample mean and variance of the slopes in each order could then be calculated and
used in IUHM. In order to test the validity of this equation it was compared (see Figure 1) to the
observed relationship between diameter and slope in a number of catchments in municipalities for
which as-built slope information was available. For each municipality the median slope for a given
diameter was plotted on the y-axis against diameter on the x-axis, indicating that with the exception of
a few outliers the theoretical relationship between conduit slope and diameter provides a good
approximation to the relationship observed in Chicago and its suburbs.

1
Calumet City Chicago Harvey Dolton Theoretical
0.9

0.8

0.7
Conduit Slope (%)

0.6

0.5

0.4

0.3

0.2

0.1

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500
Conduit Diameter (mm)

Figure 1 Comparison of observed and theoretical relationship between conduit slope and diameter

3. EXAMPLE APPLICATIONS

3.1. CDS-51
The 5th-order Calumet Drop Shaft 51 (CDS-51) catchment located in the Village of Dolton (see Figure
2) a southern suburb of Chicago, IL. The catchment contributing to CDS-51 captures storm and
sanitary flows for a 316 hectare service area. The combined sewer network feeding CDS-51 collects
inflow from in excess of 800 inlets and conveys it to CDS-51 via a network of some 722 pipes ranging
in size from 150 mm (6 inch) to 2150 mm (84 inch). All of the flow in the CSO system ends up in an
2150 mm pipe.

1884
Figure 2--Schematic diagram of CDS-51 catchment.

In order to test the proposed methodology random input parameter sub-sets were generated for
eleven different degrees of sub-sampling ranging from 5% to 100% of the conduits and
subcatchments contributing to CDS-51. The mean and variance of the overland slope, conduit slope
and imperviousness for each order were determined for each percentage of subsampling. These
newly generated input parameters were used as inputs for the IUHM model, allowing a direct runoff
hydrograph (DRH) for each degree of sub-sampling to be predicted. All degrees of sub-sampling using
30% or more of the conduits and subcatchments converged around the DRH generated using all of
the available data and fall within the 95% confidence interval predicted by IUHM for the 100% sample.
On this basis, it could be could concluded that it is possible to accurately predict the DRH for the CDS-
51 catchment using as a sub-sample that includes as little as 30% of the available input data. This
represents a significant saving in the time that would be required to develop the input parameters
required to predict the hydrologic response of the catchment.

Three scenarios were simulated for multiple storms to evaluate the different approaches to apply
IUHM for catchments with sparse input data. One approach used input parameters generated for
based on subsampling 30% of the conduits and subcatchments; the second used the ratios described
in Table 1 to generate parameters based on mean values for the higest order in the catchment; and
the third method used parameters generated using subcatchments based on Thiessen polygons and

1885
conduit slopes based on Eq. 4. These parameter sets, along with parameters generated based on
100% of the conduits and subcatchments were used to simulate a series of storms that occurred in the
greater Chicago area. For brevity and clarity of presentation, results are presented for four of the
storms simulated, representing a variety of precipitation intensities and temporal distributions.

The Nash-Sutcliffe efficiency (NSE) coefficient (Nash and Sutcliffe, 1970) was used to compare the
predicted hydrographs for each method. Nash-Sutcliffe efficiencies can range from negative infinity to
1, with 1 indicating a perfect match between the predicted and observed data, while a value of zero
indicates that the model predictions are as accurate as the mean of the observed data. NSE’s were
used to assess the predictions for DRH and are tabulated in Table 2.

Table 2 Summary of volumes and NSE’s for CDS-51


Peak Time to
Runoff Volume discharge peak
3 3
Storm Method (10 m ) NSE (m3/sec) (hours)
All data 86.0 4.50 8.2
30% Subsample 85.0 1.000 4.47 8.2
April 2007
Ratios 93.5 0.988 4.93 8.4
Theissen and Eq. 4 82.3 0.997 4.25 8.5
All data 80.4 9.46 6.1
30% Subsample 79.5 1.000 9.32 6.1
July 2007
Ratios 87.4 0.986 10.34 6.1
Theissen and Eq. 4 76.9 0.994 8.78 6.0
All data 29.0 1.79 2.2
30% Subsample 28.7 1.000 1.78 2.2
August 2007
Ratios 31.5 0.979 2.10 2.3
Theissen and Eq. 4 27.7 0.994 1.67 3.0
All data 141. 6.03 14.2
30% Subsample 140. 1.000 5.95 14.2
January 2008
Ratios 154. 0.984 6.54 14.2
Theissen and Eq. 4 135. 0.995 5.72 14.2

3.2. CDS-20
The 161 hectare service area (see Figure 3) contributing to Calumet Drop Shaft 20 (CDS-20) is
located in the City of Chicago and contributes flow to the Calumet City Leg of the Calumet TARP
system. Combined sanitary and storm flows are collected by 624 inlets and conveyed through a 4th
order network of 509 combined sewers ranging in size from 203 mm (8 in) to 1372 mm (54 in).

