United States
Environmental Protection
Agency
Report on Enhanced Framework (SUSTAIN)
and Field Applications for Placement of
BMPs in Urban Watersheds
7 Interpretation
(Postprocessor)
Office of Research and Development
National Risk Management Research Laboratory- Watei
-------
Report on Enhanced Framework (SUSTAIN) and Field Applications for
Placement of BMPs in Urban Watersheds
by
Leslie Shoemaker, Ph.D.
John Riverson, Jr.
Khalid Alvi, P.E.
Jenny X. Zhen, Ph.D., P.E.
Ryan Murphy
Tetra Tech, Inc.
10306 Eaton Place, Suite 340
Fairfax, VA 22030
In Support of
EPA Contract No. GS-10F-0268K
Project Officer
Ariamalar Selvakumar, Ph.D., P.E.
Urban Watershed Management Branch
Water Supply and Water Resources Division
2890 Woodbridge Avenue (MS-104)
Edison, NJ 08837
National Risk Management Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH 45268
September 2011
-------
Disclaimer
The work reported in this document was funded by the U.S. Environmental Protection Agency (EPA)
under Task Order 1108 of Contract No. GS-10F-0268K to Tetra Tech, Inc. Through its Office of
Research and Development, EPA funded and managed, or partially funded and collaborated in, the
research described herein. This document has been subjected to the Agency's peer and administrative
reviews and has been approved for publication. Any opinions expressed in this report are those of the
authors and do not necessarily reflect the views of the Agency; therefore, no official endorsement should
be inferred. Any mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
-------
Foreword
The U.S. Environmental Protection Agency (EPA) is charged by Congress with protecting the nation's
land, air, and water resources. Under a mandate of national environmental laws, the Agency strives to
formulate and implement actions leading to a compatible balance between human activities and the ability
of natural systems to support and nurture life. To meet this mandate, EPA's research program is
providing data and technical support for solving environmental problems today and building the science
knowledge base necessary to manage our ecological resources wisely, understand how pollutants affect
our health, and prevent or reduce environmental risks in the future.
The National Risk Management Research Laboratory (NRMRL) is the Agency's center for investigation
of technological and management approaches for preventing and reducing risks from pollution that
threaten human health and the environment. The focus of the Laboratory's research program is on
methods and their cost-effectiveness for prevention and control of pollution to air, land, water, and sub-
surface resources; protection of water quality in public water systems; remediation of contaminated sites,
sediments, and groundwater; prevention and control of indoor air pollution; and restoration of eco-
systems. NRMRL collaborates with both public- and private-sector partners to foster technologies that
reduce the cost of compliance and to anticipate emerging problems. NRMRL's research provides
solutions to environmental problems by developing and promoting technologies that protect and improve
the environment; advancing scientific and engineering information to support regulatory and policy
decisions; and providing the technical support and information transfer to ensure implementation of
environmental regulations and strategies at the national, state, and community levels.
This publication has been produced as part of the Laboratory's strategic long-term research plan. It is
published and made available by EPA's Office of Research and Development to assist the user
community and to link researchers with their clients.
Sally Gutierrez, Director
National Risk Management Research Laboratory
-------
Abstract
The System for Urban Stormwater Treatment and Analysis Integration (SUSTAIN) was developed by the
U.S. Environmental Protection Agency (EPA) to provide Stormwater managers with a decision support
system for the cost-efficient selection and placement of Stormwater best management practices (BMPs) in
urban watersheds. The SUSTAIN modeling system integrates simulation based on sound science and
engineering principles, with cost estimation tools, and optimization to support users in selecting the best
solutions on the basis of cost and effectiveness. This report documents the enhancements to the system
since the initial release of version 1.0 in October 2009 (Shoemaker et al., 2009;
http://www.epa.gov/nrmrl/wswrd/wq/models/sustain). Two case studies also provide insight into the
application of the system, and demonstrate the utility of SUSTAIN in evaluating the use of green
infrastructure (GI) in communities addressing the mitigation of combined sewer overflows (CSOs).
SUSTAINS innovative integration of optimization and simulation of BMP performance in a watershed
setting provides significant capabilities to support the evaluation of various configurations of BMPs, the
impact of local site conditions on BMP placement, and the types and relative costs of the BMPs available.
At the same time, the flexibility and range of application complexity available within SUSTAIN provide
choices for users in developing the appropriate configuration for a specific watershed application.
Through the use of case studies in Kansas City, MO, and Louisville, KY, the implications of assumptions
in the application of SUSTAIN are evaluated. The SUSTAIN model including recent enhancements was
demonstrated to perform well in predicting the hydrologic response and matched previous applications
using other modeling systems. In particular, the addition of a sub-hourly time step improved the ability of
SUSTAIN to predict hydrologic response and peak flow from design storms used as a basis for planning
many CSO and Stormwater programs. The addition of aggregate BMP tools facilitated the use of the
model in Louisville and other regions where users want to evaluate the benefits of many, in some cases
hundreds or thousands of smaller BMPs across a large catchment. The optimization process applied in
the case studies was also shown to be highly sensitive to BMP cost data in selecting the most efficient
solutions.
-------
Executive Summary
The System for Urban Stormwater Treatment and Analysis Integration (SUSTAIN) was developed by the
U.S. Environmental Protection Agency (EPA) to provide Stormwater managers with a decision support
system for the cost-efficient selection and placement of Stormwater best management practices (BMPs) in
urban watersheds. The SUSTAIN modeling system integrates simulation based on sound science and
engineering principles with cost estimation tools and optimization to support users in selecting the best
solutions on the basis of cost and effectiveness. This report demonstrates the utility of SUSTAIN and
showcases its power and flexibility as a decision making platform both for evaluating current
management decisions and for future expansion of the science.
The report provides an overview of the system and documents the system enhancements made since the
initial release of version 1.0 in October 2009 (Shoemaker et al. 2009;
http://www.epa.gov/nrmrl/wswrd/wq/models/sustain). Two regional case studies explore the use of
SUSTAIN in evaluating green infrastructure (GI) management alternatives for mitigation of combined
sewer overflows (CSOs) in Kansas City, Missouri, and Louisville, Kentucky. The case study applications
provide an overview of the user application process from start to finish, including problem formulation,
data collection, calibrating a baseline condition, BMP parameterization, and optimization. Through the
case study application process the common themes and recommendations for how SUSTAIN can be
applied and interpreted are presented.
SUSTAINhas seven components, including: (1) Framework Manager, (2) BMP Siting Tool, (3)
Watershed/Conveyance Module, (4) BMP Module, (5) Cost Module, (6) Optimization Module, and (7)
Post-Processor. The modular components are integrated under a common ArcGIS platform, which
performs hydrologic and water quality modeling in watersheds and urban streams and searches for
optimal management solutions at multiple-scale watersheds to achieve desired water quality objectives
based on cost effectiveness. SUSTAINS unique system design provides significant capabilities to support
the evaluation of various configurations of BMPs, the impact of local site conditions on BMP placement,
and the types and relative costs of the BMPs available. At the same time, the flexibility and range of
application complexity available within SUSTAIN pro vide choices for users in developing the appropriate
configuration for a specific watershed application.
Several enhancements to SUSTAIN were identified earlier in the current phase of the system development
during workshops of invited national experts in July 2010 and through development of the two case
studies. The list of planned enhancements were selected on the basis of their ability to fulfill the goal of
improving SUSTAIN by (1) enhancing existing functionality, (2) developing additional system
capabilities, (3) increasing simulation accuracy, and (4) promoting the use of SUSTAIN through regional
case studies. Enhancements were prioritized based on their potential to advance the applicability,
accuracy, and functionality of SUSTAIN and based on the needs for application to the case study areas
that emphasize CSOs. The updates to SUSTAIN included both operational and technical enhancements
affecting most of SUSTAINS core modules: Framework Manager (ArcGIS), BMP Siting Tool, and the
BMP, Watershed/Conveyance, and Optimization Modules. The first set of completed SUSTAIN
enhancements include:
IV
-------
Enhancement
Siting Tool functionality
Sub-hourly time step simulation
Morton infiltration method
Soil recovery and initial
conditions
Pump curve
Area BMP
Groundwater system
Point source time series
Description
More functionality for selection, placement, and prioritization of suitable
BMP sites
Improved performance for hydrologic and hydraulic peak flow prediction
Method added to Watershed and BMP modules as a third option for
calculating infiltration
Evapotranspiration multiplier and initial moisture conditions added to the
BMP module to provide more flexibility for model configuration and "hot"
simulation
A user-specified option (besides the available weir, orifice, or underdrain
outlets) for dewatering a storage element
New option for representing practices such as disconnected imperviousness
Accounting module for tracking and releasing infiltrated water from a BMP
into a conveyance segment
New option that allows for a user-specified inflow to a node in the routing
network (e.g. dry weather sewer flow and pollutant loads to a CSO
regulator)
The case study application of SUSTAIN in Kansas City, Missouri, demonstrates the use of GI for CSO
control through two general sets of goals: (1) management goals to inform the decision-making process
and (2) modeling goals to test system functionality and provide application guidance for the user
community. The case study examines a 480-acre sewershed in the Middle Blue River watershed. This
sewershed includes a planned 100-acre BMP implementation/monitoring site in the Battleflood Heights
neighborhood and 86 acres of an adjacent control area where only monitoring is planned. The case study
application combines a number of key outcomes from local and regional efforts, including elements of an
Overflow Control Plan and an accepted GI design plan that was being constructed in the watershed at the
time of the case study.
The model application established an existing condition baseline by using an existing model and updating
it based on more recently collected monitoring data that spanned a wider range of storms and included
larger events that were more representative of the CSO critical condition. Because of the size and
complexity of the study area, selective simplifications where used to streamline the model application
while preserving the physical responses, including: (1) grouping similar areas into Hydrologic Response
Units (HRUs), (2) consolidating the subcatchment delineation, and (3) aggregating the hydraulic network
representation. For a portion of the pilot area, SUSTAINS "aggregated BMP" approach, which represents
a coarser version of a detailed routing network, was tested against an articulated network to understand
the effects that those simplifications have on predicted storm response. Comparison of aggregated and
articulated networks showed that even though there were notable attenuation differences for small storms,
the larger critical condition storm response was well maintained. The optimization objective was to fully
contain a critical condition design storm runoff response at the lowest cost, using a mix of GI and gray
management options. Controlling the critical condition design storm achieves the regional allowable
exceedance frequency criteria for CSO.
The case study in Louisville, Kentucky also demonstrates the use of GI for CSO control. However, there
are some differences from the Kansas City case study in the specific goals and, consequently, the
application sequence. For example, one of the goals for the Louisville case study was to conduct
sensitivity tests of key BMP model parameters. The focus area for this case study is the Lousiville-
Jefferson County Metropolitan Sewer District (MSB) CSO 019 sewershed, located west of downtown
-------
Louisville. The sewershed drains 1,094 acres of mixed land use dominated by single-family residential
neighborhoods. The existing CSO 019 outfall is on the north edge of the sewershed and discharges
directly into the Ohio River. Because MSB already has a preferred model for the sewershed, another goal
of the case study is to use SUSTAIN to replicate the critical condition response of that model. The MSB
model estimates that the sewershed produces overflow volumes for a certain number of discrete events
based on a 2001 typical year precipitation record. Instead of controlling a design storm, the optimization
objective of this case study is to achieve the regional allowable exceedance frequency criteria for CSO at
the lowest cost, using continuous simulation for the 2001 typical year. Available options include a mix of
GI and gray management activities.
These case studies provide users with an overview of two urban settings and demonstrate how SUSTAIN
was used to support a cost-benefit evaluation of CSO management alternatives. The two case studies
have also shown how SUSTAIN was used to analyze, streamline, and extrapolate BMP representation
throughout the respective study areas and demonstrated how to evaluate various combinations of green
and gray management alternatives. The case study applications led to the follow general observations and
conclusions:
• SUSTAIN is a comprehensive decision support tool with many useful features and functions.
Successful and meaningful application largely depends on accurate representation of the baseline,
BMP alternatives, and the associated BMP costs.
• SUSTAIN application process is iterative and adaptive, meaning that once the SUSTAIN modeling
framework is established, it can be adapted to answer various management questions and test
underlying assumptions.
• Model simplification becomes critical when optimization is applied to a larger area or when
multiple smaller BMPs are distributed widely across a catchment. The aggregate BMP concept
and utility provided in SUSTAIN is proven to be a viable and useful technique in the evaluation of
the benefit of stormwater management practices, especially smaller GI practices, across a large
area. When the aggregate BMP tools are used, the appropriate aggregation spatial scale should be
carefully selected to maintain reasonable predictive capability and accuracy.
• The optimization process is highly sensitive to BMP cost data used in selecting solutions for each
application. As a result, performance of sensitivity analysis and evaluation of cost control
measures or economies of scale are recommended wherever SUSTAIN is applied.
VI
-------
Disclaimer i
Foreword ii
Abstract iii
Executive Summary iv
Acknowledgements xiii
Abbreviations and Acronyms xiv
Chapter 1. Introduction 1-1
1.1. SUSTAIN Enhancements 1-1
1.1.1. Enhancements to the Framework Manager (ArcGIS) 1-3
1.1.2. Enhancements to the BMP Siting Tool 1-4
1.1.3. Enhancements to the BMP Simulation Module 1-4
1.1.4. Enhancements to the Watershed Module 1-8
1.1.5. Enhancements for Optimization Efficiency 1-9
1.2. Release of SUSTAIN 1.2 1-10
Chapter 2. Case Study: Kansas City, Missouri 2-1
2.1. Background 2-1
2.1.1. Overflow Control Plan: XP-SWMM Model 2-3
2.1.2. Site-Specific Monitoring Data 2-4
2.1.3. Green Alternatives for Sewershed 059 & 069 Technical Memorandum and Overflow
Control Plan 2-6
2.1.4. Middle Blue River Green Solutions Pilot Project 2-7
2.1.5. WinSLAMM Modeling for Private Residential BMPs 2-12
2.2. Overview of Case Study Objectives 2-12
2.2.1. Establishing a Sewershed Model Baseline 2-13
2.2.2. Simplifying the Network Articulation for Large-Scale Extrapolation 2-14
2.2.3. Optimize BMP Opportunity for CSO Mitigation in the 069 Sewershed 2-14
2.2.4. Evaluate the Influence of Model Time Step on Optimization Results 2-15
2.3. Establishing a Sewershed Model Baseline 2-15
2.3.1. Development of Hydrologic Response Units 2-16
2.3.2. Subcatchment Delineation 2-17
2.3.3. Watershed Model Calibration 2-22
2.4. Simplifying the Network Articulation for Large-Scale Extrapolation 2-33
2.4.1. SUSTAIN BMP Representation 2-33
2.4.2. Articulated versus Aggregated Network 2-36
2.5. Optimizing BMP Opportunity for CSO Mitigation in the 069 Watershed 2-42
2.5.1. CSO 069 Model Configuration 2-43
2.5.2. Problem Formulation 2-46
2.5.3. BMP Cost Representation 2-47
2.5.4. Optimization Sensitivity Tests 2-49
2.5.5. Exploratory Management Scenarios 2-52
2.5.6. Validating Overflow Control Using Continuous Simulation for a Typical Year 2-54
2.5.7. Comparison of Gray versus Green Overflow Reduction Effectiveness 2-57
2.5.8. Optimization Summary and Conclusions 2-58
VII
-------
Chapter 3. Case Study: Louisville, Kentucky 3-1
3.1. Background 3-1
3.1.1. InfbWorks Model 3-3
3.1.2. Portland Wharf Storage Basin 3-3
3.2. Overview of Case Study Goals 3-4
3.2.1. Replication of an Existing Hydraulics Model 3-4
3.2.2. BMP Parameter Sensitivity Analysis 3-5
3.2.3. Cost-benefit Relationship between Gray and Green Infrastructure for Mitigating CSOs3-6
3.3. Replication of an Existing Hydraulics Model 3-6
3.3.1. Land Cover Development 3-8
3.3.2. Subcatchment Delineation 3-9
3.3.3. Review of Baseline Model Calibrations 3-14
3.3.4. CSO 019 Regulator Calibration 3-22
3.3.5. Model Run-Time Considerations 3-24
3.4. BMP Parameter Sensitivity Analysis 3-25
3.4.1. BMP Representation 3-26
3.4.2. Factorial Experiential Design 3-31
3.4.3. Results 3-32
3.5. Cost-benefit Relationship between Gray and Green Infrastructure for Mitigating CSO 3-34
3.5.1. Green Infrastructure Opportunities 3-35
3.5.2. SUSTAIN BMP Representation 3-36
3.5.3. SUSTAIN Portland Wharf Storage Basin Representation 3-39
3.5.4. BMP Cost Representation 3-39
3.5.5. Exploratory Management Scenarios 3-41
3.5.6. Optimization Problem Formulation 3-41
3.5.7. Optimization Results 3-42
3.5.8. Optimization Sensitivity Analyses 3-44
3.5.9. Optimization Summary and Conclusions 3-48
Chapter 4. Lessons Learned 4-1
4.1. Management Lessons 4-2
4.1.1. What are some of the factors that most influence cost-effectiveness of both GI and Gray
Infrastructure? 4-3
4.1.2. How does the control target affect cost-effectiveness of management alternatives? 4-6
4.1.3. Can GI be used effectively to complement existing or planned Gray? 4-7
4.2. Modeling Lessons 4-8
4.2.1. What are some of the critical data that are required for performing these evaluations?.. 4-8
4.2.2. How detailed does the model needs to be in order to properly represent the system? ..4-10
4.2.3. How can one demonstrate that a model is adequately representative of the system? ....4-10
4.2.4. How is the model applied in an iterative, adaptive process? 4-11
Chapter 5. Conclusions and Recommendations 5-1
Chapter 6. References 6-1
VIM
-------
Figures
Figure 1-1. Diagram highlighting the modules to be enhanced in SUSTAIN 1-2
Figure 1-2. Conceptual illustration of user-defined BMP initial soil moisture parameters 1-5
Figure 1-3. Conceptual flow diagram of Area BMP simulation 1-6
Figure 1-4. Conceptual illustration of the BMP pump curve 1-7
Figure 1-5. Sample data format for point source time series 1-9
Figure 2-1. Location of pilot study area within the CSO 069 sewershed boundaries 2-2
Figure 2-2. Annual precipitation for Kansas City 2-3
Figure 2-3. Storm size distributions for a typical Kansas City meteorological year 2-5
Figure 2-4. Histogram of monitored against the long-term historical precipitation record 2-5
Figure 2-5. BMP layout in the 100-acre pilot study area 2-8
Figure 2-6. Bioretention with underground storage cross-section profile 2-9
Figure 2-7. Bioswale cross-section profiles 2-9
Figure 2-8. Cascade plan view and cross-section profile 2-10
Figure 2-9. Porous sidewalk cross-section profile 2-10
Figure 2-10. Rain garden cross-section profile 2-10
Figure 2-11. Below-grade storage outlet structure 2-11
Figure 2-12. CSO 069 sewershed slope analysis 2-18
Figure 2-13. CSO 069 sewershed surface cover analysis 2-19
Figure 2-14. CSO 069 sewershed HRUs 2-20
Figure 2-15. Comparison of XP-SWMM subcatchments and subcatchment aggregation in
SUSTAIN. 2-21
Figure 2-16. Comparison of HRU distributions within XP-SWMM and sewershed boundaries 2-22
Figure 2-17. Comparison of XP-SWMM and SUSTAIN subwatershed delineations 2-24
Figure 2-18. D-storm hourly hyetographs for high, medium, and low recovery scenarios 2-26
Figure 2-19. Observed versus modeled runoff at the watershed outlet for 10 selected storms 2-28
Figure 2-20. Observed versus modeled runoff at the watershed outlet for 10 selected storms 2-29
Figure 2-21. Modeled versus observed volume and peak flow correlations, with WaPUG
criteria 2-29
Figure 2-22. Initial calibration: UMKC-01 catchment outlet (rainfall = 1.76 in.) 2-30
Figure 2-23. Intermediate calibration: UMKC-01 outlet (adjusted rainfall depth =1.13 in.) 2-31
Figure 2-24. Final calibration: UMKC-01 outlet (adjusted rainfall depth = 1.44 in.) 2-31
Figure 2-25. Example BMP renderings: Middle Blue River Green Solutions Pilot Project 2-34
Figure 2-26. SUSTAIN model representation of fully articulated model network 2-36
Figure 2-27. Conceptual diagram of a comparable aggregate BMP representation 2-37
Figure 2-28. Hourly time step, aggregated versus articulated (baseline, no BMPs) 2-39
Figure 2-29. Fifteen minute time step, aggregated versus articulated (baseline, no BMPs) 2-39
Figure 2-30. Hourly time step, aggregated versus articulated (with BMPs) 2-40
Figure 2-31. Fifteen minute time step, aggregated versus articulated (with BMPs) 2-40
Figure 2-32. Conceptual sequence of optimization scenarios relative to baseline conditions 2-42
Figure 2-33. Subwatersheds, pipe connections, and regulator assessment point for CSO 069 2-44
Figure 2-34. CSO 069 regulator schematic 2-45
Figure 2-35. Conceptual schematic for the CSO regulator 2-46
Figure 2-36. Sensitivity of model simulation time step on optimization results 2-49
Figure 2-37. Comparison of antecedent recovery conditions forthe D-storm 2-50
Figure 2-38. Sensitivity of antecedent moisture conditions on optimization results 2-51
Figure 2-39. Cost-effectiveness junctions and trajectories for exploratory optimization
scenarios 2-53
Figure 2-40. Comparison of overflow compliance costs forthe three exploratory scenarios 2-53
Figure 2-41. Hyetograph forthe March 4, 2004, storm event 2-55
IX
-------
Figure 2-42. Hyetograph for the August 27, 2004, storm event 2-56
Figure 2-43. Hyetograph of the June 9, 2004, storm event 2-56
Figure 2-44. Comparison of CSO 069 number of overflows with green versus gray storage
capacities 2-57
Figure 2-45. Comparison on annual overflow volume reduction per unit storage volume
provided 2-58
Figure 3-1. Location of CSO 019 sewershed 3-2
Figure 3-2. Proposed location of the Portland Wharf Storage Basin and Pump Station 3-4
Figure 3-3. InfoWorks model configuration exported to EPA-SWMM5 3-7
Figure 3-4. CSO 019 sewershed slope derived from topographic contours 3-10
Figure 3-5. CSO 019 sewershed InfoWorks model slope analysis 3-11
Figure 3-6. CSO 019 sewershed surface cover distribution 3-12
Figure 3-7. Comparison of InfoWorks and SUSTAIN subwatershed delineations 3-13
Figure 3-8. Distribution of typical year 2001 precipitation data by month 3-14
Figure 3-9. Monthly distribution of typical daily evaporation rates 3-16
Figure 3-10. Conceptual data flow sequence for baseline calibration and BMP scenario model
runs 3-18
Figure 3-11. Calibrated versus actual impervious flow 3-18
Figure 3-12. InfoWorks versus SUSTAIN modeled inflow volume and overflow peak 3-19
Figure 3-13. Percent difference between SUSTAIN and InfoWorks model calibration metrics 3-20
Figure 3-14. Plot of Nash-Sutcliffe by storm size for SUSTAIN versus InfoWorks regulator
inflows 3-21
Figure 3-15. Conceptual schematic for the CSO Regulator 3-22
Figure 3-16. Conceptual cross-section of the CSO 019 outfall structure 3-23
Figure 3-17. InfoWorks vs. SUSTAIN modeled overflow volume and overflow peak 3-23
Figure 3-18. Comparisons of single simulation and optimization modeling run-times 3-25
Figure 3-19. Office of Employment bioretention cell site location and drainage area 3-27
Figure 3-20. Bioretention cell subsurface cross-section from design plans (Strand Associates,
2010) 3-28
Figure 3-21. SUSTAIN surface parameter input screens for bioretention cell 3-29
Figure 3 -22. SUSTAIN substrate parameter input screens for bioretention cell 3-30
Figure 3 -23. SUSTAIN infiltration parameter input screens for bioretention cell 3-30
Figure 3 -24. Average annual reduction in total BMP outflow for all eight scenarios 3-33
Figure 3-25. Conceptual sequence of optimization scenarios relative to baseline condition 3-34
Figure 3-26. CSO 019 GI opportunities and treated drainage areas 3-35
Figure 3-27. Examples of GI practices for Louisville, Kentucky 3-36
Figure 3-28. CSO 019 GI BMP drainage networks 3-37
Figure 3-29. Portland Warf Storage Basin cost function (in 2008 dollars) 3-40
Figure 3-30. Cost-effectiveness curves for exploratory management scenarios 3-42
Figure 3-31. Four selected GI solutions along the cost-effectiveness curve 3-43
Figure 3-32. GI BMP percent utilization at four selected solutions 3-44
Figure 3-33. Sensitivity analysis for GI cost assumptions 3-46
Figure 3-34. Comparison of full build-out of GI for the three cost scenarios 3-47
Figure 3-35. Comparison of GI cost-effectiveness curves size to treat 0.75 in. and 1.00 in. of
runoff 3-47
Figure 4-1. Theoretical construct for CSO management optimization problems 4-2
Figure 4-2. Kansas City 069 land cover distribution and impervious area distribution tributary
toGI 4-4
Figure 4-3. Louisville 019 land cover distribution and impervious area distribution tributary to
GI 4-4
-------
Figure 4-4. Typical capture and dewatering modes of GI and gray BMPs in a CSO collection
system 4-6
Figure 4-5. Comparison of overflow volumes for Kansas City exploratory management
scenarios 4-8
Figure 4-6. SUSTAIN application sequence 4-11
XI
-------
Tables
Table 1-1. Summary of enhancements to the core modules in SUSTAIN 1-2
Table 1-2. Summary of completed BMP module enhancements 1-4
Table 1-3. Summary of completed watershed module enhancement 1-8
Table 2-1. Summary of design storms used for CSO control plan 2-4
Table 2-2. Gray infrastructure CSO controls for outfall 069 2-6
Table 2-3. Summary of BMP design plan components 2-8
Table 2-4. Decision-making questions and expected outcomes by study 2-13
Table 2-5. Roughness and depression storage parameters by land cover type 2-23
Table 2-6. Green-Ampt infiltration parameters 2-23
Table 2-7. Summary of antecedent recovery conditions for the D-storm 2-25
Table 2-8. Model calibration performance metrics for 10 selected storms events 2-32
Table 2-9. BMP design dimensions and specifications 2-34
Table 2-10. Subsurface layer properties for applicable BMP layers 2-35
Table 2-11. Private BMP design dimensions and specifications 2-35
Table 2-12. Reference matrix for aggregate versus articulated BMP comparison tests 2-38
Table 2-13. Comparison of model characteristics for three configurations 2-41
Table 2-14. SUSTAIN application data needs and associated data source (research study) 2-43
Table 2-15. BMP capital costs for the 069 sewershed 2-47
Table 2-16. Cost estimation for private parcel retrofit BMPs 2-48
Table 2-17. Sensitivity of cost-effectiveness to changes in antecedent moisture condition 2-51
Table 2-18. Summary and description of baseline and exploratory optimization scenarios 2-52
Table 2-19. Management component size and costs for exploratory optimization scenario 2-54
Table 2-20. Storms summary for the six largest storm events in 2004 2-54
Table 2-21. Overflow events summary 2-55
Table 3-1. Roughness and depression storage parameters for pervious land cover 3-15
Table 3-2. Area-weighted average InfoWorks model parameters for CSO 019 3-15
Table 3-3. Area-weighted subwatershed parameter ranges applied in SUSTAIN 3-15
Table 3-4. Horton infiltration parameters from Louisville Infoworks models 3-16
Table 3-5. Model calibration performace metrics for eight largest storms events causing
overflow 3-21
Table 3-6. Summary of additional predicted overflows 3-24
Table 3-7. Comparison of model representations and run-time 3-24
Table 3-8.Summary of BMP parameters used for bioretention cell configuration 3-28
Table 3-9. Suggested information sources for obtaining BMP parameters 3-31
Table 3-10. Low and high values selected for three evaluated BMP parameters 3-32
Table 3-11. Matrix of the eight designed experimental runs showing values for the three
parameters 3-32
Table 3-12. Summary of average annual BMP variation in response between low and high
conditions 3-33
Table 3-13. BMP design dimensions and specifications 3-37
Table 3-14. Distribution of impervious areas that can be treated by GI 3-38
Table 3-15. GI BMP construction cost (in 2011 dollars) 3-40
Table 3-16. Summary and description of baseline exploratory management scenarios 3-41
Table 3-17. Comparison of BMP costs per gallon of treatment capacity from various sources 3-45
Table 4-1. Comparison of case study components that influenced management alternatives 4-3
Table 4-2. Summary of factors that most influence cost-effectiveness 4-5
XII
-------
Acknowledgements
The Tetra Tech project team would like to thank EPA's project officer, Dr. Ariamalar Selvakumar, for her
active involvement in the system development process, SUSTAIN application to two case studies, and for
reviewing this report. We appreciate the continuous support and recognition of the significance of this
project by Sally Gutierrez, Director of EPA Office of Research and Development (ORD), National Risk
Management Research Laboratory, Dr. Thomas Speth, Division Director of Water Supply and Water
Resources Division, and Dr. Michelle Simon, Chief of Urban Watershed Management Branch.
We appreciate the review and insight of the internal reviewers Richard Field, Michael Borst, Dr. Lew
Rossman, and Dr. Joong Lee of EPA ORD and an external reviewer Dr. Dino Marshalonis of EPA
Region 10. We would also like to thank Kansas City, Missouri for use of data related to the Kansas City
case study. Similarly, we would like to thank Louisville Metropolitan Sewerage District for use of data
related to the Louisville case study.
We would also like to acknowledge Dr. Scott Struck and Carol Hufnagel, employees of Tetra Tech, for
their contribution to the SUSTAIN application in Kansas City and Louisville CSO sewersheds.
Finally, we would like to extend our appreciation to an EPA retiree, Dr. Fu-hsiung (Dennis) Lai, for his
insight and support to this project. He was the Project Officer during the project Phase 1, Phase 2, and
early part of Phase 3 until 2011.
-------
Abbreviations and Acronyms
BMP
CIP
CSO
CWP
CSS
DCIA
ENR
EPA
ET
GI
GIS
HRU
HSPF
IOAP
LTCP
MG
MSD
NCDC
NRMRL
O&M
OCP
ORD
PEVT
scs
SUSTAIN
SWMM
UMKC
WaPUG
WinSLAMM
WSD
Best Management Practice
Capital Improvement Plan
Combined Sewer Overflow
Center for Watershed Protection
Combined Sewer System
Directly Connected Impervious Area
Engineering News-Record
United States Environmental Protection Agency
Evapotranspiration
Green Infrastructure
Geographic Information System
Hydrologic Response Unit
Hydrologic Simulation Program—FORTRAN
Integrated Overflow Abatement Plan
Long-Term Control Plan
Million Gallons
Louisville-Jefferson County Metropolitan Sewer District
National Climatic Data Center
National Risk Management Research Laboratory
Operation and Maintenance
Overflow Control Plan
Office of Research and Development
Potential Evapotranspiration
Soil Conservation Services
System for Urban Stormwater Treatment and Analysis Integration
Storm Water Management Model
The University of Missouri-Kansas City
Wastewater Planning Users Group
Source Loading and Management Model for Windows
Kansas City, Missouri, Water Services Department
XIV
-------
Chapter 1. Introduction
The U.S. Environmental Protection Agency (EPA) initiated a research project in 2002 to develop a
decision support system for selection and placement of stormwater best management practices (BMPs) at
strategic locations in mixed land use urban watersheds. The primary objective of the system is to provide
stormwater management professionals with a BMP assessment tool based on sound science and
engineering principles that helps develop, evaluate, select, and place BMP options on the basis of cost and
effectiveness. Phases 1 and 2 of this effort culminated in the release of the System for Urban Stormwater
Treatment and Analysis Integration (SUSTAIN) version 1.0 in October 2009 (Shoemaker et al., 2009) and
is publicly available on the EPA SUSTAIN web site
(http://www.epa.gov/nrmrl/wswrd/wq/models/sustain).
Since the release of SUSTAIN version 1.0, EPA has initiated Phase 3, to further enhance SUSTAIN's
functionality, capabilities, and accuracy with the goal of releasing SUSTAIN version 2.0 in 2012 (Lai et
al., 2010). In addition to improving and enhancing the modeling system, the scope of Phase 3 includes
continued support for the use of SUSTAIN by user groups through ongoing technical support, training and
workshops, and the development of regional case studies to further demonstrate and test the model.
This report documents the enhancements and updates to the system since the release of Version 1.0, and
the results of two case study applications. Chapter 1 provides detailed documentation of the system
enhancements and their role in creating a richer, more productive tool for the user community. Chapter 2
and Chapter 3 present two regional case studies for combined sewer overflow (CSO) communities in
Kansas City, Missouri and Louisville, Kentucky, respectively. The case study applications provide an
overview of the user application process from start to finish, and the critical steps in the process including
problem formulation, data collection, calibrating a baseline condition, BMP parameterization, and
optimization. Chapter 4 discusses common themes and insights derived from the case study applications
on how SUSTAIN can be applied and interpreted when used to support decision-making in CSO and
stormwater communities. Chapter 5 provides concluding remarks and recommendations for future
research and analysis.
1.1. SUSTAIN Enhancements
Several enhancements to SUSTAIN were identified in the Phase 3 scope, during expert workshops in July
2010, and through development of case studies in Kansas City, Missouri (Chapter 2) and Louisville,
Kentucky (Chapter 3). The list of planned enhancements were selected on the basis of their ability to
fulfill the goal of improving SUSTAIN by (1) enhancing existing functionality; (2) developing additional
system capabilities; (3) increasing simulation accuracy; and (4) promoting the wide use of SUSTAIN
through regional case studies. Enhancements were prioritized based on their potential to advance the
applicability, accuracy, and functionality of SUSTAIN, consistent with the Phase 3 objectives; and based
on the needs for application to the case study areas that emphasize CSOs.
The updates to SUSTAIN under Phase 3 included both operational and technical enhancements affecting
most of SUSTAIN's core modules: Framework Manager (ArcGIS), BMP Siting tool, BMP, Watershed,
and Optimization. These enhancements, as implemented, provided improved support for the application
1-1
-------
of the model to the case studies as demonstrated in Chapter 2 and Chapter 3. Figure 1-1 and Table 1-1
provide an overview of the modules and the specific enhancements described in the following sections.
Table 1-1 identifies both planned and completed enhancements to each of the core modules.
AC Manager
Interpretation
(Post Processor)
SUSTAIN Enhancements
Figure 1-1. Diagram highlighting the modules to be enhanced in SUSTAIN
Table 1-1. Summary of enhancements to the core modules in SUSTAIN
Description
• = Enhancement completed
© = Enhancement under development
— = Enhancement does not address
Public or private land constraint for
suitable site selection of BMPs
Proximity to land features for suitable
site selection of BMPs
Rank suitable sites based on slope and
infiltration rate
Enhanced suitable site selection and
placement of BMPs on the map
Morton infiltration method
Evapotranspiration (ET) multiplier
BMP initial moisture conditions
Pump curve
Sub-hourly time step
Area BMP (new BMP)
Framework
Manager
(ArcGIS)
•
•
•
•
( )
BMP Siting
Tool
•
•
•
•
-
-
-
-
-
-
BMP
Module
-
-
-
-
Watershed
Module
-
-
-
-
•
-
-
•
•
-
Optimization
Module
-
-
-
-
-
-
-
-
-
•
1-2
-------
Description
• = Enhancement completed
© = Enhancement under development
— = Enhancement does not address
Groundwater system
Point source time series
Variable time step (dry/wet)
Improved simulation process to address
sediment and pollutant trapping in BMP
Develop templates for selection of
permeable pavement technologies
Operation and maintenance (O&M)
factors
Check dams (new BMP)
Variable number of sediment classes
Infiltration temperature correction
factor (viscosity)
Plug flow method
Enhanced BMP templates (interfaces)
Update sediment associated pollutant
loading algorithms for internal land
simulation option
Develop a user interface with CSO
models for evaluation of green versus
gray infrastructure options
Street sweeping (new BMP)
Improve infiltration recovery factor
Curve number method
Dynamic wave routing in conduits
Framework
Manager
(ArcGIS)
•
•
(V)
-
0
-
0
-
-
(0
0
-
0
0
-
(0
0
BMP Siting
Tool
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
BMP
Module
-
•
©
©
©
©
©
©
©
©
©
-
-
©
©
-
-
Watershed
Module
•
•
(V)
-
-
-
-
(V)
-
-
-
(0
0
0
(0
(0
0
Optimization
Module
-
-
(V)
-
-
(0
0
-
-
-
-
-
-
0
-
-
-
1.1.1. Enhancements to the Framework Manager (ArcGIS)
The framework manager includes the user interface and the linkage to all the core modules of the system
and is based on the ArcGIS system. The framework manager is essential to the overall operation and
application of the model. Table 1-1 shows the list of enhancements to the framework manager in Phase 3
of the project. Each change or enhancement to the various system modules, documented in this chapter,
also include related updates to system interfaces in the framework manager to link the relevant model
input parameters and data management. In this version, the framework manager was also enhanced to
provide additional support to the user in the application of the model's BMP placement function, by
alerting if the BMP placement location on the map is not suitable. The testing of suitability is based on
the information provided by the siting tool, which evaluates BMP placement based on site constraints
such as soil, slope, and land use. The siting tool is heavily dependent on ArcGIS to perform the needed
analysis on multiple geographic information system (GIS) data layers based on the user selected site
suitability criteria for the selected BMP type.
