:PA/600/R-09/095 | September 2009 | www.epa.gov/nrmrl
United States
Environmental Protection
Agency
SUSTAIN - A Framework for Placement of Best Management
Practices in Urban Watersheds to Protect Water Quality
REPORT
T - •
••I
Office of Research and Development
National Risk Management Research Laboratory - Water Supply and Water Resources Division
-------
SUSTAIN—
A Framework for Placement of Best Management Practices in
Urban Watersheds to Protect Water Quality
by
Leslie Shoemaker, Ph.D.
John Riverson Jr.
Khalid Alvi
Jenny X. Zhen, Ph.D., P.E.
Sabu Paul, Ph.D., P.E.
Teresa Rafi
Tetra Tech, Inc.
10306 Eaton Place, Suite 340
Fairfax, VA 22030
In Support of
EPA Contract No. GS-10F-0268K
Project Officer
Dr. Fu-hsiung (Dennis) Lai
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
-------
Disclaimer
The U.S. Environmental Protection Agency, through its Office of Research and Development, funded and
managed, and collaborated in the research described herein. It has been subjected to the Agency's peer
and administrative review and has been approved for publication. Any opinions expressed in this report
are those of the author(s) 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 that mandate, EPA's research program is
providing data and technical support for solving environmental problems today and building a 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
subsurface 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
ecosystems. 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 document has been produced as part of the laboratory's strategic long-term research plan. EPA's
Office of Research and Development has made it available to help the user community and to link
researchers with their clients.
Sally Gutierrez, Director
National Risk Management Research Laboratory
-------
Abstract
Watershed and stormwater managers need modeling tools to evaluate alternative plans for water quality
management and flow abatement techniques in urban and developing areas. A watershed-scale, decision-
support framework that is based on cost optimization is needed to support government and local
watershed planning agencies as they coordinate watershed-scale investments to achieve needed
improvements in water quality.
The U.S. Environmental Protection Agency (EPA) has been working since 2003 to develop such a
decision-support system. The resulting modeling framework is called the System for Urban Stormwater
Treatment and Analysis INtegration (SUSTAIN). The development of SUSTAIN represents an intensive
effort by EPA to create a tool for evaluating, selecting, and placing BMPs in an urban watershed on the
basis of user-defined cost and effectiveness criteria. SUSTAIN provides a public domain tool capable of
evaluating the optimal location, type, and cost of stormwater BMPs needed to meet water quality goals. It
is a tool designed to provide critically needed support to watershed practitioners at all levels in developing
stormwater management evaluations and cost optimizations to meet their existing program needs. Due to
the complexity of the integrated framework for watershed analysis and planning, users are expected to
have a practical understanding of watershed and BMP modeling processes, and calibration and validation
techniques.
SUSTAIN incorporates the best available research that could be practically applied to decision making,
including the tested algorithms from SWMM, HSPF, and other BMP modeling techniques. Linking those
methods into a seamless system provides a balance between computational complexity and practical
problem solving. The modular approach used in SUSTAIN facilitates updates as new solutions become
available.
One major technical requirement for SUSTAIN is the ability to evaluate management practices at multiple
scales, ranging from local to watershed applications. The local-scale evaluation involves simulations of
individual BMPs and analyses of the impact of various combinations of practices and treatment trains on
local water quantity and quality. The larger-scale evaluation could involve implementing hundreds or
thousands of individual management practices to achieve a desired cumulative benefit. The required
simulations and cost comparisons of such large-scale, distributed BMP options place significant
challenges on the computational accuracy and simulation time for system modeling. SUSTAIN
incorporates an innovative, tiered approach that allows for cost-effectiveness evaluation of both
individual and multiple nested watersheds to address the needs of both local- and regional-scale
applications.
Previously available modeling tools are significantly limited with respect to simulation of sediment
generation and its fate through natural runoff and treatment at a BMP. SUSTAIN partially resolves these
sediment routing issues by considering three sediment fractions (i.e., sand, silt, and clay), but this
IV
-------
approach remains a compromise because the state-of-the-art knowledge and the needed monitoring data
are still limited.
The SUSTAIN framework provides a comprehensive system with a modular structure that facilitates the
incorporation of improved technologies in optimization, BMP simulation, and computational efficiency.
A flexible integration and implementation of these improved methods and algorithms will be the focus of
further enhancements to SUSTAIN. Expanding the SUSTAIN capabilities will allow users to choose the
level of complexity and simulation detail that best suits project needs. EPA intends to support expansion
of the capabilities and functionalities of the system to meet continuing water quality goals and the needs
of the user community.
This document describes the rationale for developing the framework and the uses of the framework;
explains the system's design, structure, and performance; details the underlying methods and algorithms
that provide the framework's predictive capabilities; and demonstrates the framework's capabilities
through two case studies.