As with CDS-51, IUHM parameters were developed for the same eleven degrees of subsampling
ranging from 5% to 100% of the conduits and subcatchments. All degrees of sub-sampling converged
around the DRH generated using all of the available data and fell within the 95% confidence interval
predicted by IUHM for the 100% sample. On this basis, it could be could concluded that it is possible
to accurately predict the DRH for the CDS-20 catchment using as a sub-sample that includes as little
as 5% of the available input data. This represents a significant saving in the time that would be
required to develop the input parameters required to predict the hydrologic response of the catchment.

1886
Figure 3--Schematic diagram of CDS-20 catchment.

Rainfall data for simulation of CDS-20 was obtained in 10-minute intervals from a nearby Illinois State
Water Survey gage. The USGS monitored flow into CDS-20 for 12 months. These data, provided by
the USGS were analyzed and four storms were selected and simulated to allow comparison between
the different methodologies to address sparse input data. The DRH’s for the storms simulated using
the different methods described above were compared with the DRH simulated using 100% of the
conduits and subcatchments. Nash Sutcliffe Efficiencies and runoff volumes were calculated and are
included in Table 3.

Tables 2 and 3 highlight that each of the methods developed predict a DRH that is similar to the DRH
predicted by IUHM using all the available data. Using a sub-sample of the data provides the closest
prediction to using all the data, followed closely by using Eq. 4 along with subcatchments determined
from Thiessen polygons. Using parameters based on the ratios shown in Table 1 provided the poorest
simulations.

1887
Table 3 Summary of volumes and NSE’s for CDS-20

Peak Time to
Runoff Volume discharge peak
Storm Method (103 m3) NSE (m3/sec) (hours)
All data 15.5 0.95 3.5
30% Subsample 15.0 0.988 0.79 3.6
5th April 2009
Ratios 18.4 0.796 1.66 3.6
Theissen and Eq. 4 17.0 0.965 1.10 3.7
All data 9.4 0.95 3.1
30% Subsample 9.1 0.981 0.65 3.1
13th April 2009
Ratios 11.2 0.758 1.31 3.1
Theissen and Eq. 4 10.4 0.964 1.07 3.2
All data 30.1 4.42 2.6
30% Subsample 29.1 0.992 3.91 2.6
27th April 2009
Ratios 35.8 0.881 5.72 2.6
Theissen and Eq. 4 33.1 0.983 4.87 2.7
All data 4.1 0.86 3.1
30% Subsample 3.9 0.982 0.78 3.1
30th April 2009
Ratios 4.8 0.784 1.21 2.5
Theissen and Eq. 4 4.4 0.968 0.92 3.1

4. ACKNOWLEDGMENTS
This research would not have been possible without ongoing funding and support from the
Metropolitan Water Reclamation District of Greater Chicago as part of the University’s work on
Chicago’s Tunnel and Reservoir Plan project.

5. REFERENCES
Cantone, J.P., Improved understanding and prediction of the hydrologic response of highly urbanized
catchments through development of the Illinois Urban Hydrologic Model (IUHM), Ph.D. Dissertation,
University of Illinois at Urbana-Champaign, 2010.
Green, W.H. and Ampt, G.A. (1911), Studies on soil physics, Part I, the flow of air and water through
soils, Journal of Agricultural Science, 4(1), 1-24.
Han, D. and Bray, M. (2006), Automated Thiessen polygon generation, Water Resources Research,
42(1), W11502.
Horton, R.E., (1945) Erosional development of streams and their drainage basins: hydrological
approach to quantitative morphology, Bulletin of Geological Society of America, 56, 275-370.
Nash, J. E. and Sutcliffe, J. V. (1970), River flow forecasting through conceptual models part I — A
discussion of principles, Journal of Hydrology, 10 (3), 282–290.
Rodriguez-Iturbe, I. and Valdes, J.B. (1979), The geomorphologic structure of hydrologic response,
Water Resources Research, 15(6), 1409-1420.
Strahler, A.N., (1950) Equilibrium theory of erosional slope approached by frequency analysis, American
Journal of Sciences, 248, 673-696.
Strahler, A.N. (1957), Quantitative analysis of watershed geomorphology, Eos Trans. AGU, 38(6), 913-
920.
Yen, B.C. and K.T. Lee, (1997) Unit hydrograph derivation for ungauged watersheds by stream-order
laws, Journal of Hydrologic Engineering, 2(1), 1-9.

1888

Potrebbero piacerti anche