1-3
-------
1.1.2. Enhancements to the BMP Siting Tool
The BMP siting tool was developed to help users in selecting suitable locations for different types of low
impact development techniques or conventional BMPs. Site suitability is the dominant factor in
identifying potential site locations (USEPA, 1999). The siting tool provides guidance on where to place a
selected BMP on the watershed on the basis of the site suitability criteria. The BMP siting tool is for
guidance purposes only because it is highly data-driven tool. It requires field verification beyond the GIS
exercise to validate the suitable locations before using them in SUSTAIN for BMP placement.
The following enhancements were made to the publicly released version 1.0.
• Land ownership (public or private land): The user has option to limit the selection criteria to
public or private land for different selected BMP types.
• Proximity to land features (i.e., roads, streams, and buildings): The user has option to specify a
buffer size (i.e., less than, greater than, lower and upper limit) for the suitable locations.
• Prioritize the suitable locations by adding a weighting factor to the suitability criteria for slope
and hydrologic soil group. For example, a bioretention basin is best suited in areas with
hydrologic soil group A as compared to D.
• Efficient selection of appropriate sites by enhancing the code. The code logic is improved to
minimize the run-time overhead and more robust performance of the tool.
• An increased level of automation for siting and placement of BMPs on the map.
Version 1.0 of the BMP siting tool was released in September 2009, which requires ESRI's Arc View 9.3
and the Spatial Analyst extension. Version 1.1 which will be released in September 2011 is made
compatible with ArcGIS 10.0 (Service Pack 2) and is backward compatible with ArcGIS 9.3.1 (Service
Pack 2).
1.1.3. Enhancements to the BMP Simulation Module
The BMP simulation module performs process simulation of flow and water quality through BMPs. It
uses a combination of process-based algorithms, including weir and orifice control structures, flow
routing and pollutant transport, infiltration, ET, and pollutant loss/decay simulation. BMPs supported by
SUSTAIN include, but not limited to, bioretention, cistern, constructed wetland, dry pond, grassed swale,
green roof, infiltration basin, infiltration trench, porous pavement, rain barrel, sand filter, vegetated filter
strip, and wet pond. Sediment (sand, silt, and clay) settling and routing is computed using the processed
based algorithms adopted from the Hydrologic Simulation Program—FORTRAN (HSPF) (Bicknell et al.,
2001). Table 1-2 summarizes the completed enhancements to SUSTAIN^s BMP module so far as part of
Phase 3 of the project.
Table 1-2. Summary of completed BMP module enhancements
Description
• = Enhancement addresses
— = Enhancement does not address
Sub-hourly time step
Morton infiltration method
ET multiplier
BMP initial moisture conditions
Pump curve
Area BMP
Support
regional
case studies
•
•
•
•
•
•
Develop
additional
capabilities
--
-
--
•
•
•
Enhance
existing
functionality
•
•
•
•
•
•
Increase
simulation
accuracy
•
•
•
•
•
•
1-4
-------
Horton Infiltration
The SUSTAIN BMP module previously included the Green-Ampt and Holtan infiltration methods. The
Horton infiltration method is implemented in SUSTAIN using the Storm Water Management Model
(SWMM) formulation (Rossman, 2005). The Horton infiltration method is an empirically based model
parameterized by specifying an initial (maximum) infiltration rate and a final, saturated infiltration rate.
The model assumes that infiltration begins at a constant, maximum rate that decreases exponentially over
time. The shape of the curve as the infiltration rate changes from initial to final is controlled by a decay
rate specific to the type of soil (USEPA, 1998). The equation follows:
where ft is the infiltration rate at time t,f0 is the maximum infiltration rate,/c is the saturated infiltration
rate, and k is the decay constant.
Evapotranspiration (ET) Multiplier
ET rates in SUSTAIN version 1.0 were set globally for the system meaning that all BMPs used the same
evaporation rates regardless of the type of practice. A global approach requires that the same ET rates
apply to all BMPs regardless of the type of density of vegetative cover.
An evaporation multiplier was added allowing a unique multiplier value to be set for each BMP instance
in a SUSTAIN model. The multiplier is applied to the global evaporation rate (e.g., constant monthly,
time series, and so on) to account for unique evaporation conditions that are BMP specific. For example,
a multiplier greater than one can be used to parameterize an individual BMP with more abundant
vegetation to account for higher ET rates expected with that type of condition.
BMP Initial Moisture Conditions
Previously, SUSTAIN had been presented in the context of long-term, continuous simulation modeling
using months or years of runoff and pollutant time series data where optimization objectives are set on the
basis of annual flow or water quality targets. Traditional CSO applications commonly use a design storm
approach where a synthetic precipitation event is generated on the basis of a critical condition peak
intensity or rainfall depth. When performing a long-term simulation, the initial BMP conditions do not
typically affect the average annual results; however, initial conditions become critical when implementing
a design storm approach. For single storm events, it is expected that a BMP at field capacity will perform
differently from a BMP still saturated from a recent storm. Figure 1-2 illustrates the two BMP initial
condition parameters.
initial Surface
Ponding Depth
Soil Media
Moisture (%)
Figure 1-2. Conceptual illustration of user-defined BMP initial soil moisture parameters.
1-5
-------
The initial water depth is the depth of water ponding on the surface of the BMP at the start of the
simulation. The initial soil moisture (%) is the fraction of void space in the soil media occupied by
moisture. Typically that value could be set at field capacity or the wilting point.
Area BMP
While 5f/5Z4/7Vprovided an option to model directly connected impervious area (DCIA) and
disconnected impervious area using the internal land simulation option, no feature was available in the
system to model it using the external land simulation option. In order to model disconnection of
downspouts as a BMP in the case studies, a new practice, the Area BMP, was added to the BMP module.
The Area BMP is a pervious land segment over which a portion of impervious runoff, from disconnected
impervious areas like rooftops, is routed. The BMP simulation occurs only when there is no runoff from
the BMP area otherwise the total inflow to the BMP is bypassed. The Area BMP simulation is an
approximation to the reality where the runoff from the disconnected impervious area is routed to and
simulated on the pervious area. The runoff from the disconnected impervious area is captured by the
Area BMP through the infiltration (under saturated soil condition) and the surface storage. The runoff
from the BMP area (i.e., pervious area) is not simulated by the Area BMP and is always bypassed. Figure
1-3 shows the conceptual flow diagram of Area BMP simulation.
Area BMP (Pervious Land)
BMP Inflow
Runoff from pervious
area?
No
BMPSimulation
Runoff from the disconnected
impervious areas and the
BMP area
No inflow to the BMP
if bypass occurs
No evapotranspiration
Saturated infiltration rate
Nonlinear reservoir routing
BMP Outflow
Figure 1-3. Conceptual flow diagram of Area BMP simulation.
A nonlinear reservoir routing algorithm is applied to route the surface runoff from the impervious area to
the Area BMP (i.e., pervious area). This BMP does not simulate ET assuming that surface runoff already
accounts for it on both pervious and impervious land. Also because runoff occurs under saturated soil
conditions, saturated infiltration rate is used as a background infiltration rate. Surface runoff, Q, occurs
only when the surface water depth, d, exceeds the maximum surface storage depth, dp, in which case the
outflow is given by Manning's equation:
49
where
1-6
-------
Q = outflow rate (cfs),
W= pervious area width (ft),
n = Manning's roughness coefficient,
d = water depth (ft),
dp = depth of surface storage (ft), and
S = pervious area slope (ft/ft).
The pervious area width (W) can be estimated by dividing the BMP area by the length of the
representative flow path in the Area BMP.
The calibration parameters are surface storage, flow length, and slope to attenuate the flow peaks. The
feature was added to model the disconnected downspouts and to optimize the percent DCIAs as a decision
variable in the model. The option is available only to the external land simulation option in SUSTAIN.
Pump Curve
Certain management practices require an external pump to convey flow out of the BMP. In CSO
applications, storage tanks are often implemented to temporarily store excess volume until the treatment
plant has sufficient capacity and to which the volume can be pumped. SUSTAIN includes the ability to
specify unique pump curves for each BMP. Pump curves define the numeric relationship between BMP
water depth and pump flow rate, similar to the Type 4 pump curve available in SWMM (Rossman, 2005).
Figure 1-4 presents a conceptual illustration of a pump implemented in a storage tank.
QJ
4->
TO
D,
Depth
0 j •*•
Figure 1-4. Conceptual illustration of the BMP pump curve.
The curve is represented as a table of paired water depth and flow rate values. The water depths represent
the pump's operating bounds where D0 is the depth of the pump's minimum operating capacity and D, is
the depth of the pump's maximum operating capacity after which flow rate becomes constant. The pump
can be implemented in a BMP that also has orifice, weir, or underdrain; however, the pump will take
priority over the outlets.
1-7
-------
1.1.4. Enhancements to the Watershed Module
The watershed module generates runoff and pollutant loads from the land through internal land simulation
or importing calibrated land simulation time series. In the internal land simulation, surface runoff and
water quality components are provided through an internal application of EPA's SWMM (version 5)
(Huber and Dickinson, 1988) or from an external linkage to a previously calibrated watershed model.
The sediment erosion process is simulated using HSPF (Bicknell et al, 2001); the particle size
distribution for the eroded sediments is represented as fractional distribution of sand, silt, and clay. Table
1-3 summarizes the completed enhancements to the SUSTAIN s watershed module so far as part of Phase
3 of the project.
Table 1-3. Summary of completed watershed module enhancement
Description
• = Enhancement addresses
= Enhancement does not address
Groundwater system
Horton infiltration method
Point source time series
Sub-hourly time step
Support
regional
case studies
--
•
•
•
Develop
additional
capabilities
•
--
•
--
Enhance
existing
functionality
•
•
•
•
Increase
simulation
accuracy
--
•
•
•
Groundwater System
A subsurface aquifer component in SUSTAIN allows tracking water infiltrated through BMPs to the
shallow groundwater system. Each subwatershed can be linked with an aquifer, the basic delineation unit
of the shallow groundwater system. As in Haan (1972), the shallow groundwater is modeled as a simple
linear reservoir. Groundwater discharge G(t) and deep seepage D(t) from the shallow groundwater
storage S(t) at time r are calculated as
G(t) = r X 5(t)
D(t)= 5x5(t)
where rand 5are groundwater recession and seepage constants, respectively (per hour).
A recession constant can be estimated from two stream flows F(ti), F(t?) measured on hour t\ and ?2 (?2 >
?i) during the hydrograph recession as
r =
t, -t.
Recession constants are measured for a number of hydrographs, and an average value is used for the
simulation. No standard techniques are available for estimating the rate constant for deep seepage loss.
The most conservative approach is to assume that s = 0 otherwise the constant must be determined by
calibration.
Horton Infiltration
The SUSTAIN watershed module previously included the Green-Ampt infiltration method. The Horton
infiltration method is incorporated in SUSTAIN, as an alternative method for land simulation module,
1-
-------
using the SWMM formulation (Rossman, 2005). This method is based on empirical observations
showing that infiltration decreases exponentially from an initial maximum rate to some minimum rate
over the course of a long rainfall event. The equation and the input parameters are shown under the
Horton Infiltration method for BMP simulation in the previous section 1.1.3.
Point Source Time Series
While the primary boundary conditions in SUSTAIN are represented as runoff time series from the land,
specialized applications could require the representation of other flow or pollutant time series such as
flow and pollutant loading from a wastewater treatment plant. In CSO applications, dry-weather flow can
account for a significant portion of the annual flow volume and, in some cases, could affect overflow
events. SUSTIAN provides the ability to link an external time series of flow and pollutant loading to a
BMP, junction, or conduit. Figure 1-5 provides an example of the required data format for external time
series files.
154954 2001 1. L 0 0 2.ccE-04
15-3954 2001 1 1 0 15 3.05E-04
154954 2001 1. 1. 0 30 2.46E-04
154954 2001 1 1 0 45 3.3"7E-04
154954 2001 1110 2.35E-04
154954 2001 1 1 1 15 3.33E-04
154954 2001 1. '- L 30 2.c2E-04
Figure 1-5. Sample data format for point source time series.
The first column represents a station identification number and is not used by SUSTAIN. The following
five columns represent the year, month, day, hour, and minute respectively. The seventh column
represents point source flow data with units of in.-acre per time step. Subsequent columns are optional
and can be used to represent corresponding pollutant loading with units of pounds per time step.
Sub-Hourly Time Step
To improve simulations accuracy for predicting peak flows and time of concentration, a sub-hourly (1 to
60 minutes) time step option was added to input the external time series with data at temporal scales finer
than 60 minutes. Comparison of both watershed and BMP simulations using 60 minute and a smaller 15
minute time step were reviewed and are presented in the case study for Kansas City (Chapter 2).
1.1.5. Enhancements for Optimization Efficiency
Enhancements that benefit SUSTAIN's optimization module build credibility and increase computation
efficiency in the search for cost-effective solutions. Although the proposed enhancements do not change
the optimization module, they help to improve its performance and efficiency. For example,
implementing a sub-hourly time step discussed in Section 1.1.4 provides increased accuracy for
predicting time of concentration. It is expected that improved simulation accuracy will improve the
effectiveness of the optimization module. Further improvement in system application efficiency, without
a significant reduction in predicative accuracy, can also be achieved by implementing a variable time
step. In a variable time step application, during the dry periods the simulation time step is increased,
while during wet periods the time step is reduced, significantly reducing the number of operations that
need to be performed to simulate the hydrologic response over the application period of the model. This
enhancement is planned for completion as part of Phase 3 of this project.
1-9
-------
1.2. Release of SUSTAIN 1.2
An updated SUSTAIN version 1.1 compatible with ArcGIS 10.0 with an enhanced BMP siting tool will be
released in September 2011. The Version 1.1 release also includes two specific updates: (1) removing the
comma delimiter from the number format (e.g., 1,000 is modified to 1000); and (2) converting the
subcatchment ID for the internal land simulation input file to be unique by adding SC before the
catchment ID (e.g., the subcatchment ID 1 is modified to SCI). Version 1.1 also includes an updated
copy of the step-by-step user's guide which provides guidance for users in setting up the example
applications. An additional release of SUSTAIN version 1.2 is planned for the fall of 2011, which will
include the completed system enhancements discussed in Table 1-2 and Table 1-3. The release of version
1.2 will further support adoption of SUSTAIN by the user community by providing additional
functionality and will coincide with publications related to the case study applications in Kansas City and
Louisville.
1-10
-------
Chapter 2. Case Study: Kansas City, Missouri
EPA's Office of Research and Development (ORD) has conducted a pilot project to demonstrate the use
of green infrastructure (GI) for CSO control in Kansas City, Missouri. This case study effort includes two
general sets of goals: (1) management goals to inform the decision-making process; and (2) modeling
goals to test system functionality and provide application guidance for the user community. The
management goals of the case study were two-fold: (1) to quantify the benefit of a planned GI design in a
portion of the CSO basin toward overflow reduction goals; and (2) to estimate how much additional GI, if
implemented in a similar way, would be required to achieve CSO reduction goals for the basin. Likewise,
the modeling goals of this effort were to (1) highlight some of the key steps associated with problem
formulation and setup of a SUSTAIN application; (2) test the sensitivity of new SUSTAIN features and
functions; and (3) present the lessons learned through the application process to serve as guidance for the
SUSTAIN user community.
The focus area for this case study was the Middle Blue River watershed in Kansas City, Missouri, which
includes a planned 100-acre BMP implementation/monitoring site in the Battleflood Heights
neighborhood, and 86 acres of an adjacent control area where only monitoring is planned. Figure 2-1 is a
map showing the pilot and control study areas in Kansas City, Missouri, and is within the combined sewer
service area. Both pilot and control areas are tributary to CSO 069. The approximate location of the pilot
study area is between East 74th Street and East 79th Street and is generally bounded by Paseo Boulevard to
the east and Holmes Road to the west. The project combined local and regional efforts that were aimed at
collecting performance data for GI practices, assessing management performance at the sewershed scale,
and gathering stakeholder input into selection, design, and O&M of GI systems. This chapter first
provides a brief overview of past or ongoing complementary efforts and explains how those efforts were
used to inform SUSTAIN modeling. Second, it outlines the specific objectives of the study including
steps for how the SUSTAIN modeling framework was applied to the selected CSO project areas to achieve
each objective. The report also describes how SUSTAIN was used to analyze the physical system,
evaluate alternatives for BMP placement, and ultimately, refine the current knowledge and understanding
of the effectiveness of the selected management practices under certain conditions. The case study
findings provide regional insights for GI planning and implementation for CSO mitigation. Through the
case study application process, some of the key lessons learned were also summarized to provide
guidance for SUSTAIN application for the broader user community.
2.1. Background
Parallel and complementary research efforts have been conducted in Kansas City's Middle Blue River
watershed in the time leading up to this effort. This section provides (1) a brief background description of
those efforts; (2) highlighted aspects of each project that was incorporated into this analysis; and (3)
identification of areas where additional effort or information was needed. These efforts consist of the
following:
• An Overflow Control Plan (OCP) (WSD, 2009) which includes an XP-SWMM sewershed model
that was reviewed and refined to represent baseline stormwater runoff conditions;
• Project-specific monitoring data for a range of storm events flowing through the sewer network;
• Desktop analysis conducted to highlight BMP opportunity and cost estimates within the study
area;
• A siting analysis and approved design plan of a BMP implementation strategy for the 100-acre
study area; and
2-1
-------
CSO 069 Watershed
Boundaries
0 32 0.4,, N
0 0.2 0.4 i\
Figure 2-1. Location of pilot study area within the CSO 069 sewershed boundaries.
2-2
-------
• Site-specific BMP performance modeling of private land areas using Source Loading and
Management Model for Windows (WinSLAMM).
Each of those complementary research efforts is further described in the subsections below. After
summarizing the efforts, specific objectives and outcomes were formulated for this case study as outlined
in Section 2.2.
2.1.1 Overflow Control Plan: XP-SWMM Model
The Kansas City, Missouri, Water Services Department (WSD) developed an OCP to provide guidance
for managing CSOs (WSD, 2009). As part of that effort, an XP-SWMM model was developed as an
evaluation tool. The primary objective of the model was to quantify the storage capacity required to
mitigate CSOs. The XP-SWMM model was configured for the pilot study area within the CSO 069
sewershed. It was developed as an event-based model and was not intended for long-term continuous
simulation.
Synoptic statistical analyses were performed on precipitation data at (1) Kansas City International
Airport—Coop ID: 234358; and (2) Kansas City Downtown Airport—Coop ID: 234359 to characterize
the storm distribution for atypical meteorological year in the study area (WSD, 2009). A combined
precipitation time series for long-term analysis was developed for this study using the Kansas City
Downtown Airport data from January 1949 through October 1972 and Kansas City International Airport
data from November 1972 through December 2004. Long-term annual average precipitation was
calculated using the combined time series data from 1949 through 2004. This case study extended the
record through December 2009 using the Kansas City International Airport gage data. Figure 2-2 shows
variations in annual average precipitation, as well as the mean, 25 , and 75 percentiles.
70
1950
-Annual Depth 25th Percentile 75th Percentile
Mean
1960
1970
1980
1990
2000
Figure 2-2. Annual precipitation for Kansas City
A set of eight types of design rainfall events was constructed to characterize Kansas City rainfall for a
typical year. Table 2-1 presents a summary of the design storm characteristics that were determined as
2-3
-------
part of the rainfall analysis (WSD, 2006). The design storms total volume and duration were based on a
statistical analysis of precipitation events that occurred in the recreation season (April 1 - October 30)
assuming 12 hour inter-event spacing. A SCS (soil conservation service) type II distribution was applied
to the overall events, with some adjustments made to ensure that peak intensity matched historic records
(WSD, 2006). The table corresponds to the information shown in Figure 2-3. Those design storms were
used with the XP-SWMM model to predict CSO response and identify the critical condition storm size.
WSD has stipulated an allowable overflow frequency criteria of 6 events per year at CSO 069, which
corresponds to the type D design storm (D-storm: 1.4 in. depth, 0.6 peak in./hr intensity, 16.75 hr
duration).
Table 2-1. Summary of design storms used for CSO control plan
Return period
0.33 month
0.67 month
1 month
2 months
3 months
4 months
6 months
12 months
Type
A
B
C
D
E
F
G
H
Depth
(in.)
0.28
0.52
0.86
1.40
1.80
2.00
2.40
2.90
Peak
(in.)
0.16
0.25
0.38
0.60
0.73
0.82
0.95
1.20
Duration
(hr)
6.00
8.75
12.25
16.75
19.75
21.00
23.75
26.75
Annual
exceedance
frequency3
36
18
12
6
4
3
2
1
Annual
number of
events'3
18
6
6
2
1
1
1
1
Source: WSD 2006
a. Average number of events per year with total depth and peak intensity equal to or exceeding the design storm.
b. Average number of events per year with similar depth, intensity, and duration characteristics to the design storm.
A synthetic storm distribution (Table 2-1) for atypical year indicates that on average, Kansas City
experiences 78 rainfall annual events. Of those events, those with depths greater than 0.28 in. were
shown to result in overflows at the 069 outfall (WSD, 2006). Figure 2-3 is a histogram of a typical
annual storm distribution.
2.1.2. Site-Specific Monitoring Data
Additional data were collected after the development of the XP-SWMM model, which was originally
done as part of the OCP. In 2009 and 2010, a total of 20 runoff events were monitored at the outfall of
the 100-acre pilot study area in addition to the original four events from 2008. Of the 20 new events,
some showed very minimal response at the outlet (i.e., flow was less than 5 cfs), and some of the events
had no coincident precipitation data at the local gage. Nevertheless, 10 events were identified between
2008 and 2010 for which (1) coincident precipitation data existed; and (2) which generated flows greater
than 5 cfs at the monitoring site. The observed precipitation associated with the monitored events were
compared against the long-term historical precipitation record at the nearby Kansas City International
Airport gage to get an idea of how representative the monitored events were of the larger, critical
condition events associated with overflow. A storm event separation analysis was performed on the long-
term precipitation data recorded between 1949 and 2009. Storm separation assumed a 12 hr inter-event
time consistent with WSD's design storm analysis, and a minimum storm size of 0.1 in (WSD, 2006).
After storm separation, 10 equal percentile bins were established for the storms in the historical record,
according to ranges of the resulting storm event volumes. Figure 2-4 is a histogram of the rainfall
associated with the 10 monitored events against the long-term historical precipitation record at Kansas
2-4
-------
City International Airport. The graph suggests that the monitored storms (and associated pipe flow)
between 2008 and 2010 are representative of larger storm events that cause CSO in the 069 sewershed.
The data were used to calibrate and establish a model baseline, as further described in Section 2.2.1.
Storm events that result in sewer overflows
<0.28 0.28-0.52 0.52-0.86 0.86-1.4 1.4-1.8 1.8-2
Rainfall Depth (inch)
Figure 2-3. Storm size distributions for a typical Kansas City meteorological year.
Monitored Storm Events (2008-2010)
0.36-
0.45 in.
0.46-
0.60 in.
0.61- 0.80- 1.03- > 1.60 in
0.79 in. 1.02 in. 1.60 in.
50-60 60-70 70-80 80-90 90-100
Storm Volume Percentiles at Kansas City International Airport (1948 through 2009)
Figure 2-4. Histogram of monitored against the long-term historical precipitation record.
2-5
-------
2.1.3. Green Alternatives for Sewershed 059 & 069 Technical Memorandum and
Overflow Control Plan
The Kansas City, Missouri WSD conducted a desktop analysis and subsequently published a technical
memorandum, Green Infrastructure Alternatives for Outfalls 059 & 069 (WSD, 2008), which quantifies
the costs associated with modifying the CSO controls presented in the May 6, 2008, draft OCP Summary
for two outfalls in the Middle Blue River basin. That study included considerations for incorporating
both conventional gray infrastructure (i.e., intermediate underground storage) and conservational GI
technologies for mitigating CSO. Areas tributary to the two outfalls were identified as most likely to be
improved through implementing a combination of green and gray solutions.
Approximately 744 acres of the Middle Blue River Basin tributary were selected for the desktop analysis.
The desktop analysis area encompasses outfalls 059 and 069; however, the 069 drainage area was selected
as the focus of this case study. The desktop study made the assumption that a gallon of stormwater
storage was sufficient to control a gallon of CSO discharge for the selected event. The analysis did not
include modeling of the BMP and drainage network, but rather evaluated opportunities to place BMPs on
either public or private property. The desktop screening analysis included some basic volume estimates
to approximate storage requirements for the catchment, its simplified flow-accounting approach did not
attempt to represent cumulative impacts and benefits. A collection of well-distributed GI management
practices throughout the sewershed was envisioned to provide some additional storage requirements. The
desktop analysis included a basic evaluation of specific GI storage technologies that are expected to be
used in the study area, which consisted of the following:
• Inlet retrofits in road and street rights of way;
• Curb extension bioretention;
• Replacing sidewalks in road and street rights of way with permeable pavement;
• Replacing pavement outside of road and street rights of way with permeable pavement; and
• Converting roof areas to green roofs.
Gray infrastructure options included a combination of underground storage tanks with screening facilities
and outflow pumping stations. Compared to the other alternatives considered, the gray infrastructure
practices represent the highest overall capital cost, in terms of unit cost per gallon stored. For the 069
sewershed, the desktop analysis suggested that it would be more cost-effective to either replace or
supplement the gray components with GI alternatives without adversely affecting the desired level of
control at the respective outfalls. Table 2-2 is a summary of gray infrastructure controls at CSO 069.
Table 2-2. Gray infrastructure CSO controls for outfall 069
Control component
2 MG storage tank
1.5 MG per day pumping station
51 MG per day screening
100 ft, 48 in. Sewer
500 ft, 12 in. force main
Odor control
Estimated capital cost
(million dollars)3' b
$30.6
Storage provided
(MG)a
2.0
Capital cost per
gallon stored
(dollars)3
$15.30
a. Source: WSD, 2008
b. Includes a 25 percent allowance for planning, engineering and design, and an additional 25 percent contingency.
2-6
-------
Other considerations must be taken into account when comparing the cost-benefit of GI versus gray
infrastructure. For example, the gray infrastructure solution presented in Table 2-2 requires 2.0 million
gallons (MG) of storage volume, which assumes that pumping from storage occurs during the most
intense 6 hours of the design storm. A GI solution must provide storage volume greater than 2.0 MG
because of the additional pumping capacity otherwise represented in the gray storage tank. Accounting
for the additional pumping capacity, the required storage volume of GI must equal that of gray
infrastructure storage plus 6 hours' pumping volume (an additional 0.375 MG), which results in a total GI
storage capacity of 2.375 MG (WSD, 2008). According to the original desktop analysis results, the
estimated capital cost to develop 2.375 MG of GI storage in the area tributary to outfall 069 is
approximately $24.6 million—a $6 million dollar (~ 20 percent) savings (WSD, 2008). It is important to
note that cost information published in this study represented capital costs only and no O&M costs were
included.
In January 2009, the Kansas City, Missouri WSD released the full text of its OCP. The plan cites some
uncertainty associated with the performance of GI in mitigating overflow volumes at the outfalls. As a
result, the GI capital budget proposed by the desktop analysis was increased by approximately 30 percent,
bringing the original desktop analysis estimate of $24.6 million up to a value of $32 million (WSD,
2009). Following the adjustment, the updated plan suggests that gray infrastructure might be a more cost-
effective solution. Nevertheless, while the full cost of gray infrastructure is a public burden, GI offers the
possibility for cost sharing through public-private partnerships. In addition, GI provides other benefits
not offered by gray infrastructure. The OCP also proposed an annual budget of $2 million for O&M costs
associated with GI upstream of outfalls 059 & 069.
Another result of the desktop study was the selection of the 100-acre pilot study catchment that services
the Battleflood Heights neighborhood. The desktop study recommended further investigation of BMP
placement opportunity, associated costs, and a quantification of GI benefits. In addition, the pilot study
site was targeted to receive the first phase of implementation activity, for which significant pre- and post-
implementation monitoring would be performed.
2.1.4. Middle Blue River Green Solutions Pilot Project
The design plan represents a culmination of efforts. In April 2009, WSD published a Draft Conceptual
Design Report for the Middle Blue River Pilot Study (WSD, 2009). Those designs were based on a
combination of XP-SWMM hydraulic modeling analyses performed by Burns & McDonnell Engineering
Company, Inc. (for the pilot area) and HDR (for the entire Middle Blue River combined sewer area) as
part of the OCP development (WSD, 2009). URS Corporation ultimately developed the final BMP
designs for the 100-acre pilot study site, called the Middle Blue River Green Solutions Pilot Project. This
section provides a general overview of the design plan components. Section 2.4.1 describes how these
BMPs were represented in the SUSTAIN optimization model.
The BMP design plan for the 100-acre pilot study area includes 158 individual surface features, plus 21
supplemental underground storage pipe systems that were designed to retain BMP overflow and
underdrain outflow from selected bioretention and porous pavement structures. Figure 2-5 is a map
showing the locations of the surface features.
Table 2-3 summarizes the various surface and subsurface structural components from the design plan.
Figure 2-6 through Figure 2-11 are example excerpts of schematic drawings from the final BMP design
plan (URS, 2010). These schematic drawings are also cross-referenced in Table 2-3 for each of the
unique design plan component categories.
2-7
-------
Legend
Su bw ale r shed
BMPs
Bioretention
^^| Curb Extention [Bioretentionj
Curb Extention (Raingarden)
I Porous Sidewalk
Shallow Bioretention
Pilot Watershed
Design Plan BMPs
NAD 19B3 Strt^lane Mfeacui West FIP3 2403 Feet
: :- .: N
Figure 2-5. BMP layout in the 100-acre pilot study area.
Table 2-3. Summary of BMP design plan components
Design plan component
Bioretention
Bioswale
Cascade
Porous sidewalk or
pavement
Rain garden
Below grade storage
Structural description
Bioretention without curb extension
Curb extensions with bioretention
Shallow bioretention
Vegetated swale infiltrates to background
soil
Terraced bioretention cells in series
With underdrain
With underground storage cubes
Rain garden without curb extension
Curb extensions with rain gardens
Retains BMP overflow and underdrain
outflow from selected bioretention cells or
porous pavement
Number of BMPs
24
28
5
1
5
18
5
64
8
21
Figure reference
Figure 2-6
Figure 2-7
Figure 2-8
Figure 2-9
Figure 2-10
Figure 2-11
2-8
-------
4" SLOTTED DRAIN PIPE-
TOPSOIL
PLANTING MIX •
18* RISER INLET ASSEMBLY (TYP.)
ENGINEERED SOIL MIX—,
12" POOL DEPTH FROM -j
GUTTER ELEVATION AT I
INLET I
'
PONDING
-STORAGE PIPE
$&£&
i$!<£tt&^^
'M :\: \
^-CONSOLIDATED GRANULAR
BACKFILL MATERIAL
PROFILE (WELL-GRADED SAND (WS)
SECTION B-B PERASTMD2487)
NOT TO SCALE
-NATIVE SOIL
NATIVE SOIL
BACKFILL
Figure 2-6. Bioretention with underground storage cross-section profile.
4"-6" DIAMETER RIVER
ROCK SPLASH STONES
-ST-BEET FLOVV
NAVTIVE SOIL AMENDED
WITH 3" COMPOST,
ROTOTILLED, 8" MIN. DEPTH
EXISTING
CURB
12" TYP.
(VARIES)
SECTION B-B
Figure 2-7. Bioswale cross-section profiles.
2-9
-------
Figure 2-8. Cascade plan view and cross-section profile.
RAlNGARDEN OR
BlORETENTION BED
3/4" CLEAN, DOUBLE
WASHED GRAVEL
AT SELECT LOCATIONS.
ADD2"DIA. SCHEDULE
40 PVC SAMPLE PORT /
RISER. WITH THREADED
CLEANOUT PLUG
j— POROUS SIDEWALK (NO JOINTS)
2" DIA. SCHEDULE 40 PVC -
DRAIN PIPE, DAYLIGHT AT
TOE OF RAlNGARDEN OR
BIORETENTION SIDESLOPE
PERFORATED PIPE. 2" DIA,
SCHEDULE 40 PVC WITH K2
DIA. HOLES ON 6" SPACING,
UNDER SIDEWALK ONLY
END CAP
UNDISTURBED
MATERIAL
FILTER FABRIC
POROUS SIDEWALK WITH LNDERDRAIN TO B|ORETENT|OM
NOT TO SCALE
Figure 2-9. Porous sidewalk cross-section profile.
SIDEWALK
6' POOL DEPTH TYPE C-1 CURB -
FROM GUTTER
ELEVATION AT INLET
^TZLLZ:?^'^
11 —i 11—11 p-^t i i^i i i-'-'-'-i rp-1-
3" HARDWOOD MULCH
NATIVE SOIL AMENDED W|TH
3" COMPOST, ROTO-TILLEO 8"
WIN. DEPTH
RIP AND SPADE S'JBSOILTO
BREAK UP COMPACTION
SECTION A-A
NOT TO SCALE
Figure 2-10. Rain garden cross-section profile
2-10
-------
CONCRETE COLLAR FOR
STRUCTURES LOCATED
IN STREET ONLY,
- 24' NYLOPLAST GASKETED
SOLID FRAMING GRATE
24" NYLOPLAST
DRAIN BASIN
STAINLESS STEEL EYE
HOOK BOLTED TO
OUTSIDE OF PIPE OR
EQUAL TO SECURE
RISER PIPE
DO NOTWFI D RISFR PIPF
TO 90° BEND. SLIP CONNECT
SOLVENT WELD 90' BEND
TO PIPE STUB
L
4" PVC
CUT CIRCULAR HOLE IN
WALL OF MANHOLE. SEAL
PVC PIPE WITH SEALANT OR
GASKET RECOMMENDED BY
RISER MANUFACTURER
:•:;:•:•:•:•:•:•:-:• COMPACTED
SOU BACKFIII
36" PVC
STORAGE PIPE
fi' ABOVE MANHOLE BOTTOM
FLOW CONTROL STRUCTURE FOR
BELOW GRADE STORAGE SYSTEMS
NOT TO SCALE
DRAINAGE ORIFICE SIZE VARIES PER
SITE (SEE TABLE ON SHEET 1210)
Figure 2-11. Below-grade storage outlet structure.
All the BMPs were planned for construction within the public rights of way and were developed using
field surveys and feasibility assessment. The factors influencing BMP selection included the width of the
right of way, slope, soils, utility lines, physical obstructions, and public acceptance. The total estimated
storage capacity provided by all the BMPs is approximately 300,000 gallons, which corresponds to about
56 percent of the total D-design storm runoff volume. The BMPs receive runoff from the streets through
2-11
-------
road side curb and gutter, and release the flow back to the sewer network through underdrain and outlet
structures (for treated flow) or through overflow structure (for untreated bypass flow).
It is important to note that the objective of this case study is not to evaluate the effectiveness of the
proposed design. This case study recognizes that some site-specific elements of engineering design
cannot be fully addressed in a modeling application alone without verification on the ground. For that
reason, the proposed design is accepted as the best recommended solution within the pilot area because it
has been derived from on-the-ground engineering and design surveys. Nevertheless, some of the
expected outcomes of this case study are (1) to quantify the relative contribution of the proposed plan
toward CSO mitigation in the context of the larger sewershed; and (2) to evaluate the CSO mitigation
benefit of extrapolating similar design guidance throughout the remainder of the 069 sewershed.
2.1.5. WinSLAMM Modeling for Private Residential BMPs
As a parallel effort, WinSLAMM is being used to evaluate the water quality and quantity improvement
benefits of a large-scale application of GI control practice retrofits in the same pilot sewershed that was
identified by the desktop analysis study (the 100-acre catchment servicing the Battleflood Heights
neighborhood). That effort was conducted in conjunction with the previously described Middle Blue
River Pilot Study, a complementary EPA ORD demonstration project to quantify benefits associated with
advanced drainage concepts using GI for CSO control in Kansas City. The difference between the
WinSLAMM application and this case study is that the WinSLAMM effort focused on BMP practices for
private property as supplementary management to the practices designed in the public right of way. The
WinSLAMM model was used to evaluate the runoff reduction benefit from the following types of BMPs
(Pitt and Voorhees, 2010):
• Residential rain gardens;
• Rain barrels for turf irrigation;
• A combination of residential rain barrels and rain gardens; and
• Disconnected residential roof runoff controls.
