Regulatory, Technical and Modeling Challenges to Developing a Frequency Based SSO Control Project in Wayne County, Michigan

REGULATORY, TECHNICAL AND MODELING CHALLENGES
TO DEVELOPING A FREQUENCY BASED SSO CONTROL PROJECT
IN WAYNE COUNTY, MICHIGAN

Robert Czachorski, P.E., P.H., Orchard, Hiltz & McCliment, Inc. *
John Baratta. P.E., Wayne County Department of Environment
Sue Ann Hanson, P.E., Wayne County Department of Environment
Vyto Kaunelis, P.E., Orchard, Hiltz & McCliment, Inc.
Murat Ulasir, P.E., Ph.D., Orchard, Hiltz & McCliment, Inc.

* Orchard, Hiltz & McCliment, Inc.
34000 Plymouth Road
Livonia, MI 48150

ABSTRACT

Wayne County’s North Huron Valley / Rouge Valley (NHVRV) interceptor system collects
sewage from 15 communities located in Southeast Michigan and transports flows to the Detroit
Water and Sewerage Department (DWSD) for treatment and discharge. The County is
evaluating a regional approach to controlling wet weather sanitary sewer overflows (SSOs). A
new methodology called the i3D antecedent moisture (AM) model, was used to perform the
hydrologic modeling. The i3D model is a continuous model that produces a good match to
observed flow data over time. The accuracy of the model resulted in a high level of confidence
in the frequency analysis for SSOs and will serve as the basis for recommending improvements
to control wet weather SSOs. The use of the AM model combined with a frequency analysis for
sizing improvements eliminated the need to select a design storm event based on “average”
conditions. This reduced many of the conservatisms that are frequently included in event models
such as the capture coefficient and seasonal effects. The use of spatially varied rainfall also
improved the accuracy of the analysis over the use of a point rain gauge. This paper presents the
modeling and analysis innovations used and the preliminary development of a regional project.

KEYWORDS

SSO, Sanitary Collection System, Modeling, Antecedent Moisture, Frequency Analysis

INTRODUCTION

There were several regulatory and technical challenges during the development of the regional
project to control SSOs for the NHVRV Interceptor System. These have included the adoption
of a new wet weather SSO policy by the Michigan Department of Environmental Quality
(MDEQ), the hydrologic behavior of the system that is highly affected by antecedent moisture
conditions, and the complex hydraulic characteristics of the system, including the hydraulics of
the downstream DWSD system.

Because wastewater system capacity is finite, it is unrealistic to expect that all wet weather SSOs
can be eliminated. However, the State of Michigan has been very proactive in its SSO control
policy. In December of 2002, the State established a formal wet weather SSO policy. This is
one of the first SSO policies to be implemented in the Country and contains very aggressive
standards for controlling SSOs. The policy calls for elimination of SSOs up to a 25-year, 24-
hour design storm event during the growth season using average soil moisture conditions. The
policy also allows for an alternative approach with a performance standard resulting in an SSO
frequency of less than once in 10-years.

Previous event modeling had led to the conclusion that a relief/storage tunnel with an outlet to a
CSO retention treatment basin would likely be the required SSO control project. Frequency of
SSO events was explored by trying to use an event model, but the confidence in the results was
not as high as desired for the anticipated magnitude of the capital investment involved. For this
reason, the County elected to use the continuous antecedent moisture model to establish a
frequency basis for developing their SSO control strategy.

A continuous antecedent moisture model was developed to identify system improvements to
meet the 10-year frequency requirement. The continuous model was then used to perform a
frequency of use analysis on the tunnel to size a preliminary tunnel volume. Continuous model
simulations uncovered a potential hydraulic restriction in the system that had not been detected
with previous modeling. This led to the conclusion that river inflow, system blockages,
problems with downstream pump operations, and / or hydraulic gradients are likely contributors
to the observed problems in addition to high flows during infrequent events. This paper explores
these regulatory and technical challenges, presents the development of the NHVRV regional
project to control SSOs and describes the innovations that were used to overcome the challenges
facing the project.

