Combined Sewer Overflow
(CSO) Model Validation Study

NHSA 002A

NHSA006A

NHSA 012A

0 4 8
RKhmond 004

4 S 12
Richmond 005

0 4 8 12
Richmond 012

0 4 8
R»chmond 021

/.

VM

\0 •

0 4 8 12
Richmond 031

4 8

Richmond 024

4 8 12
Richmond 025

0 4 8 12
Richmond 034

/

0 4 8 12
Richmond 03S

rtk	J\ Unil,!d States

Environmental Protection
Agency

June 2022
EPA-832-B-22-001


-------
Disclaimer

The U.S. Environmental Protection Agency (EPA) has designed the Combined Sewer Overflow (CSO)
Model for Small Communities as a tool to help small CSO communities reasonably estimate CSO volume
and occurrence. EPA is not mandating the use of this model under the 1994 CSO Control Policy or the
use of the presumption approach under the 1994 CSO Control Policy. This document is not itself a
regulation, nor is it legally enforceable. Rather, it provides a guide to the CSO Model that communities
may use in analyzing combined sewer systems and reasonably evaluating the presumption approach
criteria to design or estimate sewer overflow volume and/or occurrence. Communities, small or otherwise,
might find the model useful and should consult with their National Pollutant Discharge Elimination System
permitting authorities to determine whether it is appropriate for them to use the CSO Model for Small
Communities. Any mention of trade names, manufacturers, or products in this document does not imply
an endorsement by the United States Government or EPA.

Questions regarding this document should be directed to:

Mohammed Billah

U.S. EPA Office of Wastewater Management

1200 Pennsylvania Avenue NW

Washington, D.C. 20460

(202) 564-2228

Billah. Mohammed@epa. gov


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Contents

Abbreviations	iv

Overview	1

System Characterization and Monitoring Data	2

Effect of Timestep Aggregation on Observed Data	6

Effect of Timestep Aggregation on CSO Model Output	8

Accuracy of the Modified Rational Method in Quantifying Stormwater Runoff	10

Revised CSO Model—Runoff	13

Revised CSO Model—CSO	17

Predicting CSO Occurrence	17

Predicting Event Volume and Peak Flow Rate	20

Model Validation Conclusions	25

References	26


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Abbreviations

cso

combined sewer overflow

CSS

combined sewer system

DCIA

directly connected impervious area

EPA

United States Environmental Protection Agency

HWU

Henderson Water Utility

in.

inches

LTCP

long-term control plan

MG

million gallons

MGD

million gallons per day

min

minute

NHSA

North Hudson Sewerage Authority

NPDES

National Pollutant Discharge Elimination System

tc

time of concentration

iv


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Overview

The Combined Sewer Overflow (CSO) Model for Small Communities (hereafter referred to as the "CSO
Model") is a spreadsheet-based planning tool for small communities that want a simple approach to
estimating a CSO occurrence, as well as treated or untreated CSO volume over a 24-hour period, and
have limited resources to invest in more advanced CSO monitoring and modeling. The CSO Model may
also be used to estimate the CSO controls, either green or gray, needed to meet the presumption
approach criteria (i) or (ii) in designing a CSO long-term control plan (LTCP). The CSO Model is designed
for small CSO communities that have relatively simple combined sewer systems (CSSs). However, large
CSO communities, with populations of greater than 75,000, might find the CSO Model useful if they need
to update their existing models, or as a first step before using more expensive models. CSO communities
that have many CSO outfalls and complex systems can also use the CSO Model by breaking down their
CSS into sub-sewersheds based on receiving waterbodies and sewer infrastructure.

The CSO Model is based on a modified version of the Rational Method with a computational timestep of
15 minutes. Runoff response depends on sub-sewershed impervious area and time of concentration (tc).
Routing is performed using minimal, straightforward input, including dry weather flow definition, presence
of green or gray volume controls, and regulator capacity. For additional information about the model itself,
see the CSO Model User Guide.

As part of model development, the U.S. Environmental
Protection Agency (EPA) performed a validation study to
evaluate the CSO Model and determine changes to improve
its accuracy and usability. For the validation study, EPA used
data from six communities, 28 individual sub-sewersheds
with CSSs, and 2,302 CSO events. Given the variety of data
types available for model validation, EPA used multiple
approaches divided into two major phases of testing to
evaluate different aspects of the CSO Model. The first phase
of testing used a preliminary version of the CSO Model to
test its major components, such as its timestep and its use of
percent imperviousness as a runoff coefficient. The second phase of testing used the final version of the
model, also referred to as the Revised CSO Model, which EPA revised based on findings from the first
round of testing. The main objectives of the second phase of testing were to provide an evaluation of the
level of accuracy that could be expected of the final CSO Model and to illustrate different ways in which
the CSO Model could be used.

This document summarizes data compilation, model validation, and model improvements in the following
eight sections:

1.	System Characterization and Monitoring Data

2.	Effect of Timestep Aggregation on Observed Data

3.	Effect of Timestep Aggregation on CSO Model Output

4.	Accuracy of the Modified Rational Method in Quantifying Stormwater Runoff

5.	Revised CSO Model—Runoff

6.	Revised CSO Model—Overflow

7.	Model Validation Conclusions

8.	References

Validation Study Partners

EPA worked with six partner communities
for this validation study, including:

•	Elisabeth, New Jersey

•	Henderson Water Utility, Kentucky

•	North Hudson Sewerage Authority,
New Jersey

•	Omaha, Nebraska

•	Richmond, Virginia

•	Saco, Maine

1


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

System Characterization and Monitoring Data

Six CSO communities, which range in size and complexity, provided EPA with system characterization
information and flow monitoring data. EPA worked with community staff and their contractors to identify
individual sub-sewersheds that had sufficient monitoring data to be suitable for validating the CSO Model
EPA designed the CSO Model to simulate overflows from smaller systems (ideally less than 100 acres)
with low complexity (e.g., minimal interconnections with other sub-sewersheds, simple routing, no
tailwater effects). In total, EPA selected 28 sub-sewersheds across the six communities for various
validation steps depending on the types of data provided. Test sub-sewersheds range in size from 12 to
493 acres, with a median and average size of 88 and 146 acres, respectively.

Table 1 summarizes system parameters of each sub-sewershed, including measurements in million
gallons (MG) and million gallons per day (MGD).

Table 1. Characterization of CSO systems used for CSO Model validation.