The WinSLAMM model was applied using a long-term, continuous simulation approach, which
generated time series of flow for various types of upland controls on private parcels. The goal of the
WinSLAMM study was to quantify individual private parcel BMP performance. Private parcel BMP cost
information was not a part of the initial phase of that study; therefore, there was no assessment of the
cost-benefit relationship of private parcel implementation of GI. For the purposes of this current case
study, cost estimates were derived from other sources, as described in Section 2.5.3.
2.2. Overview of Case Study Objectives
As previously noted, this case study had two general categories of objectives: (1) CSO management
objectives that inform the decision-making process; and (2) modeling objectives that test SUSTAIN
functionality and provide application guidance for the SUSTAIN user community. Section 2.1 provides a
summary of some of the recent research efforts in Kansas City. The most relevant information was
highlighted from each of those efforts; also, key areas where additional needed information was
identified. That evaluation formed the basis for the set of expected outcomes for this case study outlined
in this section. This case study aims to build on previous research efforts, while avoiding overlap or
redundancy of work products, in an effort to identify new information that will guide the decision-making
process with regard to CSO modeling and implementation of controls. Table 2-4 summarizes key
research questions and shows how the SUSTAIN case study application interfaces with each
complementary research component.
2-12
-------
Table 2-4. Decision-making questions and expected outcomes by study
Key research questions addressed by study:
• = Study directly addresses
© = Incorporated from a previous study
- = Study does not address
What is the regional rainfall-runoff response relationship?
What are suitable Gl practices within the public rights of
way for the 100-acre pilot site?
What are typical costs associated with different types of
regional management alternatives?
What are the benefits of implementing Gl practices on
private parcel in the region?
What is the cost-benefit relationship associated with: (1)
extrapolating the proposed pilot study BMP design plan
throughout the sewershed, (2) adding Gl on private parcels,
and/or (3) constructing a storage tunnel at the regulator for
CSO mitigation at the sewershed outlet?
At what level of management (1), (2), or (3) above, are CSO
069 mitigation objectives projected to be achievable?
|
5 "oi
4. °
x £
•
~
DO
£
•a •£
01 o
•S &
(0 •—
•a =
£i
•
~
00
'ui
% c
OQ "5.
-
^
—
\n
(0 flJ
z*
0?
3. -°
U) ^a
v c
O ra
~
9
—
3 °
< ro
3 -a
.= Q-
& °-
> ra
~
9
—
^.
s
3
10
•
©
©
©
•
Both (1) management and (2) modeling goals can be summarized into four focused objectives for this
effort, as listed below:
1. Demonstrate a process for establishing and confirming a model baseline condition;
2. Evaluate the computational validity and performance efficiencies associated with different
degrees of drainage network resolution and articulation;
3. Apply optimization to identify the degree(s) of management required to mitigate CSOs; and
4. Test the sensitivity of simulation time step on predicted optimization results.
The broader case study goals were first defined at the onset of the effort; but they were further refined
during the model setup, application, optimization, and results interpretation process. A strong emphasis
has been placed on describing the SUSTAIN application process, and specifically, on clearly defining the
modeling application objectives. The defined objectives directly influence (1) the direction and
complexity of each component of the analysis; and (2) the expected outcomes of the effort—how
successful achievement of the articulated objectives will be measured. Each of the refined case study
objectives is described in greater detail in the sections below.
2.2.1. Establishing a Sewershed Model Baseline
The sewershed model baseline represents the existing condition rainfall-runoff response. It characterizes
the nature of the current physical system before any new management activities are implemented. It also
represents the reference point from which any stormwater improvement will be measured, as well as the
starting point for BMP selection and placement optimization. Because it forms the basis for comparative
assessment of target achievement, establishing a representative baseline condition with a high degree of
confidence in its applicability is a critical first step in any modeling effort. It becomes especially
important where cost-benefit optimization of future management objectives is a primary focus of the
modeling effort. It is necessary to ensure that the SUSTAIN baseline representation is:
2-13
-------
• Reflective of existing landscape features and behavior;
• Adequately responsive to critical rainfall conditions;
• Able to reproduce observed flow data within accepted metrics (WaPUG, Nash-Sutcliffe, 2002);
and
• Able to be meaningfully extrapolated to areas outside the modeled 100-acre pilot project site.
As a secondary benefit of the case study (in terms of model development objectives), documenting the
step-by-step modeling process associated with establishing a SUSTAIN baseline sewershed model can
serve as a valuable reference contribution for the broader user community.
2.2.2. Simplifying the Network Articulation for Large-Scale Extrapolation
SUSTAIN provides different ways of handling issues of scale in modeling. For small-scale settings, it can
be both feasible and practical to use a fully articulated routing network, meaning that each pipe
connection, BMP size and location, and associated drainage area is explicitly defined. For larger-scale
applications, using a fully articulated approach often becomes cumbersome and impractical because of the
size and complexity of the associated network. SUSTAIN provides an aggregated BMP option that
simplifies the complexity of the specified drainage network while attempting to preserve the physical
basis of the BMP components and interactions. When the aggregate BMP approach simplifies the
complexity of the network, it sacrifices some detail of the model representation. This case study tests the
use of a simplified aggregated approach versus the fully articulated routing network. Three natural
questions arise:
• How much network simplification is tolerable without significantly compromising model
accuracy or precision or both?
• What components of a fully articulated BMP and drainage network are appropriate candidates for
aggregation, and to what degree can they be aggregated?
• How much computational advantage does the aggregate BMP approach provide?
Another ancillary product and objective of this case study through investigating those questions, is that it
provides a reference for comparative analysis configurations and performance for varying degrees of
model network articulation and complexity.
2.2.3. Optimize BMP Opportunity for CSO Mitigation in the 069 Sewershed
A third listed objective of this case study effort (though central for the local decision-making process), is
to evaluate the degree of management required to mitigate CSO throughout the larger 069 sewershed, in
which the pilot study site is located. The BMP design plan for the 100-acre pilot area focuses on suitable
GI practices in the public rights of way. The design plans were developed using on-the-ground
engineering for feasibility and best professional judgment, though it is recognized that additional physical
opportunities for implementation of BMPs exists outside those that were included in the design plan
(WSD, 2010). This case study uses the project as designed within the 100-acre pilot sewershed, as a
boundary condition in the baseline model for optimization.
At this stage of the case study application is where synthesis of the previously described components
occurs. After establishing an optimization model baseline condition (Objective 1) and proving the
validity of a streamlined spatial representation of the proposed BMPs (Objective 2), the third case study
objective builds on those efforts to quantify the degree of management that is required to mitigate CSOs
for the 069 sewershed as follows:
2-14
-------
1. Extrapolate the proposed GI design plan for public rights of way to the remainder of the
sewershed, where GI design plans have not yet been developed;
2. Expand optional GI opportunity on private parcels (as defined by additional WinSLAMM
application) as needed to supplement public BMPs (Pitt and Voorhees, 2010); and
3. Evaluate suggested gray infrastructure storage options at the sewershed outlet regulator.
It is important to note that the optimization process follows a predetermined step-wise sequence for this
study. Some inherent constraints have already been factored into developing that sequence, including
construction accessibility, O&M costs, and some of the known relative cost and implications of
constructing gray versus GI opportunity. In addition, optimization will be performed using a selected
target design storm for CSO mitigation. The optimized solution will be validated using continuous
simulation for a representative year to see if the number of overflows predicted confirms that the
proposed design objectives for the basin has been met.
2.2.4. Evaluate the Influence of Model Time Step on Optimization Results
While SUSTAIN 1.0 (USEPA, 2009) used a fixed hourly time step, version 2.0 will offer more flexibility.
The selected simulation time step can have an influence on model performance and behavior. The fourth
and final case study objective is to characterize the influence of model time step on model performance,
and ultimately, on the selected optimization results. Such a sensitivity analysis will be performed in
conjunction with the optimization modeling sequence previously outlined in Section 2.2.3, and will
provide guidance to the user community in selecting a suitable simulation time step for modeling.
2.3. Establishing a Sewershed Model Baseline
In SUSTAIN, stormwater runoff from the sewershed model is the forcing function that drives BMP
simulation. Sewershed models use site-specific spatial and temporal elements to characterize the rainfall
runoff response. The sewershed model time series represent the existing condition (or baseline), which
serves as the reference point from which stormwater improvement will be measured. A critical first step
of a SUSTAIN application is to establish or confirm a representative baseline condition with a high degree
of confidence in its applicability. That becomes especially important in the context of cost-benefit
optimization of future management objectives, because the model baseline is foundational to results
interpretation and resulting conclusions. It is important for the sewershed model baseline condition to
appropriately represent variability throughout the sewershed. It needs to consider the influence of
physical features associated with both surface and subsurface behavior.
An event-based modeling effort was conducted as part of the Middle Blue River Pilot Study by using the
XP-SWMM modeling platform version 9.50 (WSD, 2009). Review of this effort showed the model was
configured using a catchment approach where parameters like slope, flow length, and depression storage
can vary for each subcatchment. The model assumes that only runoff from DCIA reached the CSO
network. As a result, the model was primarily calibrated by adjusting the ratio of DCIA per
subwatershed. Because the model was run for a single storm event, it used fixed initial conditions with
no consideration for the influence of ET. Calibration results were presented for two storm events that
occurred in fall of 2008.
SUSTAIN provides the user an option to link to an existing sewershed model using unit-area (one acre)
time series for each land unit or hydrologic response unit (HRU) for representing land rainfall-runoff
responses as boundary conditions. The runoff time series can be generated using any watershed model
(e.g., HSPF, SWMM) that meets the temporal and spatial resolution requirements. For this application,
HRUs were developed using unique combinations of select physical features: (1) impervious elements;
2-15
-------
(2) hydrologic soil type; and (3) slope. SWMM5 was used to generate the unit-area runoff time series for
each HRU type. SUSTAIN associates the time series with the HRU distribution in the delineated drainage
area. The unit-area runoff time series for each HRU is multiplied by the HRU area within each catchment
to derive the total volume and pollutant loads boundary conditions BMP simulation. A GIS
representation of the unique HRU elements serves as the physical link that SUSTAINuses to tabulate
distributions within each catchment. For this application, one important advantage that the HRU
approach offers over the traditional catchment approach is that it provides a level of consistency that
carries across spatially to other areas outside the immediate 100-acre pilot study that were not otherwise
explicitly monitored or modeled.
Other spatial characteristics of the baseline model representation were considered. For this application,
there was a desire to simplify the size and complexity of the network within reason in a way that
minimized distortion of system behavior and response. It is important to note that the level of model
sophistication should match the required response and purpose of the application. In the context of
sewershed optimization modeling, any reduction in the model's computational time demand for a single
run will translate to potentially significant savings when thousands of runs are required for a solution;
however, care should also be given to preserve required level of model accuracy for the specific
application.
This section describes the steps taken to develop a baseline sewershed model condition. Those steps
consist of: (1) HRU development; (2) subcatchment delineation; and (3) model calibration. The
following sections describe each step in greater detail.
2.3.1. Development of Hydrologic Response Units
In a sewershed model, land unit representation should be sensitive to the features of the landscape that
most affect hydrology, including surface cover, soils, and slope. In urban areas, it is important to estimate
the division of land use into pervious and impervious components. Because the focus of this study is
volume control, it is not as important to further subdivide land use beyond pervious and impervious
cover; however, rooftop areas were distinguished from other impervious areas to facilitate rerouting of
downspout flow as a management alternative. Because the CSO 069 sewershed contains mostly older
urban soils and basic infiltration parameter guidance was available from existing XP-SWMM modes, soil
type was not used as a distinguishing element for HRUs. When hydrologic soil groups are not
homogenous in a sewershed, further subdividing pervious land cover according to hydrologic soil group
can provide a higher degree of resolution. Slope might also be an important factor in some areas,
especially where slope varies noticeably. For this case study, the combination of slope and surface cover
(pervious, impervious, and rooftop) was used to define HRUs for the CSO 069 sewershed. This section
details the HRU development processes.
Slope Analysis
For the slope analysis, a GIS data set of 2 ft topographic contours provided by WSD was used. The
contours were interpolated into a 10 ft raster representing surface elevation. Slope for each grid cell was
calculated from the digital elevation model and classified into three categories (1) low slopes less than 1.5
percent; (2) medium slopes between 1.5 and 3.5 percent; and (3) high slopes above 3.5 percent. Although
slope does not vary dramatically in the study area, catchment slope was a calibration parameter that was
varied spatially in the previous XP-SWMM application. Including slope in the HRU development
process provided a way to capture some of the spatial variation across the sewershed. A map showing the
distribution of slope categories developed for the CSO 069 sewershed is presented below as Figure 2-12.
The sewershed is fairly flat, with the highest slopes occurring at ravines adjacent to tributary banks.
2-16
-------
Surface Cover Analysis
For this analysis, GIS data sets for roads, impervious surfaces, and building rooftops provided by WSD
were used. The roads layer contained the footprint of the typical road rights of way. The impervious
surfaces layer included features such as sidewalks, driveways, parking lots, and other distributed
impervious surfaces. The building footprint layer was used to represent rooftop area in the sewershed.
Those three layers were merged into a single raster representation, with rooftops distinguished from other
types of impervious cover. The disconnected areas between impervious features were treated as pervious
land. A map showing the distribution of surface cover types for the CSO 069 sewershed is presented
below as Figure 2-13.
Hydrologic Response Units
An overlay of slope and surface cover type was created using the two raster layers described above. This
overlay resulted in a distribution of seven unique combinations of HRUs that capture both the topographic
and physical texture of the sewershed. Figure 2-14 is a map showing the resulting FiRU distribution
within the CSO 069 sewershed.
2.3.2. Subcatchment Delineation
In the original XP-SWMM model configuration for the Middle Blue River Pilot Study, the pilot
watershed was divided into 179 subcatchments ranging in size from 0.065 to 3.091 acres. For catchment
based models such as SWMM, having more subcatchments provides more latitude for creating a spatially
variable response. In other words, a higher resolution better approximates a distributed parameter
response. However, increasing the number of subcatchments and connections also increases the
complexity and run-time for a single model run. By implementing an HRU-based approach, some of the
heterogeneity of the system gets transferred away from the catchment into the land cover distribution. As
a result, the catchment resolution and the number of network connections can be judiciously aggregated
without sacrificing too much of the spatial variability of the runoff response.
The 179 subcatchments were aggregated to 85 for model calibration on the basis of coincident and nested
drainage areas, while the number of modeled pipe segments was reduced from 350 down to 36. Figure
2-15 illustrates how several XP-SWMM subcatchments were aggregated into one subcatchment. Much of
the spatial heterogeneity within the grouped subcatchment is preserved by using an HRU representation.
Note that the sewershed model boundaries provided with the XP-SWMM model were based on the
topographic boundaries, while the sewershed boundaries are based on the sewer network. The percent
difference in total drainage area associated with these two models is less than two percent. A comparison
of the HRU distributions within each of the two different delineated boundaries was reviewed and is
presented as Figure 2-16. The difference is wholly attributable to the difference in watershed versus
sewershed drainage area boundaries.
2-17
-------
Legend
Low Stope (< 1.5%)
Medium Slope (1 5% to 3.5%)
High Slope (> 3 5%)
CSO 069 Watershed
Slope Analysis
Figure 2-12. CSO 069 sewershed slope analysis.
2-18
-------
Legend
[] Pervious Surface (311 acres)
| Impervious Surface (114 acres)
I Roof (55 acres)
CSO 069 Watershed
Surface Cover Analysis
Figure 2-13. CSO 069 sewershed surface cover analysis.
2-19
-------
Pervious Low Slope (104 acres)
Pervious Medium Slope (167 acres)
Pervious High Slope (40 acres)
Impervious Low Slope (55 acres)
Impervious Medium Slope (54 acres)
Impervious High Slope (5 acres)
Roof (55 acres)
CSO 069 Watershed
Hydrologic Response Units
Figure 2-14. CSO 069 sewershed HRUs.
2-20
-------
Legend
SUSTAIN Subwatershed
XP-SWMM Subcatchment
Pervious Low Slope
Pervious Medium Slope
Pervious High Slope
Impervious Low Slope
Impervious Medium Slope
Impervious High Slope
Roof
Pilot Watershed
Subcatchment Aggregation
N«>_1SB3_StatePlane_Missaiii_We5t_FIPS_2«3_Feel
Map prod uced 0 3-21-2011
Figure 2-15. Comparison of XP-SWMM subcatchments and Subcatchment aggregation in SUSTAIN.
2-21
-------
I XP-SWMM Boundary (99.8 acres)
I SUSTAIN Sewershed Boundary (97.9 acres)
Low Slope
Medium
Slope
High Slope Low Slope Medium High Slope
Slope
Roof
Pervious Cover
ImperviousCover
Land Cover (Hydrologic Response Unit)
Figure 2-16. Comparison of HRU distributions within XP-SWMM and sewershed boundaries.
Even though the calibration was based on the XP-SWMM watershed boundaries, the sewershed boundary
was ultimately used for optimization in the 069 sewershed. Catchment boundaries for the calibration
were not changed to conform to the sewershed outline during calibration, because as suggested in Figure
2-16, doing so would only have yielded inconsequential benefits for flow calibration at the outlet of the
pilot study area. Instead, selected boundaries from the XP-SWMM delineation were dissolved to form
larger catchments. Figure 2-17 compares the original XP-SWMM subcatchment boundaries, the
collapsed calibration boundaries used for the baseline model calibration, and the SUSTAIN sewershed
boundary.
After collapsing some of the subcatchments, there was also no need to explicitly model pipes smaller than
one foot in diameter. The contributing areas for the pipes were directly routed to the next largest pipe size
in the network. Although this section describes how the model was spatially reconfigured for model
calibration purposes, Section 2.4 evaluates the larger implications associated with model spatial
resolution, simulation time, and predictive precision.
2.3.3. Watershed Model Calibration
During model calibration, parameters are expressed uniquely for each HRU. The objective of the
calibration process is to identify a unique set of parameters for each HRU that remain constant for all
instances of that HRU within the study area, such that the spatial variation of the watershed response
becomes only a function of the HRU distribution within each subarea. Parameters from the XP-SWMM
modeling effort were used as a starting point for calibration. Of the four available 2008 storm events, the
previous XP-SWMM calibration presented results for two that were much smaller than the critical
condition design storm. Given that new monitoring data were available for this effort, the calibration
objective became to characterize model performance for the wider range of storm conditions. As
previously described in Section 2.1.2, 10 storms of various sizes were selected, spanning 3 years and two
different seasons: fall 2008, fall 2009, and spring 2010. Some of the calibrated storms had rainfall
volumes that were higher than the critical condition design storm, while others had comparable peak
2-22
-------
intensities. Calibration parameters were adjusted during the process until an acceptable match of
benchmark calibration metrics was achieved. Some of the key parameters were those associated with (1)
depression storage and overland flow; (2) infiltration; and (3) DCIA. The earlier parts of this section
describe the three general aspects of model parameterization and time series generation, while the later
summarizes model testing, output summarization, time series comparisons, and calculation of calibration
indicator metrics.
Depression Storage and Overland Flow
Depression storage describes the depth of surface ponding. The subcatchment roughness coefficient
describes Manning's roughness coefficient N for overland flow. Both the roughness coefficient and
depression storage parameters are set independently for pervious and impervious areas. A summary of
those parameters for the CSO 069 watershed model is presented below as Table 2-5. The values of the
parameters do not vary by slope, but only by land cover type.
Table 2-5. Roughness and depression storage parameters by land cover type
Land cover type
Pervious areas
Impervious areas
Parameter
Roughness coefficient (unitless)
Depression storage (in.)
Roughness coefficient (unitless)
Depression storage (in.)
Value
0.1
0.2
0.02
0.1
Subcatchment width was set by calculating the area-weighted average of the widths that were used in the
XP-SWMM model. On the basis of that area-weighted average methodology, the subcatchment width
was set to 100 ft. The percentage of impervious surfaces with zero depression storage was also set
consistently with the XP-SWMM model at zero percent (USEPA, 2010).
Infiltration
The Green-Ampt infiltration method assumes that a sharp wetting front exists in the soil column, which
separates the un-wetted zone of soil with some initial moisture content below and the wetted zone of soil
above (Rossman, 2005). The infiltration rate is calculated as a function of soil moisture, saturated
hydraulic conductivity, and average wetting front suction head, and is based on Darcy's law and the
principle of mass conservation (Huber and Dickinson, 1988).
A major advantage of the Green-Ampt method is that the input parameters can be determined from
physical measurements. The XP-SWMM model for much of the Middle Blue River watershed used the
following parameters for the Green-Ampt infiltration method presented below in Table 2-6 (Burns &
McDonnell, 2009). Those parameters were used as the starting values when developing time series for
each of the seven HRUs using SWMM5, and were ultimately left unchanged.
Table 2-6. Green-Ampt infiltration parameters
Parameter
Average capillary suction head (in.)
Initial soil moisture deficit (unitless)
Saturated hydraulic conductivity (in./hr)
Value
6.75
0.37
0.134
2-23
-------
XP-SWMM Subcatchment
| | SUSTAIN Subwatershed
Sewershed Boundary'
Pilot Watershed
Subwatershed Delineation
Figure 2-17. Comparison of XP-SWMM and SUSTAIN Subwatershed delineations.
2-24
-------
Design Storm Time Series
In its long-term control plan (LTCP) update, WSD identified the type D design storm (D-storm) as the
critical event that must be controlled to achieve CSO reduction objectives. The D-storm is a 1.4 in.
rainfall event with a peak intensity of 0.6 in. per hr and an event lasting 16.75 hours. Runoff time series
for the D-storm were developed using the calibrated baseline model.
A single-event design storm approach like the modeling effort conducted previously must specify the
initial condition of antecedent soil moisture. Another reason for testing across a range of antecedent
moisture conditions is because the observed storms were only monitored between April and October,
which do not fully represent all possible conditions. Recognizing the potential variability of antecedent
moisture conditions, three sets of the time series were developed for the D-storm to represent low,
medium, and high ET antecedent recovery conditions. The low recovery time series represents winter
conditions, when evaporation is low with a 12 hr inter-event time. The medium recovery time series
represents average conditions of the spring or fall, with moderate evaporation and 3 days between storm
events. The high recovery time series represents dry conditions during the summer, with a high
evaporation rate and a longer inter-event time of 7 days. These three conditions are meant to bracket the
uncertainty associated with antecedent conditions inherent in design storm modeling. Table 2-7 presents
a summary of the three recovery conditions. Three sets of time series, corresponding to the three
antecedent recovery conditions, were developed for each of the seven HRUs.
Table 2-7. Summary of antecedent recovery conditions for the D-storm
Antecedent condition
Low recovery
Medium recovery
High recovery
Interpretation
Most conservative
Average condition
Least conservative
Dry time
(days)
0.5
3
7
Evaporation
(in./day)
0.001
0.030
0.106
Hyetographs of the three D-storm scenarios are presented as Figure 2-18 to better illustrate the antecedent
recovery condition concept. The time series were developed with two consecutive D-storms, with the
first serving as a seed to initialize system storages for recovery. In the high recovery scenario, the storms
are separated by 7 dry days subject to a high evaporation rate, which would provide ample time for the
BMPs to recover storage capacity. For the medium recovery scenario, dry time is decreased to 3 days,
and evaporation rate decreased to 0.03 in. per day. For the low recovery scenario, the two events are
separated by 12 hours with a low evaporation rate. Previous estimates calculated impervious runoff from
this storm at 1.25 in. versus the total rainfall amount of 1.4 in. (WSD, 2008). The time series generated
with SWMM5 for the D-storm show runoff depth from the medium slope impervious HRU at 1.23 in.
Directly Connected Impervious Area (DCIA)
DCIA refers to the fractions of impervious surfaces that are connected directly to the combined sewer
system (CSS) without first draining to any pervious surfaces or buffers. In addition to primary roads,
DCIA often incorporates features in the right of way such as sidewalks, private parking lots, and private
driveways all of which might be continuously connected to the CSS.
As mentioned previously, the existing XP-SWMM model of the pilot study area reviewed for this case
study varied DCIA by subcatchment as a calibration parameter. The new SUSTAIN baseline watershed
model incorporated the previous DCIA parameters (% DCIA) as area-weighted values by subwatershed.
Runoff from DCIAs was routed first to an area-BMP (described in Chapter 1) before being conveyed to a
downstream junction.
2-25
-------
0.7
~ 0.6
c
" n ^
c u.o
o
ra 0.4
:& no
it 09
n 1
On _
j
A
t- CM
>s >s
CO CO
Q Q
^•Precipitation (in.) Evaporation Rate (in./day)
L
c
c
H
ah
b' '
Rea
TVf
arxy
-' Y
4_
O ^~ LO CD l^ OO O)
>»>»>»>»>»>»>»
"0 CO CO CO CO CO CO
D Q Q Q Q Q Q
c
T
A
D •<
— T
[0
j
>s
[0
CM
CO
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
-t
_
/le
^•Precipitation (in.) Evaporation Rate (in./day)
rfii
JIT
R(
BCOV(
?n
A
I
: '-.:.
A
"~r ~
.
0.14 _
0.12 I
0.10 i,
0.08 |
0.06 |
0.04 |
0.02 |
LLJ
0.00
t- CM
>s >s
CO CO
Q Q
0.7
n R
f 0.5
o
i5 n 4
jS u-^
!§• no
0
CL 09
n 1
co^-Lncor^oooo-*-
>s>s>s>s>s>s>s'^T~
COCOCOCOCOCOCO>s>%
QQQQQQQtoro
^^Precipitation (in.) Evaporation Rate (in./day)
.0
t- CM
>s >s
CO CO
Q Q
Low Recoven
/
Jk
co ^- LT> CD h- oo o>
CO CO CO CO CO CO CO
Q Q Q Q Q Q Q
c
T
c
- J
D •!
— T
>s
3 c
CM
CO
Q
|
0
CM
>s
CO
Q
0.14 _
0.12 |
0.10 i,
0.08 |
0.06 |
0.04 |
0.02 |
LLJ
0.00
Figure 2-18. D-storm hourly hyetographs for high, medium, and low recovery scenarios.
Stormwater Runoff Calibration
Updated watershed monitoring data from the Middle Blue River Pilot Study area were used to recalibrate
the SWMM5 baseline watershed model. The calibrated SWMM5 model was then used to generate unit
area (one acre) HRU time series for SUSTAIN. SUSTAIN was setup to represent the BMP and pipe
routing network for the entire study area; however, it was driven by the SWMM5-generated runoff
boundary conditions. Because of the updated watershed monitoring data available for this effort, the
newly calibrated model reflected a wider range of storms than previously. The flow meter number 01 at
The University of Missouri-Kansas City (UMKC-01) was located at the outlet of the pilot watershed. The
data were compared against modeled pipe outflow at the same location. Coincident 5 minute and 15
minute rainfall data collected at a locally operated precipitation gage in the watershed were used as the
forcing functions to drive the watershed model. Because the model was run on a continuous simulation
2-26
-------
basis, ET data were also required to provide recovery of soil moisture between storm events. ET was
computed in two steps. First, evaporation was computed as a function of temperature, solar radiation,
dew point, and wind speed using the Penman energy balance method (Penman, 1948). Quality controlled
National Climatic Data Center (NCDC) surface airways data from the Kansas City International Airport
and Kansas City Downtown Airport were used for this computation. Second, a locally referenced crop
factor for turf grass of 0.85 (KSU, 2010) was applied to convert the evaporation time series to ET.
Certain calibration benchmarks were also used to confirm goodness of fit for model prediction. The
Wastewater Planning Users Group (WaPUG) is a technical group that operates under England's
Chartered Institution of Water and Environmental Management. WaPUG is internationally recognized
for its promotion of best practice standards and for publishing related technical guidance as industry
standards. The WaPUG Code of Practice for use in hydraulic sewer system modeling provides
acceptable calibration criteria for observed versus modeled time series comparison. Those criteria include
acceptable percent error ranges for both peak flow and total volume. Modeled wet-weather peak flow
should be no more than 15 percent below or 25 percent above metering data, while modeled wet-weather
flow volume should be no more than 10 percent below or 20 percent above metering data (WaPUG,
2002). Those metrics provide acceptable ranges in recognition of the fact that a certain amount of
inherent error associated with rainfall gages, model parameterization, or flow measuring gages exist, and
are propagated through modeling.
Another metric that is commonly used for assessing the performance of continuous simulation hydrology
models is a model efficiency metric, E, developed by Nash and Sutcliffe (1970). They interpret the model
efficiency metric E as follows:
• Values below zero suggest that the mean of observed data is a better predictor than the model;
• A value of 0 indicates that the observed data mean is equally as good a predictor as the model;
and
• The closer the model efficiency is to 1, the better it predicts observed data.
For example, a Nash-Sutcliffe value of 0.70 indicates that the mean square error of the difference between
observed data and model prediction is 1.0-0.70, or 30 percent of the variance in the observed data. A
common rule of thumb in hydrology modeling practice suggests that obtaining a value of 0.7 or better
generally indicates adequate model fit. The WaPUG criteria and the Nash-Sutcliffe metric were used as
the bases for evaluating model performance and confirming the watershed model calibration for the pilot
watershed.
As previously noted, 10 selected storms were calibrated using observed data at the UMKC-01 flow
monitoring station at the outlet of the pilot watershed. For each storm event, the observed and modeled
flow in at the pilot area outlet was converted to inches of runoff by dividing the flow by the drainage area.
Such normalization provided a convenient and consistent approach for comparing the modeled versus
observed runoff yields for storms of different sizes. The average projected runoff volume for the CSO
critical condition D-storm was also computed using the calibrated model for relative comparison. For
each storm event, antecedent potential evapotranspiration (PEVT) was computed by summing the total
PEVT from the end of the previous event to the start of the event. All those values were computed and
plotted together during model calibration to better visualize and understand the behavior of the natural
system and modeled systems versus precipitation and PEVT. Figure 2-19 shows the area-normalized
observed versus modeled runoff depths, total depth antecedent PEVT, and projected runoff depth for the
D-storm.
One might expect that storms with lower antecedent PEVT would generate more runoff. While that was
observed in some instances, there was no clear and consistent trend between antecedent PEVT and model-
predicted volume alone because other factors such as rainfall intensity and the spatial/temporal variability
2-27
-------
of precipitation can cloud interpretation of a perceived response. That fact further demonstrates the need
to look at multiple metrics (graphical and numerical) during model calibration. A weight-of-evidence
approach, comparing multiple evaluations such as time series plots, correlation plots, and a number of
calculated indicator metrics, was used to confirm model calibration.
2.0
1.8
1.6
f "
<" 1 n
Q 1.0
5 0.8 -
2 0.6
0.4
0.2
0.0
Precipitation (in.)
Observed Outflow (in.)
Modeled Outflow
Antecedent PEVT (in.)
D-Storm Runoff (in.)
1.10
1.14
1.56
1.44
0.96 0.96
0-59
0.60
n 71
UJJ.
0 72
_
__^B
^
oo 'xf Is* ^B ro Iso *3~
«HI-J ""j1"! f^0! fNifNi
do do do do
10
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
7
I
-2
>
^
c
ii
123456789
Storms (Ranked by Precipitation Volume)
Figure 2-19. Observed versus modeled runoff at the watershed outlet for 10 selected storms.
Using the normalized outflow depths, runoff ratios for observed and modeled runoff yield was computed
by dividing the normalized depth runoff by the precipitation depth for each of the ten events. The average
projected runoff ratio for the D-storm was also computed using the calibrated model. Figure 2-20 shows
observed versus modeled runoff ratios for each of the 10 selected storms and the average projected runoff
ratio for the D-storm.
2-28
-------
2.0
1.8 -
Precipitation (in.) -•-Runoff-Ratio (Observed)
, ,. -A-Runoff-Ratio (Modeled) —Runoff-Ratio (D-Storm)
l.b
1.4
0.5
0.4
0.0
123456789 10
Storms (Ranked by Precipitation Volume)
Figure 2-20. Observed versus modeled runoff at the watershed outlet for 10 selected storms.
The calibrated model predicted runoff averaged around 0.2 in. of runoff per in. of precipitation, with
values ranging between 0.1 and 0.3 in. for different storms. That predicted range is corroborated by those
predicted by the WinSLAMM study, which estimated an annual runoff coefficients around 0.3 for this
watershed (Pitt and Voorhees, 2010). In addition to volume, the model predicted peak flow was also
evaluated for goodness of fit. Modeled versus observed volume and peak flow were correlated for all 10
storms, as shown in Figure 2-21. The WaPUG criteria bands and the regression equations are also shown
in both panels of the figure. Red points are storm that fell outside the targeted calibration ranges.
0.4
= 0.992x + 0.007
R2 = 0.892
y=1.243x-1.343
R2 = 0.964
0.0
0.0 0.1 0.2 0.3
Observed Flow Volume (in.)
0.4
5 10 15
Observed Peak Flow (cfs)
20
Figure 2-21. Modeled versus observed volume and peak flow correlations, with WaPUG criteria.
So what constitutes a suitable calibration? The calibration effort focuses on refining certain physical
parameters of the model to characterize spatially dispersed features, processes, and responses. The
2-29
-------
computed calibration metrics provide some guidance for how much difference is acceptable between
modeled and observed time series. Nevertheless, occasionally some unfactored conditions cause
differences between modeled and observed responses, which might need to be either identified or
reconciled. One example is localized rainfall events, which are common occurrences. Because rainfall
time series are typically applied uniformly per model segment, the true heterogeneous nature of localized
events cannot always be accurately represented in a model. Of the 10 selected calibration storms, 8 were
relatively easy to calibrate; however, 2 events did not seem to realistically match the observed response in
the pipe network. For example, initial runs for the April 22, 2010, event gave results that (1) were outside
the acceptable WaPUG criteria; and (2) had a negative Nash-Sutcliffe E value. Figure 2-22 shows the
initial hydrograph comparison and computed metrics for the April 22, 2010 storm event.
The underlying rainfall data were investigated for clues as to the source of the discrepancy. The local
rainfall gage data for the storm event had a total rainfall depth of 1.76 in. Daily rainfall data for the same
event at the nearby NCDC Kansas City International Airport gage (COOPID 234358, about 20 miles
away from the watershed gage) was also reviewed for cross comparison. The NCDC gage reported a total
rainfall of 1.13 in., compared to 1.76 at the local gage. That represented a 0.63 in. difference in rainfall
volume, which is not an unusual amount of variation between localized events. The ratio of rainfall
volumes was used to normalize the 5 minute rainfall distribution to conform to the NCDC measured
depth. Figure 2-23 shows the resulting hydrograph and calibration metrics for the adjusted volume event.
Modeled (Pilot Area Outflow)
• Observed (UMKC1)
"3
O
4/22/2010
10:00
4/22/2010
12:10
4/22/2010
14:20
4/22/2010
16:30
4/22/2010
18:40
4/22/2010
20:50
Figure 2-22. Initial calibration: UMKC-01 catchment outlet (rainfall = 1.76 in.).
2-30
-------
Modeled (Pilot Area Outflow)
Observed (UMKC1)
16
14
12
10 -
6 -
Storm Peak: -21.5%
Storm Volume: -25.2%
Nash-Sutcliffe: 0.478
0
4/22/2010
10:00
4/22/2010
12:10
4/22/2010
14:20
4/22/2010
16:30
4/22/2010
18:40
4/22/2010
20:50
Figure 2-23. Intermediate calibration: UMKC-01 outlet (adjusted rainfall depth = 1.13 in.).
The tests revealed that the model responded dramatically to the adjusted storm depth. In fact, the
adjustment resulted in an under-prediction for both peak flow and total volume, but it improved the Nash-
Sutcliffe E value. The test confirms that a better representation of the rainfall volume lies somewhere
between 1.13 and 1.76 in., and it probably follows a slightly different temporal distribution. For the sake
of expediency, the two rainfall values were averaged to produce a value of 1.44 in. for this event. A
straight multiplier of 0.818 (1.44/1.76) was applied to the original 1.76 in. storm distribution for a third
test run for the event. Figure 2-24 shows the final calibration hydrograph comparison and associated
metrics for the event.
Modeled (Pilot Area Outflow)
Observed (UMKC1)
16
14
12
10 -
8 -
6 -
Storm Peak:
19.3%
Storm Volume: 9.4%
Nash-Sutcliffe: 0.344
0
4/22/2010
10:00
4/22/2010
12:10
4/22/2010
14:20
4/22/2010
16:30
4/22/2010
18:40
4/22/2010
20:50
Figure 2-24. Final calibration: UMKC-01 outlet (adjusted rainfall depth = 1.44 in.).
2-31
-------
The second adjusted storm event occurred on October 8, 2009. It also had a reported rainfall depth of
1.93 in. at the local gage, with 0.90 and 1.38 in., respectively reported at the NCDC Kansas City
International Airport (20 miles away) and the Kansas City Downtown Airport gages (12 miles away).