METHODOLOGY

Several innovative techniques were utilized to perform the frequency analysis for the NHVRV
SSO control project. An antecedent moisture model was developed to provide a highly accurate
continuous estimate of system flows for a long period of record. A frequency analysis was used
to analyze the statistical frequency of SSOs. A hydraulic process, called “null modeling”, was
used to isolate the hydraulic performance of the system. This process uncovered a potential
hydraulic restriction in the interceptor system. The technical background and description of
these innovative approaches is described in this section.

The Antecedent Moisture Model

Antecedent moisture is a term that describes the relative wetness or dryness of a sewershed,
which is a function of the distribution of recent rainfall, temperature, seasonal variations and
other dynamics. Because of the complex transport mechanisms of these flow sources, they are
heavily dependent on variables that are not normally included in runoff models such as soil

moisture conditions and seasonal effects. These variables continuously change during and in
between storms in response to antecedent moisture conditions.

Figure 1 shows antecedent moisture effects from observations from the Novi subarea of the
NHVRV system from the year 2000. The measured flow is comprised of a diurnal sanitary
component and an I/I component. The I/I has been separated from the total flow by utilizing a
time series extraction technique that does not alter the I/I information content within the data.
This figure shows how antecedent moisture conditions cause a large variation in the observed
capture coefficients (percentage of rainfall captured by the sewer system) from storm-to-storm.
This effect is demonstrated by comparing the April 9, 2000 and May 31, 2000 storms. Both
storms have rainfall volumes of approximately 1.0-inch, but have significantly different capture
coefficients of 4.4% and 0.6%, respectively. This variation in capture coefficient is caused by
the antecedent moisture conditions preceding each storm event. The May 31, 2000 storm occurs
after a long dry period, whereas the April 9, 2000 storm occurs during the spring period, which is
wet due to the cool, rainy spring season as well as the winter snowmelt. This demonstrates the
significant impact that antecedent moisture has on system response to storm events.

Figure 1 – Antecedent Moisture Effects on I/I
The effects of antecedent moisture can clearly be seen in the I/I flow data. Note the high
capture coefficients in the spring and the low values in the summer. The capture
coefficients are also affected by the amount of recent rainfall.

The state-of-the-art in modeling I/I has been dependent on physically based modeling derived
from stormwater runoff simulations. Historically, collection system models were limited in their

ability to account for antecedent moisture conditions. The antecedent moisture conditions that
occurred during the period of the flow monitoring or for the “design event” were implicitly
incorporated into the collection system model, essentially as a constant condition.

An alternative to purely physical modeling is the use of a modeling procedure known as system
identification. The basis of this approach is to allow measured data (in this case flow, rainfall
and temperature) to guide the modeling process. It does not assume that the model must adhere
to a specific preconceived notion of the physical system, but rather arrives at a statistically
relevant description based on the data alone. Accordingly, this procedure differs from purely
black-box modeling since it provides a model that offers insight into the underlying mechanics of
the system. This is contrary to existing methods, which unduly enforce a predetermined set of
physical principles on the data that may or may not be present in the actual system.

A new modeling technique called inflow and infiltration identification (i3D) based on system
identification theory was developed specifically to simulate the impacts of antecedent moisture
on I/I [Czachorski, 2001]. The i3D model incorporates nonlinear dynamics to account for short-
term antecedent rainfall conditions and long-term seasonal variation in the wet weather response.
Unlike regression techniques or complex physically based models, the resulting system
identification model for antecedent moisture is uniquely identified for each sewershed, is
parsimonious in nature (containing relatively few modeling parameters) and predicts the amount
of antecedent moisture within the sewershed as a continuous variation in the capture coefficient.
A block diagram of the i3D model structure is depicted in Figure 2.

Inflow

Lf ( rk ; pf )

temp tk
pf
flow qk
rain rk
N ( rk , tk )
ps
Antecedent
Moisture
Ls ( rk ; ps )

Infiltration
Figure 2 – i3D Model Structure
This figure shows a block diagram of the i3D model. Note that the antecedent moisture
block alters the parameters within the separate inflow and infiltration blocks.