Sub-sewershed/
CSO ID

Sub-basin
area
(acres)

Average
impervious
surface (%)

CSO hydraulic
control capacity
(MGD)

Total CSO
volume
control (MG)

Dry weather
flow rate
(MGD)

EPA Region 1: Saco, Maine

001

18

69%

11.4

0

1.107

EPA Region 2: Elizabeth, New Jersev3

001

439

58%

4.16

0.32

1.37

031

59.5

68%

3.00

0

0.24

036

210

46%

6.62

0

1.06

039

245

69%

21.5

0

0.70

040

34.9

63%

2

0

0.24

EPA Region 2: North Hudson Sewerage Authority, New Jersey

002A (H1)

276

69%

0.1

0

0.16

006A (H5)

151

98%

2.4

0

0.67

012A (18PS)

85.9

47%

1.3

0

0.030

015A (W5)

36.9

80%

1

0

0.035

EPA Region 3: Richmond, Virginiab

004

91.6

22%

1.4

0

0.22

005

11.8

31%

7.5

0

0.02

012

90.0

17%

1.3

0

0.17

014

394

36%

60

0

0.87

021

493

30%

20

0

0.26

024

197

14%

2.5

0

0.23

025

65.7

21%

2.4

0

0.18

026

101

15%

1.6

0

0.15

031

176

21%

13

0

0.35

034

63.0

35%

12.8

0

0.42

035

30.9

32%

5.2

0

0.16

039

174

22%

1.9

0

0.38

2


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Sub-sewershed/
CSO ID

Sub-basin
area
(acres)

Average
impervious
surface (%)

CSO hydraulic
control capacity
(MGD)

Total CSO
volume
control (MG)

Dry weather
flow rate
(MGD)

EPA Region 4: Henderson Water Utility, Kentucky0

003 - Ragan St.

347

20%

2.7

0

1.39

004 - Jackson St.

43.4

38%

5.6

0

0.17

007 - Powell St.

27

26%

3.4

0

0.11

EPA Region 7: Omaha, Nebraskad

110

72.0

50%

Inflow only

NA

0.055

114

80.3

25%

Inflow only

NA

0.039

203

70.5

44%

Inflow only

NA

0.14

a Hydraulic control capacity is inferred from maximum observed regulator flow, or the difference between observed inflow and
overflow. Dry weather flow is calculated as the average regulator flow for time steps in which no rain had occurred for at least
three hours.

b Contractor provided impervious surface in the form of directly connected impervious area (DCIA). Therefore, EPA made no
corrections for larger sub-catchments.

0 CSO hydraulic control capacity is estimated based on values provided in the 2009 LTCP and known up-sizing of pipes to 18
inches. Dry weather flow is estimated using sub-basin acres provided by Henderson Water Utility's contractor and the method
used in the 2009 LTCP, which allocated peak dry weather flow of 11.5 MGD to sub-basins by area. Average dry weather flow is
assumed to be 50 percent of peak dry weather flow.

d Dry weather flow calculated as sum of 1) inflow and infiltration, 2) sanitary flow, and 3) commercial/industrial flow, as provided
by contractors in Omaha, Nebraska.

EPA determined model inputs for the test sub-sewersheds using a variety of approaches that depended
on the type of data provided by each community. In some cases, communities provided model inputs
directly, whereas other communities provided spatial data and system reports that were used to define
model inputs. In addition to files provided directly by the communities, the following resources—with URLs
provided where available—were used for various aspects of system characterization:

•	City of Elizabeth System Characterization Report (Mott MacDonald, 2019).

•	North Hudson Sewerage Authority Selection and Implementation of Alternatives for the Adams
Street Wastewater Treatment Plant (Jacobs Engineering Group, 2020).

•	Richmond VA Wastewater Utility Website (City of Richmond, 2022).

•	Henderson Water Utility Long Term Control Plan webpaqe and report (HWU and Strand
Associates, Inc., 2009).

•	City of Omaha Long Term Control Plan (City of Omaha. 2014).

EPA obtained CSO hydraulic control capacity (i.e., regulator capacity) directly from community personnel
when possible. For many sub-sewersheds, these data were not available and EPA either estimated
regulator capacity from design drawings or inferred it through analyzing the monitoring data. Similarly,
when not provided directly, EPA estimated dry weather flow rate from design reports or calculated it from
the monitoring data by averaging data for total inflow during periods three hours before or three hours
after any recorded rainfall.

In addition to the basic characterization data provided in Table 1, EPA needed sewer network layout and
elevation data to calculate tc within the CSO Model. When not provided directly, EPA used network
layouts in shapefile format, as well as publicly available digital elevation models, to identify the longest
flow path length, upstream elevation, and downstream elevation. Table 2 presents these data.

3


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Table 2. tc inputs.

Sub-
sewershed/CSO
ID

Length of
longest flow
path (feet)

Elevation at
upstream end
of main flow
path (feet)

Elevation at
downstream
end of main
flow path (feet)

Slope (%)

tc (hour)

EPA Region 1: Saco, Maine

001

<1000

NA

2-4%

0.25

EPA Region 2: Elizabeth, New Jersey

001

7281

33.6

10.9

0.3%

1

031

4644

28.6

12.5

0.3%

0.75

036

4725

41.9

24.1

0.4%

0.75

039

5798

29.2

11.5

0.3%

1

040

2928

21.1

7.9

0.5%

0.5

EPA Region 2: North Hudson Sewerage Authority, New Jersey

002A (H1)

8269

Topography indicates minimal

slope; no data available on
subsurface pipes. Assume 0.5%
slope.

0.5%

1

006A (H5)

5018

0.5%

0.75

012A (18PS)

4671

0.5%

0.75

015A (W5)

2660

0.5%

0.25

EPA Region 3: Richmond, Virginia

004

tc estimated directly by contractor to Richmond, VA.

0.25

005

0.25

012

0.25

014

0.5

021

0.5

024

0.25

025

0.25

026

0.25

031

0.25

034

0.25

035

0.25

039

0.25

EPA Region 4: Henderson Water Utility, Kentucky

003

8925

367

354

0.5%

1

004

4536

426

354

1.6%

0.5

007

2217

423

354

3.1%

0.25

EPA Region 7: Omaha, Nebraska

110

4524

1168

968

4.4%

0.5

114

6590

1204

958

3.7%

0.75

203

5167

1242

1100

2.7%

0.5

4


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Communities also provided rainfall and flow monitoring data in a range of formats and temporal
resolutions. Accordingly, EPA performed different types of model validations for each community,
supported by the available data. For example, Elizabeth, New Jersey, provided data on continuous
rainfall, runoff, and overflow time series. EPA used this data to compare modeled hydrographs to
observed hydrographs at multiple points within a single sub-sewershed, allowing for detailed evaluation of
runoff response and overflow hydrographs produced by the CSO Model. In addition, North Hudson
Sewerage Authority (NHSA) provided data on rainfall and the number of CSO events per month, which
allowed for an evaluation of the CSO Model's ability to predict the presence or absence of a CSO event
from three years of historic rainfall patterns. Table 3 provides a characterization of the monitoring data
available from each community, as well as the number of individual storm events used for validation
purposes.

Table 3. Characterization of storm events used for CSO Model validation.

Community

Number

of
basins

Number

of
events

Rainfall

per
event
(in.)