The Normal Ratio Method (Dunne and Leopold, 1978) was used to estimate a rainfall total for the
watershed, resulting in a total estimated rainfall value of 1.56 in. For that storm event, there was a 30 to
40 percent difference in rainfall depths among the three measuring gages.
Table 2-8 presents a summary of model performance as defined by the selected calibration metrics for the
ten selected storm events. The table, whose data are also plotted in Figure 2-19 through Figure 2-21,
identifies three of the storms as having one WaPUG calibration metric that is outside the recommended
range. For the October 8, 2009, event, no additional effort was spent on further adjustments because the
measured rainfall distribution at the gage appeared to be fundamentally different than what probably fell
throughout the watershed. That suspicion arose because model calibration would consistently cause
either one metric or another to fall out of range. The event was the largest recorded storm among the
selected calibration storms. The other two events with metrics that fell outside the range were the two
smallest events on record. Although the peak flow criterion was not met for the October 29, 2009 event,
it had the highest Nash-Sutcliffe E value of all the storms that were evaluated, indicating a high degree of
model efficient at replicating the observed temporal runoff distribution. Because percent error for both
peak flow and volume were on the lower end of acceptable, it probably would have been possible to
adjust the rainfall totals to bring that metric into compliance without violating the peak flow criteria.
However, model testing has already proven the variable nature of rainfall and localized event differences.
Doing so would have been an exercise that did not advance the ultimate objective of establishing a model
baseline. The weight-of-evidence for the remaining storms that were within the criteria was deemed to be
sufficient grounds to verify good systematic model performance. No further adjustments were made to
any of the other observed rainfall records. In general practice, adjusting measured data should not be
done unless there is clear and corroborative justification like the body of evidence presented above.
Table 2-8. Model calibration performance metrics for 10 selected storms events
Start date
09/24/08 04:00
10/15/08 03:00
10/21/08 16:00
11/05/08 22:00
10/08/09 00:00
10/21/09 23:00
10/29/09 06:00
04/22/10 10:00
04/30/1007:00
05/19/10 18:00
End date
09/24/08 12:00
10/15/08 12:00
10/21/08 22:00
11/06/08 04:00
10/09/09 00:00
10/23/09 00:00
10/30/09 05:00
04/22/10 21:00
04/30/10 17:00
05/20/10 03:00
Precipitation
(in.)
0.71
1.10
0.39
1.14
1.56
0.96
0.60
1.44
0.72
0.96
Volume3
(percent error)
-4%
6%
-14% c
23%
-14% c
8%
-10%
19%
18%
-4%
Peak"
(percent error)
18%
18%
-8%
-9%
20%
-12%
-23% c
9%
18%
-4%
Nash-
Sutcliffe £
0.62
0.50
0.67
0.19
0.51
0.67
0.76
0.34
0.32
0.50
a. Calibration target for percent difference in volume: +20 percent to -10 percent (WaPUG, 2002)
b. Calibration target for percent difference in peak flow: +25 percent to -15 percent (WaPUG, 2002)
c. Highlighted metrics are outside the recommended range.
2-32
-------
2.4. Simplifying the Network Articulation for Large-Scale Extrapolation
In the context of optimization, where thousands of individual model runs are often needed as the system
searches for the optimal solution, any amount of computational time that can be saved per model run
quickly translates into significant computational time savings over the course of an optimization.
SUSTAIN provides an aggregated BMP option that facilitates simplification of the representation of the
physical system. Network simplification reduces computation time; however, it also simplifies the
representation of the physical process and has an impact on the accuracy of the model prediction. The
model user must consider the balance between efficiency and accuracy in the selection of the appropriate
level of simplification for the particular application and decision support.
Section 2.3 describes the process of establishing a representative watershed model baseline condition. An
important distinction between the model calibration baseline and the optimization baseline is that for the
latter, the proposed BMP design plan is assumed to be fully deployed throughout the drainage network.
The aspect of the system that is being simplified is the BMP drainage network as overlaid on the
watershed or catchment areas. The aggregate BMP network was compared against a more detailed
articulated network with BMPs and evaluated for predictive accuracy. Through that process, the
following questions are addressed and discussed:
• How much network simplification is tolerable without significantly compromising model
accuracy and/or precision?
• What components of a detailed or articulated BMP and drainage network are appropriate
candidates for aggregation, and to what degree can they be aggregated?
• How much computational advantage does the aggregate BMP approach provide?
This section begins with a discussion of how BMPs were configured for both the aggregate and
articulated configurations, followed by a sensitivity analysis with results and conclusions.
2.4.1. SUSTAIN BMP Representation
The BMP design plan consisted of 179 structural components (158 surface features and 21 subsurface
storage components), as described in Section 2.1.4. All the BMPs have specific design configurations
associated with the functions they provide. Figure 2-25 presents conceptual renderings of three proposed
BMP project sites in the Middle Blue River Green Solutions Pilot Project: bioretention/rain garden,
porous sidewalk, and porous parking on cube storage. Each of the 179 unique surface and subsurface
structures was evaluated and grouped into 11 generalized categories according to similar configurations,
as outlined in Table 2-9. For each BMP type, SUSTAIN provides three optional vertical layers—(1)
surface ponding depth; (2) soil media depth; and (3) underdrain layer (as needed), to characterize the
actual BMP physical response.
2-33
-------
Figure 2-25. Example BMP renderings: Middle Blue River Green Solutions Pilot Project.
Table 2-9. BMP design dimensions and specifications
BMP categories
Bioretention
type 1
type 2
type 3
shallow
Bioswale
Cascade
Porous
pavement
Rain garden
on sidewalk
on storage cube
type 1
type 2
Storage pipes
BMP component depths
(ft)
Ponding
0.83
0.75
0.167
0.83
0.5
0.93
0.01
0.01
0.5
0.5
3
Soil media
0.83
1
0.83
0.83
0.83
0.83
1
0.3
0.83
0.83
--
Underdrain
0.33
0.33
0.33
0.1
--
--
0.33
0.33
--
--
--
Outlet type
weir
weir
weir
weir
weir
weir
weir
weir
weir
2 in. orifice
orifice (variable)
The detailed BMP design dimensions were estimated on the basis of the 100 percent design plans. In the
course of deriving the design dimensions, simplifications were made to accommodate the SUSTAIN BMP
configuration requirements. For example, the designed bioretention, rain gardens, and bioswale have a
side-slope of 2:1; however, in SUSTAIN, the ponding pool of BMPs were configured with vertical sides.
To preserve the ponding pool volume, the ponding depth was adjusted lower while holding the surface
area constant, as shown in the design plan. Another example is the representation of the subsurface
storage pipes, the actual storage pipes are 3 ft in diameter with an outlet structure in a manhole. The
outlet orifices of the storage pipes are at the same elevation as the invert elevation of the horizontal
storage pipes. In SUSTAIN, the storage pipes are represented using cubical storage tanks that were 3 ft
tall, with bottom orifices. The cylindrical pipe volume was maintained by calculating the surface areas of
the tank equal to the pipe volume divided by the assumed depth of 3 ft. In terms of BMP subsurface soil
layers, the associated properties are specified in Table 2-10.
2-34
-------
Table 2-10. Subsurface layer properties for applicable BMP layers
Soil layer
Engineered soil media
Underdrain layer
Native background soil
Property
Porosity
Field capacity
Wilting point
Holtan vegetation parameter
Saturated infiltration rate
Void fraction
Saturated infiltration rate
Value
0.4
0.3
0.1
0.6
2
0.4
0.1
Units
--
--
--
--
in./hr
--
in./hr
A conservative background infiltration rate of 0.1 in/hr was used in this study, for consistency with the
value reported in the WinSLAMM application (Pitt and Voorhees, 2010). The report states that native
undisturbed soils had infiltration rate of 0.2 in./hr, and loam soil fill had an infiltration rate of 0.15 in./hr;
however, in the WinSLAMM application it is noted that disturbed urban soils, such as those typical for
the urbanized parts of Kansas City, can have greatly reduced infiltration rates compared to non-
compacted soils (Pitt and Voorhees, 2010).
For the articulated BMP network representation, individual BMPs are simulated. The only exception
applied in this application, is that if there is more than one unit of the same BMP type in a subwatershed,
the unique units are simulated as a single object with an adjusted volume that accounts for the combined
storage benefit of the unique components. That happened in 14 instances, bringing the number of unique
BMP units in the articulated representation down from 158 to 144. For the aggregate BMP network, the
same 11 BMP types listed in Table 2-9 were aggregated into a single representative volume, while
preserving the relative position in the generalized BMP network. The sizes and drainage areas of each
component were calculated as the total of all the individual BMPs of each type. The outlet structures of
all BMP types were kept consistent. Therefore, in the aggregate representation, the unit surface areas of
each BMP component are estimated as the average surface area of BMPs of the same type. In general,
the outlet structure dimensions were maintained; however, for the below-grade storage pipes, the outlet
orifice sizes in the articulated network of the actual design plan BMPs varied significantly, from 0.375 in.
to 1.25 in. The aggregate BMP needs to assume one representative size to approximate the articulated
variability. Recognizing that the release rate is largely dependent on the orifice diameter, the
representative storage outlet orifice diameter was calibrated (within the 0.375 in. to 1.25 in. range) to
obtain a close match of the aggregate BMP outflow hydrograph with that of the articulated representation.
For private parcel BMPs, the WinSLAMM application indicates that rain barrel and rain gardens are
suitable alternatives. The BMP dimensions are listed in Table 2-11. The private BMP rain gardens were
modeled with the same subsurface parameters as the public green BMPs, as summarized in Table 2-10.
Table 2-11. Private BMP design dimensions and specifications
BMP
categories
Rain garden
Rain barrel
BMP dimensions
Surface area
200 sq ft per house
(1,000 sq ft roof)
Ponding
(ft)
1
Soil media
(ft)
2
Underdrain
No underdrain
35-gallon tank
Outlet type
Weir
Weir and orifice
2-35
-------
2.4.2. Articulated versus Aggregated Network
For this application, the baseline condition in Figure 2-26 below shows the fully articulated model
representation of BMPs in the pilot watershed, where distributed collection points and conveyance
segments are explicitly defined. Model configuration included 85 subwatersheds and 36 unique pipe
segments. Only pipes with diameters larger than 12 in. were included in the model. Within each
catchment, the amount of storage associated with each practice from the design plan was added to the
network. Flow is routed from each catchment through the network to the watershed outlet.
Legend
—i-— Conduit
* Stormwaterinlet
X Flow monitoring station
® Assessment point
* Junction
• Basin routing
( BMPs
Subwatershed
Fully Articulated Network
0 0.035 0.07
0 0.025 0.05
INJ
A
Figure 2-26. SUSTAIN model representation of fully articulated model network.
In this aggregate BMP network configuration, the routing network is simplified in recognition of the fact
that the time of concentration for the watershed is on the order of about 15 minutes or less, which is
similar to the simulation time step. The time of concentration was estimated by comparing the average
amount of time between the peak of the rainfall event and the peak observed flow at the outlet gage.
Figure 2-27 presents a conceptual diagram of the aggregate BMP representation of the same network
presented in Figure 2-26. Each BMP component is represented as an object; however, the typical
situation of that BMP relative to others within close proximity is preserved by the network. Because
aggregate BMP application is intended for places with short times of concentration, it assumes that all
intermediate connections are instantaneous routing elements where no pipe conveyance is simulated.
Runoff generated from the 100-acre pilot is proportionally distributed among the various pathways
according to the typical orientation and distribution of land cover upstream of each type of practice. The
land cover distribution was surveyed and estimated using GIS spatial analysis of aerial photographs. A
portion of runoff can also be assigned as untreated, which would allow it to flow directly to the outlet
without passing through the BMP network.
2-36
-------
Drainage Area Land Distribution
(Proportional to articulated network contributing areas)
Untreated
Land
Porous
Sidewalk
Shallow
Bioretention
Porous
Pavement on
Storage Cube
outlet
Aggregate BMP Network
0 0.035 0.07 0.14
0 0.025 0.05 0.1
I Miles
N
A
Figure 2-27. Conceptual diagram of a comparable aggregate BMP representation.
Notwithstanding the inherent simplification involved, some of the major advantages of the aggregate
network representation is that it is (1) easier to implement in the modeling environment because fewer
individual BMPs need to be configured; (2) scalable for other areas, allowing for extrapolation to a larger
watershed; and (3) computationally less demanding because detailed simulation of all pipe segments is
not necessary. Those advantages become critical when performing optimization where thousands of
consecutive model runs are performed.
Although all models are simplified version of a natural system, the aggregate BMP is significantly less
detailed than the network typically used to represent watershed systems with BMPs. The aggregate BMP
representation assumes that the dominant function of BMPs is water storage, evaporation, and infiltration,
and that the impacts of transport through a network, especially timing and attenuation are less significant.
Model testing (i.e., calibration/validation), and an understanding of the local hydrologic response, can be
used to evaluate to what degree that the various levels of simplification are both valid and representative
of the local conditions. Four comparison simulation sets for the 100-acre pilot site were developed to test
the model (1) at two different time steps; and (2) with and without BMPs. For those runs, the D-storm
was applied, assuming average antecedent moisture conditions. Table 2-12 is a matrix of figure
references for the comparison runs. The figures and associated observations and conclusions are
subsequently presented.
2-37
-------
Table 2-12. Reference matrix for aggregate versus articulated BMP comparison tests
Model scenario
Baseline condition runs
Proposed BMP design plan
Simulation time step
Hourly
Figure 2-28
Figure 2-30
15 minute
Figure 2-29
Figure 2-31
Baseline Condition Runs
Figure 2-28 shows a hydrograph comparison for the aggregate approach versus the articulated routing
networks for the baseline watershed model scenario (without BMPs). Both were run on an hourly time
step for this comparison. The aggregate hydrograph (solid line) shows a strong match with the fully
articulated hydrograph (discrete points). Total outflow volumes show 100 percent agreement (0 percent
difference), while peak flow shows a 1.4 percent higher peak for the aggregate scenario. The aggregate
representation forgoes detailed routing through pipes and instead uses an instantaneous flow routing to the
outlet. This is consistent with expectations since simplification of the routing reduces attenuation and
associated time of concentration.
Similarly, Figure 2-29 shows a hydrograph comparison for the baseline scenario; however, the models
were run with a 15 minute time step. Overall, the aggregate hydrograph (solid line) shows a strong match
with the fully articulated hydrograph (discrete points). Once again, the total outflow volumes show 0
percent difference, but the peak flow for the aggregate scenario is 1.1 percent higher than for the
articulated scenario. Although both the hourly and 15 minute time step simulations show good overall
agreement for both the aggregate and articulated network representations, it is interesting to note the
difference in shape of the hydrographs. The 15 minute hydrographs for both the aggregate and articulated
networks (Figure 2-29) provide a higher resolution of the timing of the response. The timing of the peak
is modeled at 9:00 a.m. for the hourly scenarios (which interpreted, means during the 9:00 hour), but is
plotted at 9:45 a.m. for the 15 minute scenarios (i.e., between 9:45 a.m. and 10:00 a.m.).
Proposed BMP Design Plan
The effect of time step on model precision is even more pronounced when BMPs were incorporated into
both the aggregate and articulated scenarios. Figure 2-30 shows a hydrograph comparison for the
aggregate approach versus the fully articulated routing networks for the proposed BMP design plan. At
an hourly time step, the aggregate BMP produces a peak flow response that is 11.4 percent lower than the
articulated response; however, it gives a total flow volume that is 7.5 percent higher than the articulated
network, which equals a 5 percent lower volume reduction relative to the baseline condition). The lower
peak in the aggregate hydrograph in Figure 2-30 translates to more attenuation than the articulated
hydrograph.
2-38
-------
Aggregate ( Design Storm Medium ET)
Fully Articulated ( Design Storm Medium ET)
Storm Volume: 0%
Nash-Sutcliffe: 0.958
9/10/2008
0:00
9/10/2008
4:00
9/10/2008
8:00
9/10/2008
12:00
9/10/2008
16:00
9/10/2008
20:00
Figure 2-28. Hourly time step, aggregated versus articulated (baseline, no BMPs).
Aggregate ( Design Storm Medium ET)
Fully Articulated ( Design Storm Medium ET)
o
5=
O
torm Volume: 0%
Nash-Sutcliffe: 0.926
0 4
9/10/2008
0:00
9/10/2008
4:00
9/10/2008
8:00
9/10/2008
12:00
9/10/2008
16:00
9/10/2008
20:00
Figure 2-29. Fifteen minute time step, aggregated versus articulated (baseline, no BMPs).
2-39
-------
Aggregate ( Design Storm Medium ET)
Fully Articulated ( Design Storm Medium ET)
30
25
Storm Peak: -11.4%
i
o
Storm Volume: 7.5%
Nash-Sutcliffe: 0.732
9/10/2008
0:00
9/10/2008
4:00
9/10/2008
8:00
9/10/2008
12:00
9/10/2008
16:00
9/10/2008
20:00
Figure 2-30. Hourly time step, aggregated versus articulated (with BMPs).
Figure 2-31 shows a hydrograph comparison for the proposed BMP design plan scenario; however, the
models were run with a 15 minute time step. The aggregate hydrograph (solid line) shows a fairly good
match with the fully articulated hydrograph (discrete points). For the BMP scenario, the total outflow
volumes show 4.2 percent higher flow volume than the articulated, and a 0.1 percent lower peak. Once
again, the shape of the hydrograph shows more attenuation of the water from the start of the storm, as
seen in Figure 2-31.
Aggregate ( Design Storm Medium ET)
Fully Articulated ( Design Storm Medium ET)
30
25
o
1
O
Storm Peak: -0.1%
Storm Volume: 4.2%
Nash-Sutcliffe: 0.78
9/10/2008
0:00
9/10/2008
4:00
9/10/2008
8:00
9/10/2008
12:00
9/10/2008
16:00
9/10/2008
20:00
Figure 2-31. Fifteen minute time step, aggregated versus articulated (with BMPs).
There are a few notable observations from these sensitivity analyses. First, simulation time step seems to
have the most influence on temporal resolution of peak flow. In both the baseline and the BMP scenarios,
2-40
-------
the peak flow occurs at 9:00 a.m. for the hourly time step and at 9:45 a.m. for the 15 minute time steps.
The second notable observation is the effect of the modeled BMPs on both volume and peak flow. The
BMP scenarios at both time steps predicted more storm volume for the aggregate relative to the
articulated configuration. Although both networks contain the same kinds and total storages for the
various BMPs in plan, the aggregate network generalizes the connectivity of the routing network. The
higher outflow volume predicted by the aggregate BMP network means that the modeled BMP network
provides a slightly lower volume reduction than the fully articulated network (5 percent for the hourly
time step, versus 3 percent for the 15 minute time step). Detailed routing provides greater attenuation of
the flows. Although the baseline run did not show any difference in the responses, the BMP scenario
accentuates the effects of flow attenuation. Because the optimization objective is volume control and not
peak flow, the selected simulation time step is not as much of a concern. However, because the 15 minute
time step gives a better overall agreement between the aggregated and articulated networks, it is preferred
over the hourly time step for this pilot study modeling. In addition, the fact that BMP performance is
slightly under-predicted with the aggregate BMP configuration provides an additional margin of safety
for the predicted volume control benefit.
Model Run-Times
One of the most notable advantages of the aggregate BMP is the time savings for each model run. These
reductions in time have a large impact during optimization runs which typically require in excess of
10,000 iterative model runs over the simulation time period. Table 2-13 shows a comparison of model
characteristics for three configurations: (1) the original XP-SWMM model configuration; (2) the fully
articulated SUSTAIN baseline; and (3) the simplified aggregate SUSTAIN baseline models. The model
comparisons presented were all run using the same weather boundary condition: D-storm with average
antecedent recovery conditions at a 15 minute simulation time step.
Table 2-13. Comparison of model characteristics for three configurations
Characteristics
Number of catchments
Number of pipes
D-storm peak (cfs)
D-storm volume (cubic feet)
Single run-time (sec.)
10,000 runs (hr)
Configuration
XP-SWMM
179
350
N/A
N/A
25
69.5
SUSTAIN articulated
85
36
24.52
46.38
1
2.75
SUSTAIN aggregate
1
1
24.80
46.38
<1
<1
The run-time values presented above for the XP-SWMM model were derived from simulating a single
storm event (9/24/2008) using the version of the model obtained from Kansas City that was exported into
an SWMM5 format. Estimates of run-times for 10,000 runs were calculated by multiplying a single-
event run-time by 10,000 (which represents a possible number of iterations associated with an
optimization run).
2-41
-------
2.5. Optimizing BMP Opportunity for CSO Mitigation in the 069
Watershed
The previous objectives of this effort have established a calibration baseline condition and modeled the
BMP network associated with the approved design plan for the 100-acre pilot site. Model testing also
demonstrated the validity of a streamlined spatial representation for model representation. Because WSD
has committed to implement the design plan for the 100-acre pilot site, it was included as part of the
model baseline for optimization. The previous analyses presented thus far represent the foundational
elements on which optimization scenarios are based. The central question within the minds of regional
policy makers is what degree of management is required to mitigate CSO throughout the larger 069
sewershed? Exploratory management alternatives include (1) extending the proposed GI design plan (GI
on public rights of way) to the remainder of the 069 sewershed; (2) expanding the scope of GI to include
implementation of certain practices on private land; and (3) exploring gray infrastructure options for
supplemental CSO storage at the regulator outlet. The objectives for optimization are to (1) maximize
runoff volume control; and (2) minimize the total capital cost of implementation, as needed, to satisfy the
allowable exceedance criteria for CSO (i.e., treating 100 percent of D-storm runoff). Figure 2-32
conceptually illustrates the development sequence of exploratory optimization scenarios relative to
established baseline conditions.
Allowable Exceedances
c
CD
o
00
u
CD
_Q
E
Calibration
Baseline
\ \
Public Green
(Pilot Area)
Optimization
Baseline
Public Green
(Other 069 areas)
Public Green
(Pilot Area)
Optimization
Scenario 1
MaxA/Min$
Gray at
Regulator
Private Green
(All 069 areas)
Public Green
(Other 069 areas)
Public Green
(Pilot Area)
Optimization
Scenario N
_o
Q.
X
T3
CD
E
E
o
u
Figure 2-32. Conceptual sequence of optimization scenarios relative to baseline conditions.
SUSTAIN provides a platform for synthesizing information derived from various independent but
complementary research efforts in Kansas City. The respective conclusions of the studies discussed in
Section 2.1 are maintained during the synthesis because of the important assumptions associated with
each component. For example, the BMP design plan for the 100-acre pilot area incorporates site
feasibility surveys of public rights of way in the BMP designs; therefore, the relative BMP sizing and
placement rules were maintained both within the pilot area and during extrapolation to other parts of the
069 sewershed. The WinSLAMM application quantifies benefits for management alternatives on private
property parcels throughout the sewershed. Table 2-14 maps out certain key SUSTAIN optimization
requirements with the relevant study from which they were based. The updated SWMM model refers to
the reconfigured SWMM5 baseline calibration model described in Section 2.3, for which development
was partially based on the XP-SWMM model.
2-42
-------
Table 2-14. SUSTAIN application data needs and associated data source (research study)
SUSTAIN requirements:
• Available
- Not available
Calibration Baseline
Optimization baseline
CSO control targets
Pipe routing information
Dry-weather sewer flow
BMP specifications
BMP capital cost
Updated
monitoring
data
•
-
-
-
-
-
-
Updated
SWMM
model
•
•
•
•
•
-
-
BMP design
plan
-
•
-
-
-
•
•
Desktop
analysis
-
-
•
-
-
-
•
WinSLAMM
application3
•
-
-
-
-
•
-
a. For private property
The remainder of this section (1) describes how the spatial extent of the model was expanded to the 069
sewershed and its overflow regulator; (2) defines the objective functions and constraints for the
optimization problem formulation; (3) describes how BMP costs were derived from local contractor bid
data; and (4) presents the SUSTAIN optimization results and sensitivity analysis.
2.5.1. CSO 069 Model Configuration
The CSO 069 watershed (480 acres) was subdivided into five subwatersheds ranging in size from 58.9 to
139.7 acres, with an average size of 96.2 acres. As previously described, the aggregate BMPs within the
100-acre pilot study area were maintained as constant for the optimization baseline condition. For the
remaining areas outside the 100-acre pilot study area, the relative BMP to area ratio was prescribed, with
the optimization decision variable defined as the percentage of the remaining 069 outfall drainage area
receives GI practices according to the same proportions applied within the pilot study area. In addition,
the volume of the gray storage basin was also defined as a decision variable for the scenarios when a
supplemental storage basin was an available alternative.
Because model testing showed that the aggregate BMP configuration was valid and representative for
subwatersheds around 100 acres in size, the remainder of the area within CSO 069 was delineated into
subwatersheds of similar size in which aggregate BMP rules could be applied. In each subwatershed, the
main sewer truck line was retained to create a composite network of primary pipe segments with
aggregate BMP drainage areas. This network representation allowed for two key advantages for the
model, namely, (1) the ability to use a simplified aggregate BMP approach in each of the five
subwatersheds; and (2) the ability to preserve the primary flow network that ultimately deliver the water
volume to the regulator. Figure 2-33Error! Reference source not found, is a map showing the
subwatersheds, pipe connections, and the relative location of the regulator assessment point for CSO 069.
2-43
-------
SUD watershed
Junction
Assessment Point
SUSTAIN Conduit
Existing Storm Sewer
Land-to-BMP Routing
CSO 069 Watershed
SUSTAIN Model Setup
Figure 2-33. Subwatersheds, pipe connections, and regulator assessment point for CSO 069.
2-44
-------
For the baseline 069 sewershed model, the calibrated time series from the 100-acre pilot site were
extrapolated throughout the 069 sewershed on the basis of HRU distribution. Because the HRU runoff
time series characterize the rainfall-runoff response, the HRU is a convenient and consistent basis for
extrapolating outside of the calibrated pilot watershed. The simplified drainage network was modeled as
a series of junctions and circular pipes. One key element of the 069 model is the regulator at the
sewershed outlet. As shown in original the regulator design schematic (Figure 2-34), the structure
behaves more like a run-through device with a depressed area in the middle of the device where water can
accumulate. The top panel is the view from above while to bottom panel is the side view. Figure 2-35 is
a conceptual schematic of water movement through the regulator.
PLAN e/- JUNCTION
gmgjv-
L ^^S
'• *. 0»* flo/VoT
*^!^_^"" Cl"
Figure 2-34. CSO 069 regulator schematic.
2-45
-------
Combined Sewer Overflow
Figure 2-35. Conceptual schematic for the CSO regulator.
Water in the storage depression is diverted to the wastewater treatment plant through an 18 in. diameter
pipe that runs perpendicular to the inflow direction of the water. During dry-weather flow, the sanitary
flow exits the chamber through the 18 in. pipe. During wet-weather flow conditions, the increased
velocity of the water will bypass the sanitary pipe outlet and continue downstream to an overflow point.
In SUSTAIN, the total storage volume of the regulator is represented as a vertical box structure with a
weir at the height of 1.5 ft, width of 7.8 ft, and an 18 in. bottom orifice. The baseline overflow frequency
was confirmed by running the model with continuous time series from the 2004 typical year. Because of
slight differences in regulator geometry and the head loss associated with the sanitary pipe outlet
perpendicular to the line of flow, the orifice discharge coefficient was adjusted to a value of 0.24 to
maintain the expected 33 overflow events for the 2004 typical year.
2.5.2. Problem Formulation
As previously conceptualized in Figure 2-32, the optimization baseline included the BMP designs being
implemented within the 100-acre pilot area. Using the CSO 069 model configuration presented in Section
2.5.1, the optimization problems were formulated for the selected management options. The generalized
multi-objective functions and constraints are presented as follows:
Minimize
Minimize
Subject to
X BMP Capital Costs
Regulator overflow volume (for the D-storm)
Final design plan for 100-acre pilot site
• Exploratory Management Options:
o GI on 069 public rights of way (outside of pilot area)
o GI on 069 private parcels
o Supplemental gray infrastructure at the regulator
Background analysis from the OCP suggests that adequately controlling runoff volume such that no
overflow occurs during the D-storm (D-storm: 1.4 in. depth, 0.6 in./hr intensity, 16.75 hr duration) would
achieve the CSO allowable exceedance objective. Considering the 2004 typical year, controlling the D-
storm reduces the number of overflows from 33 to 2 or 3, as further described in Section 2.5.6. Because
antecedent conditions can have a significant influence on runoff generation potential and BMP
performance, three sets of optimization runs were executed to provide a range of results corresponding to
low, medium, and high antecedent moisture conditions. The three sets of time series are described in
Section 2.3.3, under Design Storm Time Series.
2-46
-------
2.5.3. BMP Cost Representation
BMP cost information is a critical element of cost-benefit optimization. Local sources were used to
derive capital cost data for GI on public rights of way and gray infrastructure components while the costs
of BMPs on private parcels were estimated from local and literature values as detailed below.
GI Costs
The BMP cost data used in this study were derived using March 8, 2011, contractor bid data provided by
WSD for the Middle Blue River Green Solutions Pilot Project. There were both general site preparation
and specific BMP-associated costs provided in the contractor bid data. The general cost components
included the following items:
• Preconstruction costs (mobilization, traffic control, erosion and sediment control, surveying and
construction staking);
• Tree removal and utilities relocations;
• Street and sidewalk improvements;
• Landscape restoration; and
• Mulch, plants, and other miscellaneous landscape materials.
The specific BMP-associated costs included the following items:
• Below-grade storage system structures and general backfill; and
• BMP construction for various surface BMP types (rain gardens, shallow bioretention, porous
pavement, cascades, bioretention, and grass swale).
Consistent with the cost module input format used by SUSTAIN, the general cost components from the
contractor estimate were converted to area-based BMP-associated costs. Those costs were proportionally
distributed among the BMPs according to the total number of BMP units in the design plan, with the
exception of mulch, plants, and other miscellaneous landscape materials items. Those costs were evenly
divided among the BMP types that incorporate vegetation (i.e., rain garden, bioretention, cascade, and
bioswale). The BMP-specific cost items were then averaged by BMP type to derive a unit cost. As
summarized in Table 2-15, the total unit cost for each type of BMP was calculated by adding the
distributed general site preparation costs with the BMP-specific costs.
Table 2-15. BMP capital costs for the 069 sewershed
BMP types
Bioretention
Bioswale
other
shallow
Cascade
Porous sidewalk
Porous pavement on cube
Rain garden
Storage
BMP cost
Site preparation
$19,616
$19,616
$19,616
$19,616
$16,163
$16,163
$19,616
$19,616
BMP-specific costs
$1,938
$3,247
$2,923
$3,383
$13.1 per square foot
$10.7 per cubic foot
$1,249
$59,048
Total cost per unit
$21,554
$22,863
$22,539
$22,999
Varies by surface area
Varies by volume
$20,865
$75,210
The functional effects of GI applied to private property were modeled as (1) disconnected downspouts
with rain barrels; and (2) on-site rain gardens. The WinSLAMM application did not include a phase
which provided cost estimates for these BMPs; therefore, these costs were derived from local applications
2-47
-------
and literature sources. Schueler et al. (2007) published a manual through the Center for Watershed
Protection (CWP), which provided construction cost estimates for both rain garden and rain barrels
retrofits, and design and engineering cost estimates of 5 to 40 percent of the construction cost.
Table 2-16. Cost estimation for private parcel retrofit BMPs
BMP type
Rain garden
Rain barrel
BMP cost ($ per gallon of runoff treated)
Construction cost
Literature range
$0.40-$0.67
$1.67-$5.35
Median
$0.53
$3.34
Design and engineering
(40% of construction costs)
$0.21
Not applicable
Total cost
$0.75
$2.81
Source: Schueler et al., 2007
Costs were presented in terms of runoff volume treated; however, for rain gardens, an additional step was
required to translate the cost data into a convenient basis for SUSTAIN because the volumetric cost
component in SUSTAIN is based on excavation volume instead of storage volume. Nevertheless, the rain
gardens were parameterized with a constant uniform soil column depth, as previously shown in Table
2-10. Because the uniform soil media column has a predefined void space, there are two options for
representing its cost in SUSTAIN: (1) calculate an equivalent excavation depth from the treatment depth;
or (2) calculate a surface area equivalent cost. Using an equivalent surface area basis, the computed $5.6
per cubic feet of void space translated to an area-based cost of $10 per square foot. For the rain barrel,
the unit cost of a 3 5-gallon unit and downspout connection was estimated to be $100 by considering a
rain barrel cost of approximately $2 per gallon (Woodland Direct, 2011), which falls within the published
literature range presented in Table 2-16.
Gray Infrastructure Costs
As presented in the OCP, the total capital cost of a 2 MG storage facility was estimated to be $30.6
million. The facility includes a 2 MG storage tank, 1.5 MG per day pumping station, 5 IMG per day
screening, a 100 ft 48 in. sewer pipe, and 500 ft 12 in. force main, and an odor control facility. The
estimated capital costs include an allowance of 25 percent of the total estimated construction cost for
planning, engineering and design, and an additional contingency cost of 25 percent. This cost estimate is
based on 2006 data and has been updated for this case study using a multiplier of 1.163 (20 city
Engineering News-Record (ENR) index value of March 2011/2006 Annual Average) to reflect a 2011
cost of $35.6 million.
To represent the optimization cost function for the storage facility, the separate fixed cost elements were
calculated separately from those that varied with size. The selection of a storage facility initially involved
costs including upfront planning, mobilization, and design costs, regardless of the facility's size. The
resulting cost function also implicitly accounts for economies of scale as the storage capacity increased
beyond the initial cost investment. An initial fixed cost of $11.63 million (about one-third of the
literature-based cost value, or $10 million x 1.163) was approximated to be a reasonable amount on the
basis of inference from local contractor bids for certain components. The remainder of the gray
infrastructure cost was approximated as a linear function of storage capacity by back-calculating the rate
as follows:
Storage Cost = ($35,600,000 - $11,630,000) - 2 MG = $12 per gallon = $89.76 per cubic foot
Total Capital Cost ($) = $11,630,000 + $89.76 x (storage volume in cubic feet)
2-48
-------
2.5.4. Optimization Sensitivity Tests
While developing the optimization baseline model, sensitivity testing for the articulated versus the
aggregated network configuration demonstrated that the simulation results were sensitive to the
simulation time step (Section 2.4.2). Furthermore, because SUSTAIN optimization was to be run on an
event basis for the D-storm instead of on a continuous simulation basis where moisture recovery is
dynamically accounted for, the user-specified antecedent moisture condition represented a potentially
significant unknown variable. Two sets of sensitivity tests were performed to test the influence of (1)
model simulation time step; and (2) antecedent moisture conditions on the predicted cost-effectiveness
curve derived through optimization.
Simulation Time Step Sensitivity Analysis
Sensitivity testing (low, medium and high recovery conditions) of overflow volume reduction at the
regulator outlet revealed a surprising trend that initially seemed counterintuitive. The analysis presented
in Section 2.4.2 suggests that because more runoff volume was associated with the 15 minute time step
compared to hourly, using the smaller time step would result in a more conservative overflow reduction
estimate during optimization. However, the opposite appeared to occur. Figure 2-36 summarizes the
influence of model simulation time step on the optimization results and suggests that the 15 minute time
step consistently has a higher percent volume reduction along the entire cost-effectiveness curve than the
hourly time step. On further investigation it was discovered that it was actually the difference in resulting
baseline volume associated with the different time steps that was responsible for the higher percent
volume reduction as shown.
OU 70
C cno/ .
°ercent Reductio
o js>. c
DOC
o ^p v
^ 0^ 0
ono/, .
^ zu /o
0
t
**"* *
'ercent Reduction (60-min)
'ercent Reduction (1 5-min)
Volume Reduction (60-min)
Volume Reduction (1 5-min)
O.3U
- 3.25
- 2.60
- 1.95
- 1.30
- 0.65
n nn
U 70 n l \J.\J\J
0 5 10 15 20 25
CO
D)
O
I
o
T3
-------
2-36 as a function of the associated BMP capital costs. Up until around the $10 million point, the
absolute volume reductions for both time steps seem to track very closely; however, after that point, the
15 minute time step curve tends to follow a lower trajectory than the hourly simulation time step.
Another factor at play behind the different responses is attenuation associated with routing through the
pipe network to the regulator. In general, the smaller 15 minute time step does a better job of predicting
the peak attenuation than the hourly time step simulation because it has less temporal averaging;
consequently, the resulting peak attenuation at a 15 minute time step can be slightly higher. That is why
the overall volume reduction for the 15 minute time step trails the hourly time step even with increasing
GI spatial extent.
Antecedent Moisture Condition Sensitivity Analysis
Optimization runs were generated for three antecedent moisture conditions to provide a range of response
for the D-storm. Figure 2-37, which was previously presented as Table 2-7 during the discussion about
how the D-storm time series were developed, compares antecedent moisture conditions for the D-storm.