In order to gain confidence in the model flow projections and account for antecedent moisture
effects, a continuous modeling simulation was performed for the NHVRV system using the i3D
model. The model was developed using data from the County’s extensive permanent metering

system, which contains a period of record of up to 11 years. This metering system provided an
excellent long-term flow record that was used to calibrate and validate the model, and evaluate
the accuracy of the statistical predictions of the model. For the Wayne County NHVRV project,
the i3D model was successfully calibrated and validated using the flow metering data available.

Frequency Analysis

Rather than using an event based modeling approach to size system improvements, a frequency
analysis was performed for the NHVRV system. The frequency analysis used the i3D model in a
continuous simulation for 18 years of available rainfall records to determine the statistical
probability of peak flows. Because the i3D model incorporates the variations in antecedent
moisture and predicts the variations in capture coefficient, the resulting frequency analysis
incorporated those effects and provided confidence in the results. This eliminates the need to
establish a design event and to estimate “average conditions” for use in the event model.

The use of a frequency analysis to design wet weather improvements for sanitary systems is a
new and innovative approach. Frequency analyses are very commonly used for river peak flows
and flood stage analysis. For those systems, a Log-Pearson Type III analysis is commonly used
to represent the statistics of peak flows, including analysis for FEMA Flood Insurance Studies
[FEMA, 2002]. To our knowledge, the use of a Log-Pearson Type III analysis is the most
applicable technique for performing a statistical frequency analysis on sanitary sewers flows.

The continuous prediction of I/I from the i3D model was used to analyze the statistics of
exceeding certain peak flows to assess system performance. This results in a design recurrence
interval flow event being used to analyze collection systems, as opposed to a design rain event as
is typically used. This design flow event implicitly incorporates many hydrologic variables
including rainfall frequency, antecedent moisture and seasonal effects [Van Pelt, 2002]. It is
interesting to note that the Michigan Department of Environmental Quality (MDEQ), in
establishing a Sanitary Sewer Overflow Policy in December, 2002, allowed for consideration of
a design rain event (25-year, 24-hour storm) and also a design recurrence interval event [MDEQ,
2002] by specifying an SSO frequency of less than once every 10 years.

For the NHVRV system, a frequency analysis was performed by utilizing the i3D continuous
model output from 18 years of rainfall and temperature data. A Log Pearson Type III statistical
analysis was then be used to determine the 10-year recurrence interval peak flow for this system.

Hydraulic Null Modeling

Comparison of the continuous modeling results to observed flow and depth data revealed that
there may be a hydraulic restriction affecting system performance in the NHVRV system. A
methodology that we call “null modeling” was used to isolate the system model to only the
hydraulics in order to evaluate this potential system bottleneck. The null modeling procedure is
shown schematically in Figure 3.

Figure 3 – Hydraulic Null Modeling
This figure shows a schematic of the null modeling procedure used to isolate the model
with respect to hydraulics and evaluate potential system hydraulic restrictions.

In mathematics the null set is a set or matrix having no elements. Similarly, in collection system
modeling, null modeling is the process of identifying an incremental flow hydrograph without
the use of a hydrologic model. When attempting to isolate the hydraulics, a hydrologic model is
not needed because the incremental hydrographs (input between flow meters) can be determined
from the downstream hydrograph, upstream hydrograph and piping characteristics. The process
requires iterations to solve because of the routing effects of the piping system. However, this
process can be automated using a hydraulic model such as the EXTRAN block of EPA SWMM.

The process is simple in that it relies on observed metering data and known characteristics of the
piping system to estimate the incremental flow. An initial estimate of the incremental
hydrograph is determined by first routing the observed upstream hydrograph through the piping
system by itself, and then subtracting the routed hydrograph from the observed downstream
hydrograph. However, the key to the null modeling process is that this initial estimate of the
incremental hydrograph is corrected and rerouted in the model to improve the incremental
hydrograph until an acceptable match is achieved at the downstream end. This iterative
correction implicitly incorporates the hydraulic routing effects of both the upstream hydrograph
and the incremental hydrograph into the estimate of the incremental hydrograph, which results in
improved accuracy.