-------
Chapter 1 Introduction
1.1. Proj ect Rationale ..
1.2. Overview of SUSTAIN.
1.2.1. Structure of SUSTAIN
1.2.2. Multiple Scale Application Features.
1.3. The Role of SUSTAIN in Watershed Applications...
1.3.1. TMDL Development and Implementation
1.3.2. Evaluation of GI Practices as Part of a CSO Control Program.
Table of Contents
-1
-1
-3
-3
-6
-7
-8
-9
1.4. SUSTAIN Application Process 1-10
1.5. About this Report 1-13
Chapter 2 SUSTAIN Design and Structure 2-14
2.1. Framework Manager 2-14
2.2. Simulation Modules 2-16
2.3. Optimization Module 2-20
2.4. Post-Processor 2-22
2.5. Summary 2-24
Chapter3 Simulation Methods and Algorithms 3-25
3.1. Framework Manager 3-27
3.1.1. Data Management Component 3-27
3.1.2. BMP Site Selection 3-28
3.1.3. Routing Network 3-30
3.2. Land Module 3-31
3.2.1. Weather Component 3-33
3.2.2. Hydrology Component 3-35
3.2.3. Water Quality Component 3-41
3.2.4. Important Considerations and Limitations: Land Module 3-47
3.3. BMP Module 3-49
3.3.1. BMP Simulation Component 3-52
3.3.2. Overland Flow Routing and Pollutant Interception 3-59
3.3.3. Aggregate BMP Component 3-63
3.3.4. BMP Cost Database Component 3-65
3.3.5. Summary of Management Practices and Treatment Processes in SUSTAIN 3-72
3.3.6. Important Considerations and Limitations of the BMP Module 3-72
3.4. Conveyance Module 3-74
3.4.1. Methodology 3-74
3.4.2. Important Considerations and Limitations of the Conveyance Module 3-82
3.5. Optimization Module 3-83
3.5.1. Problem Formulation 3-83
3.5.2. Optimization Algorithms 3-86
3.5.3. Regional Application 3-90
3.5.4. Important Considerations and Limitations: Optimization Module 3 -92
3.6. Post-Processor for Results Interpretation 3-93
3.6.1. Storm Event Classification 3-93
3.6.2. Storm Event Viewer 3-95
3.6.3. Storm Performance Summary 3-97
VI
-------
3.6.4. Cost-effectiveness Curve 3-100
3.6.5. Important Considerations and Limitations: Post-Processor 3-103
Chapter 4 Case Studies 4-104
4.1. Upper North Branch Oak Creek Watershed 4-104
4.1.1. Project Setting 4-105
4.1.2. Data Collection and Analysis 4-108
4.1.3. Project Setup 4-108
4.1.4. Optimization and Results Analysis 4-112
4.1.5. Summary 4-120
4.2. Little Rocky Run Watershed 4-121
4.2.1. Project Setting 4-121
4.2.2. Data Collection and Analysis 4-122
4.2.3. Project Setup 4-122
4.2.4. Optimization and Results Analysis 4-128
4.2.5. Summary 4-137
Chapters References 5-138
Appendices
Appendix A. Needs Analysis and Technical Requirements A-143
Appendix B. Model Evaluation and Selection B-151
Appendix C. Summary of the Optimization Technical Panel Meeting C-177
Appendix D. Appendix References D-185
VII
-------
List of Figures
Figure 1-1. Overview of SUSTAIN components
Figure 1-2. Watershed assessment points
Figure 1-3. Sample cost-effectiveness curve
Figure 1-4. SUSTAINS multiple scales of application
Figure 1-5. Tiered application of SUSTAIN for developing cost-effectiveness curves....
-4
-5
-5
-6
-7
Figure 1-6. Using SUSTAIN in the watershed planning process
Figure 1-7. SUSTAIN application process 1-10
Figure 2-1. SUSTAIN components and flow chart 2-15
Figure 2-2. Comparison of runtime for various simulation units and periods 2-19
Figure 2-3. Comparison of runtime for 1-yr simulation of various units 2-19
Figure 2-4. Scatter Search evaluation scenario results 2-21
Figure 2-5. NSGA-II evaluation scenario results 2-21
Figure 2-6. Example storm event classification graph 2-22
Figure 2-7. Example storm event viewer graph 2-23
Figure 2-8. Example performance summary report graph 2-23
Figure 2-9. Example cost-effectiveness curve 2-24
Figure 3-1. The routing network showing the connections among the simulation components 3-30
Figure 3-2. Schematic showing the land simulation processes 3-31
Figure 3-3. Conceptual view of surface runoff. 3-37
Figure 3-4. Two-zone groundwater model adapted from SWMM 3-39
Figure 3-5. Two-zone soil moisture storage under low water table condition 3-40
Figure 3 -6. Two-zone soil moisture storage under high water table conditions 3-41
Figure 3-7. Schematic of sediment production and removal processes 3-42
Figure 3-8. A schematic showing the BMP simulation processes modeled in SUSTAIN. 3-49
Figure 3-9. Wetland/lake/reservoir weir and orifice outflow 3-52
Figure 3-10. Processes considered in anunderdrain structure 3-56
Figure 3-11. Conceptual pond shapes simulated by Persson etal. (1999) 3-58