Rainfall data
description

Runoff data
description

Overflow data
description

Saco, ME

1

22

0.31-4.3

15-minute timestep/
per event

N/A

Event size (MG)
and magnitude
(MGD)

Elizabeth, NJ

5

7

0.31-1.9

5-minute timestep/
continuous time
series

5-minute
timestep/
continuous time
series

5-minute timestep/
continuous time
series

North Hudson
Sewerage
Authority, NJ

4

259

CO
I

o

15-minute timestep/
continuous time
series

N/A

Number of
overflows per
month

Richmond, VA

12

79

CM

c\i
I

o

15-minute timestep/
continuous time
series

N/A

15-minute
timestep/
continuous time
series

Henderson
Water Utility,
KY

3

78

0.05-2.3

15-minute timestep/
continuous time
series

N/A

5-minute timestep/
continuous time
series

Omaha, NE

3

9

0.13-1.6

60-minute timestep/
continuous time
series

15-minute
timestep/
continuous time
series

N/A

5


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Effect of Timestep Aggregation on Observed Data

CSO events, especially in small communities, can be caused by fast, intense storms that often last under
one hour. Models that average rainfall and runoff response over an hour or longer may underestimate
short-term events, or peak flows, which can ultimately lead to an underestimation of CSO volumes.
Conversely, many communities do not have access to rainfall data that are recorded more often than
every 15 minutes, or even every hour. For the CSO Model, it is therefore important to incorporate a
simulation timestep that balances model accuracy with data availability.

To test how the timestep can produce sufficiently accurate flow data, EPA reproduced the original data at
different levels of aggregation using Elizabeth and Omaha as test cases. Elizabeth sub-sewersheds
provided data at a five-minute timestep, so EPA produced 15- and 60-minute aggregations. Omaha sub-
sewersheds provided data at a 15-minute timestep, so EPA only produced 60-minute aggregations.
Figure 1 illustrates a selection of these comparisons.

Results in Figure 1 demonstrate how data displayed on a 60-minute timestep can reduce peak flows by
up to 50 percent. The top tile, from sub-sewershed 114 in Omaha, clearly illustrates the difference
between 15-minute and 60-minute levels of aggregation. Results from sub-sewersheds 031 and 036 in
Elizabeth show that differences between a five-minute and 15-minute timestep are minimal, most likely
because the tc for each of these sub-sewersheds is greater than 15 minutes. These conclusions were
generally consistent across all sub-sewersheds and events.

6


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Basin 114, Event 4: Effect ofTimestep on Calculation of Observed Flow

Stormwater - 0_60
Stormwater - 0 15

o

I5



	***

• *

I ; * /

; • i /

\* ; ;



H IA

v V ,

/

/

/

/ I

* \ 1 ;

V

v

•V v

	• % 1



/

/

/





	nr, — .. m «.

Q » 	' •' ¦	I	I	I	I	I	'	'	'	I	I	I	I	I

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Event Hour

Basin 114, Event 4: Effect ofTimestep on Calculation of Observed Flow

12

10

8

o

<3

a 6

|
u_

4

2
0

1

Basin 036, Event 5: Effect ofTimestep on Calculation of Observed Flow

35

: A

§ 20
5

Figure 1. Example illustrations showing effect of timestep on observed flow calculation.



Stormwater - O_60
Regulator - O_60
CSO-O 60





Stormwater - 0_15
Regulator - O 15
CSO - 0_15

	Stormwater - 0_5

Regulator-O 5



1 • M <

'SI in

i

* '
/ :

•n | Itl

' ;
' :
i j

A * / •!

/\ # / »1

JL /m

// j/



	

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Event Hour

	Stormwater - O_60

Regulator - O_60
CSO -O_60
Stormwater - 0_15
Regulator - 0_15
CSO -0_15

	Stormwater - 0_5

	Regulator - Q_5

7


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Effect of Timestep Aggregation on CSO Model Output

Based on the results described above, EPA created a revised version of the CSO Model using a 15-
minute timestep and compared it with model output that used a 60-minute timestep, again using Elizabeth
and Omaha as test communities. EPA ran each version for each Elizabeth sub-sewershed listed in Table
1 and the associated events listed in Table 3. EPA only simulated Omaha sub-sewersheds using a 15-
minute timestep. For Elizabeth sub-sewershed simulations, EPA aggregated the original five-minute
rainfall data to 15-minute averages. For Omaha sub-sewershed simulations, EPA distributed the original
60-minute rainfall data evenly across each 15-minute interval of each simulation hour. In other words, for
an observed record of 0.1 inches over one hour, EPA used a model input of 0.025 inches per 15 minutes
instead.

Figure 2 illustrates results for the same sub-sewershed and event combinations used in Figure 1. In the
legend, "M" and "O" are used to denote modeled and observed, respectively. For sub-sewershed 114,
observed results are at a 15-minute timestep, while observed results for sub-sewersheds 031 and 036
are at a five-minute timestep.

Again, results show that simulation on a 60-minute timestep results in a significant loss of detail in terms
of peak flow rate prediction. Results from sub-sewersheds 031 and 036 show that by decreasing the
model timestep from 60 minutes (M_60) to 15 minutes (M_15), the ability to reproduce the timing of the
peak flows is improved. In other words, the timing of runoff response appears to be as dependent on
model timestep as on tc.

Although reducing the timestep from 60 to 15 minutes improves the detail and timing of model outputs,
model accuracy still has limitations. First, the top tile in Figure 2 shows that although model timestep
improves runoff response detail, certain hydrograph peaks are not reproduced due to differences
between the resolution of rainfall data input (hourly) and actual rainfall variability. The ability to reproduce
any fluctuations in flow due to fluctuations in rainfall at less than an hourly timestep is limited by using an
hourly average rainfall input.

Next, results from sub-sewershed 036, which is 210 acres in size, show that even with a 15-minute
timestep, peak flows are overestimated, sometimes by a factor of two or more. Conversely, results from
sub-sewershed 031, which is just 60 acres, are reasonably accurate. A qualitative review of results
across other simulations shows the same pattern, whereby simulation results for larger sub-sewersheds
(e.g., greater than 100 acres) are much higher than observed results. This difference is due to the
interaction of impervious area, sewershed size, and runoff response, and is evaluated quantitatively in the
next section.

8


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Basin 031, Event 5: Preliminary Modeled (M) vs. Observed (O) Flow

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Event Hour

Basin 036, Event 5: Preliminary Modeled (M) vs. Observed (O) Flow

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Event Hour

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Event Hour

Stormwater- M_60
Regulator - M_60
CSO - M_60
Stormwater - M_15
Regulator - M_15
CSO - M_15
Stormwater- 0_5
- Regulator - O 5

60
50

_ 40

o
u

30

5
o
u-

20

	Stormwater - M_60

Regulator - M_60
CSO - M_60
Stormwater - M_15
Regulator - M_15
CSO - M_15

	Stormwater - 0_5

	Regulator - Q_5

Basin 114, Event 4: Preliminary Modeled (M) vs. Observed (O) Flow

	Stormwater - M_60

Stormwater - M_15
	Stormwater - O 15

S" 10

— 8
5

O

u. 6

Figure 2. Example illustrations of preliminary results using hourly (M_60) and 15-minute (M_15)
models.