The null modeling has the added benefit of producing an accurate flow estimate of the system for
use in hydraulic model calibration. Because the system flows very accurately match observed
flows, and are not assumed based on a hydrologic model, the resulting model performance is
isolated to the hydraulics. This means that any discrepancy between the model depths and
observed depths is the result of hydraulic effects and not caused by the inaccuracies of
hydrologic models. This isolation of the hydraulic effects allows for a detailed hydraulic model
calibration to be performed and for an evaluation of any potential system hydraulic restrictions.

RESULTS

Several innovative analysis techniques were used to evaluate the required recommendations for
the NHVRV SSO control project. These included the antecedent moisture model, a statistical
frequency analysis and null modeling. This section describes how each of these techniques were
applied to the NHVRV modeling project and discusses the results achieved.

i3D Model Calibration and Validation

A system schematic was developed to simplify the system and provide a visual explanation of
the hydrologic models that were developed. This schematic can be seen in Figure 4. On this
figure, as well as throughout the paper, a “+” is used with meter names to denote when the
addition of meter data was used. Numbers, such as 1P, 26P, etc. are used to denote meters. The
system was divided into 12 sub-areas, two of which are combined systems. i3D models were
developed for 11 of these 12 sub-areas and the existing SWMM RUNOFF model was used for
the remaining combined area (Sub-area 4). Initially it was planned that i3D models would only
be developed for separated areas. However, since the Redford combined area is metered, and the

SWMM runoff model was not matching accurately under some conditions, an i3D model was
developed for this area as well.

Figure 4 – System Schematic
The NHVRV system is shown schematically in red. There are parallel interceptors for
much of the system. i3D models were created for each of the separated areas shown in
red using the historic meter data from the meters denoted with green circles.

Sub-areas were chosen primarily based upon available metering data, but community boundaries
and previous work were also taken into account. In general, far less sub-areas were included in
this modeling effort when compared to previous event model efforts. Existing flow metering data
was used to develop hydrologic i3D models. Missing data and meter subtraction were initially

concerns. However, use of the continuous model allowed for a large number of comparisons
between metered and modeled data, and provided confidence in the model.

Once the observed data for each sub-area were obtained, the daily diurnal flow pattern was
filtered so that the resulting observed flow signal only contained inflow and infiltration. The
diurnal flow pattern was filtered by determining the daily nighttime minimum flows. This was
assumed to be the base infiltration. It is acknowledged that there may be some nighttime users or
minimal nighttime sewage usage; however, this is accounted for in the hydraulic model through
the average base sewage flow. The base infiltration was then subtracted from dry weather flow
data to estimate a dry diurnal pattern. For wet days, this dry diurnal pattern is replicated from
dry days based on the day of the week. The inflow and infiltration flow signal is then estimated
by subtracting the continuous estimate of the diurnal flow from the total metered flow. All of the
flow graphs shown in this paper are the I/I flow signal after the diurnal flows have been
subtracted using this methodology.

Two years (2000 and 2001) of meter data were used initially to build the i3D models. The
models for these years were input into the hydraulic model for routing and then compared to
actual meter data. A sample from the two-year calibration fit is shown in Figure 5. This figure
shows the model prediction and the observed I/I flow for a two-month period in the fall of 2001.
The figure shows the resulting model fits at three meter locations along the interceptor system:

1. Meters 6P+7P+8P measure the upstream areas of the system, which contains drier,
separate areas.
2. Meters 9P+10P+11P includes the central area of the system, which is a wetter,
separate area.
3. The outlet includes the combined areas near the downstream portion of the system.

The time period in Figure 5 contains an excellent variation in antecedent moisture for calibration,
as nearly 8-inches of rain fell in a 15 day period. This is over 5 times the normal precipitation
for this time of year. This series of back-to-back rain events produced a very large variation in
capture coefficient as the system became wetter. The i3D model was calibrated to replicate this
relationship between the system wetness and the resulting capture coefficients. As shown in the
figure, the i3D model does an excellent job of accurately simulating these dynamics.