Note the differences in dry days and evaporation rates associated with each condition.
0.16
0.14 -
-§ 0.12 -
0.10
0.08 -
IB 0.06 -
LU
0.04 -
0.02 -
0.00 -
i Evaporation (in./day)
Dry Time (days)
U.D
0.001
Low Recovery
(High Moisture)
Most Conservative
Medium Recovery
(Medium Moisture)
Average Condition
High Recovery
(Low Moisture)
Least Conservative
Figure 2-37. Comparison of antecedent recovery conditions for the D-storm.
For this analysis, the cost-effectiveness curves associated with optimization of GI in public areas served
as the basis for comparison. Figure 2-38 illustrates the influence of antecedent moisture conditions on
optimization results.
2-50
-------
g
"o
^
T3
O
O
t
6
E
£
"oo
Q
60%
50%
40% -
30% -
Public Green (Pilot Area)
—Public Green (069, Low Recovery)
—Public Green (069, Medium Recovery)
—Public Green (069, High Recovery)
0 Public Green (069 Max)
20% -
10%
10 15
BMPCapitalCost( Million $)
Figure 2-38. Sensitivity of antecedent moisture conditions on optimization results.
Comparing those curves reveals a number of interesting patterns. One of the first noteworthy
observations is how close the medium recovery curve is to the low recovery curve. The nonlinear
relationship between the product evaporation rate and the number of days is evident in the spread between
the three graphs. Looking at the point on Figure 2-38 labeled Public Green (069 Max) reveals that at a
fixed cost interval, the spread in volume reduction between the three scenarios is relatively small (-1.1 to
+2.4 percent) compared to the range of cost variation around a fixed volume reduction (+3 to -9.9
percent). The respective ranges of variability are plotted as vertical and horizontal error bars in Figure
2-38. Table 2-17 summarizes the sensitivity of optimized treatment cost and associated overflow
reduction for the three levels of antecedent moisture conditions.
Table 2-17. Sensitivity of cost-effectiveness to changes in antecedent moisture condition
Antecedent moisture
condition
Low recovery
Medium recovery
High recovery
Reduction variation (fixed cost)
Cost
($ million)
$19.26
$19.26
$19.26
Overflow
reduction
48.3%
49.4%
51.8%
Percent
difference
-1.1%
--
2.4%
Cost variation (fixed reduction)
Overflow
reduction
49.4%
49.4%
49.4%
Cost
($ million)
$19.83
$19.26
$17.36
Percent
difference
3.0%
--
-9.9%
The range of the horizontal and vertical error bars around the point associated with maximum projection
of the GI design plan for the 100-acre pilot site to all public rights of way throughout the 069 sewershed
represent the range of uncertainty associated the selected initial condition associated with moisture
storage, and its influence on optimization results. For subsequent analyses, cost-effectiveness curves for
the medium recovery condition are plotted with the associated horizontal and/or vertical error bars to
illustrate the possible range of variability associated with optimization results.
2-51
-------
2.5.5. Exploratory Management Scenarios
Three optimization scenarios were developed to evaluate the cost-benefit of the exploratory management
alternatives. Although the optimization objective is 100 percent containment of the Type-D storm,
plotting the cost-effectiveness curves associated with each scenario provides insight into the cost-benefit
trajectory toward achieving the management goal of controlling the Type-D storm. The baseline
condition and a predetermined sequencing of three exploratory optimization scenarios were simulated.
Table 2-18 provides a summary and description of the baseline and the three exploratory optimization
scenario sequences.
Table 2-18. Summary and description of baseline and exploratory optimization scenarios
Optimization scenario
Baseline
Exploratory
Public green (pilot
area)
Gray only
Public green +
gray
Public + private
green + gray
Description
Full adoption of the BMP design plan within the 100-acre pilot study area
Baseline + supplemental gray storage at the 069 regulator outlet
Baseline + public green expanded to other 069 areas + gray supplemental
storage
Baseline + public green expanded to other 069 areas + private green
opportunity + gray supplemental storage
The optimization scenarios were run using the D-storm time series, with the medium antecedent moisture
conditions; however, the projected range of variation associated with low and high antecedent moisture
conditions was also plotted using error bars around key junctions along the trajectories. Figure 2-39
shows the key junctions and cost-effectiveness curve trajectories for the three exploratory optimization
scenarios. Because the optimization target is 100 percent containment of D-storm overflows, Figure 2-40
zooms into the Optimization Target box in Figure 2-39 to show a comparison of overflow compliance
costs for the three exploratory scenarios, with error bars denoting the range of variation associated with
high and low antecedent moisture conditions.
2-52
-------
100%
90% -
80% -
70% -
60% -
50% -
40% -
30% -
20% -
10% -
0%
Target
Public Green (Pilot Area)
—Public Green (Other069 Areas)
O Public Green (069 Max)
^Public + Private Green
• Public + Private Green (Max)
•—•Gray only
^Public Green + Gray
Private Green + Gray
0
10 15
45
50 55
60
20 25 30 35 40
BMPCapitalCost( Million $ )
Figure 2-39. Cost-effectiveness junctions and trajectories for exploratory optimization scenarios.
$54
m
$0
Gray Only
Public Green + Private Green +
Gray
Public Green + Gray
Optimization Scenario
Figure 2-40. Comparison of overflow compliance costs for the three exploratory scenarios.
Among the exploratory optimization scenarios, proposed GI options for both public and private land were
maximized, with the exception of the gray only scenario, where GI was not considered. The difference in
cost is attributable only to the size of the gray supplemental storage associated with each of the three
scenarios. Table 2-19 shows the component sizes and costs from the exploratory optimization scenarios.
2-53
-------
Table 2-19. Management component size and costs for exploratory optimization scenario
Scenario
Public
green
Private
green
Gray
Management component
other
shallow
Bioswale
Cascade
Porous sidewalk
Porous pavement on cube
Rain garden
Pipe storage
Rain barrels
Rain gardens
Gray only
Public green + gray
Public + private green + gray
Total storage
capacity
(gallons)
520,023
82,109
44,313
64,188
59,301
11,404
474,081
915,905
14,662
950,443
2,778,970
1,819,660
1,617,700
Total cost
IS)
$4,310,671
$519,610
$102,447
$522,682
$381,698
$180,020
$6,069,606
$7,178,998
$41,480
$706,000
$44,880,833
$31,681,177
$32,651,283
Unit storage
volume cost
($/gallon)
$8.29
$6.33
$2.31
$8.14
$6.44
$15.79
$12.80
$7.84
$2.83
$0.74
$16.15
$17.41
$20.18
2.5.6. Validating Overflow Control Using Continuous Simulation for a Typical Year
Optimization was performed using the D-storm series as the driver for rainfall and runoff for a range of
antecedent moisture conditions. However, the design storm approach does not necessarily tell us which
storms are controlled under the context of a continuous simulation run. For continuous simulation, an
overflow event can be caused by a smaller event that occurs immediately after a series of events that have
saturated both the ground and the BMP storage capacity. For this validation effort, BMP selections
corresponding to the three points crossing the 100 percent D-storm containment threshold (as summarized
in Table 2-19) were tested on a continuous simulation basis using the 2004 typical rainfall and ET time
series as the driver. Storm separation was performed on the 2004 year by dividing the time series into
discrete storm events for comparison. The storm separation process assumed a minimum inter-event time
of 12 hours, with a minimum storm size of 0.1 in. Of the 50 discrete storm events that resulted for 2004,
6 had an overall rainfall depth greater than 1.4 in. Those storms (as summarized in Table 2-20)
theoretically represent events that would otherwise be allowed to overflow the regulator.
Table 2-20. Storms summary for the six largest storm events in 2004
Storm start
time
Storm end
time
D-storma
8/27/04 17:00
3/4/04 4:00
8/23/04 8:00
5/18/04 6:00
9/5/04 17:00
6/9/04 3:00
8/28/04 18:00
3/5/04 12:00
8/24/04 21:00
5/19/04 17:00
9/6/04 8:00
6/10/04 22:00
Rainfall depth
(in.)
1.4
1.570
1.680
1.750
1.840
1.980
2.090
Peak intensity
(in./hr)
0.6
0.65
0.180
0.790
0.280
0.870
0.520
Average ET rate
(in./hr)
0.03
0.092
0.018
0.092
0.081
0.066
0.108
Antecedent dry
hr
72
49
12
79
94
156
221
a. Assumes average antecedent moisture conditions (3 dry days, 0.03 in./day ET)
2-54
-------
The validation test supported confirmation that the design objectives had been met. The simulation
results revealed that three of the six events overflowed under the Gray Only solution; whereas only two of
the six events overflowed for both of the Green + Gray scenarios. Both solutions exceeded the design
objective of six overflow events per year. The storms dates and the overflow volume and peak flow rate
are summarized in Table 2-21. Note that the June 9 event barely crested the regulator storage facility
with a marginal peak discharge rate of 0.0157 cfs.
Table 2-21. Overflow events summary
Scenario
Baseline (No BMPs)
Near optimal
solutions
Gray only
Public green +gray
Public + private
green + gray
Overflow events
Overflow occurs for all storms > 0.28 in., with a total of 33 overflow events in 2004.
Storm start
time
6/9/043:00
8/23/04 8:00
9/5/04 17:00
8/23/04 8:00
9/5/04 17:00
8/23/04 8:00
9/5/04 17:00
Storm end time
6/10/04 22:00
8/24/04 21:00
9/6/04 8:00
8/24/04 21:00
9/6/04 8:00
8/24/04 21:00
9/6/04 8:00
Overflow peak flow rate
(cfs)
0.0157
39.0
119.0
31.0
110.0
26.5
95.7
Overflow volume
(MG)
0.001
3.185
7.487
2.638
6.772
2.216
5.837
Multiple factors affect the occurrence of overflow events; with rainfall intensity and antecedent condition
the dominant factors. Of the six rainfall events with the total depth greater than 1.4 in. (D-storm rainfall
depth), only three have peak intensities greater than 0.6 in./hr (D-storm peak intensity). Two of the three
storms produced overflow in all three scenarios.
Take for example, the 1.68 in. storm event for March 4, 2004, shown below as Figure 2-41. That storm
did not cause an overflow at the regulator storage facility. The event has a total rainfall depth of 1.68 in.,
but it had a peak intensity of only 0.18 in. per hour. The rainfall volume was well distributed over a
relatively longer time than the other storms presented in Table 2-19.
0 o
CM 0
o m
CO •<"
o
•sr
0 o
c\i °
O 03
CO •<"
o
0 o
CM 0
<2 co
CO
-------
The August 27, 2004, storm, shown below as Figure 2-42, also did not produce overflow. That event had
a peak intensity of 0.65 in./hr; however, when assessed in the broader context of average ET rate
(relatively high summer rate) and longer antecedent dry hours, the results suggest that GI was able to
provide a little more control benefit for the event than the Gray Only solution could provide. The
combined effect of higher ET and longer antecedent dry period resulted in (1) less overall runoff from the
watershed; and (2) better GI BMP performance eliminated the overflow.
Figure 2-42. Hyetograph for the August 27, 2004, storm event.
Another interesting event is the June 9 event shown below as Figure 2-43. That storm has the highest
total rainfall depth (2.09 in.); however, the peak intensity was only 0.52 in./hr (lower than the 0.6 in. peak
of the D-storm). The graph shows that storm had a pattern of smaller intensity rainfall for several hours
before the most intense peak. Although its peak intensity was less than the D-storm intensity, this storm
still produced a marginal overflow under the Gray Only scenario. A closer look at the storm reveals that
that 53 percent of the total volume (1.11 in.) fell gradually over the course of a full day before the arrival
of the 0.52 in./hr peak. The conditions on the ground were primed for an overflow, even with a peak that
was lower than the one associated with the design storm.
Figure 2-43. Hyetograph of the June 9, 2004, storm event.
2-56
-------
Figure 2-41 through Figure 2-43 demonstrate the important role of continuous simulation at capturing the
dynamic nature of storm-related phenomena. Observations like those gained from continuous simulation
investigation can paint a better overall picture of natural system and provide some meaningful
information to inform the decision-making process.
2.5.7. Comparison of Gray versus Green Overflow Reduction Effectiveness
Although gray storage has a much higher unit storage volume cost (as previously shown Table 2-19), the
total cost of gray solution for meeting the control target is lower than the cost of green alternatives
because more GI is needed to achieve the same level of gray performance. In other words, not all storage
is created equal. GI tends to reduce volume reduction from the bottom of the hydrograph, whereas the
supplemental gray storage directly treats the top of the hydrograph because of its physical location
immediately downstream of the regulator overflow. There are two other phenomena at play in GI worth
noting. First, small storms may saturate GI storage, depleting the available storage capacity during
consecutive storms. Second, large intense storms may fall at a rate that is higher than the infiltration rate
into GI, which also may tend to diminish their effectiveness. For these reasons centralized gray storage
facilities, if measured on the basis of number of overflows is more cost-effective in this case for
controlling overflow. Exploratory scenarios were simulated to further compare the green and gray
alternatives in reducing overflows. Figure 2-44 presents the number of overflows in a weather typical
year (2004) with various storage capacity provided by green versus gray facilities. The graph shows that
given the same storage capacity, the number of overflows associated with gray solutions was lower than
that of the green alternatives.
Green Infrastructure
Gray Storage
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
Storage (Million Gallon)
1.8
2.0
2.2
2.4
Figure 2-44. Comparison of CSO 069 number of overflows with green versus gray storage capacities.
Using the four storage volume points on the plot above, overflow volumes for both green and gray storage
were normalized to present an expected annual overflow volume reduction per unit volume storage
provided. These results are presented below as Figure 2-45. The plot suggests that on the basis of unit
storage volume provided, in this case the gray solution consistently outperforms the green storage
scenario. This trend also appears to increase with increasing storage volume.
2-57
-------
c
o
£ ">
-
~
il
c
c
60
50 -
40 -
30 -
20 -
10 -
0
Gray Storage Effectiveness
I Green Storage Effectiveness
0.26
2.17
0.65 1.20
Storage Volume (Million Gallons)
Figure 2-45. Comparison on annual overflow volume reduction per unit storage volume provided.
It is important to note that this analysis only considers physical volume of each treatment type, and not
the treatment volume. For GI, storage capacity is recovered by means of gravity outflow through
underdrains, infiltration, and ET, depending on the practice. For gray infrastructure, flow in the storage
basin is pumped out of the system to create more capacity. The rate of recovery associated with
infiltration and ET is lower than the pumping rate of gray infrastructure. Therefore, the potential
treatment volume of gray infrastructure per unit of actual volume is greater than that of GI.
2.5.8. Optimization Summary and Conclusions
The optimization analysis of BMP opportunity for CSO mitigation in the 069 watershed yielded some
interesting findings. The findings can be summarized in terms of (1) implications for planning and
management decisions; and (2) implications on modeling approach development and assumptions.
Optimization results had certain management and planning implications for the study area. First,
extrapolating the proposed design plan from the 100-acre pilot study site to the remainder of the 069
sewershed suggested that CSO mitigation objectives could not be achieved by only implementing GI on
public rights of way, as defined by the design plan—more is needed. Second, adding GI on private
parcels provided an additional 6 to 8 percent volume reduction. The most notable observation about GI
on private parcels was demonstrated by the slope of the associated cost-effectiveness curve. This curve
had a steep slope indicating that the additional benefit realized by including private property opportunities
came at lower cost relative to when GI was limited to public rights of way or if only gray infrastructure
was applied. It is important to note that the GI alternatives proposed for this study are urban retrofit
projects. For this reason, they are likely more expensive than new GI construction costs because of
significant overhead costs associated with site preparation, reconstruction of curbs and sidewalk system,
and the traffic control measures needed during construction. In addition, this project is a pilot based on an
approach not typically used in this area. The uncertainty and risk with constructing the GI in this area
2-58
-------
therefore likely includes a higher bid cost than if this were a regular practice within Kansas City. As a
result, GI on private parcels (or for that matter, GI in new development areas) would likely become less
costly as the technology and understanding matures. New construction GI costs would probably be
integrated into the overall construction and planning cost, making GI under those circumstances much
more cost-effective. Third, the results suggest that the combination of GI (as defined by the proposed
design plan) plus supplemental gray storage at the regulator costs more than implementing only a slightly
larger gray supplemental storage at the regulator. It is important to note that the management conclusions
should be interpreted in the context of the associated modeling assumptions, as further described below.
This study also evaluated the sensitivity of certain modeling assumptions and configurations on the
results. The three modeling elements evaluated and tested for sensitivity included (1) model simulation
time step; (2) antecedent moisture conditions; and (3) the use of a design storm for optimization versus a
continuous simulation. The first element evaluated was the influence of the model simulation time step
on the optimization cost-effectiveness curve. Two parallel runs were performed using 15 minute and 60
minute time steps. There were only slight differences in the resulting cost-effectiveness curves associated
with the two different simulation time steps. Because both runs generated differing baseline runoff
conditions, the resulting BMP performance and regulator response gave mixed results. Nevertheless,
because time of concentration estimates for the sewershed suggested that the travel time of peak flow was
less than one hour, the 15 minute time step was used as the basis for the remainder of the analysis. The
second element evaluated was the influence of antecedent moisture conditions on model predictions. The
D-storm was tested under dry, average, and wet soil moisture conditions. The model showed varied
responses under the three antecedent moisture conditions (i.e. drier conditions produced better
performance at a lower cost than wetter conditions); however, the relatively large size of the D-storm
tended to contain variability within a narrow range. The third element evaluated was the use of a design
storm to drive the optimization instead of an observed weather time series. The D-storm was used for
optimization because it was previously identified as the critical condition for CSO. In other words,
controlling the D-storm was expected to result in attainment of CSO mitigation objectives. To validate if
the BMP sizes derived from optimization would perform as designed, the resulting BMP configurations
were also run using continuous simulation for a weather typical year (2004). The test showed that the
recommended cost-effective solutions were able to reduce CSO from 33 overflows per year (under
baseline conditions) to fewer than the 6-overflow allowance (2 to 3 overflows). The typical year
validation run also revealed some interesting observations about the nature of overflows. One of the
overflow events was caused by a 2.09 in. storm, where 53 percent (1.11 in.) fell gradually over the course
of a 24 hour period, followed by a second burst of the remaining 0.98 in. over only 7 hours. Although the
rainfall sequence causing the overflow had a smaller volume and peak than the D-storm, its relatively
short duration, along with the fact that it occurred under saturated watershed conditions, resulted in an
overflow at the regulator storage facility. Conversely, another storm larger than the D-storm did not
cause an overflow because it occurred in July after a long dry antecedent period; therefore, the storm did
not yield as much runoff. The model sensitivity analyses provided some good insights and understanding
about factors that most influence CSOs.
This study has demonstrated that SUSTAIN can provide a versatile platform for (1) integrating
multidisciplinary data and methods, and (2) evaluating multiple competing factors toward achieving
stormwater and CSO management goals. The results demonstrate how the model can integrate modeling
and management assumptions to evaluate the implications on the complex behavior of GI and gray
infrastructure solutions. In the future additional analysis and expansion of model capabilities could be
used to explore other aspects of the management of CSOs in Kansas City. For example, the modeling
performed in this study limits GI in public rights of way as defined by the design plan. A broader
application of GI technologies could be evaluated to see how much additional benefit would be derived.
In addition the cost-benefit analyses could be performed with O&M costs in addition to the capital cost of
construction, contingencies, and design fees considered in this study. Evaluation of the long-term life-
2-59
-------
cycle costs could result in a different optimization result. Finally the SUSTAIN formulation could be
expanded to consider the other benefits of GI beyond the driving factors of overflow frequency and cost,
such as aesthetic improvement benefit, community educational opportunity, increased property value,
volume reduction of treatment plant inflow, carbon sequestration, possible reduction in heat island effects,
or other potential benefits. When GI benefits are being evaluated and quantified in a triple-bottom-line
context (environmental, social, economic), certain factors may be prioritized over others even though they
are not the lowest cost options available. For example, GI implementation and maintenance could
possibly be a way of creating employment opportunity for a municipality (e.g., green jobs), which can
ultimately contribute to sustaining a local economy. Furthermore, GI may also provide aesthetic appeal
to a community, which increases property values, which in turn, may increase tax revenue. If a portion of
the new tax revenue is directed towards O&M of GI facilities, the management cycle becomes self-
sustaining.
2-60
-------
Chapter 3. Case Study: Louisville, Kentucky
EPA's ORD conducted a pilot project demonstrating the use of GI for CSO control in Louisville,
Kentucky. The primary purpose of this case study was to demonstrate the tradeoffs between green and
gray infrastructure alternatives. This case study effort was designed to address three goals: (1) test the
sensitivity of key BMP hydrologic input parameters in SUSTAIN using local monitoring data to provide
guidance for model calibration; (2) demonstrate replication of an existing hydraulics model of storm drain
network and CSO regulator; and (3) characterize the cost-benefit relationship between a number of green
and gray infrastructure options for mitigating CSO. This case study also provides the SUSTAIN user
community with a demonstration of the application of the model.
The focus area for this case study is the Lousiville-Jefferson County Metropolitan Sewer District (MSB)
CSO 019 sewershed west of downtown Louisville, Kentucky, which is bounded by the Ohio River and
Interstate 1-64 to the north and 1-264 to the west (Figure 3-1). The sewershed drains 1,094 acres of mixed
land use dominated by single-family residential neighborhoods. A large rail yard operated by Norfolk
Southern Corporation is also adjacent to North 30th Street. The existing CSO 019 outfall is on the north
edge of the sewershed along North 34th Street between Rudd Avenue and 1-64. Overflows discharge
directly to the Ohio River. An Info Works hydraulic model estimated the sewershed produced 297.91 MG
of overflow volume as a result of 60 discrete overflow events based on the 2001 typical year precipitation
record (MSB, 2008). Later refinements to that model using recent monitoring data were provided by
Thomas Waters, from O'BRIEN & GERE, for this case study effort. The refined model resulted in the
output that estimated a fewer number of overflows for the base scenario.
One of the project goals at the onset of the effort was to test the sensitivity of BMP parameters during
calibration to monitored inflows and outflows from various BMP types that are built and monitored by
MSB at local demonstration projects. However, such data were not available for use in this effort. In lieu
of observed BMP monitoring, the sensitivity analysis tested and quantified the sensitivity of key input
against their impact on predicted BMP outflow volumes.
This chapter presents (1) a summary of background supporting information; (2) the case study goals; (3)
the methodology and findings of a BMP model sensitivity analysis for key input parameters; (4) a
discussion about the replication of the existing condition Info Works watershed model baseline in an
SWMM5 modeling environment and the subsequent derivation of an optimization baseline condition; and
finally (5) the BMP selection and placement optimization analysis of green and gray infrastructure
opportunities for CSO mitigation in the MSB service area.
3.1. Background
The MSB serves the Louisville metro area, which includes all of Jefferson County, with a 2009
population of roughly 722,000 (MSB, 2010). MSB serves a tributary area of approximately 385 square
miles of which the CSS area encompasses 37 square miles (approximately 10 percent of the total area).
The agency was established in 1946 and charged with the responsibility to manage the city's sanitary
sewer and drainage system. In 2004, the EPA filed legal enforcement actions against MSB. In August
2005, MSB entered into a consent decree with EPA and the Kentucky Environmental and Public
Protection Cabinet. The consent decree required compliance with the clean water act by the end of 2020
for CSOs, and 2024 for SSOs. The consent decree required the development of a Final CSO LTCP and
the Final Sanitary Sewer Bischarge Plan by Becember 31, 2008. These documents were incorporated
into an Integrated Overflow Abatement Plan (IOAP). The IOAP was finalized and incorporated into an
3-1
-------
amended consent decree in April 2009. The approved plan includes control of CSO discharges to levels
prescribed in the CSO Control Policy by December 31, 2020 (MSD, 2011).
Legend
Outfall
| | CSO 019Boundary
Com Dined Sewer ServiceArea
Louisville, Kentucky
CSO 019 Location
Figure 3-1. Location of CSO 019 sewershed.
3-2
-------
The Final CSO LTCP uses conventional infrastructure (gray) projects such as storage facilities, system
optimization projects including real-time controls to maximize in-line storage and ability to shift flow
within the system, and GI. Gray infrastructure projects were defined without considering the potential
beneficial performance of GI to achieve CSO control. At the time (2006 - 2007) GI facilities were
viewed by many regulators as unproven and the overflow reduction benefits not quantifiable over the
long-term. As a result, GI is incorporated into the I OAP such that MSB assumes the risk for the
effective performance of GI, and must demonstrate its effectiveness. If shown effective, Louisville will
have an opportunity to resize gray infrastructure on the basis of the documented benefits of the GI. To
promote the use of GI to achieve necessary reductions, MSB committed to spending approximately $6
million per year for the first six years of LTCP implementation, followed by an allocation of $ 1 million
per year for the nine subsequent years, for a total GI budget of $47 million (MSB, 2009). GI
implementation would increase if additional implementation of GI was demonstrated to be sufficient to
replace or downsize gray infrastructure cost-effectively.
3.1.1. InfoWorks Model
MSB's initiatives to develop and refine both separate sanitary and CSS models date back to the early
1990s. Most recently, MSB developed a baseline runoff and hydraulic model of its combined sewer area
using the InfoWorks modeling platform, a proprietary watershed and hydraulics modeling system.
Output from the originally developed version of the CSO 019 InfoWorks watershed model was initially
provided as the baseline model for this case study. However, when a revised version became available,
the model baseline was revised to reflect those changes. The latter model ultimately became the version
that was integrated with SUSTAIN for this case study effort.
The IOAP model, developed in 2008, is a one-year continuous simulation performed using the 2001
selected typical precipitation year. The model predicted 60 annual overflow events at the outfall to the
Ohio River and was the basis for initial conceptual designs of the Portland Wharf Storage Basin (Section
3.1.2). No overflow monitoring data were available at the time when this model was calibrated.
An update to the 2008 IOAP model calibration was being developed concurrent with this case study in
2011. However, the advantage of the latter model revision over the former was that it was based on
recently collected monitoring data that captured overflow events between January and June of 2010.
Modifications were made to the outfall configuration and some subcatchment properties. The updated
model calibration was rerun using the same typical year 2001 precipitation time series, and predicted a
decrease in the number of overflow events relative to the original version.
3.1.2. Portland Wharf Storage Basin
MSB commissioned conceptual designs and capital funding for constructing the Portland Wharf Storage
Basin to reduce the number and accumulative volume of CSOs from the CSO 019 sewershed. That basin
was envisioned as a 6.37 MG concrete storage basin and pump station. The storage basin was expected to
reduce the annual average overflow volume to 52 MG resulting with a CSO target of eight overflow
events per year using the 2001 rainfall time series. The proposed location of the storage basin was just
north of the CSO 019 regulator between 1-64 and the Ohio River, as shown in Figure 3-2. In 2008
dollars, the expected capital cost of the project was estimated to about $20 million.
3-3
-------
Volume 2 - Final CSO Long-Term Control Plan
Ohio River
SolmlonlD*LJ>R_MF_0(9_S_l3_B_A_8
Portland Wharf Storage Basin
Preliminary • For Budget Development Only
Legend
Proposed Flow Control Solu
Pn^xj&ed Pump Stetwn Soi
' -Underground, Covered. Off-line Storage Basin - ' '
^ Proposed Pipe SokjKm
•F Fore* Main
-*. Coilectof f 12"
--- fnieraeplof -> 12"
•-» Combined Sew Pipe
— Flood Wall
Proposed Storage Stfu
p~] Floxway
Metro Parks
General representation of
overflow abatement solutions
are for preliminary planning
purposes. Alignments and
locations may be altered
during design.
m IM MSP
sod
Figure 3-2. Proposed location of the Portland Wharf Storage Basin and Pump Station.
3.2. Overview of Case Study Goals
The three goals for this case study are to (1) demonstrate applying SUSTAINto replicate an existing
Info Works hydraulics model of a storm drain network and CSO regulator; (2) perform a BMP modeling
analysis to test and quantify the sensitivity of key input parameters versus their impact on predicted BMP
outflow volumes; and (3) characterize the cost-benefit relationship between a number of green and gray
infrastructure options for mitigating CSO. These goals were first defined at the onset of the effort; but
they were further refined during the model setup, application, optimization, and results interpretation
process. Throughout this chapter, a strong emphasis was placed on describing specific aspects of the
SUSTAIN application process, and relating it back to the case study objectives. The following sections
further elaborate on each of the three case study goals.
3.2.1. Replication of an Existing Hydraulics Model
The MSB has invested in an Info Works runoff and hydraulic model to characterize the dynamics of its
combined sewer area. The purpose of the Info Works model is to characterize the detailed connectivity of
the drainage network and represent existing conditions as closely as practical, whereas for SUSTAIN, the
primary modeling objective is to characterize the critical condition associated with the baseline model as
closely as possible, while making every effort to gain computational efficiency wherever possible. To
provide consistency between SUSTAIN solutions and the Info Works baseline model, it was important to
ensure that the SUSTAIN replica of the network and regulator was indeed representative of the existing
InfoWorks model, especially for CSO critical conditions.
3-4
-------
The watershed model baseline represents the existing condition rainfall-runoff response. It characterizes
the nature of the current physical system before any additional management activities are implemented. It
also represents the baseline from which any stormwater improvement will be measured, and the starting
point for BMP selection and placement optimization. Because it forms the basis for comparative
assessment of alternatives, establishing a representative baseline condition with confidence is a critical
first step in any modeling effort. It becomes especially important where cost-benefit optimization of
future management objectives is a primary focus of the modeling effort.
This application also provides SUSTAIN users with an example of how scale and resolution affect the
model results, efficiency, and accuracy. This case study application also examines the use of a simplified
routing network that preserves the essential features of the system response, while improving
computational efficiency. SUSTAIN provides the user with a range of options for handling spatial scale
and resolution. For smaller watershed or study areas, it is both feasible and practical to use a more
detailed or articulated routing network, meaning that smaller pipe networks, specific BMPs and
associated drainage areas, are explicitly defined. For larger-scale applications, using a fully articulated
approach can becomes cumbersome, impractical, and resource intensive because of the size and
complexity of the associated network. To provide an alternative, simplified approach, SUSTAIN provides
an aggregated BMP option that reduces the complexity of drainage network while preserving the
dominant physical basis of the BMP performance. Although the aggregate BMP approach significantly
reduces the network complexity, it also sacrifices some details of the model network and routing. This
case study tests the performance of a simplified aggregated approach versus a higher resolution routing
network. Three natural questions arise:
• How much network simplification can be introduced without significantly compromising model
accuracy or precision?
• What components of a fully articulated drainage network are appropriate candidates for
aggregation, and to what degree can they be aggregated?
• How much computational advantage does aggregation provide?
Because MSB has accepted the Info Works model calibration as representative of existing conditions, this
case study evaluates the ability of SUSTAIN to replicate the calibrated Info Works model. In testing
model performance and accuracy, SUSTAIN's ability to mirror the Info Works model response was
evaluated. The Info Works application used a higher resolution network, while the SUSTAIN replica used
an aggregated, computationally streamlined drainage network. Successful model replication was
measured by (1) evaluating the percent difference between the SUSTAIN and InfoWorks model results;
and (2) computing the efficiency of the streamlined SUSTAIN network at replicating results generated by
the higher resolution InfoWorks network, especially for critical conditions associated with CSO.
3.2.2. BMP Pa ram eter Sensitivity An a lysis
An initial objective of this case study was to validate the BMP performance in SUSTAIN using
monitoring data collected at local demonstration projects. However, monitoring data were not available
at the time of this case study; therefore, that objective was later refined to test the relative sensitivity and
response of key BMP calibration parameters in SUSTAIN.
Of the various BMPs considered for use in the study area, bioretention cells are the practice that provides
the most flexibility in how it is represented. The sensitivity analysis used a factorial experimental design
approach to test various hydrologic parameters in a single bioretention cell. Three input parameters were
varied. They included (1) the vegetation-dependent multiplier for estimating ET; (2) the Horton saturated
infiltration rate of the bioretention cell media; and (3) the Horton maximum infiltration rate of the
bioretention cell media. High and low values for each of these three parameters were applied to show the
3-5
-------
range of influence on predicted BMP outflow volume, for eight different scenario combinations. The
factorial experiment design approach applied in this context, together with the analysis results, provide
examples of how to manage predictive uncertainty associated with BMP model parameterization in
SUSTAIN.
3.2.3. Cost-benefit Relationship between Gray and Green Infrastructure for Mitigating
CSOs
The third case study objective is to investigate cost-benefit relationships between green and gray
infrastructure for mitigation of overflows in the CSO 019 sewershed in light of the planned Portland
Wharf Storage Basin. The objective builds on outcomes from the previous objectives by using the
calibrated watershed optimization baseline model as the basis. The analysis considers a series of
individual green and gray infrastructure implementation scenarios to mitigate CSOs in the watershed, as
well as different combinations of integrated green and gray solutions. The analysis will hinge on
integrating five different implementation scenarios:
1. Use of gray infrastructure only, specifically the Portland Wharf Storage Basin. The tank volume
is set as the optimization decision variable, and cost data are derived from Louisville's cost versus
size relationships provided by MSB in a spreadsheet format;
2. Use of downspout disconnections only. Downspout disconnection has been identified as a
relatively low cost way of reducing stormwater runoff. This option evaluates the cost-benefit
impact of fully implementing a downspout disconnection incentive program throughout the
sewershed. It is assumed that full adoption of downspout disconnection occurs before additional
structural GI measures are adopted;
3. Use of GI only. This scenario reflects management by GI only. It explores the cost-benefit
impact of GI beyond full implementation of the downspout disconnection program;
4. Use of gray infrastructure in combination with downspout disconnections; and
5. Use of gray infrastructure, downspout disconnections, and GI in combination. This scenario
reflects a full build-out condition of GI with supplemental gray infrastructure required to meet
increasing reduction intervals.
The results from the first three scenarios are evaluated independently to determine the cost-effectiveness
of each practice in achieving CSO mitigation objectives. The last two scenarios evaluate a combination
of green and gray solutions. The optimization objective is to identify the point at which CSO mitigation
objectives will be achieved at the lowest cost. Sensitivity testing of both cost and sizing assumptions will
be conducted to provide ranges of predicted management outcomes. The cost effectiveness curve will
also be evaluated to show percent utilization of each practice at each solution. GI utilization results will
also be mapped by subwatershed to gain insight into the optimal spatial placement of these practices
derived under the defined objective and constraints.
3.3. Replication of an Existing Hydraulics Model
In SUSTAIN, modeled stormwater runoff is the forcing function that drives BMP simulation. Watershed
models use site-specific spatial and temporal elements to characterize the rainfall runoff response. The
watershed model runoff time series represent the existing condition (or baseline), which serves as the
reference point from which stormwater management effectiveness will be measured. A critical first step
of a SUSTAIN application establishes or confirm a representative baseline condition with a high degree of
confidence in its applicability. The baseline becomes especially important in the context of cost-benefit
optimization of future management objectives, because the model baseline is foundational to results
interpretation and resulting conclusions. The watershed model baseline condition must represent
3-6
-------
variability throughout the watershed, including the influence of physical features associated with both
surface and subsurface behavior.
MSB developed a baseline runoff and hydraulic model of its combined sewer area using the proprietary
modeling platform Info Works. Although it is based on same underlying equations as SWMM, Info Works
uses a different approach to solve the equations. Info Works has the ability to export the input
configuration file to an SWMM5 compatible file format. The portion of the Info Works model
representing the CSO 019 sewershed was exported to SWMM5 format and reviewed for key hydrologic
parameters. This section summarizes the findings of the model reviews. Two model calibrations were
available for review: (1) the original 2008 IOAP model; and (2) the refined 2011 model which was
recalibrated using observed monitoring data collected from January through June of 2010. The model
review revealed an inter-basin transfer occasionally occurs between CSO 019 and adjacent CSO 190 for
interceptor relief. Because this effort focuses on management objectives within CSO 019, it was
necessary to isolate only runoff contributed from CSO 019 as the baseline for subsequent optimization.
The Info Works model represents a combined sewer area with 203 subcatchments and 647 pipes or
connections as shown in Figure 3-3. The model uses a kinematic wave routing method and does not
account for backwater flow. Each subcatchment can vary slope, depression storage, overland flow
coefficients, widths, and impervious cover. The Info Works model was used to inform selected
parameters for SUSTAIN, including roughness coefficient, pervious depression storage, and infiltration.
The subcatchment and routing network were also simplified for SUSTAIN from the original Info Works
configuration to reduce run-time while preserving model responsiveness, which is demonstrated in
section 3.3.5.
*
*BO 7 & B-.70QO T
* * Si
m*~''flt? ' » '- »~* »T S -.''I-
S^^^^fcA,
a^4>%t C*v:J-.--:^x * 'N
'Ml^rj^^S^-^
tj J^i^fe^
/ ;-]:.'V "•'. ' -.f •i-,*«iV:-T%
:^;\,
JtV.iaPHat«.arT«g1014
Figure 3-3. InfoWorks model configuration exported to EPA-SWMM5.