9


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Accuracy of the Modified Rational Method in Quantifying Stormwater
Runoff

EPA designed the CSO Model to quantify the CSO volume that results from a given storm event. While
depicting realistic hydrographs is an important factor, CSO volume is the main model output. As shown
above, however, there are instances where the 15-minute timestep still appears to be over- or under-
predicting stormwater runoff, which directly affects the model's ability to quantify CSOs. Because CSOs
are so closely related to wet weather flow, EPA first evaluated the accuracy of the modified Rational
Method in quantifying stormwater runoff in detail. Again, EPA used data from Elizabeth and Omaha, the
two communities that provided runoff data (Table 3).

The Rational Method is not recommended for larger basins. Often, stormwater practitioners cite 200 acres
as a hard cutoff, though the actual cutoff can be much smaller and variable depending on the desired
degree of accuracy and site-specific conditions (Thompson, 2006). In addition, the appropriateness of
impervious area alone as a runoff coefficient surrogate is questionable at larger scales. To evaluate the
predictive power of the modified Rational Method across all available monitoring records, EPA
aggregated total flow volume and peak flow rate over each 24-hour simulation period and compared to
modeled results. Linear regressions were fitted through each sub-sewershed data set, using observed
data as the predictor. Equations for each data set help show the degree to which volumes or flow rates
are overpredicted (slope >1, assuming an intercept of 0) or underpredicted (slope <1, assuming an
intercept of 0).

Figure 3 shows 15-minute modeled results plotted against observed results. Results are separated
according to sub-sewershed size and total volume or peak flow rate. The tiles on the left show results for
sub-sewersheds smaller than 100 acres, while the tiles on the right show results forsub-sewersheds
larger than 100 acres. The top two tiles display 24-hour volume totals, while the bottom two tiles display
peak flow rates observed over the 24-hour simulation period.

Forsub-sewersheds smaller than 100 acres, slopes for total runoff volume (top left) range from 0.5 to 1.4,
with an average of 0.9. Sub-sewershed 040, which has the smallest slope but largest intercept, is tidally
influenced. This tidal influence has a noticeable effect on flow records, especially for smaller events, as
higher tailwaters limit the ability of pipe networks to convey stormwater. Figure 3 illustrates this effect, with
smaller events being relatively more overpredicted than what was observed, resulting in a larger intercept
and flatter slope than would be expected without tidewater effects.

For sub-sewersheds greater than 100 acres, slopes for total runoff volume (top right) range from 1.4 to
2.9, with an average of 2.1. In other words, the CSO Model overpredicted runoff volume for these sub-
sewersheds by an average factor of approximately two. In larger watersheds, a directly connected
impervious area (DCIA) is often a more appropriate indicator of runoff-generating potential than total
impervious area (Sutherland, 1995). However, DCIA is more complex than impervious area, as it refers to
impervious areas directly connected to stormwater drainage infrastructure, and can therefore be hard to
measure at the landscape scale. A set of equations exists to calculate DCIA from impervious area,
referred to as the "Sutherland Equations" (Sutherland, 1995). For high-density land uses, the equations
predict DCIA to range from 25 to 40 percent of total impervious area. As a rough approximation, if DCIA
of sub-sewersheds 001, 036, and 039 were 40 percent of the current impervious area and used as model
input instead of the values shown in Table 1, the average slope of the resulting regressions would likely
be closer to 1.

Peak flow rate results, as the bottom two tiles of Figure 3 illustrate, are similar to total volumes but with
greater variability, especially for smaller sub-sewersheds (bottom left). The slopes forsub-sewersheds
110 (0.2), 114 (0.5), and 203 (0.1) are all well below 1, indicating considerable underprediction of peak
flow. However, these simulations use average hourly rainfall data, which dampens sub-hourly fluctuations
in actual rainfall patterns and limits the ability of the revised CSO Model to capture that variability. This
effect is illustrated for the M_15 series in the top tile of Figure 2, where—despite having a shorter model

10


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

timestep—peak flows were still underpredicted by about half (7.6 MGD observed, 4.2 MGD modeled),
owing to the use of hourly rainfall. By comparison, observed and modeled total volumes for that same
event were closer (0.81 MG observed, 0.72 MG modeled).

Figure 3 also shows that, in almost all cases, linear regressions result in positive y-intercepts due the
current version of the CSO Model not including initial abstraction, or the initial volume ofwaterthat must
be "abstracted" before runoff is generated. Initial abstraction is the result of factors like vegetation
interception and small depressional storages (e.g., parking lot puddles) scattered throughout a
watershed. Based on the results in Figure 3, initial abstraction has an appreciable effect on modeled flow
rates, particularly for small events. EPA compiled the results separately to determine initial abstraction by
regressing observed runoff depth (inches) to observed rainfall depth (inches) for each sub-sewershed.
Intercepts of the resulting linear regression equations provide an estimation of initial abstraction across
the events considered. For the sub-sewersheds included in this study, initial abstraction ranged from 0.1
to 0.19 inches, with an average of 0.14 inches.

11


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

2.0

1.5

13

Stormwater Runoff: 24-Hour Total (MG)
Basins < 100 acres

y = 1.0x + 0.3 •
•

y = 1.0x + 0.2

1.0

Q>
"O

o

0.5

0.0

y = 1.4x + 0.2 #•;.

• f;

¦its''

f- y = 0.5x + 0.5
y = 0.7x+ 0.1



0.0	0.5	1.0	1.5

Observed (MG)

14
12
10

•	031 £.8

•	040 1

"8 6

•	110 5

•	114

•	203

2.0

Stormwater Runoff: 24-Hour Total (MG)
Basins > 100 acres
y = 2.9x+ 1.2.

y = 1.4x+ 1.1

• /* •

/ • /*
y = 2.1x+ 1.0

«•

_j	i	

•	001
036

•	039

0 2 4 6 8 10 12 14
Observed (MG)

20
18
16
14

Q

u 12
2

J 10

a>

"8 8
5

6
4
2
0

Stormwater Runoff: Peak Flow (MGD)
Basins < 100 acres

y = 2.Ox -1.0

*•

	

• • •
	*



y = 0.6x + 4.4

y = 0.1x+ 7.0

y = 0.2x + 3.7 •

• y = 0.5x+ 0.8

Stormwater Runoff: Peak Flow (MGD)
Basins > 100 acres

120

100

o 80

"g 60

o

5 40

20

y = 2.8x + 4.4 •

y = 2.3x+5.9

y = 1.7x+ 2.6



5	10	15

Observed (MGD)

20

•	001
036

•	039

20 40 60 80 100 120
Observed (MGD)

Figure 3. Summary of modeled (15-minute) versus observed runoff for Elizabeth and Omaha sub-
sewers heds.