Once calibrated, the models were then validated. Validation tests the model performance for
years not included in the calibration. Validation is necessary because it is possible for an
incorrect model to be calibrated such that it fits specific observed data to a reasonable degree of
accuracy, yet in general is an inaccurate model of the system. Consequently, the ability to
calibrate a model to a given data set does not solely validate the model. A more confident
evaluation lies in the ability of a model to fit a set of data that was not used for calibration. Two
years of additional data not used in calibration (2002 and 2003) were used to validate the models.
A sample from the two-year validation is shown in Figure 6. This validation was successful and
was critical to providing confidence in the model for both Wayne County and the MDEQ.

Figure 5 – Model Calibration Example
The model was calibrated to the years 2000-2001. This figure shows a portion of the
calibration with a 15-day period with nearly 8-inches of rain, which produced highly
varied AM conditions. The model accurately predicted these variations.

Figure 6 – Model Validation Example
The model was validated to the years 2002-2003. Validation was performed to test the
model with no further manipulation of the model by the user. Rainfall and temperature
were entered into the model and the output was compared against observations.

An accuracy of fit analysis was performed to quantify the performance of the model. The
accuracy of fit analysis evaluated the ability of the model to predict peak flows and volumes for
the four largest storms in each of the four years from the calibration/validation period. Figure 7
depicts a sample of the accuracy of fit results for the flow meter in Novi. Note that the i3D
model accurately predicts the volumes, even for significant variations in capture coefficient.
This is evident by examining the fit for the April 20 and July 30 storms, which have similar
rainfalls, but different capture volumes. The overall model performance for the entire system
resulted in less than 1% error in net peak flows and less than 5% error in net volumes for the four
year period. Individual years had a greater variation in the accuracy than those for the entire four
year period. For example, figure 7 shows that for the single year shown for Novi, the net error in
peak flow and volume was 0.3% and 9.6%, respectively. This is viewed as excellent model
performance, especially considering that, once calibrated, the model is predicting the variations
in capture coefficient with no additional input from the modeler.

Comparing these accuracies to those from other studies should be done very carefully. Other
studies that we examined used calibration data rather than validation data, used observed capture
coefficients for simulations, selectively included or excluded events for analysis, or included
base flows in the volume, all of which tend to artificially minimize error percentage.

System Hydraulic Restriction Findings

The EPA Storm Water Management Model (SWMM) was used to perform the hydraulic
simulations for this study. Previous event-based modeling was done by another consultant and
was performed in SWMM Version 4. For this project, the existing model was converted to
SWMM 5.0 for use in the continuous simulation. The EXTRAN block of SWMM 5.0 was run
using Dynamic Wave routing, which takes into account both the continuity and momentum
equations for conduits, and a volume continuity equation at nodes. Data produced by the
hydrologic models were used as input hydrographs in the hydraulic model.

The hydraulic model was first used in conjunction with the hydrologic model for calibration.
This was mainly done to get a good match between modeled and observed flows. Because of the
reported SSOs that have occurred, depths were also checked against observed data for storms
that had known surcharging from 2000-2003. These comparisons were made at meters 9P, 10P,
and 11P, which are located nearest to the locations of reported overflows. These comparisons
revealed that the depths at 9P, 10P, and 11P were much less in the model than recorded in the
observed data. The model showed that if the piping system functioned as modeled, no SSOs
would have occurred from 2000-2003.

To validate these findings and ensure that the model accurately reflected the system, several
steps were taken. First, the model was checked with the previous design event EXTRAN model
(from SWMM Version 4) to ensure that translation of the hydraulic model was not causing the
discrepancy. This work was performed by the same consultant that originally developed the
SWMM 4 hydraulic model. The verification was done successfully and the resulting depths
matched, indicating that hydraulic model translation was not the cause for the discrepancy.

Figure 7 – Sample of Accuracy of Fit Analysis
An accuracy of fit analysis was performed for the four largest storms in each year. The
overall net accuracy of fit was < 1% for peaks and < 5% for volumes. This is an
excellent fit considering that the model predicts the variations in capture coefficients.