3-7
-------
SUSTAIN provides the user an option to link to an existing sewershed model using unit-area (one acre)
runoff time series for each land unit or hydrologic response unit (HRU) for representing land rainfall-
runoff responses as boundary conditions. When linking to an existing watershed model, SUSTAIN uses
unit-area runoff time series files as input and associates those with the land cover distribution present
within the delineated drainage area boundaries to drive the routing and BMP simulations. A GIS
representation of the unique land use types serves as the physical link that SUSTAIN uses to tabulate area
distributions within each catchment.
Other spatial characteristics of the baseline model representation were considered. For this application,
there was a desire to simplify the size and complexity of the network, within reason, in a way that
minimized distortion of system behavior and response. The level of model detail was selected to match
the required response and purpose of the application and management questions under consideration. For
the purposes of watershed optimization modeling, reduction of computational time results in a more
efficient optimization process and the ability to explore a wider range of management alternatives.
This section describes the steps taken to develop a baseline watershed model condition. Those steps
include (1) land cover development; (2) subcatchment delineation; and (3) model calibration. The
following sections describe each of those steps in greater detail.
3.3.1. Land Cover Development
In a watershed model, land unit representation must be sensitive to the features of the landscape that most
affect hydrology, including surface cover, soil type, and slope. Experiences have shown these three
watershed features have the most impact on hydrology. In urban areas, it is important to estimate the
division of land use into pervious and impervious components. Because the focus of this study is volume
control, it is not necessary to further subdivide land use beyond pervious and impervious cover; however,
rooftop areas were distinguished from other impervious areas to facilitate rerouting flow from
downspouts as a management alternative. Slope might also be an important factor in some areas. In a
commonly used watershed model land unit characterization approach, the unique combination of land
cover, soil type, and slope form HRU. This section looks at each of these three components in an effort to
characterize an appropriate basis for representing runoff boundary conditions.
Soil Type
The available soil survey GIS information (NRCS, USDA, 2006) suggested that soil type was fairly
homogenous throughout the study area. There was no extensive soil infiltration testing data available to
either refine or refute the validity of the GIS soil surveys. When soil hydrologic groups are not
homogenous in a watershed, further subdividing pervious land cover according to soil hydrologic group
can provide improved resolution. However, for this application, soil type was not used as a distinguishing
land element.
Slope Analysis
Slope can play an important role in watershed modeling because it controls the magnitude and, to a lesser
degree, the timing of peak flows. GIS coverage of slope in the CSO 019 sewershed was derived from a
data set of 2 ft elevation contours (LOJIC, 2003) and is presented as Figure 3-4. Slopes in the watershed
are generally less than 1 percent. Areas of high slope tend to closely trace and highlight building features
and highway embankments. While those slopes appear prominent, they are more associated with
structural features within the watershed rather than the actual topographic configuration of the watershed.
Further review of the existing Info Works model configuration also suggested that slope does not vary
greatly within the CSO 019 sewershed. The 203 subcatchments defined in the Info Works model were
3-8
-------
delineated in GIS and joined with a table of their physical properties, including slope, width, and
depression storage. A spatial distribution of subcatchment slope as defined in the Info Works model is
presented as Figure 3-5. The map confirms that slope does not vary widely in the watershed. The slope
of most subcatchments is less than 0.15 percent.
Surface Cover Analysis
This analysis used data sets in GIS format for roads, impervious surfaces, and building rooftops (LOJIC,
2003). The roads layer contained the footprint of the road rights of way. The impervious surfaces layer
included sidewalks, driveways, parking lots, alleyways, and other distributed impervious surfaces. The
building footprint layer was used to represent rooftop area in the watershed. Those three layers were
merged into a single raster representation, with rooftops distinguished from other types of impervious
cover. The void space between impervious features was defined as pervious area. A map showing the
distribution of surface cover types for the CSO 019 sewershed is presented below as Figure 3-6. That
overlay resulted in a distribution of three unique combinations of HRUs that capture both the physical
texture of the watershed.
In summary, on the basis of these analyses, it was determined that not enough variability exists to warrant
additional spatial resolution by explicitly incorporating either soil type or slope into the delineation of the
subcatchements. Instead, only the three land cover types presented in Figure 3-6 were used to represent
the physical texture of the watershed surface.
3.3.2. Subcatchment Delineation
The original Info Works model configuration for the CSO 019 divided the sewershed into 203
subcatchments with areas ranging from 0.61 acre to 53.86 acres. For lumped parameter models such as
SWMM, having more subcatchments provides more latitude for creating a spatially variable response. In
other words, a higher resolution better approximates a distributed parameter response. However,
increasing the number of subcatchments and routing connections also increases the complexity and run-
time for a single model run. By making land cover the smallest modeling unit, some of the heterogeneity
of the system is transferred from the catchment into the land cover distribution. As a result, the catchment
resolution, and the number of network connections, can be judiciously aggregated with acceptable losses
of the spatial variability of the runoff response.
The 203 subcatchments defined in Info Works were aggregated into 20 subwatersheds for model
calibration on the basis of the larger delineation provided by MSB. The number of modeled pipe
segments was also reduced from 647 in the Info Works model to 24 in the SUSTAIN model. Figure 3-7
compares the original Info Works subcatchment boundaries with the subwatershed boundaries used in
SUSTAIN. After aggregating some of the subcatchments, it was appropriate to only explicitly model
pipes greater or equal to 3 ft in diameter, because 3 ft is the smallest pipe size connecting aggregated
subcatchments to the pipe network. In two instances, pipes smaller than 3 ft (2 ft and 2.5 ft) diameter
were modeled to complete necessary routing connections. Although this section describes how the model
was spatially reconfigured for model calibration purposes, Section 3.3.5 evaluates the larger implications
associated with model spatial resolution, simulation time, and predictive precision.
3-9
-------
Legend
"5^ Outfall
Percent
,
• >
'•
Louisville, Kentucky
CSO 019 Slope Analysis
NADJ 56 3_StBtePtare_Ken1u di j_N uth_FI PSJE 31 Jeet
Figure 3-4. CSO 019 sewershed slope derived from topographic contours.
3-10
-------
Legend
Infoworks Subcatch merit
Infoworks Slope
015% -0.52%
0.53% -1.1 4%
1.15% -2.5%
> 2.5%
Louisville, Kentucky
CSO 019 Infoworks Slope Analysis
Figure 3-5. CSO 019 sewershed InfoWorks model slope analysis.
3-11
-------
Impervious Cover
Pervious Cover
Roof
Louisville, Kentucky
CSO 019 Sewershed Surface Cover
Figure 3-6. CSO 019 sewershed surface cover distribution.
3-12
-------
SUSTAIN Subwatershed
| | InfoWorks Subcatchment
Louisville, Kentucky
CSO 019 InfoWorks Subcatchments
HM>_1SS2_St3t=Fla-5_Kaitucky_Norlh_FIF3Sje!)1JeEl
Mac crnaLosa 06-20-2011
Figure 3-7. Comparison of InfoWorks and SUSTAIN subwatershed delineations.
3-13
-------
3.3.3. Review of Baseline Model Calibrations
Buring the model calibration process, parameters are expressed uniquely for each land cover type. The
objective of the calibration process is to identify a unique set of parameters that remain constant for all
instances of that land cover in the study area, such that the spatial variation of the sewershed response
becomes a function of only the land cover distribution in each subarea. Parameters from the MSB 2011
InfoWorks modeling effort were used as source for SUSTAIN model parameter values.
For this effort, the calibration objective was to characterize model performance for the typical Louisville
precipitation year 2001. Figure 3-8 shows the monthly precipitation distribution for the typical year 2001,
also shows the monthly number of overflow events out of the eight largest storm events. MSB
established the typical precipitation year through a statistical analysis of the historical rainfall record from
1949 through 2002 (MSB, 2007). The 2001 precipitation year rainfall record consists of 62 storm events
ranging in depth from 0.1 in. to 3.15 in. assuming 12 hr inter-event time and a minimum storm size of 0.1
in.
8
7
'•=• 6
Typical Year Precipitation
8 Largest Overflow Events
7 i/)
7 =
0 tt
01
S?
5 ™
S2.
4 1
•E
3 g
O
M-
o
2 S
_Q
E
Jan
Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Figure 3-8. Distribution of typical year 2001 precipitation data by month.
Calibration parameters were adjusted during the process until an acceptable match of benchmark
calibration metrics was achieved. The calibration process is elaborated later in this section. Some of the
key parameters were those associated with (1) depression storage and overland flow; (2) infiltration; and
(3) BCIA. The earlier parts of this section describes those three general aspects of model
parameterization and time series generation, while the later summarizes model testing, output
summarization, time series comparisons, and calculating calibration indicator metrics.
Depression Storage and Overland Flow
Bepression storage describes the depth of storage available for surface ponding. The subcatchment
roughness coefficient describes Manning's N for overland flow. Both the roughness coefficient and
depression storage parameters are set independently for pervious and impervious areas. The values of the
parameters were held constant between the 2008 and 2011 model calibrations across all subcatchments.
The values are presented blow in Table 3-1.
3-14
-------
Table 3-1. Roughness and depression storage parameters for pervious land cover
Land cover type
Pervious areas
Parameter
Roughness coefficient (unitless)
Depression storage (in.)
2008
Info Works
0.2
0.2
2011
Info Works
0.2
0.2
Depression storage and roughness coefficient for impervious land cover varied by subcatchment for both
the 2008 and 2011 Info Works model calibrations. Table 3-2 presents a comparison of the area-weighted
average Info Works model parameters for the 2008 and 2011 CSO 019 model calibration.
Table 3-2. Area-weighted average InfoWorks model parameters for CSO 019
inno
Land cover type
Impervious areas
Other
Parameter
Roughness coefficient (unitless)
Depression storage (in.)
Slope (%)
Width (ft)
Percent zero (%)
2008
InfoWorks
0.013
0.059
0.230%
55.050
40.960%
2011
InfoWorks
0.013
0.058
0.200%
62.910
42.070%
Values for the impervious cover type in SUSTAINwere estimated by calculating the area-weighted
average of the values in the InfoWorks model using impervious area as the weighting factor. The same
approach was also applied to estimate values for slope, width, and percent of impervious cover with zero
depression storage. To capture spatial resolution in the sewershed, a set of impervious time series was
developed for each subwatershed rather than applying a single impervious time series uniformly across
the basin using area-weighted parameter values. Catchments from the InfoWorks model were grouped by
SUSTAIN subwatershed (Figure 3-7). A set of impervious cover parameters was then calculated by
subwatershed using an impervious area-weighted average of the parameters for catchments grouped in
that sewershed. The ranges of the area-weighted parameters by subwatershed are presented below as
Table 3-3.
Table 3-3. Area-weighted subwatershed parameter ranges applied in SUSTAIN
Parameter
Depression storage (in.)
Slope (%)
Width (ft)
Percent zero (%)
Minimum
value
0.04
0.00%
37.82
1.94%
Median
value
0.05
0.08%
60.15
45.30%
Maximum
value
0.10
1.20%
99.64
55.19%
Infiltration
The Horton infiltration method is an empirically based model parameterized by specifying an initial
(maximum) infiltration rate and a final, saturated infiltration rate. The model assumes that infiltration
begins at a constant, maximum rate that decreases exponentially over time. The shape of the curve as the
infiltration rate changes from initial to final is controlled by a decay rate specific to the type of soil
(USEPA, 1998). The relationship is commonly presented as
3-15
-------
where ft is the infiltration rate at time t,f0 is the initial maximum infiltration rate,/c is the saturated
infiltration rate, and k is the decay constant.
The MSB modeling guidelines document for hydraulic and hydrologic modeling provides suggested
Horton infiltration values on the basis of hydrologic soil type (MSB, 2007). Parameter value consistent
with Type B soils were set for all subcatchments in Info Works and were left unchanged in the SUSTAIN
model configuration. The Horton infiltration values used are presented in Table 3-4.
Table 3-4. Horton infiltration parameters from Louisville Infoworks models
Parameter
Maximum infiltration rate (in./hr)
Saturated infiltration rate (in./hr)
Infiltration decay rate (hr"1)
2008
IOAP model
3.00
0.30
2.00
2011
calibration
3.00
0.30
4.14
Evaporation
While evaporation is often considered negligible for single storm or design storm events, it is an
important part of the annual water balance when performing long-term, continuous simulation modeling.
MSB developed a distribution of constant daily evaporation rates by month for use in hydraulic and
hydrologic modeling applications in the county (MSB, 2007). The distribution of evaporation rates by
month is presented below in Figure 3-9.
ro
0.14
0.12
o.io
2 o.os
re
tc
.1 0.06
re
+j
re
! 0.04
re
LLJ
0.02
0.00
Jan Feb Mar Apr May Jim Jul Aug Sep Oct Nov Dec
Figure 3-9. Monthly distribution of typical daily evaporation rates.
Dry-Weather Flow
Bry-weather flow is defined as flow through the sewer network when there is no precipitation. While
dry-weather flows often show little or no effect on storm peaks, they can account for a sizable percentage
of flow volume and use a large portion of the system capacity. Bry-weather flows are also described
3-16
-------
using monthly, daily, or hourly diurnals that represent patterns of system water use. Info Works
distributes dry-weather flow from each catchment using population data and a per capita flow rate.
The representation of dry-weather flow in SUSTAIN was developed using the population data and per
capita flow rates used in the Info Works model to calculate daily dry-weather flow from each
subcatchment. The daily dry-weather flows were added to calculate a total daily dry-weather flow
volume for the CSO 019 sewershed. Because SUSTAIN requires time series in a unit-area format, the
total daily dry-weather flow volume was divided by the total acres in the CSO 019 sewershed.
Stormwater Runoff Calibration
A set of unit area (one acre) time series was generated using the EPA SWMM5 modeling platform to
represent the 20 unique impervious series and 1 representative pervious land use for each subwatershed in
the CSO 019 sewershed. The runoff time series were then used to drive the routing and BMP simulation
in SUSTAIN. The pervious land cover parameters were previously shown in Table 3-1. Impervious land
cover parameters, summarized in Table 3-3, varied by subwatershed. They were computed as area-
weighted composites from the individual Info Works subwatersheds contained within the 20 aggregated
SUSTAIN subwatershed boundaries. The SWMM model calculated runoff time series using hourly
precipitation and evaporation time series for representative year 2001. Figure 3-10 conceptually
illustrates the data flow sequence for both Info Works baseline model (1st Pass) and SUSTAIN model
configuration (2nd Pass). The key difference between the two passes is how the diffuse runoff losses are
represented in the model. The 1st pass, i.e., InfoWorks baseline model, represents the diffuse runoff
losses by reducing the impervious area, and the 2nd pass, i.e., the calibrated SUSTAIN model, represent the
losses using the regression relationship described later in this section.
Although the InfoWorks and SWMM5 have differences in terms of their specific computational methods,
hydrographs generated by the models should be comparable for the same conditions and model
configurations. As previously noted, an SWMM5 export of the InfoWorks model configuration has
served as the only available documentation of model parameters for the recently updated calibration.
Review of the model files indicated that runoff volumes were calibrated in part by adjusting the
impervious area footprint in the model. While the impervious surfaces identified in Figure 3-6 are the
actual impervious footprint, it commonly recognized that not all of the impervious runoff reaches the
regulator (or even the collection system inlets). As illustrated in Figure 3-10, there are diffuse runoff
losses throughout the system from the time precipitation hits an impervious surface to the time resulting
runoff reaches the regulator. Examples of these losses include things like disconnected imperviousness,
surface ponding or flooding, or even pipe exfiltration.
Although modeling effective impervious area is sufficient for calibrating a model baseline for InfoWorks,
the SUSTAIN baseline configuration needs to explicitly account for all of the water in the system. This is
because BMP selection and placement involves physical changes to the landscape of the model. For
example, BMPs are often designed on the basis of the contributing impervious drainage area footprint.
Two SWMM models were run to reconcile the difference. Each model used identical parameters, but
varied the impervious area distribution. The first model used the calibration-adjusted effective impervious
area from InfoWorks, while the second used the physical impervious footprints as characterized in the
land cover GIS layer (LOJIC, 2003). The difference in runoff between the two runs represents diffuse
runoff losses upstream of the regulator overflow. To quantify the flow difference between the two
models, a regression relationship between the two runs was developed, as presented in Figure 3-11.
3-17
-------
Actual Impervious Cover (From GIS)
SUSTAIN:
Impervious Area Used
for BMP Selection,
Sizing,and Placement
Pervious
Impervious
1s'
Pass
Diffuse Runoff Losses (due to):
• Disconnected Imperviousness
•Surface Pondingor Flooding
• Pipe Exfiltration
•Other Unknown
BMP Scenarios Preserve
Runoff Losses from
BaselineCalibration
Baseline Model Calibration
Effective Impervious Area (InfoWorks)
Calibrated Response
based on Observed
Ove rf I ow Ti mese ri es
Figure 3-10. Conceptual data flow sequence for baseline calibration and BMP scenario model runs.
500
o
— 400
O
• Modeled Flow
One-to-OneLine
Linear(Modeled Flow)
0 100 200 300 400
Actual Impervious Flow (cfs)
Figure 3-11. Calibrated versus actual impervious flow.
500
3-18
-------
The plot shows a strong linear trend between inflows from the two model representations. It also shows
that only about 31 percent of the actual impervious area runoff reaches the regulator. This suggests that it
is reasonable to simply remove a fixed percentage of the water from the system before it reaches the
regulator. This approach provided a consistent way to quantify the observed relationship associated
diffuse losses during baseline calibration and preserve the response to the BMP scenarios.
The runoff-loss-adjustment described above was applied to the actual impervious area model
configuration, as shown in Figure 3-11, to represent the SUSTAINbaseline calibration. This
configuration was compared against exported time series from the Info Works model calibration. A one-
to-one plot of SUSTAINversus Info Works regulator inflow volume and peak flow (for discrete storm
events) was used to test the quality of the baseline model replication. These plots are shown as Figure
3-12. A perfect replication of the Info Works baseline would plot along the dotted one-to-one line. Figure
3-12 show a strong fit between the SUSTAIN and Info Works calibrations for both total inflow volume and
peak flow. The regression lines report R2 values of 0.999 for total inflow and 0.990 for peak flow. The
value of R2 varies between 0 and 1, where higher values indicate a stronger correlation.
250
y=1.025x +0.914
R2 = 0.999
150
2" 125
u
I 100
y=1.066x +0.652
R2= 0.990
0 50 100 150 200 250 o 25 50 75 100 125 150
InfoWorks Inflow Volume (MGal) InfoWorks Peak Flow (cfs)
Figure 3-12. InfoWorks versus SUSTAIN modeled inflow volume and overflow peak.
Because CSO are mostly associated with larger storm events, it was important to further investigate the
goodness of fit, i.e., percent difference in flow volume, peak, and timing—as measured by Nash and
Sutcliffe (1970), for larger storms using additional validation metrics. First, the 62 discrete storm events
were also categorized into ten percentile-bin intervals (two bins have seven storms, and eight bins have
six storms) to discern the goodness of fit metrics variability by storm size. Second, the percent error was
calculated between the SUSTAIN and InfoWorks baseline models. Percent difference between the two
model configurations was calculated for both volume and peak as follows:
Percent Difference =
(SUSTAIN - InfoWorks)
InfoWorks
where the InfoWorks result served as the replication target. Positive values of a given metric indicated
that the 5£/$TAZ7Vbaseline over predicted the InfoWorks result, whereas negative values indicated that it
under predicted the InfoWorks result. The range of computed metrics within each percentile bin were
summarized and presented as the box-and-whiskers graph shown in Figure 3-13. As shown in Figure
3-12, the correlations between InfoWorks and 5£/$TAZ7Vmodeled inflow and overflow peak have positive
intercepts. The positive intercept indicates there is a constant difference, which explains the trend that as
3-19
-------
storm size increases, the percent difference between the SUSTAIN and Info Works configurations
decreases for both flow peak and volume. This trend is more visible with the percent difference
comparison as shown in Figure 3-13. The effects of model aggregation become evident in the SUSTAIN
configuration because of the tendency to over predict volumes and intensities. For example, model
aggregation reduces time of concentration and lowers routing precision, resulting higher peak flow and
less opportunity for water losses. Nevertheless, percent differences decreases with increasing storm size,
and is the lowest for the largest storms, which are more critical for accurately replicating because they are
the ones associated with overflows.
IS] UU/0
O
c
I 40%
3 30%
i£.
S 20%
c
"o
Q-
_»
"o
t
•
^
03
OJ
c
L
OJ
E
^
0
•
i
_
• 1
^
03
OJ
Q-
OJ
E
^
0
s
.
±
Kj
-:
03
OJ
Q.
OJ
E
^
0
_v^
03
OJ
Q-
OJ
E
^
0
Overpred
.1
i
/rts
i
1
^
03
OJ
Q-
Underpredicts
-------
calculations. The log-space Nash-Sutcliffe is another way to measure model performance for high flow
events because it emphasizes model prediction of the peak flows.
OJ
'o
OJ
o
^
I
o
4-J
LO
If]
03
10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-100
Percentile Magnitude of Storm Size
Figure 3-14. Plot of Nash-Sutcliffe by storm size for SUSTAIN versus InfoWorks regulator inflows.
Once again, Figure 3-14 confirms the goodness of fit across a range of hydrologic conditions. It suggests
that the efficiency of the SUSTAIN model representation for matching InfoWorks model time series is
generally at or above 0.9 for storm events above the 30th and 100th percentile. Model replication
efficiency begins to degrade rapidly for storm events below the 30th percentile as evident from the lower
average values and higher variability. Nevertheless, these smaller events are well below the target
containment values for optimization. Finally, Table 3-5 presents a summary of model performance as
defined by the selected calibration metrics for the 8 largest calibrated storm events on record.
Table 3-5. Model calibration performace metrics for eight largest storms events causing overflow
Start date
5/7/01 14:00
9/9/01 10:00
10/5/01 13:00
10/11/01 5:00
10/13/01 6:00
10/23/01 14:00
11/28/01 4:00
12/16/01 8:00
End date
5/8/01 14:00
9/10/01 9:00
10/6/01 10:00
10/12/01 18:00
10/14/01 21:00
10/25/01 8:00
11/30/01 13:00
12/18/01 8:00
Precipitation
(in.)
2.03
1.57
1.43
1.58
1.60
1.56
3.13
2.11
Volume
(percent
difference)
4%
7%
10%
3%
-1%
3%
2%
2%
Peak
(percent
difference)
5%
2%
6%
7%
2%
0%
2%
7%
Nash-
Sutcliffe £
0.97
0.97
0.97
0.98
0.97
0.95
0.97
0.99
3-21
-------
3.3.4. CSO 019 Regulator Calibration
After characterizing the stormwater inflow boundary condition, the next objective of the baseline
calibration was replicating the Info Works regulator response in a SUSTAIN environment. As previously
noted, 62 storms were evaluated during watershed calibration for the 2001 precipitation year. Properly
characterizing what happens at the regulator junction is as important as characterizing the stormwater
inflow boundary condition because combined stormwater and dry-weather sewage sometimes result in a
regulator overflow. Figure 3-15 is a conceptual schematic of the CSO regulator activity. As was done for
the stormwater inflow validation, similar metrics were used to validate the quality of the regulator
replication.
c;
'ombined Sewer Overflow
Dry-Weather
Sewage Inflow
CSO
Regulator
Outflow to the
'reatment Plant
Figure 3-15. Conceptual schematic for the CSO Regulator.
The physical outfall configuration for the CSO 019 sewershed consists of a weir located at the end of an
11.5 ft diameter pipe. Flow from the pipe terminates at the weir and is diverted through a 24 in. orifice
leading to the 38th Street pump station. At the pump station, it is transferred to an interceptor and
ultimately to the wastewater treatment plant. During events that exceed the capacity of the weir, excess
volume crests the 2.75 ft weir and continues down the outfall pipe to where it discharges into the Ohio
River. Figure 3-16 shows a schematic cross-section of the 11.5 ft circular pipe where it meets the outfall
weir.
The actual model representation of the outfall structure in SUSTAIN differs slightly from the schematic
shown in Figure 3-16. The SUSTAIN regulator object is modeled as a box with two outlet structures: an
orifice and a weir. The orifice is located at the bottom of the box whereas in reality the orifice is
physically located in the side of the pipe. To account of this difference in representation the orifice
discharge coefficient was varied as a calibration parameter at ultimately set at 0.155. In the 2011
Info Works model the height of the weir was also increased from 2.75 ft to 4.25 ft. The physical size of
the regulator box used in SUSTAIN (25 ft x 25 ft x 4.25 ft) was also subtracted from the 160 ft pipe.
Discussion with model developers from MSD revealed that this change was imposed to account for
uncertainty in model elevations caused by several different surveying datums referenced over a number of
decades. For consistency, the SUSTAIN model incorporated this change in weir height.
3-22
-------
CSO 019 Regulator and Pump Station
Figure 3-16. Conceptual cross-section of the CSO 019 outfall structure.
Figure 3-17 plots SUSTAIN versus Info Works total overflow volume and peak flow rate for water
overflowing the regulator. The eight largest overflow events identified by the 2008 IOAP model are
highlighted as diamonds on both panels of the graph. Also, eight new overflow events occurred in
SUSTAIN that did not occur in Info Works—this will be discussed later. Notice that neither volume nor
peak flow was distinctly predictive of overflow. In other words, the eight largest volumes are not the
eight largest peaks. Instead, overflow occurs under a critical condition caused by a combination of
factors.
100
01
E
in
=>
in
Modeled Overflows
8 Largest Overflows
11 New Overflows
012345 0 20 40 60 80
InfoWorks Inflow Volume (MGal) InfoWorks Peak Flow(cfs)
Figure 3-17. InfoWorks vs. SUSTAIN modeled overflow volume and overflow peak.
100
3-23
-------
During the SUSTAIN calibration, there were 11 additional overflow events observed that did not occur in
Info Works. A closer look at these events reveals that all had short durations and relatively small in
overflow volume and flow rate. Aggregation of the 203 Info Works subcatchments into 20 subwatersheds
in the SUSTAIN model and the simplification of the routing network from 647 conduit segments to 26
offers an explanation for predication of these additional events. The detailed Info Works model included
conduits ranging in size from 1.5 ft to 11.5 ft while the SUSTAIN representation mostly excluded pipes
smaller than 3 ft in diameter. Consequently, the SUSTAIN model representation provides less attenuation
capacity in its conduit network than the original Info Works, which also tends to make it slightly more
conservative in its prediction.
All 11 new events listed in Table 3-6 are in the lowest fifth percentile for both overflow volume and peak
flow rate. Because these events will be easily captured under any BMP scenarios, it is expected that they
will have no consequential impact for achieving the overflow target during optimization.
Table 3-6. Summary of additional predicted overflows
Start time
1/19/20010:15
4/1/2001 5:00
4/3/2001 4:00
5/24/20016:15
7/3/2001 17:45
7/24/2001 14:45
8/19/20011:15
8/23/2001 15:45
8/31/200123:00
10/16/2001 2:45
12/8/2001 5:00
End time
1/19/2001 14:00
4/1/2001 17:45
4/3/200116:15
5/24/2001 18:30
7/4/2001 6:45
7/25/20014:15
8/19/2001 14:00
8/24/2001 4:45
9/1/2001 11:30
10/16/2001 15:45
12/8/2001 19:15
Overflow volume
(MG)
0.027
0.022
0.002
0.003
0.057
0.084
0.022
0.075
0.013
0.036
0.095
Peak flow
(cfs)
1.02
1.61
0.32
0.44
3.66
4.01
1.45
4.53
1.09
2.52
2.37
3.3.5. Model Run-Time Considerations
The SUSTAIN replica of the CSO 019 sewershed incorporates considerable simplification from the
Info Works version that used as a basis for model setup and testing. The resulting model can be used to
examine the trade-off between model performance and computational efficiency. Savings in model
computation time become significant when performing optimization analysis. Table 3-7 summarizes the
model simplification and the computational savings in model run-time for a single continuous simulation
of the typical precipitation year 2001.
Table 3-7. Comparison of model representations and run-time
Model characteristics
Number of subcatchments
Number of pipe segments
Estimated run-time (minute)
InfoWorks
calibration
203
647
60 a
SUSTAIN
configuration
20
26
0.5
Percent
reduction
90%
96%
99%
a. InfoWorks run-time based on conversations with modeler running only the CSO 019 sewershed.
3-24
-------
Both single run-times presented in Table 3-7 may seem acceptable considering the complexity of a
hydraulics model, the scale of the watershed, and the goals of the model application; however, the
benefits come to light when thousands of iterative simulations are needed during optimization runs.
Figure 3-18 shows the estimated run-times required to perform a 10,000 run optimization using the model
configurations described in the table above compared to a single run.
1000
0.0001
SUSTAIN
InfoWorks
Single
Simulation
Optimization
(10,000 Simulations)
Figure 3-18. Comparisons of single simulation and optimization modeling run-times.
When performing a single simulation the difference between waiting one hour or half a minute may be
beneficial for gaining the additional accuracy. However, it is important to keep in mind the optimization
objectives, which in this case, are to minimize the number of CSO regulator overflows. The model has
been shown to perform very well for the largest events associated with overflows, despite the relatively
coarse spatial resolution. When a large number of runs are needed for the optimization process, shorter
run-times can support the practical application of the system within realistic time frames of hours or days.
Since optimization scenarios are typically run with various objectives, assumptions, and management
scenarios, a shorter run-time also facilities the use of the system for exploratory analysis. With careful
examination of the tradeoff between accuracy and simplification an appropriate level of resolution can be
identified consistent with the management questions under consideration.
3.4. BMP Parameter Sensitivity Analysis
Both the geometric representation and the parametric representation of BMP properties have an influence
on the way a BMP responds in SUSTAIN. Sometimes an irregularly shaped BMP must be simplified as a
rectangular or square box in the model. At the same time, some BMP calibration parameters are more
influential on how the BMP responds than others. The first part of this section demonstrates how an
actual BMP plan was translated from construction drawings into a BMP configuration in SUSTAIN. The
second part shows how a traditional laboratory analytical approach (full factorial experimental design)
3-25
-------
was adapted and applied to study the sensitivity of key BMP configuration parameters in SUSTAIN.
Finally, this section concludes with quantifying the range of the response variations for the sensitivity
analysis. As previously noted, this analysis was conducted using a single bioretention cell.
3.4.1. BMP Representation
MSB participated in the design, construction, and current monitoring of a GI demonstration project at the
Office of Employment and Training at 600 Cedar Street in downtown Louisville, Kentucky. The project
is adjacent to a 3-acre parking and that drained to inlets directly connected to the storm sewer. Three
bioretention cells and 2,000 square feet of porous asphalt were installed along with several bioinfiltration
areas and porous paver features to decrease the volume of stormwater runoff to the city's CSS from the
parking lot. A single bioretention cell in the southwest corner of the parking lot was selected to evaluate
the sensitivity of BMP simulation parameters.
A map of the project location and bioretention cell site are shown in Figure 3-19. The map shows an
overlay of the construction drawings on an aerial photo of the site. An image of the site schematic was
extracted from the construction drawings and geo-referenced to a current aerial photo in ArcGIS. The
drainage area and BMP location were then delineated using the geo-referenced construction plans.
3-26
-------
Legend
Bioretention
Drainage Area
ji.fcriK-w
4:u ">T=
1
"'
\ttJE$l&»
. ^*r*Mtmj IF -
. , (A«TO Tig r
_ LOT eSSML
*"-.Vw M'rvc'
fWK*^
— -_
PUT*. .1-
e::> IKW
CF 91-5 s?
;• JOT
Louisville Office of Employment
Bioretention Demonstration Project
MAC 1962 StatePlans Missouri WES! FIPS 1601
Figure 3-19. Office of Employment bioretention cell site location and drainage area.
3-27
-------
A SUSTAIN model representation of the bioretention cell was constructed using drainage area and
dimensional information from the design plans in Figure 3-20. The bioretention cell is designed to
receive runoff from a 0.3-acre section of the parking lot. In SUSTAIN, the bioretention cell was
configured using a length and width of 30 ft for a total surface area of 900 square feet. The typical
bioretention cell cross section presented shown in Figure 3-20 was used to construct the BMP vertical
profile (Strand Associates, 2010).
BARRIER CURB
FOR CORNER BIOSWALE
SEE SHEET C-13 FOR
ADDITIONAL CURB DETAILS
EXISTING SIDEWALK
m
BARRIER CURB FOR CORNER BIOSWALE
SEE SHEET C-13 FOR
ADDITIONAL CURB DETAILS
EXISTING PARKING LOT PAVEMENT
HARDWOOD MULCH
(2 INCH DEPTH)
~BIOINFILTRATION SOIL MIX
(24 INCH DEPTH)
SEE NOTE 1.
CLEAN, DOUBLE WASHED NO. 57
AGGREGATE (18 INCH DEPTH)
SEE NOTE 1.
NON-WOVEN GEOTEXTILE FILTER FABRIC
ALONG SIDES. AND TOP OF
NO. 57 AGGREGATE
PERFORATED PIPE UNDERDRAIN
WITH GEOTEXTILE FABRIC WRAPPED
AROUND TOP HALF OF PIPE
(SLOPED 0.2% MIN.)
UNDISTURBED SOIL
NOTES:
NORTHWESTERN CORNER BIOSWALE SHALL BE CONSTRUCTED WITH
22 INCH DEPTH BIOINFILTRATION SOIL MIX, AND 20 INCH DEPTH
CLEAN. WASHED NO. 57 AGGREGATE.
Figure 3-20. Bioretention cell subsurface cross-section from design plans (Strand Associates, 2010).
The figure shows a soil media depth of 24 in. and an underdrain depth of 18 in. A 6 in. ponding depth was
used in the SUSTAIN model configuration. A complete list of the physical BMP parameters used for
model setup is presented in Table 3-8. Each parameter in the respective SUSTAIN BMP interfaces, shown
in Figure 3-21 (surface), Figure 3-22 (substrate), and Figure 3-23 (infiltration), is also highlighted in the
figures (as A-F), corresponding to how they are labeled in Table 3-8.
Table 3-8.Summary of BMP parameters used for bioretention cell configuration.
Figure group
A
B
C
D
E
Parameter
Length (ft)
Width (ft)
Orifice diameter (in.)
Ponding depth (ft)
Soil media depth (ft)
Soil media porosity
Underdrain depth (ft)
Underdrain porosity
Value
30
30
0
0.5
2
0.3
1.5
0.4
3-28
-------
Figure group
F
Parameter
Maximum infiltration rate (in./hr)
Decay constant (1/hr)
Drying time (days)
Maximum volume (in.)
Value
5
0.2
3
48
Define BMP Parameters
Dimensions Substrate Properties Infiltration Parameters Water Quality Parameters Cost Factors Sediment
General Inforr
Name
Basic Dimens
Length (ft)
Number of
Surface Stora
I ;
A L
BioReterrtionBasin2
30 _J A Width (ft) 30
Units 1 . i Design Drainage o
Area (ad
j
:":'-//^
•'; J
- • . /' X"
w
D Orific
Orific
~
e Diameter (in) o
e Height (Ho, ft) o
,— -t -y-
rJ^S3 °"ei '1 0'61 °5 I
r P 1 ^
« r
Release Option
(~ Cistern Number of People
Number of Dry Days
(B None
L
ttiete
• ' . NV£
Weir Height (Hw, ft) Q.5 _j
^S»V Weir Crest Width (B, ft) (*
Vertex Angle (theta, deg)
a c
\
^
OK
Cancel
Figure 3-21. SUSTAIN surface parameter input screens for bioretention cell.
3-29
-------
Define BMP Parameters
Dimensions Substrate Properties | Infiltration Parameters | Water Quality Parameters | Cost Factors] Sediment]
depth (Du)
Depth of Soil, Ds (ft).
Soil Porosiiy (0-1):
Soil Reid Capacity U
Soil Wilting Point
Saturated Soil Infiltration
(in/hr):
Aquifer
k Consider Underdrain Structure:
Storage Depth (Du, ft) Media Void Fraction 10-1 )• l» da -y era mrl livfihation (m/hr)
Figure 3-22. SUSTAIN substrate parameter input screens for bioretention cell.
« Define BMP Parameters [x]
Dime
t
nsons] Substrate Properties Infiltration Parameters Water Qua ity Parameters | Cost Factors] Sedime
Suction Head (in) nitial Deficit (fract on)
3
0 0.3
^Morton Infiltration Parameters "^
Maximum infiltration (in/hr) [5
Decay Constant (1 /hr) [o~2
Drying Time (day) |3
Maximum Volume (in) ;48
-- -. • -i r.
Vegetative Parameter A: rj g
Monthly Growth Index
|>
Month Value
January 0 55
February 0.6
March Q.EE
Arjril 0.85
May 0.95
June 1
July 1
Auqust 1
September 1
October 0.95
November 0.75
December 0 E
t
OK
Cancel
Figure 3-23. SUSTAIN infiltration parameter input screens for bioretention cell.