12


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Revised CSO Model—Runoff

Based on the results presented above, the 15-minute model was further updated to address identified
shortcomings. The revised CSO Model includes the following updates relative to the original CSO Model:

•	Simulation timestep of 15 minutes, reduced from 60 minutes.

•	Ability to use hourly or 15-minute rainfall time series as input.

•	Recommendation that sub-sewersheds greater than 100 acres use a value of "0.5*impervious
area" as model input.

•	Incorporation of initial abstraction, set at a default of 0.1 inches, with the ability to enter a custom
value when known.

o Within the revised CSO Model, initial abstraction modifies the rainfall time series so the
first 0.1 inches (or other, if custom value is used) of rainfall is effectively removed.

Figure 4 illustrates results for the same sub-sewershed and event combinations used in Figure 1 and
Figure 2. The leading edge of the first hydrograph peak shows the effect of incorporating initial
abstraction. For each simulation, initial abstraction results in a more realistic lag between rainfall and
runoff initiation. In each case (and across other simulations not shown), the observed lag is greater than
the modeled lag, indicating that actual initial abstraction may be greater than 0.1 inches. Additionally,
qualitative review of event hydrographs shows that initial abstraction may be "recharged" multiple times
within a 24-hour time period. In other words, the storage that contributes to initial abstraction (e.g.,
interception, small depressional storages) can dry out in less than 24 hours. However, this is a highly
variable process and depends on local weather conditions such as temperature and humidity. The CSO
Model assumes that initial abstraction only occurs once during each simulation period as input of
additional weather data and is beyond the scope of the CSO Model.

The bottom tile of Figure 4 contains a comparison of the 15-minute timestep (M_15) and revised model
(M_15_imp), which shows the effect of using "0.5*impervious area" as input for sub-sewersheds greater
than 100 acres. As shown, this input achieves much better agreement between revised (M_15_imp) and
observed (0_5) runoff results.

13


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Basin 036, Event 5: Improved Modeled (M) vs. Observed (O) Flow

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Event Hour

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Event Hour

60
50
40

o

is

— 30

3

o

U-

20

	Stormwater - M_15

Regulator - M_15
CSO - M_15

Stormwater- M_15_imp
Regulator - M_15_imp
CSO - M_15_imp

	Stormwater - 0_5

Regulator - Q_5	

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Event Hour

Basin 114, Event 4: Improved Modeled (M) vs. Observed (O) Flow

Basin 031, Event 5: Improved Modeled (M) vs. Observed (O) Flow

Stormwater - M_15
Stormwater- M_15_imp
Stormwater-O 15

	Stormwater - M_15

Regulator - M_15
CSO - M_15

Stormwater - M_15_imp
Regulator - M_15_imp
CSO - M_15_imp

	Stormwater - 0 5

	Regulator-O 5	

Figure 4. Example illustrations of 15-minute (M_15) and revised (M_15_lmp) model results.

14


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Figure 5 shows revised CSO Model results for runoff plotted against observed results for all monitoring
events, similar to Figure 3. Table 4 compares regression statistics, including slope and y-intercept,
between revised CSO Model results (i.e., 15-minute timestep, initial abstraction of 0.1 inches, and
0.5*impervious area for basins >100 acres) and 15-minute CSO Model results. As summarized in Table
4, the adjustments made to the 15-minute CSO Model, including incorporating initial abstraction and
modifying impervious area input for large sub-sewersheds, results in improved accuracy.

Incorporating initial abstraction reduced the average y-intercept from 0.65 to 0.25 for total volume
regressions and from 3.0 to 0.94 for peak flow rate regressions. These values suggest that using a
default initial abstraction of 0.1 inches is a significant improvement, but may still be an underestimation.
Incorporating a modified impervious area input for larger sub-sewersheds reduced the average slope
from 1.4 to 0.87 for total volume regressions and from 1.4 to 0.95 for peak flow rate regressions.

Table 4. Comparison of the 15-minute and revised CSO Model results for the prediction of
stormwater runoff.

Sub-
basin
ID

Number
of events

Sub-basin
area
(acres)

Initial

Total volume

Peak flow rate

SI

ope



SI

ope



abstraction
(in.)a

15-min

Revised

15-min

Revised

15-min

Revised

15 min

Revised

Elizabeth, New Jersey

001

7

439

0.14

2.9

1.4

1.2

0.38

2.8

1.4

4.4

0.85

031

7

59.5

0.14

1.0

1.0

0.31

0.20

0.59

0.64

4.4

3.6

036

7

210

0.13

2.1

0.77

0.95

0.21

2.3

0.83

5.9

1.6

039

7

245

0.12

1.4

0.69

1.1

0.33

1.7

0.95

2.6

-1.66

040b

7

34.9

-0.015

0.50

0.50

0.49

0.43

2.0

2.1

-1.0

-1.9

Omaha, Nebraska

110C

9

72.0

0.18

1.0

1.0

0.21

0.11

0.21

0.21

3.7

3.5

114°

9

80.3

0.19

0.71

0.71

0.11

0.059

0.51

0.53

0.76

0.56

203c

9

70.5

0.10

1.4

1.4

0.18

0.10

0.13

0.13

6.7

6.7

Average:

1.4

0.87

0.62

0.25

1.4

0.95

3.0

0.94

a Calculated from regression of observed runoff volume to observed rainfall volume.
b Observed data wastidally influenced.

0 Hourly rainfall data.

15


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

2.0

Stormwater Runoff: 24-Hour Total (MG)
Basins < 100 acres

y = 1.0x+ 0.2

1.5



 100 acres

2.0

y = 1.4x + 0.4
.•

•/	y = 0.7x+ 0.3

¦0

• .

y = 0.8x+ 0.2

•	001

•	036

•	039

2 4 6 8 10 12 14
Observed (MG)

20
18
16

~ 14

o

^ 12

"a 10

aj

Stormwater Runoff: Peak Flow (MGD)
Basins < 100 acres

y = 0.6x+ 3.5

y = 2.1x-1.9

1

• ..-•
a.

y = 0.2x + 3.5 •

^	y = 0.5x + 0.6

120

100

S80

y = 0.1x + 7.0 -o go

	...w	oi

40

20

Stormwater Runoff: Peak Flow (MGD)
Basins > 100 acres

y = 1.4x + 0.9

y = 0.8x+ 1.6

y = 1.0x-1.7

5	10	15

Observed (MGD)

20

•	001
036

•	039

20 40 60 80 100 120
Observed (MGD)

Figure 5. Summary of modeled (revised CSO Model) versus observed runoff results for Elizabeth
and Omaha sub-sewersheds.