The next step taken was to input the actual observed flows and boundary conditions into the
hydraulic model to eliminate the possibility of the discrepancy being caused by AM model
hydrologic errors. This was done by following the null modeling procedures described earlier.
This isolated the model to just the hydraulic simulation, because actual observed flows were
used, and not the AM models. These models also showed that the system did not surcharge at
9P, 10P, and 11P, as the observed data indicated. Figure 8 depicts the results of this model
simulation. Note that the depths match well during low flows, but during high flows there is a
significant discrepancy between the modeled and observed depths.

Figure 8 – Modeled Versus Observed Depths
Observed depths show significant back-ups for the largest storms, whereas the model
predicted only moderate increases. From the information available at this time, the
hydraulic discrepancy is thought to be the result of river inflow, a hydraulic restriction,
or downstream hydraulic gradients. (Observed depth above 105-inches are not recorded
due to meter limitation. Actual depth may have been higher.)

These results lead to the belief that there is possibly an influence on the system from the adjacent
Rouge River, a blockage somewhere in the system, or influence from downstream flow levels.
Since it is difficult to determine from existing data whether a system blockage or downstream
levels may be the source of the discrepancy, further investigation is required. There may be open

river flap gates in the system that may be the sources of river inflow. Yearly peak flow data for
the nearest known USGS stream gauge (Middle Rouge River at Dearborn Heights) was gathered.
The dates of reported SSOs all were determined to be peak yearly flows for the stream gauge,
supporting the idea that river influence may explain or partly explain the elevated levels in the
system. For the remainder of the study, the conditions causing the hydraulic discrepancy were
assumed to be correctable and therefore were not integrated in further modeling. This
assumption was also made for the null modeling described above, which may have slightly
biased the estimation of the incremental flow hydrographs by lumping the hydraulic restriction
effects into the null model.

This finding resulted in a recommendation to perform a physical investigation of the interceptor
to locate and remove the source of the hydraulic discrepancy. The model results suggest that if
the hydraulic discrepancy is located and removed, no SSOs would have occurred during the
2000-2003 period. This is a significant finding, as observations from the system suggest that
several SSOs occurred during this period.

Continuous 18-Year Simulation

The calibrated and validated model was used to perform an 18-year continuous simulation for
use in the frequency analysis. The existing EXTRAN model was used for all simulations, which
assumes that the source or sources of hydraulic restriction are identified and corrected. Rainfall
data was used from a series of seven (7) gauges located throughout the system to approximate
the spatial variation in rainfall. Spatial variation of rainfall is a critical input to this type of
analysis, and results are enhanced the more accurately it is represented in model runs. The 18-
year period was used for the continuous simulation because that represented the period of record
for which the spatially varied rainfall was available (1988-2005). The 18-year continuous
simulation was used to represent how the current system would respond to a variety of rainfall
events, providing data for use in the frequency analysis.

Frequency Analysis for Peak Flows

Based upon prior modeling work, frequency analyses were performed in order to determine the
10-year peak flow in the NHVRV system. The frequency analysis focused on meters
9P+10P+11P since this is near where SSOs have occurred. April 1 – October 31 was defined as
the period of concern by MDEQ, so the analysis was limited to this period. A Log Pearson Type
III probability function was used to describe the peak flow data. In addition to predicting the
peak flow with a frequency of once every ten years, the probability distribution was used to
estimate the frequency of exceeding the existing system’s design capacity, which was
determined to be approximately 200 cfs at meters 9P+10P+11P, based on the threshold that
produced surcharging in the model.

Figure 9 shows the frequency analysis that was performed for existing conditions. This figure
shows the peak flow versus the annual exceedance probability of that flow. As shown on this
figure, the Log-Pearson Type III distribution can be used to determine the 10-year frequency

peak flow (indicated by the dashed black line) and can also be used to determine the probability
of exceeding the 200 cfs capacity of the system at meters 9P+10P+11P (indicated by the dashed
blue line).