3-30
-------
Some additional observations about BMP configuration and setup are as follows:
• The bioretention cell configuration from the design plans did not show an orifice for outflow of
water that ponds on the surface; therefore, the orifice diameter was set to zero. In this case, the
orifice height was also set to zero. Other orifice related parameters such as the exit type were not
used in this case;
• The background infiltration rate should be set as the infiltration rate for the native soils into which
the bioretention cell is installed. The parameter was treated as a variable in Section 3.4.2 and was
left blank accordingly in Figure 3-21 (near figure group E); and
• Typically, not all BMP parameters are used in every simulation. In this case, the Horton
infiltration method was used to simulate the BMP media infiltration process. Therefore, only the
Horton infiltration parameters were used. .
Not all BMP configuration parameters are found on the design or construction plans. If site specific data
was not collected in the design study, values for soil field capacity, wilting point, and the infiltration
parameters are best evaluated through other local data sources, such as geotechnical reports, or academic
literature reviews. In this application, porosity and underdrain void fraction were assumed. A list of
BMP parameters and suggested information sources is presented in Table 3-9.
Table 3-9. Suggested information sources for obtaining BMP parameters
Key information sources:
• = Primary source
© = Secondary source
- = Not applicable
BMP dimensions
Infiltration rates
ET multiplier
Design plans
•
~
~
Geotechnical
report
~
•
~
Academic
literature
~
©
•
3.4.2. Factorial Experiential Design
A factorial experiment was designed to test the sensitivity of three independent BMP parameters.
Factorial experiments are designed to evaluate multiple responses from three or more independent
variables and quantify the magnitude of each response (Berthouex and Brown, 2002). This type of
experiential design limits the number of experiments needed to evaluate multiple independent variables.
The factorial framework is often implemented in laboratory settings to minimize the use of material
resources, lab time, and person-hours expended while maximizing the utility of data collected.
Three BMP parameters were evaluated, that included the ET multiplier and two Horton infiltration
parameters, the saturated infiltration rate (fc) and maximum infiltration rate (/"<,). The ET multiplier can
vary for each BMP and is applied to the ET rate at each time step. Low and high ET multiplier values
were selected to represent turf grass at low and high ET conditions (Bedient and Huber, 1992). A
monthly distribution of constant daily ET rates was used for the simulation as referenced in Section 3.3.3.
The saturated infiltration rate (fc) is the rate at which water can infiltrate under saturated soil conditions.
The initial infiltration rate (f0) is the infiltration rate for unsaturated soil. Low and high infiltration
parameters were selected on the basis of MSB suggested values for hydrologic soil group Type C and D
soils (MSB, 2007). Those are suggested parameter starting values by MSB's hydraulic modeling
guidelines document. The rate of change between the maximum and saturated infiltration rate is
3-31
-------
controlled by an infiltration decay coefficient (k) that was held constant at 0.2 hr"1. Selected values for
each of three parameters are listed in Table 3-10.
Table 3-10. Low and high values selected for three evaluated BMP parameters
Parameter
ET multiplier
Saturated infiltration rate (in./hr)
Maximum infiltration rate (in./hr)
Low
0.35
0.10
3.00
High
0.85
0.25
5.00
The number of experimental runs incorporated into a full factorial analysis is a function of the number of
variables being tested and can be calculated as 2X, where 2 is the number of conditions (low and high) and
x is the number of variables. For this analysis, eight simulations were constructed to test three key
variables, although the sensitivity of more than three variables could be explored. A matrix outlining the
variables for each of the eight simulation runs is presented below in Table 3-11.
Table 3-11. Matrix of the eight designed experimental runs showing values for the three parameters
Simulation number
1
2
3
4
5
6
7
8
ET multiplier
0.35
0.85
0.35
0.85
0.35
0.85
0.35
0.85
Saturated infiltration, fc
(in./hr)
0.10
0.10
0.25
0.25
0.10
0.10
0.25
0.25
Maximum infiltration,/,,
(in./hr)
3
3
3
3
5
5
5
5
The 2001 precipitation time series was used as the boundary condition for each of the eight simulations.
The •S't/.ST^/jVbioretention cell configuration described previously was used for draining 0.3 acre of
impervious parking lot. Other than the variables outlined in Table 3-11, all simulation parameters were
held constant. Each simulation was evaluated for total outflow volume from the bioretention cell.
3.4.3. Results
Figure 3-24 shows the average annual reduction in BMP total outflow for all eight scenarios as a
percentage of the baseline condition where no BMP is present. The sensitivity of each parameter is
evaluated as the difference in average response (outflow volume) between the high condition and low
condition. Average BMP outflow percent reduction versus low and high conditions for the three tested
variables is shown in Figure 3-24.
3-32
-------
ET
Multiplier
Saturated
Infiltration Rate
High
Maximum
Infiltration Rate
Figure 3-24. Average annual reduction in total BMP outflow for all eight scenarios.
The error bars, represented as lines at the top of each bar in Figure 3-24, show the range of variability
(minimum and maximum) for the averaged outflow values. Wide bands for a given parameter suggest
that variability is controlled not by that parameter but by one of the other two. For the saturated
infiltration rate, the very narrow bands on the error bar suggest that variability in BMP outflow is
controlled foremost by controlling the value of that parameter.
The variation in average response between low and high conditions can also be calculated by subtracting
the average annual BMP outflow for the low condition from the high condition for each of the three
parameters. Table 3-12 presents the average variation in total BMP outflow expected on the basis of the
range of BMP parameters presented in Figure 3-24.
Table 3-12. Summary of average annual BMP variation in response between low and high conditions
Parameter
ET multiplier (unitless)
Saturated infiltration rate (in./hr)
Maximum infiltration rate (in./hr)
Average
variation
(gal. per year)
267.68
13,211.92
14.34
Percent
of baseline
0.13%
6.46%
0.01%
The saturated infiltration rate is the most sensitive of the three parameters evaluated, followed by the ET
multiplier, and the maximum infiltration rate. The range of BMP responses in the table above varies by
three orders of magnitude. When setting individual BMP parameters, the range of expected responses can
be interpreted from the table above as follows:
• Increasing the ET multiplier from 0.35 to 0.85 will decrease the average annual flow volume by
56 cubic feet, or 0.13 percent of the baseline annual flow volume;
• Increasing the saturated infiltration rate from 0.10 to 0.25 in. per hr will decrease the average
3-33
-------
annual flow volume by 2,674 cubic feet, or 6.46 percent of the baseline annual flow volume; and
• Increasing the maximum infiltration rate from 3 to 5 in. per hr will decrease the average annual
flow volume by 3 cubic feet, or 0.01 percent of the baseline annual flow volume.
Results of the sensitivity analysis suggest prioritizing research and interpretation of model values for the
saturated infiltration rate, which shows a 6.5 percent variation in annual average outflow when selecting
an/c value suggested for Type C versus Type D soils. While it is important for the purpose of accurate
model representation, the ET multiplier and maximum infiltration rate each show a less than 1 percent
variation in annual average outflow when selecting representative low and high parameter values.
3.5. Cost-benefit Relationship between Gray and Green Infrastructure
for Mitigating CSO
The central question in the minds of regional policy makers is how might the cost of planed gray
infrastructure be offset through the use of GI alternatives? Exploratory management alternatives relevant
to the case study area include (1) a downspout disconnection program that redirects rooftop runoff to rain
barrels and existing pervious land; (2) implementing green street and green parking practices in
conjunction with a downspout disconnection program; and (3) exploring the supplemental gray
infrastructure necessary to satisfy the optimization objectives when GI is completely built-out. The
objectives for optimization are to (1) minimize annual number of overflows and volume; and (2)
minimize the total capital cost of implementation, as needed, to satisfy the allowable CSO exceedance
criteria (eight events per year). Figure 3-25 illustrates the development sequence of exploratory
optimization scenarios relative to the established baseline condition.
IS)
-M
CD
o
CO
u
CD
_Q
E
CSO 190
Intercepto
Relie
Allowable Exceedances
ptoA
>^A
GSQ
Runoff
Only
Original
Baseline
Optimization
Baseline
\ \
Gray
Infrastructure
Only
Optimization
Scenario 1
MaxA/ Min $
Gray
+
Green
Infrastructure
Downspout
Disconnection
Optimization
Scenario N
_o
Q.
X
Figure 3-25. Conceptual sequence of optimization scenarios relative to baseline condition.
3-34
-------
3.5.1. Green Infrastructure Opportunities
MSB performed an analysis of potential BMP opportunities in the CSO 019 sewershed. Those data sets
were available as GIS shapefiles and represent a screening level analysis of possible BMP opportunities
in the watershed. This study was not a comprehensive cost feasibility or on-site assessment of limiting
factors such as conflicting utility infrastructure, hardscaping, or other landscape features. The types of
BMPs evaluated in this analysis include (1) bioinfiltration; (2) downspouts disconnection; (3) green alley;
(4) green parking lot; (5) reforestation; (6) tree lawn retrofit; and (7) intersection bump out. A map
highlighting the possible BMP opportunities and drainage areas is presented in Figure 3-26.
Biointiltraoon
Y//A Downspout Disconnection
Green Parking Lot
' BMP Drainage Area
Subwatersned
Louisville, Kentucky
CSO 019 Green Infrastructure Opportunities
Figure 3-26. CSO 019 Gl opportunities and treated drainage areas.
The BMP opportunities presented in Figure 3-27 are indicative of common BMPs that are considered
acceptable practices in MSB's service area. When represented in the SUSTAIN model, each BMP has an
associated treatment capacity (ponding volume + substrate volume + underdrain volume). In a
subwatershed, the collective volume of individual BMPs represents the total treatment capacity for the
subwatershed. It is likely that during implementation, other suitable types of BMPs could be identified
that were not included in this study. Treatment capacity provides a standard basis of comparison when
implementing other types of BMPs. This case study focused on the following BMP types that directly
control impervious runoff, including:
• Rain barrel and downspout disconnection;
3-35
-------
• Green street and green alley related opportunities, mainly bioinfiltration facilities, including
bioretention, curb extension bioretention, and bioswale; and
• Green parking lot using combination of pervious pavement and bioinfiltration.
Figure 3-27 illustrates examples of downspout disconnection to a rain barrel, typical green street
bioinfiltration, and typical green parking with bioinfiltration and pervious pavement. Rain barrels collect
rooftop runoff and drain the water to pervious land during dry days. Green street bioinfiltration practices
are along the streets to collect and treat runoff from impervious road, sidewalk, and driveways. Green
parking lots adopt pervious pavement in the parking areas and use bioinfiltration practices in medians and
islands to reduce the runoff.
Rain Barrel / Downspout
Disconnection
fAuCtr
Figure 3-27. Examples of Gl practices for Louisville, Kentucky.
3.5.2. SUSTAIN BMP Representation
An aggregate BMP approach was implemented to represent the BMP opportunities in the CSO 019
sewershed model. The aggregate BMP consists of five components including (1) rain barrels; (2)
downspout disconnections; (3) green street bioinfiltration; (4) green parking pervious pavement; and (5)
green parking bioinfiltraiton. The modeled GI BMP drainage pathways are illustrated in Figure 3-28.
The figure is a conceptual diagram of the treatment pathways showing the relationship between different
tributary land cover types for the potential BMPs. Disconnected rooftops represent downspouts that are
no longer connected directly to the sewer main via a lateral. Instead, runoff from downspouts is directed
to a rain barrel for use as a non-potable water supply. Both outflow and overflow from the rain barrel is
directed to adjacent pervious area before being routed to the outlet.
3-36
-------
Treated Drainage Area Land Distribution
Disconnected
Rooftop
Road/Street
Sidewalk
Driveway
Parking
Untreated
Land
Green Parking
Green Street
Bioinfiltration
Pervious
Pavement
Bioinfiltration
outlet
Figure 3-28. CSO 019 Gl BMP drainage networks.
Table 3-13 lists the BMP design dimensions and specifications of the BMP types represented. Two sets
of BMP design capacity scenarios were explored in this study to capture (1) 0.75 in. of runoff; and (2) 1.0
in. of runoff. The 0.75 in. capacity is considered as a level acceptable to MSB (MSB, 2011). The 1.0 in.
capacity scenario explored the implication on cost-effectiveness by applying a more rigorous design
capacity. The vertical designs of the BMPs are held constant for the two design capacities, and the
various design capacities are obtained by varying the BMP surface areas. The BMP units are sized
assuming a 0.25 acre drainage area.
Table 3-13. BMP design dimensions and specifications
Parameter
Rain barrel
Bioinfiltration
Pervious pavement
Surface parameters
Unit size (0.75 in. runoff) (gal)
Unit size (1.00 in. runoff)
Design drainage area (acre)
Ponding depth (ft)
ET multiplier
60
60
0.005
N/A
0
350
470
0.25
1
1.5
450
600
0.25
0
0.5
Substrate parameters
Substrate depth (ft)
Substrate porosity
Substrate field capacity
Substrate wilting point
N/A
N/A
N/A
N/A
2
0.35
0.3
0.1
2
0.35
0.2
0.05
3-37
-------
Parameter
Rain barrel
Bioinfilt ration
Pervious pavement
Substrate infiltration parameters
Maximum rate (in./hr)
Minimum rate (in./hr)
Decay constant (1/hr)
Dry time (day)
Maximum volume (in.)
N/A
N/A
N/A
N/A
N/A
10
1
1
3
48
10
2
2
3
48
Underdrain parameters
Underdrain depth (ft)
Underdrain porosity
Infiltration rate (in./hr)
N/A
N/A
N/A
1
0.4
0.3
2
0.4
0.3
The substrate layer infiltration parameters were selected on the basis of MSB guidance for well-draining,
hydrologic soil group Type A soils. Those parameters control the rate at which water passes through the
substrate layer into the underdrain. The background infiltration parameter controls the rate at which water
passes from the underdrain into native soils. Section 3.4 discussed the sensitivity of this background
infiltration rate on the total annual outflow from BMPs and suggests that this is the limiting parameter in
the BMP simulation related to infiltration. To address this key parameter, the decision was made to use a
background, or final, infiltration rate in the BMPs consistent with the value used in the InfoWorks
watershed model and the calibrated SUSTAIN model, which was 0.3 in. per hr.
The amount of upstream impervious area drainage to each BMP was also essential to evaluating the
overall effectiveness. The baseline model integration presented in Section 2.3 was used as the basis for
configuring BMP tributary area. In each of the subwatersheds, one aggregated BMP was configured to
represent the GI practices described in Section 3.5.1. The number of units of each BMP type was
calculated by dividing the corresponding drainage area by the unit's design drainage area. The
distribution of treatable impervious areas among subwatersheds and land use types was estimated on the
basis of analysis of the GIS coverage of BMP opportunities provided by MSB and presented in Figure
3-26. The distribution of treatable impervious area by subwatershed is shown in Table 3-14.
Table 3-14. Distribution of impervious areas that can be treated by GI
Subwatershed
1
2
3
4
5
6
Untreated
impervious
(acre)
7.6
27.6
14.5
11.9
7.0
10.8
Treated impervious area
Rooftop
disconnection
(acre)
7.4
9.1
1.2
6.9
5.9
0.0
Green
street
(acre)
20.2
5.1
6.0
6.2
3.8
0.0
Green
parking
(acre)
2.0
1.1
13.2
0.7
0.8
0.0
Total treated
impervious
(acre)
29.6
15.3
20.5
13.8
10.5
0.0
Total treated
impervious
(%)
79.6%
35.7%
58.6%
53.7%
59.9%
0.0%
3-38
-------
Subwatershed
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Total
Untreated
impervious
(acre)
9.5
7.8
5.8
19.9
8.5
12.5
18.8
9.7
8.0
10.8
12.0
0.6
45.1
6.8
255.1
Treated impervious area
Rooftop
disconnection
(acre)
2.9
1.7
2.6
13.5
6.1
8.5
11.7
12.2
5.5
8.9
6.8
5.2
12.5
7.0
135.8
Green
street
(acre)
4.4
15.4
0.8
5.3
1.1
6.3
6.3
7.6
1.6
10.8
3.7
9.7
19.3
3.1
136.9
Green
parking
(acre)
0.0
0.5
0.0
0.3
0.2
2.6
1.5
2.2
0.1
1.0
1.2
1.9
4.4
0.0
33.7
Total treated
impervious
(acre)
7.4
17.6
3.4
19.0
7.4
17.4
19.6
22.0
7.3
20.6
11.7
16.9
36.3
10.1
306.3
Total treated
impervious
(%)
43.7%
69.4%
37.1%
48.9%
46.6%
58.2%
51.0%
69.3%
47.7%
65.6%
49.3%
96.6%
44.6%
59.7%
54.6%
3.5.3. SUSTAIN Portland Wharf Storage Basin Representation
For the purposes of this study, the gray infrastructure considered was the Portland Wharf Storage Basin
(Section 3.1.2). The basin was represented in the SUSTAIN model as an impervious storage unit,
simulated in the model with a zero infiltration rate and ET multiplier. It was placed in the network
downstream of the CSO 019 regulator and receives overflow from the regulator weir. A constant
pumping rate was applied to the storage unit designed to empty the 6.37-MG tank within 24 hours when
water level is greater than zero. Volume exceeding the sum of the storage and pumping capacity
discharges from the storage basin and was considered a system overflow.
3.5.4. BMP Cost Representation
GI Costs
The cost for the GI BMPs is listed in Table 3-15. The GI cost calculation spreadsheet provided by MSB
was used to estimate the unit cost of bioinfiltration and pervious pavement. The cost calculation
spreadsheet was developed on the basis of actual construction bid data submitted for the GI demonstration
project at the Office of Employment and Training (Section 3.4.1). The cost per downspout disconnection
is obtained through verbal and email communication with MSB (MSB, 2011). The cost of a 60 gallon
rain barrel is approximately $120 when considering a rain barrel unit cost of $2 per gallon (Woodland
Birect, 2011).
3-39
-------
Table 3-15. Gl BMP construction cost (in 2011 dollars)
BMP types
Bioinfiltration
Pervious pavement
Rain barrel (60 gallon)
Downspout disconnection
(per 200 sq ft of rooftop)
Sized to control 0.75 in. runoff
Unit surface area
(sqft)
350
450
Unit cost
$14,842
$17,082
Sized to control 1 in. runoff
Unit surface area
(sqft)
470
600
Unit cost
$17,398
$19,506
$120
$100
Gray Infrastructure Costs
MSB provided estimates of construction cost of the Portland Warf Storage Basin at five intervals as
percent of the tank's total capacity. A mobilization cost of $3 million was assumed. A plot of the cost
estimates is shown in Figure 3-29.
$18
$16
$14
$8
$6
$4
$2
$0
• SUSTAIN cost function
--•-• MSD-Provided
Construction Cost ($ Million) = $2.0 x Volume in Million Gallon+$3.0
12345
Storage Volume (million gallons)
Figure 3-29. Portland Warf Storage Basin cost function (in 2008 dollars).
The cost values were converted from gallons to cubic feet, which is consistent with the standard units in
SUSTAIN. On the basis of the cost estimates presented in Figure 3-29, a cost function for the Portland
Wharf Storage Basin was expressed in 2008 dollars using the following equation:
Total Construction Cost ($) = $3,000,000 + $14.96 x (storage volume in cubic foot)
For the SUSTAIN model configuration, the cost function was converted to 2011 values by applying ENR
construction cost index values, i.e., 8,310 for 2008 and current value of 9,104 for June 2011. The cost
function in 2011 used the following equation:
Total
Construction Cost ($) = $3,288,000 + $16.40 x (storage volume in cubic foot)
3-40
-------
3.5.5. Exploratory Management Scenarios
Five exploratory management scenarios were developed to evaluate the cost-effectiveness of (1) gray
infrastructure only; (2) downspout disconnection only; (3) gray infrastructure with downspout
disconnection; (4) GI only; and (5) maximum build-out of GI with supplemental gray storage. Table 3-16
is a summary and description of the optimization scenarios.
Table 3-16. Summary and description of baseline exploratory management scenarios
Optimization scenario
Gray only
Downspout
disconnection only
Downspout
disconnection + gray
Downspout
disconnection + green
Maximum
green + gray
Description
Exploratory runs that includes only the Portland Wharf Storage Basin and varies
the tank volume to control the number of annual overflows.
Exploratory runs using only the extent of downspout disconnection identified
Exploratory run that optimizes using a mix of downspout disconnections and
storage in parallel.
gray
Exploratory runs with only green option, including downspout disconnections
Fixed maximum green options, and exploratory gray options
Those five management scenarios are designed to answer the following key questions:
• What is the cost-effectiveness of using gray infrastructure only to reduce annual overflow?
• What is the cost-effectiveness of using GI only to reduce annual overflow?
• How much annual overflow volume can be reduced by implementing GI?
• What is the optimal combination of green and gray infrastructure for reducing the annual
overflow volume?
3.5.6. Optimization Problem Formulation
Using the CSO 019 baseline model configuration presented in Section 2.3 with the BMP representations
presented in Sections 3.5.2 and 3.5.3, optimization problems were formulated for the exploratory
management options.
The generalized multi-objective functions and constraints are presented as follows:
Minimize £ BMP construction costs
Minimize Regulator overflow count (eight allowable overflows for 2001 simulation)
Subject to
• Maximum extent of identified GI opportunities
• Maximum size of Portland Wharf Storage Basin
• Combinations of exploratory management options:
o Gray Infrastructure Only
o Downspout Disconnection Only
o Downspout Disconnection + Gray infrastructure
o Downspout Disconnection + GI
o Green build-out + Gray Infrastructure
3-41
-------
MSB set an overflow target of eight events per year for the CSO 019 sewershed. The Portland Wharf
Storage Basin was sized to meet that target on the basis of the 2008 IOAP model (MSD, 2008). GI
practices could reduce the size of the tank if they prove a more cost-effective measure. During the
optimization process, the decision variables were (1) the percentage of area treated by the various GI
practices, as listed in Table 3-14, and (2) the size of the Portland Wharf Storage Basin for supplemental
gray storage. Because BMP construction costs for GI practices heavily influence the resulting solutions, a
sensitivity analysis was performed using additional BMP cost literature values to demonstrate a range of
expected construction costs and discuss the uncertainty associated with costing GI practices. The cost for
both storage basin and GI practices in this analysis were based on construction cost; O&M cost is not
considered due to lack of local data. It is recognized that O&M cost is an important factor in the
assessment of the total long-term cost and the comparison of gray storage and GI options.
3.5.7. Optimization Results
Optimization was performed for the five management scenarios discussed in Section 3.5.5. Figure 3-30
shows optimization results as cost-effectiveness curves for the five exploratory management scenarios.
Effectiveness is plotted in terms of overflow counts (points - read on the left axis), and overflow volume
reduction (lines - read on the right axis).
• Gray Only (Count)
• Green Infrastructure (GI)
^—Downspout Disconnection + Gray
Downspout Disconnection
GI Buildout + Gray
-Green Infrastructure (GI)
• Downspout Disconnection + Gray
—Gray Only (Volume)
^—GI Buildout + Gray
0
8 Overflows
(~ 90% Volume
50
0
8
10
12
14
16
18
20
22
24
26
28
BMP Construction Cost ( $ Million )
Figure 3-30. Cost-effectiveness curves for exploratory management scenarios.
The following observations are made in the interpretation of the results. First, any selection of gray
infrastructure solution results in an immediate cost of $3.29 million for mobilization and other activities
associated with building the storage basin. In the figure, the Gray Initiation Cost is always represented as
a dashed arrow in each scenario that includes a gray component. Those costs are incurred regardless of
the size of the storage basin and do not directly correspond to any reduction in overflow volume. As the
3-42
-------
tank increases in size, the trajectory of the cost-effectiveness curve maintains a steep slope that flattens
only when approaching an overflow volume reduction of 100 percent.
For each optimization scenario the curve of overflow volume reduction tracks consistently above the
discrete points representing the number of overflows at varying levels of implementation. Using the
number of annual overflows as an objective presents a slightly diminished view of overall performance
when compared to a true overflow volume reduction. Because overflow events are discrete points, it is
possible for BMPs to provide additional volume reduction without affecting the overflow count. Flow
attenuation associated with GI provides reduction in volume that does not translate directly into
reductions in the number of overflows. That is clearly shown at the highest extent of GI, where a 50
percent overflow volume reduction still causes 37 overflows. On the other hand, the smallest gray storage
provides about 65 percent volume reduction and only allows 21 overflows. Because the optimization goal
was to reduce the number of overflows, together with the fact that the gray alternatives directly address
overflow containment, the gray alternative seems to be more cost-effective. However, it is important to
remember that the optimization objective drives the optimization result.
Third, GI costs seem disproportionally large compared to gray costs. Two aspects associated with how
GI representation is worth noting: (1) the cost assumptions; and (2) the sizing criteria. Just as the
optimization results are driven by the specified optimization objective, cost-effectiveness is driven by the
associated cost assumptions and modeled BMP performance. The next section further explores the
sensitivity of those BMP characteristics on model results.
Four solutions from the Gl-only cost-effectiveness curve were selected for further detailed evaluation.
Each point along the cost-effectiveness curve presented in Figure 3-31 corresponds to a unique
combination of BMP selections from a single simulation run. For a given solution, the selection of BMPs
can be (1) quantified in terms of the magnitude of build-out; and (2) analyzed spatially in terms of BMP
selections throughout each subwatershed.
55%
50% -
T3
-------
The utilization percentage of each practice for the four solutions is plotted in Figure 3-32. Percent
utilization for each solution is defined as the ratio of how much of the available opportunity was used
divided by the total available opportunity. Percent utilization is computed as follows:
Percent Utilization =
(Maximum Available GI Storage Volume — GI Solution Storge Volume)
Maximum Available GI Storage Volume
iGreenSolution #1
iGreenSolution #2
GreenSolution #3
I GreenSolution#4
4-i C
OJ O
cu £
I . CD
1/1 i:
>- o
(3 a
QD c
.£ O
-* +3
>- ro
ro i_
CL ±;
c *;
QD
at
tc
CL
S o3
oj CL
5
=> OJ
•i £
£ a
O u
Q -!2
Figure 3-32. GI BMP percent utilization at four selected solutions.
Figure 3-32 shows that using downspout disconnections is selected for maximum implementation in all
four solutions. That is because of the user-imposed assumption that structural GI practices are
implemented only after full adoption of the downspout disconnection program, which was also shown to
be the most cost-effective practice modeled. Of the structural BMPs, rain barrels are first selected only in
Solution #3 and maximized only in Solutions #4, suggesting that those practices were considered least
cost-effective for reducing volume. Rain barrels were configured in the model as an intermediate storage
that receives flow from disconnected downspouts and then conveys outflow to pervious land. A low
utilization percentage suggests that allowing water to route from disconnected downspouts directly to
pervious land is more cost-effective in terms of achieving overflow volume reduction than adding rain
barrels as an intermediate storage (with its additional associated cost). It is only fully implemented after
all other options have been exhausted. Notice that the utilization of all five GI practices is 100 percent for
Solution #4 where maximum build-out is achieved.
3.5.8. Optimization Sensitivity Analyses
Section 3.5.7 presents the optimization results and cost-effectiveness curve for the five exploratory
management scenarios discussed in this case study. In the discussion it notes that the cost of GI is
disproportionately high compared to the modeled gray components. Two key aspects of the GI
parameterization are identified as heavily influencing the cost and modeled performance of the practices:
(1) the BMP cost assumptions; and (2) the BMP performance sizing criteria.
3-44
-------
This section explores sensitivity testing associated with those key aspects of the model configuration and
discusses the implications of parameter assumptions on the model results. In addition to the GI cost
provided by MSB, two alternative cost scenarios are presented, which use literature cost value and
remove retrofit costs from the original MSB cost estimates. The results show how cost assumptions at
the BMP unit level can affect the total cost of achieving the optimization target. The influence of BMP
size on BMP performance and total construction cost is evaluated by considering two alternative criteria
sizing BMPs to capture and treat (1) 0.75 in. of runoff; and (2) a more stringent target of 1.00 in. of
runoff.
BMP Cost Sensitivity Testing
The cost-effectiveness curves for Gl-only and the integrated green and gray infrastructure options cost are
much higher than the gray-only scenario. Further review of those costs suggests that the Louisville GI
costs might be higher than BMP construction costs cited in literature. The unit costs presented in Section
3.5.4 fall at the upper end or beyond the range of values typically expected from literature. Table 3-17
presents a comparison of local GI cost estimates against available bid costs from a GI project in Kansas
City, Missouri, and literature values published by the CWP.
Table 3-17. Comparison of BMP costs per gallon of treatment capacity from various sources
BMP type
Bioinfiltration
Pervious pavement
Rain barrel
Louisville MSD ($/gallon)a
GI retrofit
2.91
3.36
2
New
construction
1.94
2.24
2
CWPa
report
($/gallon)
1.16
1.34
2
Kansas City
bid cost
($/gallon)
9.73
6.44
2.83
Kansas Cityb
modified
bid cost
($/gallon)
4.18
2.77
2.83
a. Construction cost only.
b. On the basis of construction bid cost, excluding 5 percent contingency cost, 3 percent tree removal and utility relocation cost,
and 43 percent sidewalk and street improvement cost.
Two main factors identified that account for the high cost estimates are (1) the cost estimates were
derived on the basis of demonstration project bids without a local precedent on which to base the costing;
and (2) retrofit components for demolition, curb and sidewalk replacement, and repaying were included in
the cost estimates, which are only indirectly related to BMP construction. To test the sensitivity of BMP
cost and examine the uncertainty in the cost estimate data, two additional GI scenarios were run that use
alternative cost data. Those alternative cost scenarios are plotted in Figure 3-33 for comparison with the
original GI cost-effectiveness curve. Figure 3-33 shows only the volume reduction curves. Note that the
optimization target of eight overflow events per year corresponds to about a 90 percent annual overflow
volume reduction.
Cost Alternative 1 is based on the BMP costs provided by MSB but removes the retrofit aspects of those
cost estimates to represent the cost for implementation as part of new construction. CWP suggested that
retrofit base construction costs generally exceeded the cost of new construction BMPs by a factor of 1.5
to 6 (CWP, 2007). In this alternative, the factor of 1.5 was applied to the MSB cost values to estimate
new construction cost for bioinfiltration and pervious pavement. The cost of rain barrels and downspout
disconnection remained the same.
Cost Alternative 2 is a low-end estimate based solely on published literature values. Because BMP
retrofit costs are extremely variable depending on site conditions and retrofit design complexity (CWP,
3-45
-------
2007), the low-end literature BMP retrofit cost values were examined. CWP listed a low-end larger
bioretention retrofit BMP of $7.5 per cubic foot of runoff treated in 2006 dollars, which is equivalent to
$8.72 in current dollar value. The MSB costing spreadsheet indicates a value of $21.75 per cubic foot of
runoff treated in current dollars, which is 2.5 times higher than the low-end literature values. As a result,
the MSB costs were divided by 2.5 to derive the low-end retrofit BMP cost values.
c
o
••§
T3
0
or
0
t
-------
0
MSDGI
Cost
MSDGI
Cost Without
Retrofit Components
Literature
Gl Cost
Figure 3-34. Comparison of full build-out of Gl for the three cost scenarios.
During optimization, the construction cost of each Gl practice is a key parameter that strongly influences
the modeled cost-effectiveness of annual overflow reduction. Assumptions regarding BMP construction
costs could vary the total cost of full implementation, without O&M, by up to $10 million.
Gl Sizing Criteria
The amount of treatable area, or area routed to BMPs, is another key BMP sizing assumption that
influences the modeled performance of Gl practices. BMPs were sized according to two different
performance criteria (Section 3.5.2) intended to treat 0.75 in. and 1.00 in. of runoff were compared. The
cost-effectiveness curves for the two scenarios are presented below as Figure 3-35.
55%
Downspout Disconnection
—Gl sized for 0.75 in. of runoff
^—Gl sized for 1.00 in. of runoff
0 2 4 6 8 10 12 14 16 18 20 22
BMP Construction Cost( $ Million )
Figure 3-35. Comparison of Gl cost-effectiveness curves size to treat 0.75 in. and 1.00 in. of runoff.
Both Gl cost-effectiveness curves tracked almost identically until about 35 percent overflow reduction
when the 1.00 in. treatment scenario begins to show marginally better performance. The curve for the
3-47
-------
1.00 in. treatment scenario maintains the same final trajectory tracking slightly above the 0.75 in.
treatment scenario and to a higher final cost with marginal benefit in terms of overflow volume reduction.
In the two scenarios, only the physical BMP footprint was changed. The amount of treatable impervious
area in the watershed was fixed at 54.6 percent (Table 3-14). While increasing the size of the BMPs
increases the opportunity to provide overflow volume reduction, the achievable overflow volume
reduction remains limited by the amount of impervious area routed to the BMP.
3.5.9. Optimization Summary and Conclusions
The optimization analysis of gray and green BMP opportunities for CSO mitigation in the 019 sewershed
yielded some unique insights in terms of the implication of the optimization problem formulation and
objective function on model results, and the influence of BMP cost and sizing assumptions on simulation
results. This case study specifically (1) demonstrates techniques for replicating a baseline hydraulic
model used in CSO sewersheds; (2) highlights the sensitivity of key BMP hydrologic parameters in a
factorial experimental design framework; and (3) evaluates the cost-effectiveness of gray infrastructure
versus GI for CSO management while testing modeling assumptions related to cost functions and BMP
sizing.
In Section 3.5.7 the importance of the optimization objective on the optimization result is observed. For
this case study, the optimization objective was to reduce the number of annual overflow events. The
results were also summarized in terms of percent overflow volume reduction for reference purposes. The
cost-effectiveness curves presented highlighted that there was not necessarily a one-to-one relationship
between overflow volume and the number of discrete overflow events. Downspout disconnections and
the introduction of GI practices initially appear to provide a more cost-effective means of achieving a 30
to 35 percent reduction in overflow volume. However, the practices provide marginal benefit in terms of
meeting the objective target of eight overflow events per year.
Nevertheless, there are some indirect benefits associated with total volume reductions that were not
specifically addressed in this case study because they were not considered during the optimization. Two
immediate examples are apparent. First, water quality improvement was not an optimization objective.
GI might provide significant water quality benefits that might not directly affect CSO reduction. The
volume reduction associated with GI use in conjunction with gray could actually have beneficial impact
downstream in the receiving water even though it might not reduce the number of overflows. Given the
same downstream assimilative capacity, reduction in overflow volume can translate into lower pollutant
concentrations in the receiving waterbody for the overflow discharge. Second, recall that flow volume
that does not crest the weir at the regulator is directed through the 24 in. orifice (Section 3.3.4) to the 38th
Street pump station where it is conveyed to a wastewater treatment plant. This study did not attempt to
quantify the cost-benefit relationship between decreased flow volumes and decreased operating costs at
the treatment plant. It is expected that additional savings could be realized in terms of reduced pumping
costs and reduced treatment costs at the wastewater treatment plant.
This case study also demonstrated the flexibility that the SUSTAIN modeling framework provides for both
formulating and investigating management questions in the context of CSO mitigation planning.
Similarities between commonly accepted modeling approaches allow for easy replication of existing
sewershed models. Network simplification can substantially improve model run-time for optimization
while maintaining accuracy over critical conditions. As highlighted in Section 3.3.4, the network
simplification resulted in additional overflow events for the SUSTAIN baseline mode as compared to the
Info Works baseline model. However, review of those events revealed that each fell below the fifth
percentile overflow volume and would likely not affect optimization results, which are driven mainly by
the largest overflow events. Understanding the critical modeling condition, which in this case study is
3-48
-------
complete capture of the ninth largest overflow event, was an important consideration during problem
formulation. This understanding helped to balance setup considerations associated with preserving model
accuracy while significantly reducing model run-time. Section 3.3.5 presented estimated model run-times
for the simplified SUSTAIN model and the Info Works model used as a baseline for replication.
Sensitivity analyses were performed to bracket the range of uncertainty associated with certain important
modeling assumptions and their effect of optimization results. As described in Section 3.5.8, sensitivity
analyses were performed on assumptions related to both BMP cost and sizing criteria (represented by
amount of treatable impervious area). Sensitivity testing of both BMP costs and treatable drainage area
suggest that assumptions made regarding those parameters were highly influential on the optimization
modeling results, with cost assumptions having the greatest observable impact. Varying the cost between
literature values and local bid data showed that GI construction cost could vary by a factor of two or
more.
Thinking about treatable impervious area in the context of the Portland Wharf Storage Basin provides
additional insight regarding the favorable cost-effectiveness of this practice. The storage basin is at the
most downstream point in the network beyond the CSO 019 regulator and receives only overflow volume
from the regulator weir. Essentially, the Portland Wharf Storage Basin can receive flow from 100 percent
of impervious areas in the watershed as compared with only 54.6 percent for GI practices. Its position
beyond the regulator means that while runoff from 100 percent of impervious areas have the opportunity
to reach the storage basin, only overflows will ever reach that point in the network, making the storage
basin a highly specialized practice. In contrast, GI practices are subjected to all runoff volume from
treatable impervious area with no distinction between flows to the treatment plant or overflows.