16


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Revised CSO Model—CSO

Using the revised CSO Model, EPA compared model predictions to observations using two main
approaches that depend on the type of data provided by the test communities. The different approaches
also demonstrate the different ways that the CSO Model may be used, including:

•	Predicting CSO occurrence: For communities with minimal knowledge of their combined sewer
system, simply knowing whether a CSO occurred for a given storm can be helpful. Here, we ran
the CSO Model using a full year of rainfall data for Richmond, Virginia, and Henderson Water
Utility (HWU) in Kentucky, as well as three years of rainfall data for NHSA in New Jersey (for a
total of 2,218 individual simulations), to evaluate the ability of the CSO Model to predict the
number of events that would occur during a given month.

•	Predicting CSO volume (MG) and peak flow rate (MGD): Using the model as a typical
community would use it, we ran simulations of 21 sub-sewersheds across four communities (for a
total of 1,239 simulations) to evaluate the ability of the CSO Model to predict the total volume and
peak flow rate of a CSO event.

Predicting CSO Occurrence

The communities of NHSA, Richmond, and HWU each provided EPA with continuous time series rainfall
data and some measure of CSO occurrence, including either outfall flow time series data or monthly
overflow reports. To simulate monthly events, EPA ran the CSO Model for all sub-sewersheds on days
when the total rainfall exceeded 0.1 inches, which is equivalent to the default initial abstraction.

For each community, EPA ran multiple rounds of simulations using different inputs for initial abstraction
and impervious surface area based on the preliminary findings discussed earlier in this report. For NHSA
and HWU, results of these iterations showed better agreement between modeled and observed results
when using an initial abstraction value of 0.2 inches instead of 0.1 inches. In comparison, for Richmond, a
value of 0.1 inches resulted in better agreement between modeled and observed results when
considering monthly events, total CSO volume, and peak flow. Therefore, EPA updated the guidance for
this model to recommend a range of 0.1 to 0.2 inches, with a minimum default of 0.1 inches.

Impervious surface area was also varied for sub-sewersheds greater than 100 acres, which include 002A
(69 percent impervious) and 006A (98 percent impervious) from NHSA, as well as 003 (20 percent
impervious) from HWU. Several Richmond sub-sewersheds were greater than 100 acres; however, the
sewershed characterization data were already in terms of DCIA, so they did not require correction.
Simulation results showed that for NHSA sub-basins 002A and 006A, using the full impervious area
resulted in a significant overprediction of CSO events, while using a value of "0.5*impervious area"
resulted in a much better agreement. For HWU sub-sewershed 003, the correction of "0.5*impervious
area" actually resulted in an underprediction of CSO occurrence. Although drawn from a small sample
size, EPA suggests that the "0.5*impervious area" correction is more suitable for sewersheds with an
initial percent impervious greater than 20 percent. EPA has also updated guidance for this model input
accordingly, and recommended caution when using the correction for large sewersheds with low initial
impervious area.

Table 5 summarizes and Figure 6 illustrates the results of the CSO occurrence testing for NHSA,
Richmond, and HWU.

The data provided in Table 5 summarize the results in terms of residuals, which refer to the deviation of
individual modeled results from observed data. Residuals are described for the full period of record of
each sub-sewershed. The period of record bias quantifies the average deviation and the direction of that
deviation overall simulation events. For example, for NHSA sub-sewershed 002A, a bias of-0.47 events
means that over the 36-month simulation period for that sub-sewershed, which included 259 individual
events, the CSO Model predicts an average number of monthly events that is 0.47 events less than the
actual average of 4.5 events per month. The average residual, in comparison, is the average of the

17


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

absolute value of all monthly residuals and can be interpreted as the average monthly deviation. Using
NHSA sub-sewershed 002A as an example again, an average residual of 1.5 events per month means
that over the period of record, the CSO Model predicts the number of monthly events to within an average
range of 1.5 events above or below the actual value. Across the 19 sub-sewersheds and 2,218 individual
simulations, the CSO Model output results in an average period of record bias of 0.26 events per month
(or +8 percent compared to the average 3.11 events per month) and an average monthly residual of 1.12
events per month (or 36 percent of the average 3.11 events per month). The overall positive bias
indicates that the CSO Model is slightly conservative, in that it tends to estimate more overflows per
month than are observed.

Table 5. Comparison of modeled (revised CSO Model) to observed CSO events per month for
NHSA, Richmond, and HWU.







Sub-basin

Initial

Average

Events per month residuals

Sub-basin/
CSO ID

Number of
events3

Number of
months

area
(acres)

abstraction
(in.)

events per
month

Period of record

bias
(events/month)0

Average
residual
(events/month)d

EPA Region 2: North Hudson Sewerage Authority, New Jersey

002A (H1)

259

36

276

0.2

4.5

-0.47

1.5

006A (H5)

259

36

151

0.2

3.3

1.4

2.1

012A
(18PS)

259

36

85.9

0.2

4.1

0.44

1.6

015A (W5)

259

36

36.9

0.2

4.2

0.53

1.5

EPA Region 3: Richmond, Virginia

004

79

12

91.6

0.1

5.2

0.17

1.0

005

79

12

11.8

0.1

1.8

-1.8

1.8

012

79

12

90.0

0.1

4.0

1.2

1.2

014

79

12

394

0.1

1.1

0.0

0.33

021

79

12

493

0.1

4.3

-1.0

1.2

024

79

12

197

0.1

4.1

0.58

0.58

025

79

12

65.7

0.1

0.8

2.3

2.3

026

79

12

101

0.1

2.9

1.7

1.7

031

79

12

176

0.1

2.0

-0.25

0.25

034

79

12

63.0

0.1

0.8

0.25

0.42

035

79

12

30.9

0.1

1.8

-0.42

0.42

039

79

12

174

0.1

4.8

1.0

1.0

EPA Region 4: Henderson Water Utility, Kentucky

003

78

12

347

0.2

5.0

-0.33

0.50

004

78

12

43.4

0.2

1.3

0.42

0.75

007

78

12

27

0.2

3.1

-0.75

1.25

Average:

3.11

0.26

1.12

a Number of simulated events based on rainfall amounts that cause one or more outfalls to overflow.

b Refers to the observed average events per month. For NHSA, EPA determined these values from monthly NPDES monitoring data from
2017-2019, obtained for Facility ID NJ0026085 from https://echo.epa.gov/. EPA calculated observed events from Richmond and HWU based
on flow records provided to EPA from each community.

0 Calculated as the average of all monthly residuals.
d Calculated as the average of the absolute value of all monthly residuals.

18


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

NHSA 002A

c
o

r

	I	I	I

12
8
4
0

12
8
4
0

12
8
4
0

NHSA012A

• S./
• • • •

0 4 8 12
Richmond 012

•jr

12
8
4
0

12

NHSA 015A

• • r'

• J» •

• y*

0 4 8 12
Richmond 014

V'

_l	I	

0

-J	1—

0 4 8 12
Richmond 025

• •

0 4 8 12
Richmond 026



12

•:/
4 *• '/

« y

_i	I	

0 4 8 12
Richmond 034

0 4 8 12
Richmond 035

•/

_i	i	i

12
8
4
0 ¥



-j	i	

12
8
4
0

0 4 8 12
Richmond 036

* •'
•

_j	j	

0 4 8 12
HWU 004

0 4 8 12
HWU 007

8 12

y.	1	1	1

0 4 8 12 0 4 8 12

12
8
4

' /
0 >—	1-

0 4

• ••

_i	i

8 12

Observed CSOs per Month

Figure 6. Summary of modeled (revised CSO Model) versus observed monthly CSO events for
NHSA, Richmond, and HWU. The x- and y-axes for each plot range from 0-12 events per month,
so that the dashed line represents a 1:1 slope, or perfect agreement.