Figure 9 – Existing Peak Flow Statistics
This plot shows the results of the Log-Pearson Type III analysis for existing peak flows.
Note that the observed points match very well with the modeled points. The plot was
used to determine the 10-year flow and the frequency of exceeding the system capacity.

Each red point in Figure 9 represents the highest peak flow modeled for each of the 18-years.
The black points show the results from observations for each of the 8-years available. The good
match between the model points and observed points is evidence of the accuracy of the model.
This provided confidence in the modeling results and statistical analysis of peak flows.

The frequency analysis demonstrated that the capacity of the existing system will be exceeded
approximately once every 4.3 years (23% annual exceedance) and that the 10-year peak flow rate
is 224 cfs (10% annual exceedance). The 4.3 year frequency for exceeding system capacity
seems reasonable considering the observed frequency of SSOs. These results also produced
confidence in the modeling and statistical analysis for peak flows.

Tunnel Volume Sizing

The final step in the process assumed that construction of a transport / storage tunnel would be
required to augment system capacity to minimize SSOs. To account for future conditions
adjustments for growth, future flows were entered into the model and the frequency analysis was
revised. Additionally, the future conditions continuous model was modified to include a
transport/storage tunnel to handle flows that exceed the system capacity. Figure 10 shows the
results of the Log-Pearson Type III analysis for future conditions. The plot also contains the
volume that entered the tunnel in the model for each of the events that exceeded existing system
capacity.

Figure 10 – Tunnel Volume Sizing
This plot shows the results of the Log-Pearson Type III analysis for future peak flows.
Events that exceeded system capacity are labeled with the volume discharged to the
tunnel. A 6.4 MG tunnel is required to capture all historic storms up to a 10-year flow.

There were six (6) events in the 18-year simulation that exceeded the existing system capacity of
200 cfs. Four (4) of these events had peak flows that were less than the 10-year peak flow rate
estimate from the Log Pearson Type III analysis, and two (2) events had peak flows that were
greater than the 10-year peak flows. The tunnel volume was selected to completely capture the
volume generated from storms in the model that produced up to the 10-year peak flow. This
resulted in a recommended 7.5-foot diameter, approximately 20,000 feet in length tunnel with an
effective volume of 6.4 MG. It should be noted that the MDEQ appears comfortable with the
modeling and peak flow statistics, but has not concurred with the volume recommendation at this
time. For peak flows above a 10-year frequency, opportunities may exist to route tunnel
overflow through a nearby CSO basin for treatment.

DISCUSSION

Previous event modeling had led to the conclusion that a relief/storage tunnel with an outlet to a
CSO retention treatment basin would likely be the required SSO control project. Previous
analysis also estimated that the largest constructable tunnel was 9.5-feet in diameter, which
would result in a total volume of approximately 10 MG. Frequency of SSO events was explored
with this tunnel size by trying to use an event model, but the confidence in the results was not as
high as desired for the anticipated magnitude of the capital investment involved.

The use of the i3D AM model resulted in a continuous model prediction that closely matched
system observations. This match was achieved for both continuous flow predictions and for the
system peak flow statistics. This provided confidence in the model, the predicted frequency
analysis, and resulting tunnel sizing for the historic peak flow events.

The results from the antecedent moisture modeling revealed a potential system hydraulic
restriction. Additional sewer system physical evaluation is required to determine the cause of the
restriction, but the EXTRAN model was felt to accurately represent the hydraulic characteristics
of the system with the bottleneck removed. Though the cause of the hydraulic restriction is
unknown, the current model was used to size the tunnel improvements by assuming that the
source or sources of the restriction can be located and corrected.

Using this methodology, an effective tunnel size of 6.4 MG was recommended to capture the
flow for up to a 10-year frequency event. The tunnel size will be revisited once the source or
sources of the hydraulic restriction is located and removed, and it can be verified that the
hydraulic model accurately represents the system after rehabilitation. The following
recommendations have been made as a result of the study:

• A Sewer System Evaluation Survey (SSES) was recommended to determine the cause
and solution for the hydraulic restriction identified as a result of modeling work.