Finally, it is crucial to interpret the optimization results in light of the findings about problem
formulation, model assumptions and sensitivity of BMP parameters. The results suggest that gray
infrastructure is more cost-effective than GI at achieving the objective of eight overflow events per year
and that it is infeasible to achieve the objective without the use of gray infrastructure. If another objective
was formulated, for instance total annual volume reduction, the expected results and BMP selection could
be substantially different. GI cost functions based on direct BMP construction costs only or derived from
bid data not related to a retrofit demonstration project but rather from new construction with local
precedent for GI might yield more cost-effective GI solutions.
3-49
-------
Chapter 4. Lessons Learned
The case studies presented in this document highlight the power and utility of SUSTAIN as a decision
making tool. By applying the system to the conditions in Kansas City and Louisville, the application
explored hydrologic response, management options, and their effects on a highly urbanized system. The
resulting study provides useful and practical information that can help managers to understand, measure,
and evaluate the benefits of GI in urban watersheds. The lessons learned from these case studies benefit
two audiences -managers/decision makers and practitioners:
• Management Lessons: These case studies were used to address locally derived questions
regarding the selection, placement, and strategy for the use of BMPs to mitigate CSOs. The
results provide a template for the development of similar decision making frameworks in other
regions outside of the case study application area.
• Modeling Lessons: The development of the application framework and execution for the case
studies in Kansas City and Louisville provides a template for similar applications in other
regions. Although there were similar goals, each case study demonstrates how the model can be
adapted and configured to represent the system and formulate the optimization problems based on
local constraints including (1) the available supporting information; and (2) the specified control
targets. The modeling examples presented provide meaningful guidance for the SUSTAIN user
community.
The SUSTAIN model developed for these CSO case studies was based on the existing municipal
collection system models. Efforts included developing a SUSTAIN model which replicated the response
of the municipal model hydrologic response, calibration of a baseline runoff model to match either an
existing model or available flow monitoring data, and the development of a lumped or aggregated model
to better manage the computational run-times required for optimization. In order to address the case
study questions, the SUSTAIN model then incorporated the collection system hydraulic elements at the
regulator and any existing or proposed controls were incorporated into the model. The model was used to
evaluate the case study questions, which focused on optimizing GI and gray infrastructure to achieve a
CSO control target, and comparing the performance and cost of GI to conventional storage facilities at the
downstream end of the systems.
The existing system model was designed to represent a series of scenarios that provide the context for
decision making. Figure 4-1 is a schematic of the general sequence of scenarios used for CSO
management optimization using GI and gray infrastructure.
First, there needs to be a baseline model that represents the existing condition. This model provides the
basis for calibration. Second, it is important to recognize that the baseline for optimization may differ
from the existing condition baseline if there are existing or planned management activities for which
commitments have already been made. In such a case, these activities must be appropriately reflected in
the model and considered as part of optimization baseline. Third, exploratory management scenarios for
optimization should include a mix of different alternatives, each with its respective cost assumptions.
Fourth, the control target must be clearly defined. Finally, optimization is then formulated with the
objectives of achieving the stated control target while minimizing the cost of implementation. The
iterative interpretation of the results of each step will help to guide and refine the optimization objectives.
4-1
-------
Allowable Exceedances
c
CD
O
oo
CD
-Q
E
Existing or
Planned Mgmt
Figure 4-1
Calibration
Baseline
Theoretical construct for CSO management optimization problems
Optimization
Baseline
Gray
Infrastructure
Only
Existing or
Planned Mgmt
Optimization
Scenario 1
Gray
Green
Infrastructure
Existing or
Planned Mgmt
Optimization
Scenario N
Q.
X
LLJ
T3
CD
E
E
o
u
The Louisville case study tested the ability of GI to cost effectively complements or replaces a CSO
storage basin for CSO area 019, a 1,094 acre project area. The criterion for overflow control was
specified as a control level of 8 overflows per year. In Louisville, the alternate gray infrastructure control
was identified as a 6.37 MG storage basin. Opportunities for GI placement were defined by Louisville
MSB and were situated to manage approximately 55% of the tributary area. The case study evaluated the
level of control that could be achieved with GI, its associated cost, and how green and gray infrastructure
could be collectively optimized to accomplish a cost effective solution.
In Kansas City, a 480 acre tributary to CSO 069 was identified by the City as an opportunity for control
using GI. The control target was the elimination of overflows during a specific design storm. In the
development of the City's LTCP, this design storm condition was anticipated to correlate to 6 overflows
per year. Kansas City is currently implementing GI controls in 100 acres of the tributary area. This
condition was reflected in the baseline for the optimization scenario.
For both the Kansas City and Louisville studies, the exploratory scenarios included both green and gray
infrastructure and the associated cost considerations. Constraints and cost of both green and gray
infrastructure were applied consistent with constraints as identified by the municipalities.
4.1. Management Lessons
In both Kansas City and Louisville, the primary management objective was to reduce the frequency of
overflows. There were differences in the problem formulation dependent on site conditions, the current
adoption of BMPs, and the alternate structural (gray infrastructure) solutions identified. Table 4-1
summarizes and compares the key management considerations that influenced the results from the Kansas
City and Louisville case studies. A number of these components are worth noting. First, GI in Kansas
City was constrained in the analysis on the basis of the committed design plan that was under
construction. In other words, GI approaches were already established for the 100 acre pilot area. For this
reason, the implementation of the plan was assumed as part of the baseline for optimization. Future GI
placement in the balance of the tributary area was then extrapolated consistent with the approach applied
4-2
-------
in the initial 100-acre area. The number of new opportunities for BMPs was constrained by the design
plan template, which resulted in only 57 percent of the impervious area runoff reaching any BMP. In
contrast, Louisville had identified potential new GI sites which could be applied throughout the study
area. Second, the control targets for the two case studies differed. The control target for optimization in
the Kansas City case study was 100 percent capture of a critical condition design storm, which was
intended to correlate to 6 overflows per year. Attainment of the allowable exceedance criteria of 6
overflows was also tested using continuous simulation for the typical year represented by 2004. For this
precipitation record there were only 2 or 3 overflows, because the D-storm itself was a conservative
design storm. For the Louisville case study, the target was an allowable exceedance frequency for a
statistically-derived typical year.
Table 4-1. Comparison of case study components that influenced management alternatives
Key management
components
Control target
Optimization baseline
(committed activity)
Exploratory management
alternatives
Spatial constraints for GI
Physical constraints for
gray infrastructure
Case study application
Chapter 2: Kansas City, MO
Design Storm: 100% capture of the critical
condition D-storm
Validation: Continuous simulation for
statistically identified 2004 typical year
Allowable Exceedance: 6-overflows/year
The pilot area (which is about 25% of the
069 sewershed area) has a committed GI
design plan. Actual BMP construction was
already underway at the time of study.
1. Extrapolate GI design plan template to
the remainder of the watershed
2. GI opportunity on private parcels
3. Determine supplemental gray storage
capacity downstream of regulator
City GI design plan constrained to public
rights-of-way as specified by engineering
plan. Based on the configuration of the GI
practices as designed, about 57% of the
total impervious area is tributary to GI
opportunity.
Proposed storage tank and pump station
downstream of regulator outfall. Reduce
the size
Cha pter 3: Louisville, KY
Typical Year: 2001 statistically identified as
a typical year for management
Allowable Exceedance: 8-overf lows/year
Isolate runoff to regulator from immediate
study area for baseline runoff. Remove
diverted interceptor inflow to regulator.
1. Downspout disconnection
2. GI opportunity in sewershed
3. Consider proposed gray storage
design downstream of regulator
GIS map of all potential GI opportunity in
the sewershed. About 55% of the total
impervious area is tributary to GI
opportunity.
Proposed storage tank and pump station
downstream of regulator outfall. Verify
required size to meet management
objectives.
4.1.1. What are some of the factors that most influence cost-effectiveness of both GI
and Gray Infrastructure?
One of the most influential factors affecting the ability of GI to reduce CSO frequency toward an
overflow frequency target is the fraction of the CSO catchment area that can be routed for treatment to the
various BMPs. In Kansas City, the proposed design plan for the pilot area limited GI facilities to public
rights-of-way. Because GI was constrained to public rights-of-way in Kansas City, the potential tributary
area managed is likewise constrained to area that readily drains to those facilities. Figure 4-2 summarizes
the impervious land cover distribution upstream of private and public GI facilities in the Kansas City 069
sewershed. The graph shows that full build-out of proposed GI opportunities would be able to treat 57
percent of the total impervious area.
4-3
-------
* Treated rooftop area is 79% of the total rooftop area.
t Other treated impervious is 63% of impervious area.
Total Area
Impervious Area
Other
Impervious
Untreated
ImperviousArea Distribution
Figure 4-2. Kansas City 069 land cover distribution and impervious area distribution tributary to Gl.
Similarly, Figure 4-3 summarizes the impervious land cover distribution upstream of different GI
facilities in the Louisville 019 sewershed. The GI opportunities that were identified by Louisville MSB
were suitable to manage 55% of the area in the sewershed.
70%
60%
I Total Area
ImperviousArea
J-
* Treated rooftop area is 79% of the total rooftop area.
Green
Parking
Untreated
Area
ImperviousTributary Area Distribution
Figure 4-3. Louisville 019 land cover distribution and impervious area distribution tributary to GI.
Because CSO control targets are typically defined based on overflow frequency, it is important to
maximize the total area controlled. Even a small amount of uncontrolled area can result in CSO
discharges. If CSO control targets also consider annual volume, the total area controlled may not be as
4-4
-------
significant. However with the current case studies, which focus on a frequency target, this hypothesis
was not specifically tested.
As the focus of the case study application is the definition of cost effective solutions for CSO control, the
analysis is highly sensitive to costs of construction as well as life cycle costs. Costs used in these case
studies were based on local municipal data associated with material and construction costs. (Broader
costs, such as O&M, were not available from the municipalities for either GI or alternate gray
infrastructure facilities.) The material and construction costs were found to be highly influenced by the
means of implementation. The Kansas City project was implemented as a retrofit project which resulted
in a number of other project elements being included in the construction contract. This had the impact of
increasing the cost associated with GI implementation. In contrast, projects on private property may have
limited cost to the municipality. For example, roof disconnection was by far the most cost-effective
practice in both the Louisville and Kansas City case studies. In Louisville, the only cost included for this
practice was the labor for a trained technician to certify proper disconnection of a downspout. In fact,
Louisville includes downspout disconnection as part of an incentives program that provides a $100 credit
to individual homeowners for each disconnected downspout. The reason why this practice was shown to
be so cost effective is that for the price of the incentive and/or certification costs, the entire associated
roof area now has an opportunity to infiltrate into the lawn or garden instead of directly draining into the
collection system along with driveway and road runoff.
Table 4-2 is a list that summarizes some of the most influential factors for predicting cost-effectiveness.
Table 4-2. Summary of factors that most influence cost-effectiveness
List of factors
Case study application
Chapter 2: Kansas City, MO
Chapter 3: Louisville, KY
Treated tributary area
(GI vs. gray)
GI constrained to public rights of way (57%
treatment potential for impervious area).
Gray infrastructure is physically located
after the CSO regulator; therefore, volume
provided is applied to 100% of the
tributary area.
Full-GI build-out opportunity results in
55% of impervious area being treated.
Gray infrastructure is physically located
downstream of the regulator and sized
based on overflow volume for intended
control target.
Disconnected
rooftops
Private parcel rooftop area is subject to
capture by rain barrels as specified by the
WinSLAMM study. A very cost-effective
practice; however, it is opportunity limited
(only 15% of private parcel rooftop area
remains that has not been disconnected).
Incentive program includes payment to
private parcel owners for each
disconnected downspout. Only cost
includes incentive payments or
performance certification expenses. Also
limited by opportunity.
Site preparation costs
(retrofit vs. capital
improvement plan (CIP)-
integrated)
Much of the high GI cost was attributable
to Site Preparation costs including curb
installation, traffic control and other costs
included in the construction project but
which may not be directly associated with
BMP construction. As retrofit project,
these costs become associated with GI
implementation.
Sensitivity analyses on GI cost
assumptions show a swing of about 50%
in cost-effectiveness between full-retrofit,
CIP-integrated, and other literature-based
cost figures for GI.
Time-dependent cost
efficiencies
This case study assumed constant cost
components with time. It did not account
for cost efficiencies overtime because of
contractor experience, reduced project
uncertainty, and economies of scale.
Sensitivity analyses provided a range of
cost-effectiveness as a function of
different cost assumptions. This analysis
can provide some indirect insights about
the possible range of efficiency associated
with time-dependent design refinements.
4-5
-------
List of factors
O&M costs
Case study application
Chapter 2: Kansas City, MO
This case study focused on capital costs for
optimization purposes. It did not consider
O&M costs. Other studies have shown
that wide-spread adoption of Gl has a job
creation benefit.
Cha pter 3: Louisville, KY
This case study also focused on capital
costs for optimization; however, it is
recognized that long-term O&M can also
have an impact on cost-effectiveness
comparisons between BMPs.
4.1.2. How does the control target affect cost-effectiveness of management
alternatives?
Flows discharged by green and gray infrastructure affect different portions of the hydrograph. As a result
of this difference, the amount of control accomplished by the same volume removal differs. First,
management capacity placed near the origin of runoff will tend to capture the rising limb of the
hydrograph, whereas capacity placed downstream of a CSO regulator is primarily focused on the flow
which cannot be captured by the wastewater collection system. Second, the mode by which the volumes
are dewatered affects the amount of water that can be potentially managed by the facility. GI practices
that rely on infiltration may require more time to regain capacity than gray practices, although this is
dependent on the characteristics of each system. Figure 4-4 shows typical capture and dewatering modes
of GI and gray infrastructure capacity in the context of a CSO collection system.
Overflow Target
Overflow Target
Infiltration/ET
Compare
Overflow Target
'ompat
T
Dewatering
Figure 4-4. Typical capture and dewatering modes of GI and gray BMPs in a CSO collection system.
The selected control target for optimization has the greatest impact on the mode of treatment that is
considered cost-effective within the context of GI versus gray infrastructure in a CSO collection system.
If the control target is to minimize overflow frequency, gray infrastructure may be able to be applied more
strategically to address the top of the hydrograph, which is typically the critical condition associated with
overflows. In a frequency context, an overflow event still counts as one overflow regardless of amount of
4-6
-------
water that actually discharges. On the other hand, when the control target is overflow volume, the volume
controlled and associated with the GI practice will more directly correlate to overflow volume. The GI
practices may have additional benefits when considered in the context of the broader interceptor system
and wastewater treatment plant, although this was beyond the scope of these case studies.
In the Louisville case study, flow attenuation associated with GI provides a reduction in volume that does
not translate directly into a reduction in the number of overflows. With full build-out of GI (within the
placement limitations previously discussed), a 50 percent reduction in the overflow volume was
projected, although this scenario only reduced 12 out of 49 overflows (about 24 percent reduction in the
number of overflows). For GI, volume reduction still outpaced frequency reduction; however, the
difference in the percentage reduction between volume and frequency were not as disparate for gray
infrastructure as they were for green.
Current trends in the industry, as reflected in recent consent decrees and proposed plans, are to focus to a
greater degree on volume control as opposed to frequency targets. This is the control approach in both
Philadelphia and Cincinnati, which have approved LTCPs (PADEP, 2011). Similarly; the Northeastern
Ohio Regional Sewer District is looking to couple green with gray to get additional, cost-effective volume
reduction in their collection system (USEPA, 2011). Their management questions are more related to
whether adding green or adding more gray is a better investment for achieving this goal. The industry
trends and rationale for adopting volume-based standards is understandable given what is known about
collection system responses to different modes of treatment.
4.1.3. Can GI be used effectively to complement existing or planned Gray?
GI may be used to complement the benefits of gray infrastructure for CSO control. Because GI captures
the rising limb of the hydrograph, it tends to reduce the overall volume of runoff that reaches the
regulator. In these case studies, coupling gray controls with GI also tended to reduce the overflow
volumes more effectively than the addition of gray storage capacity. In Kansas City various scenarios
reflecting gray infrastructure or a mix of gray with green were defined that had comparable anticipated
cost. The performance was tested in the continuous simulation validation in the Kansas City study. For
the two largest allowable overflow events, an interesting trend was observed when comparing
management scenarios using only gray infrastructure versus supplementing with different amounts of GI.
For the two largest allowable exceedance events, Figure 4-5 presents a comparison of overflow volume
for (1) gray only; (2) GI on public rights-of-way + gray; and (3) GI on public rights-of-way and on private
parcels + gray. Recall that the number of overflow events for typical year 2004 was fewer than the six
allowable overflow events. However, applying more GI resulted in (1) a smaller required gray capacity;
and (2) a progressively smaller amount of water overflowing the storage facility. The lesson learned is
that in places where gray infrastructure controls already exist, adding GI to supplement controls upstream
in the collection system can potentially be an effective way of reducing overflow volumes from the
system. The effectiveness of such an approach in other areas will depend on the local setting, including
the physical constraints, management opportunities, cost information, and other site-specific
considerations.
4-7
-------
_ 8
V)
C
= 7
(0
c 6
o
i 5
g 4
1 3
S !
O
0
Gray Only
Public Green +Gray
Public + Private Green + Gray
[-17%]
[-30%] I
[-9.5%] [-22%]
Event 1: (1.75 in., 0.79 in./hr)
8/23/048:00 - 8/24/0421:00
Event 2: (1.98 in., 0.87 in.hr)
9/5/0417:00 - 9/6/048:00
Figure 4-5. Comparison of overflow volumes for Kansas City exploratory management scenarios.
4.2. Modeling Lessons
SUSTAIN uses large numbers of model runs in order to evaluate the optimization scenario. Therefore,
model run-times are a critical concern in the application of SUSTAIN. The best practice in the application
of SUSTAIN mode ling is to rely on the simplest approach that is able to adequately represent the system at
hand, in light of the application objectives. Because hydrology and hydraulics models like those used in
SUSTAIN are a theoretical construct of natural/man-made hybrid systems, they need to be tested to verify
that they provide a reasonable representation of the proposed system. The model should be sensitive to
and reflective of the key processes that most influence the management decisions that will be explored
using the model. The model should also strategically simplify the problem, taking into consideration the
relevant information that is needed, while recognizing and quantifying any error propagation in as much
as it affects management decisions that are to be made using the model.
This subsection summarizes some of the modeling lessons learned through the setup and execution of the
Kansas City and Louisville CSO optimization models.
4.2.1. What are some of the critical data that are required for performing these
evaluations?
This case study effort highlighted the importance of a representative baseline model. The model baseline
is the foundational element upon which all subsequent analyses depend. It also tends to be the place
where the largest body of supporting data exists for characterization. For both case studies, there were
good precipitation time series, flow monitoring data for a number of storms, and good spatial data to
characterize land use and impervious cover. A significant amount of effort was invested to ensure that the
baseline watershed model could (1) adequately and consistently predict the temporal patterns (volume,
4-8
-------
peak flow, and timing) of coincident observed historical records; and (2) be shown to be representative of
a wide range of storm conditions, including critical condition CSO events. The model was tested and
confirmed using a weight-of-evidence approach that compared both modeled versus observed time series
plots and computed statistical metrics.
One of the more significant hydrologic factors in determining the impact of various BMPs is the fate of
precipitation that is generated from impervious areas. In all drainage systems there is a reduction in
runoff losses throughout the system that occurs from the time precipitation hits an impervious surface to
the time it reaches the regulator. These diffuse losses may include things like disconnected
imperviousness, surface ponding or flooding, or possibly even pipe exfiltration. It is important to
recognize how these are reflected in the baseline model because improvements to the collection system to
mitigate this behavior may tend to influence the amount of water delivered to the sewer system, and the
CSO regulator. As BMP projects are implemented, some of these drainage inefficiencies may be
corrected—this must be considered in sizing the BMP network. In other words, fixing the baseline
drainage problems could increase the rate and volume of runoff, resulting in more flow to be managed
than currently predicted.
The original Kansas City and Louisville model configurations dealt with impervious areas differently,
which in turn impacted the ability to consider potential drainage changes as BMPs were implemented. In
the Kansas City baseline model, DCIA in rights of way and for adjacent driveways, etc. treated all
imperviousness as connected. For parcels (particularly rooftop areas), DCIA was estimated and applied
based on surveys of disconnected downspouts. Therefore, the effective impervious area was physically
represented at the source, and this representation could be maintained in the SUSTAIN model. In the
Louisville case study, the baseline model originally defined an effective imperviousness approach,
whereby the total physical footprint distribution of pervious and impervious area was adjusted to account
for a reduction in the impervious area that was effectively connected to the system. However, there was
no specific identification of which areas were directly connected. Therefore, the optimization baseline
was modified to explicitly account for runoff origination by adding a reduction term to account for diffuse
losses throughout the network. Careful attention was paid to how the baseline models were configured
and applied because these assumptions will often play a significant role in how results are interpreted.
The placement opportunities for BMPs define the extent to which GI can beneficially impact flow volume
and overflow frequency. In each of the case study communities, limitations were placed on the locations
available for GI placement, which in turn led to a definition of the maximum potential effectiveness of the
GI in controlling CSO discharges. Some of these limitations were physical constraints of the landscape
that were derived during the engineering and design process. Other limitations were defined based on
land use or ownership criteria resulting from the local decision making process. The restrictions placed
on GI must be understood in order to evaluate the management scenarios.
Because optimization measures cost-effectiveness, there is a strong dependence on the available BMP
cost information. These case studies included costs associated with both GI and gray infrastructure.
Uncertainties in the local cost data used can strongly influence the management conclusions. Cost data
may reflect different levels of precision (i.e. planning, engineering estimate based on design, bid prices),
the implementation year (affecting the cost index), or the types of costs included in the data presented
(construction, engineering, contingencies, etc.). The experience with BMP retrofit projects is often that
other infrastructure improvements are incorporated into these projects, which influence the cost basis for
comparison. Similarly, gray infrastructure costs need to be defined over the full potential range of
application in order to fully assess the tradeoff between green and gray approaches. Decisions on how
costs will be applied for GI on private property likewise need to be addressed.
4-9
-------
The Louisville case study was used to compare the potential impact of various cost methodologies for GI.
Three different cost scenarios for GI were evaluated based on (1) data from retrofit projects; (2) CIP
projects that included GI; and (3) literature values. Total costs varied widely based on the costing
methodology used.
Each of these issues needs to be identified and an approach selected. The best way to constrain
propagation of uncertainty in this type of modeling is to constrain uncertainty associated with the key
building blocks of the optimization model. Certain steps were taken during model development to
establish a consistent basis for model extrapolation to other areas that were not monitored.
4.2.2. How detailed does the model needs to be in order to properly represent the
system?
Another key aspect of the study involved managing model complexity. It is important to keep in mind at
all times the purpose of the modeling study, as well as the questions that need to be answered. The model
should only be as complex as necessary to address modeling objectives and answer the management
questions. Especially in the context of optimization model development, there is a fine balance to be
struck between model complexity and run-time efficiency. SUSTAIN provides the aggregate BMP
approach as a way to simplify the complexity of the network while preserving the robust responsiveness
of the system being modeled; however, as with any model, the burden of proof falls on the modeler to
prove its validity. The Kansas City study tested the sensitivity and behavior of an aggregated routing
representation alongside a fully-articulated network. The Louisville case study model included a
simplified drainage network that replicated the Info Works baseline model in SUSTAIN. Both case studies
demonstrated examples of model simplification that were robust enough to adequately address the volume
reduction optimization objective, while significantly reducing simulation time.
4.2.3. How can one demonstrate that a model is adequately representative of the
system?
Sensitivity tests provide an informative way of showing a range of responsiveness associated with key
assumptions and processes. After going through the process of baseline model testing and confirmation
with observed data, sensitivity analyses were performed to bracket the range of uncertainty associated
with certain important modeling assumptions, and their effect of optimization results. For the Louisville
case study, sensitivity tests were performed on common BMP model parameters to identify which
assumptions had the most influence on model results. A range of GI cost assumptions were also applied
to show the resulting impact on predicted GI cost-effectiveness. For the Kansas City case study,
sensitivity analyses were performed on both model simulation time and antecedent moisture conditions
associated with the critical condition design storm. Cost-effectiveness curves for optimization solutions,
and associated capital cost-benefit estimates, were presented as a range of variability attributable to the
bands of uncertainty associated with the underlying modeling assumptions. As a final test, the
optimization solutions were also validated using continuous simulation for an average precipitation year
(2004). Model validation showed that the use of a critical condition storm also inherently provided some
margin of safety for optimization, because the D-storm is a conservative representation of a frequency
target. The validation test confirmed that CSO mitigation objectives had been achieved by optimizing to
the design storm. All of these independent tests and validations were performed as part of a weight-of-
evidence approach to establish model defensibility.
4-10
-------
4.2.4. How is the model applied in an iterative, adaptive process?
The iterative, cyclical nature of the model application process was highlighted through these case studies.
Two formative drivers frame the typical SUSTAIN application process. The first driver is the set of
management questions to be addressed. Thoughtfully outlining the management questions is essential to
the development of the appropriate model application, and the selection of the appropriate model
complexity, processes to be simulated, and required testing and analysis. Essential the understanding of
management questions is the financial implications of the decision process. Second, the management
questions must be translated into numeric objectives that are used for optimization. These two formative
drivers inform all subsequent decisions regarding the model setup including data, complexity, and
interpretation of results. As part of the application process is the identification of data collection and
monitoring needs to support both the application and future testing of the model results.
Figure 4-6 is a conceptual flowchart of the SUSTAIN application process. This flowchart shows two
feedback loops. If and when new information becomes available that better characterizes the baseline or
critical conditions, the model can be updated to incorporate new information. If new management
questions arise, or if the results provide new insights into the management options and new formulation
can be tested.
Management
Objectives
Numeric Control
Targets
Insufficient Data
Modeling
Objectives &
Problem
Formulation
Data
Collection
and Analysis
Applying SUSTAIN
Optimization
Baseline Calibration
Figure 4-6. SUSTAIN application sequence
During the case study applications, this feedback process was illustrated. For example, for the Kansas
City Case Study, an XP-SMMM model was provided as the baseline model for optimization; however, a
revised calibration was necessary because new and improved monitoring data became available that better
reflected the critical conditions associated with management objectives.
4-11
-------
Chapter 5. Conclusions and Recommendations
SUSTAIN is a comprehensive modeling system that provides users with the ability to evaluate the cost
effectiveness of urban storm water management techniques across a wide range of conditions - various
urban densities, climate, and geologic settings. As the SUSTAIN modeling system begins to be applied to
address management questions, the range of applicability and functionality can be demonstrated. As
illustrated by the case study applications included in this report, optimization tools can be very powerful
when combined with hydrologic modeling and cost analysis in the SUSTAIN modeling framework.
The recent enhancements to SUSTAIN documented in this report and applied during the case study
development process, provided selective improvements to the functionality and flexibility of the modeling
system. In particular, the addition of a sub-hourly time step improved the ability of SUSTAIN to predict
hydrologic response and peak flow from design storms used as a basis for planning many CSO and
stormwater programs. Verification of the aggregate BMP approach supported the use of the model in
Kansas City and Louisville. It also provided guidance for other regions where users want to evaluate the
benefits of many, in some cases hundreds or thousands, of smaller BMPs across a large catchment.
Even with the new enhancements and tools provided in SUSTAIN, applying the system to a catchment
should not be considered an automated or simple task, instead it requires careful formulation of the
management questions and the optimization objectives. Set up of the model, as demonstrated in the case
study applications, also requires deciding on the appropriate level of detail, such as the number of sub-
catchments and resolution used to represent BMPs, as well as the associated data collection, model
testing/calibration, and development of the baseline condition. Application of the model optimization
tools is iterative, and users may want to consider testing multiple cost and management assumptions
before developing their recommendations.
These case studies have provided users with an overview of two urban settings in Kansas City and
Louisville, and demonstrated how SUSTAIN was used to support a cost-benefit evaluation of CSO
management alternatives. The two case studies have also shown how SUSTAIN was used to analyze,
streamline, and extrapolate BMP representation throughout the respective study areas; and demonstrated
how to evaluate various combinations of green and gray management alternatives. The case study
applications led to the follow general observations and conclusions:
• SUSTAIN is a comprehensive decision support tool with many useful features and functions.
Successful and meaningful application largely depends on accurate representation of the baseline,
BMP alternatives, and the associated BMP costs.
• SUSTAIN application process is iterative and adaptive, meaning that once the SUSTAIN modeling
framework is established, it can be adapted to answer various management questions and test
underlying assumptions.
• Model simplification becomes critical when optimization is applied to a larger area or when
multiple smaller BMPs are distributed widely across a catchment. The aggregate BMP concept
and utility provided in SUSTAIN is proven to be a viable and useful technique in the evaluation of
the benefit of stormwater management practices, especially smaller GI practices, across a large
area. When the aggregate BMP tools are used, the appropriate aggregation spatial scale should be
carefully selected to maintain reasonable predictive capability and accuracy.
• The optimization process is highly sensitive to BMP cost data used in selecting solutions for each
application. As a result, performance of sensitivity analysis and evaluation of cost control
measures or economies of scale are recommended wherever SUSTAIN is applied.
5-1
-------
Chapter 6. References
Bedient, P. B. and Huber, W.C. 1992. Hydrology and Flood Plain Analysis. Addison-Wesley, New
York.
Berthouex, P.M., and L.C. Brown. 2002. Statistics for Environmental Engineers. 2nd ed. CRC Press,
Boca Raton, FL.
Bicknell, B.R., J.C. Imhoff, J.L. Kittle, Jr., T.H. Jobes, and A.S. Donigian, Jr. 2001. Hydrological
Simulation Program—FORTRAN, Version 12, User's Manual. U.S. Environmental Protection
Agency, National Exposure Research Laboratory, Athens, GA., in cooperation with U.S.
Geological Survey, Water Resources Division, Reston, VA.
Burns & McDonnell. 2009. Draft Conceptual Design Report: Middle Blue River Pilot Study. Prepared
for Water Services Department, Kansas City, MO, by Burns & McDonnell Engineering
Company, Inc., Kansas City, MO.
CWP (Center for Watershed Protection). 2007. Urban Subwatershed Restoration Manual Series,
Manual 3 - Urban Stormwater Retrofit Practices. Center for Watershed Protection, Ellicott City,
MD.
Dunne, T., and L.B. Leopold. 1978. Water in Environmental Planning. W.H. Freeman and Company,
San Francisco, CA.
Haan, C.T. 1972. A water yield model for small watersheds. Water Resources Research 8(1): 58-69.
Huber, W.C., and R.E. Dickinson. 1988. Storm Water Management Model Version 4, User's Manual.
EPA 600/388/OOla (NTIS PB88-236641/AS). U.S. Environmental Protection Agency, Athens,
GA.
KSU (Kansas State University). 2010. Measuring Evapotranspiration in Urban Irrigated Lawns:
Current Findings and Future Research. K-State Turfgrass Research 2010, Report of Progress
1035. Kansas State University, Agricultural Experiment Station and Cooperative Extension
Service, Manhattan, KS.
Lai, F., L. Shoemaker, K. Alvi, J. Riverson, and J. Zhen. 2010. Current Capabilities and Planned
Enhancements of SUSTAIN. In Proceedings, World Environmental & Water Resources
Congress 2010, Challenges of Change, Providence, RI, May 16-20, 2010. Environmental &
Water Resources Institute (EWRI) of ASCE, Reston, VA.
LOJIC (Louisville/Jefferson County Information Consortium). 2003. GIS Database.
Louisville/Jefferson County Information Consortium, Louisville, KY.
MSD (Louisville-Jefferson County Metropolitan Sewer District). 2007. Hydraulic Sewer System
Modeling Guideline Manual. Louisville-Jefferson County Metropolitan Sewer District,
Louisville, KY.
MSD. 2008. Combined Sewer Overflow Fact Sheet: CSO 019. Louisville-Jefferson County
Metropolitan Sewer District, Louisville, KY.
6-1
-------
MSB. 2009. Integrated Overflow Abatement Plan, Final CSO Long-term Control Plan Volume 1 and 2
of 3. September 30, 2009. Available at: http://www.msdlouky.org/projectwin/ioap.htm. Accessed
June 13th, 2011.
MSB. 2010. Comprehensive Annual Financial Report. Louisville-Jefferson County Metropolitan Sewer
Bistrict, Bivision of Budget and Finance, Louisville, KY.
MSB. 2011. Preliminary Resolution by the Board of the Louisville and Jefferson County Metropolitan
Sewer District Amending Its Schedule of Rates, Rentals and Charges for Wastewater and
Drainage Services Pursuant to KRS Chapter 76.
Nash, J.E., and J.V. Sutcliffe. 1970. River flow forecasting through conceptual models part I—A
discussion of principles. Journal of Hydrology 10(3):282-290.
NRCS (Natural Resources Conservation Service), USBA (United States Bepartment of Agriculture).
2006. Soil Survey Geographic (SSURGO) Batabase for Louisville KY.
PABEP (Pennsylvania Bepartment of Environmental Protection). 2011. Consent Order and Agreement.
June 1, 2011, with appendices.
. Accessed July 29, 2011.
Penman, H.L. 1948. Natural evaporation from open water, bare soil, and grass. In Proceedings of the
Royal Society London A193:120-146.
Pitt, R., and J. Voorhees. 2010. Modeling Green Infrastructure Components in a Combined Sewer Area.
In Dynamic Modeling of Urban Water Systems, W. James ed., Monograph 18. CHI, Guelph, ON
Canada.
Rossman, L.A. 2005. Stormwater Management Model User's Manual, Version 5.0 EPA/600/R-05/040.
U.S. Environmental Protection Agency, Water Supply and Water Resources Bivision, National
Risk Management Research Laboratory, Cincinnati, OH.
Schueler, T., B. Hirschman, M. Novotney, and J. Zielinski. 2007. Manual 3: Urban Stormwater Retrofit
Practices. Center for Watershed Protection, Ellicott City, MB.
Shoemaker, L., J. Riverson, K. Alvi, J. X. Zhen, S. Paul, and T. Rafi. 2009. SUSTAIN—A Framework
for Placement of Best Management Practices in Urban Watersheds to Protect Water Quality.
EPA/600/R-09/095. U.S. Environmental Protection Agency, Water Supply and Water
Resources Bivision, National Risk Management Research Laboratory, Cincinnati, OH. <
http://www.epa.gov/nrmrl/pubs/600r09095/600r09095.pdf>. Accessed August 13, 2011. <
http://www.epa.gov/nrmrl/pubs/600r09095/600r09095app.pdf>. Accessed August 13, 2011.
Strand Associates, Inc. 2010. Office of Employment and Training Green Demonstration Project. Besign
Plans 100% Submittal. Prepared by Strand Associates, Inc., Louisville, KY, for Louisville-
Jefferson County Metropolitan Sewer Bistrict, Louisville, KY.
URS (URS Corporation). 2010. Middle Blue River-Green Solutions Pilot Project WO%Plan. Prepared
for Water Services Bepartment - Engineering Bivision, Kansas City, MO, by URS Corporation,
Overland Park, KS.
6-2
-------
USEPA (U.S. Environmental Protection Agency). 1998. Estimation of Infiltration Rate in the Vadose
Zone: Compilation of Simple Mathematical Models Volume I. U.S. Environmental Protection
Agency, Washington, DC.
USEPA. 1999. Preliminary data summary of urban storm water best management practices. EPA
821/R-99/012. U.S. Environmental Protection Agency, Office of Water, Washington, DC.
USEPA. 2009. Self-extracting installation program for SUSTAIN 1.0 (EXE).
. Accessed July 30, 2011.
USEPA. 2010. Economic Benefits of Runoff Controls. U.S. Environmental Protection Agency,
Washington, DC.
USEPA. 2011. United States and State of Ohio v. Northeast Ohio Regional Sewer District.
. Accessed June 14,
2011.
WaPUG (Wastewater Planning Users Group). 2002. WaPUG Code of Practice for the Hydraulic
Modelling of Sewers, 3rd ed. Watershed Planning Users Group, Wallingford, U.K.
Woodland Direct. 2011. 50-100 Gallon Rain Barrels. . Accessed May 31, 2011.
WSD (Water Services Department). 2006. Summary of Design Storms for CSS Areas. Water Services
Department, Kansas City, MO.
WSD. 2008. Green Alternatives for Outfalls 059 & 069. Water Services Department, Kansas City, MO.
WSD. 2009. Overflow Control Plan. Water Services Department, Kansas City, MO.
WSD. 2010. Middle Blue River Green Solutions Pilot Project. 100% Design Plans: Contract No. 1036.
Water Services Department, Kansas City, MO.
6-3
------- |