19


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Predicting Event Volume and Peak Flow Rate

The communities of Saco, Maine; Elizabeth, New Jersey; Richmond, Virginia; and HWU in Kentucky each
provided EPA with sufficient rainfall and overflow time series data to compare CSO Model output to
observed CSO volumes and peak flow rates for individual events. To simulate each event, EPA ran the
CSO Model for all days in which a CSO occurred (Saco and Elizabeth) or all days in which the daily
rainfall exceeded 0.1 inch (Richmond and HWU).

Similar to the prediction of monthly events, multiple rounds of simulations were run using different inputs
for initial abstraction and impervious surface area based on the preliminary findings discussed earlier in
this report (see discussion on page 19). The determination of final model inputs for each community is
provided in Table 6.

The comparison of modeled to observed CSO volumes and peak flow rates are described in Table 6 and
illustrated in Figure 7. For each metric (total volume in terms of MG or peak flow rate in terms of MGD),
results are described in the same way as the runoff results presented in Table 4, where values of slope
and intercept can be used to characterize the general deviation of CSO Model results from observed
characteristics.

Given the intentional simplicity of the CSO Model and the difficulty of reproducing complex flow regimes
that dictate CSO characteristics (e.g., assuming a static value for regulator capacity), the data described
in Table 6 and illustrated in Figure 7 have considerable variability. For example, slopes of total volume
plots for each sub-sewershed range from 0.13 (Saco 001) to 8.6 (HWU 004), which translates to an
underestimation of event volume of 87 percent to an overestimation of nearly nine times what was
observed. However, these extreme examples highlight what are likely unique scenarios.

First, Saco sub-sewershed 001 is one of the smallest systems evaluated (18 acres), yet has one of the
highest regulator capacities (11.4 MGD), suggesting a highly "flashy" system where rainfall is converted to
short but intense flows through the sewer system. The 0.74 slope of the peak flow rate regression is
much closer to 1, meaning that peak flow rates are less underestimated than CSO volumes. The
difference between the 0.13 and 0.74 slopes suggests that although the CSO Model can reasonably
predict peak CSO flow rate, it does not capture the sustained high flow rate that exists in this sub-
sewershed. Although the underprediction of total volume is not ideal, the ability for a screening tool to
predict the occurrence of a CSO, which is more dependent on the ability to reliably predict peak flows, still
provides value.

At the other end of the spectrum is HWU sub-sewershed 004. Based on conversations with HWU
personnel, their community's CSO outfalls are mostly within a single corridor that runs along the Ohio
River, and all regulators are simple 18-inch drop pipes, meaning that when the capacity of the 18-inch
pipe is exceeded, a CSO occurs. This description would seem to imply that all regulators function
similarly. However, as shown in Table 6, predictions of total volume from HWU sub-sewersheds 003 and
007 result in slopes of 0.9 and 0.92, respectively, which is much closer to 1 than the slope of 8.6 from
HWU sub-sewershed 004. In other words, there appear to be nuances in sub-sewershed 004 that result
in far less CSO volume than predicted by the CSO Model. These nuances could be due to unaccounted
storage capacity in the conveyance system or complex flow regimes that unintentionally limit peak flows
within the system.

The presence of these hidden complexities is further evidenced by flow records from three other sub-
sewersheds in the HWU system: 005 (Towles Street), 008 (Washington Street), and 009 (First Street) that
were not included in this evaluation due to the presence of only one CSO event between all three sub-
sewersheds over the entire 12-month period of record. Irregular flow observations such as these further
reinforce the conclusion that the CSO Model's effectiveness as a screening tool can be greatly improved
by coupling the model with basic field monitoring techniques, such as chalking at a CSO outfall as a
means of determining whether a CSO occurred (see Section 3.1.3 of EPA's CSO Guidance for Monitoring
and Modeling for additional discussion of simple field monitoring techniques).

20


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Despite cases where complex systems resulted in poor predictive performance of the CSO Model, the
CSO Model behaves reasonably well when judged by the average slopes and intercepts determined
across all 21 sub-sewersheds included in Table 6 and illustrated in Figure 7. For the prediction of CSO
volume, regressions of modeled to observed results yield an average slope of 1.29 and an average
intercept of 0.05. These results suggest that, on average, the CSO Model slightly overpredicts total CSO
volume (based on a slope >1) and, for very small events, may predict the occurrence of a CSO event
when there was no CSO (based on a small intercept >0). In terms of peak flow, the resulting average
slope of 0.91 suggests that the CSO Model slightly underpredicts larger peak flows (based on a slope
<1), while the peak flow intercept of 1.08 implies that, for small events, the CSO Model overpredicts peak
flows and may predict the occurrence of a CSO event when there was no CSO.

Table 6. Comparison of modeled and observed CSO event total volume (MG) and peak flow rate
(MGD) for Saco, Elizabeth, Richmond, and HWU.

Sub-basin/CSO

Number
of

Sub-
basin

Initial
abstraction

Total volume

Peak flow rate

ID

events3

area
(acres)

(in.)b

Slope

Intercept

Slope

Intercept

EPA Region 1: Saco, Maine

001

22

18

0.1

0.13

0.03

0.74

0.76

EPA Region 2: Elizabeth, New Jersey

001

7

439

0.1

1.1

0.36

1.4

0.87

031

7

59.5

0.1

1.2

0.21

0.62

2.9

036

7

210

0.1

0.23

-0.27

0.28

0.51

039

7

245

0.1

1.5

-0.03

1.6

0.54

040

7

34.9

0.1

3.1

0.14

1.8

2.2

EPA Region 3: Richmond, Virginia

004

79

91.6

0.1

0.49

0.04

0.64

0.98

005

79

11.8

0.1

Model results indicate 0 overflow.

012

79

90.0

0.1

0.40

0.04

0.58

0.87

014

79

394

0.1

0.79

0.01

0.79

1.2

021

79

493

0.1

0.39

0.11

0.58

0.82

024

79

197

0.1

0.35

0.07

0.45

2.1

025

79

65.7

0.1

0.96

0.03

0.92

1.5

026

79

101

0.1

0.82

0.03

0.71

0.74

031

79

176

0.1

0.60

0.02

0.67

0.40

034

79

63.0

0.1

1.8

0.01

1.3

0.83

035

79

30.9

0.1

0.89

0.00

0.74

0.17

039

79

174

0.1

0.70

0.08

0.75

1.5

EPA Region 4: Henderson Water Utility, Kentucky

003 - Raqan St.