• Several of the existing system meters are not as effective as desired in obtaining good
system data during critical conditions (especially at the outlet). It was recommended that
the metering system be modified to better capture this data. This can be accomplished by

moving meters 4P and 8P (shown on figure 4) to downstream locations near the outlet to
better record downstream hydraulic gradients.

• A relief tunnel was recommended with a diameter of approximately 7.5 feet with one
intermediate control point to provide an effective storage volume of approximately 6.4
million gallons.

• The relief tunnel would ideally be connected to a local CSO basin to accept flows beyond
the 10-year frequency event to provide some degree of treatment and protect water
quality in the river to the maximum degree practical.

• An option that was suggested by MDEQ was to consider adjusting the regulators for the
combined sewer areas to a minimal setting to allow for additional flow from separate
sewer areas. The model is planned for expansion to the Lower Branch of the system,
which will provide better information to adequately examine this alternative.

CONCLUSIONS

The frequency analysis for the NHVRV system was performed using a new and innovative
hydrologic modeling approach that incorporates antecedent moisture effects. We offer the
following conclusions regarding the use of this approach:

• The use of the antecedent moisture model produced output that accurately matches
system flow observations, even for highly varied rainfall and system capture coefficients,
providing confidence in the resulting analysis, design simulations and recommendations
for improvements.
• Spatially varied rainfall had a significant impact on the observed flow in the system. It is
critical to account for spatially varied rainfall to perform an accurate frequency analysis.
• A hydraulic restriction in the system was uncovered through a process called null
modeling that isolated the model performance to the hydraulics. This restriction was not
found during the event modeling because hydrologic model imprecision masked the
hydraulic effects of the restriction. It is important to isolate the hydraulic model to
accurately assess the system hydraulic performance.
• A frequency basis may be preferable to a design storm event for regulatory agencies. The
use of a frequency analysis eliminates the need to develop “average conditions” or to
assume a capture coefficient for design event simulations.
• Use of the antecedent moisture model as opposed to a design storm event eliminated
many of the conservatisms that are frequently included in design event models, can result
in the identification of cost effective projects to address capacity deficiencies, and
provide a greater degree of confidence that the system will perform as predicted.

ACKNOWLEDGEMENTS

We would like to acknowledge the contributions of the following people for their hard work,
support and collaboration during the development of this unique and innovative project:

• The staff of Wayne County’s Department of the Environment, especially Sue Hanson and
John Baratta
• Tobin Van Pelt of i3DLab
• Tom Knueve, Phil Argiroff and Doug Early of the MDEQ
• Steve Kalinowski of Wade-Trim
• Karen Ridgway of Applied Science

REFERENCES

R.S. Czachorski and T.H. Van Pelt, On the Modeling of Inflow and Infiltration within Sanitary
Collection Systems for Addressing Nonlinearities Arising from Antecedent Moisture Conditions,
WEFTEC 2001 (poster symposium), Atlanta, GA.

T.H. Van Pelt and R.S. Czachorski, The Application of System Identification to Inflow and
Infiltration Modeling and Design Storm Event Simulation For Sanitary Collection Systems,
WEFTEC 2002, Chicago, IL

Michigan Department of Environmental Quality, Sanitary Sewer Overflow Policy Statement,
December 27, 2002

Michigan Department of Environmental Quality, Sanitary Sewer Overflow Policy Clarification
Statement, 2003

Federal Emergency Management Agency, Guidance and Specifications for Flood Hazard
Mapping Partners, February, 2002

Regulatory, Technical and Modeling Challenges to Developing a Frequency Based SSO Control Project in Wayne County, Michigan

More Related Content

What's hot (10)

Viewers also liked (20)

Similar to Regulatory, Technical and Modeling Challenges to Developing a Frequency Based SSO Control Project in Wayne County, Michigan (20)

More from OHM Advisors (20)

Recently uploaded (20)

Regulatory, Technical and Modeling Challenges to Developing a Frequency Based SSO Control Project in Wayne County, Michigan