78

347

0.2

0.90

0.22

1.14

2.3

004 - Jackson St.

78

43.4

0.2

8.6

0.00

2.1

0.16

007 - Powell St.

78

27

0.2

0.92

0.00

0.36

0.17

Average:

1.29

0.05

0.91

1.08

a Number of simulated events based on rainfall amounts that cause one or more outfalls to overflow.

b Initial simulations run with values ranging from 0.1 to 0.2 inches to determine the most suitable input based on the agreement
between modeled and observed results. Values shown here resulted in the best agreement.

21


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Saco 001

(3

(L>

E

5

"D
O

Elizabeth 001

o e&-

~ i

1.0

0.5

~

Elizabeth 031
~

~

CD
~

Elizabeth 036

§ 15

§

"O

o

10

Saco 001
A

A A

A

If*

0 As—

-1 0 O

2 0

60
40
20
0 A

2 4
Elizabeth 001
A

0.0 Q-

30
20
10
0

0.0 0.5 1.0
Elizabeth 031

-o-o-

. rP

~
a

Obs. (MG)

A

4a

100

50

0	2

Elizabeth 036

£

13 1.0

CL>

J 0.5

o
>

T3

° 0.0 fc>-

0 5 10
Elizabeth 039

15

0.6
0.4
0.2
—1 0.0

0 20 40
Elizabeth 040

60

0 10 20 30
Richmond 004



n



o

(3 30

0.0 0.5
Elizabeth 039

1.0


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

w 3

qj

£ 2
5 1

"D

O n

Richmond 012

it

~ in

Richmond 014	Richmond 021

20

10

0 &a-

cP
~

U 60

40
20

"D

° 0

Richmond 012

V

0 20 40 60

(5

	10
xs

o	o

Richmond 025

50

Richmond 026

A A

M

0

£ A

A

30
20
10
—1 0

0	2

HWU 003
A

¥

4 0.0 0.1
HWU 004

10

A
^A
£

0.2

Obs. (MGD)




-------
Validation Study: Combined Sewer Overflow Model for Small Communities

The range of slopes, particularly for total volume regressions, is wider for CSO regressions (Table 6) than
runoff regressions (Table 4), which illustrates a limitation of the revised CSO Model in its ability to capture
realistic hydraulic control capacities. The revised CSO Model uses a single input for hydraulic control
capacity, which is mainly dependent on regulator capacity, and calculations assume all incoming flow up
to that capacity diverted to the interceptor. However, not only is this capacity difficult to accurately predict,
but it is variable across a range of incoming flows and responds more as a rating curve than a single rate.
While CSO Model input for regular capacity could be modified to include a rating curve, this may also
present a usability challenge as the intended user is not expected to have access to the type of detailed
monitoring or modeling data necessary to define an accurate rating curve.

24


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

Model Validation Conclusions

Results presented above yield several conclusions that EPA used to improve CSO Model accuracy.

These conclusions include:

•	Aggregating five-minute time series to a 15-minute timestep causes minimal loss of detail, while
aggregating to a 60-minute timestep causes significant loss of detail.

•	Decreasing the CSO Model timestep from 60 minutes to 15 minutes greatly improves the
accuracy of peak flow prediction and improves runoff response timing.

•	The resolution of rainfall data has a large effect on the ability to accurately simulate peak flows,
especially for smaller sub-sewersheds.

o EPA recommends using rainfall data collected at a 15-minute interval or shorter.

•	Using impervious area as a runoff coefficient, the CSO Model overpredicts stormwater runoff
volume for sub-sewersheds larger than 100 acres by a factor of approximately two, unless the
percent imperviousness of the sub-sewershed is 20 percent or less.

•	Initial abstraction is an important term, even for simulation of larger storm events. A value of 0.1
to 0.2 inches is recommended based on validation results.

•	The CSO Model predicts runoff volumes and rates better than CSO volumes and rates, owing to
the difficulty in estimating a system's hydraulic control, or regulator, capacity.

•	Despite a variable ability to predict CSO volumes and flow rates, the CSO Model performs well in
its ability to evaluate the presence or absence of a CSO event.

•	Combined with simple, low-cost field monitoring techniques, the CSO Model can serve as a
powerful screening-level tool to help communities better understand their combined sewer
systems and reduce the need to monitor every rain event.

Based on these conclusions, EPA made the following improvements to the CSO Model:

•	The model timestep has been reduced from 60 minutes to 15 minutes.

•	The user guide instructions for obtaining rainfall data have been updated to reflect a
recommendation that 15-minute data be obtained wherever possible.

•	An initial abstraction term has been incorporated, set at a default of 0.1 inches with the ability to
update based on local conditions.

•	Text in the user guide has been added to recommend the following: For larger sub-sewersheds
(generally greater than 100 acres), the model tends to overpredict peak runoff flow rates and total
runoff volumes (and CSOs by extension) when using total percent impervious area as a model
input. This overprediction is likely due to the influence of directly connected and disconnected
impervious surfaces—as discussed in EPA's factsheet on Estimating Change in Impervious Area
CIA') and Directly Connected Impervious Areas CDCIA') for Massachusetts Small MS4 Permit—
especially as drainage areas increase in size. If modeling larger sub-sewersheds, EPA therefore
encourages the user to use the model input for percent imperviousness as a calibration
parameter, reducing the value until reasonable results are obtained. Based on general model
validation performed by EPA, a reduction of percent imperviousness by up to 50 percent was
found to better predict runoff rates and volumes for larger sub-sewersheds.

25


-------
Validation Study: Combined Sewer Overflow Model for Small Communities

References

City of Omaha. 2014. Update to the Long Term Control Plan for the Omaha Combined Sewer Overflow
Program. Prepared by Clean Solutions for Omaha. Retrieved from https://omahacso.com/about-
proqram/lonq-term-control-plan.

HWU and Strand Associates, Inc. 2009. Henderson Water Utility combined sewer overflow long-term
control plan, http://psc.ky.gov/pscscf/2010%20cases/2010-00223/20100914_Appendix%20D.pdf

Jacobs Engineering Group. 2020. Selection and implementation of alternatives for the Adams Street
Wastewater Treatment Plant. Prepared for North Hudson Sewerage Authority. Draft June 2020.
Retrieved from https://www.ni.qov/dep/dwq/cso-ltcpsubmittals.htm.

MacDonald, M. 2019. System characterization report: Combined sewer overflow long term control
program. Prepared for City of Elizabeth, New Jersey.

City of Richmond. 2022. Wastewater Utility Website, https://www.rva.oov/public-utilities/wastewater-utilitv.

Sutherland, R.C. 1995. Methodology for estimating the effective impervious area of urban
watersheds. Watershed Protection Techniques, 2(1), 282-284.

Thompson, D.B. 2006. The Rational Method. David B. Thompson, Civil Engineering Department, Texas
Tech University.

26


-------