Figure 3-12. Filter description for the sediment transport algorithm 3-61
Figure 3-13. Generic aggregate BMP schematic 3-64
Figure 3-14. Illustration of land use area assignment to aggregate BMP components 3-64
Figure 3-15. Aggregate BMP testing configuration 3-65
Figure 3-16. Schematic of sediment transport, deposition, and scour in conduits 3-76
Figure 3-17. Conceptual overview of the optimization module 3-84
Figure 3-18. Illustration of assessment points 3-84
Figure 3-19. Comparison of Scatter Search andNSGA II optimization techniques 3-87
Figure 3-20. Tiered application of SUSTAIN for developing cost-effectiveness curves 3-91
Figure 3-21. Construction of the tier-2 search domain using tier-1 results 3-91
Figure 3-22. Simulation process for each iteration run 3-92
Figure 3-23. Number of storm events and total precipitation and peak intensity as a function of
dry hours between storms 3-94
Figure 3-24. Precipitation events sorted by total precipitation volume and peak intensity 3-95
Figure 3-25. Storm event 1: August 9, 2001 6:00 p.m. to August 9, 2001 8:00 p.m. (0.66 in. to
2 wet hours, peak: 0.48 in.) 3-96
Figure 3-26. Storm event 2: May 14, 2001 11:00 a.m. to May 14, 2001 1:00 p.m. (0.95 in. to 4
wet hours, peak: 0.38 in.) 3-96
Figure 3-27. Storm event 3: April 20, 2001 2:00 a.m. to April 21, 2001 5:00 a.m. (1.03 in. to 12
wet hours, peak: 0.30 in.) 3-97
VIM
-------
Figure 3-28. Storm event 4: February 8, 2001 12:00 a.m. to February 9, 2001 4:00 p.m. (2.34
in. to 25 wet hours, peak: 0.61 in.) 3-98
Figure 3-29. Sediment removal summary for 34 storm events arranged by baseline sediment
load 3-98
Figure 3-30. Peak flow reduction summary for 34 storm events arranged by peak flows 3-99
Figure 3-31. Peak flow comparison by storm event arranged by post-developed peak flows 3-99
Figure 3-32. EMC of sediment for the corresponding storms 3-100
Figure 3-33. Example cost-effectiveness curve for sediment load reduction 3-100
Figure 3-34. BMP cost distribution by effectiveness for sediment load reduction on the cost-
effectiveness curve 3-101
Figure 3-35. Example cost-effectiveness and cost-distribution pair 3-102
Figure 3-36. BMP storage distribution by cost-effectiveness interval 3-103
Figure 4-1. Oak Creek Watershed and Upper North Branch Oak Creek 4-106
Figure 4-2. Upper North Branch Oak Creek Watershed case study road map 4-108
Figure 4-3. Average annual precipitation volume at Milwaukee Airport for water years 1988-
2002 4-109
Figure 4-4. Rainfall volume and intensity wet-interval distribution for water year 2001 4-109
Figure 4-5. Comparison of daily flow at Model Outlet 58 with USGS 04087204 at South
Milwaukee 4-110
Figure 4-6. Comparison of monthly flow at Model outlet 58 with USGS 04087204 at South
Milwaukee 4-110
Figure 4-7. Modeled vs. observed TSS (mg/L) at Oak Creek gage OC-05, water years 1995-
2001 4-111
Figure 4-8. Modeled vs. observed TSS (mg/L) at Oak Creek gage OC-05, water year 2001 4-111
Figure 4-9. Land use distribution in the modeled subwatersheds 4-112
Figure 4-10. Subwatershed grouping for two-tiered optimization 4-113
Figure 4-11. Aggregate BMP schematic 4-114
Figure 4-12. Aggregate BMP land use distribution 4-114
Figure 4-13. Aggregate BMP arrangement in tier-1 subwatershed A 4-116
Figure 4-14. Tier-1 cost-effectiveness curve for subwatershed A with the selected solutions 4-117
Figure 4-15. Composition of best solutions on tier-1 cost-effectiveness curve for subwatershed
A 4-118
Figure 4-16. Schematic oftier-2 analysis network 4-119
Figure 4-17. Tier-2 cost-effectiveness curve 4-119
Figure 4-18. Little Rocky Run watershed in Fairfax County, Virginia 4-122
Figure 4-19. Selected study area in Little Rocky Run watershed 4-123
Figure 4-20. Little Rocky Run case study road map 4-123
Figure 4-21. SWMM calibration results at junction 3 of the Little Rocky Run watershed 4-125
Figure 4-22. Representation of three subareas 4-126
Figure 4-23. SWMM-generated hydrograph at the outlet of the study area 4-127
Figure 4-24. SWMM versus SUSTAIN-genemted hydrograph comparison 4-127
Figure 4-25. BMP placement schematic in the Little Rocky Run case study area 4-131
Figure 4-26. Annual precipitation at the Washington Dulles International Airport 4-134
Figure 4-27. Volume and intensity distribution of storm events in 1994 4-134
Figure 4-28. Results showing the benefits of BMPs 4-135
Figure 4-29. Domain of optimization searches and identified best solutions 4-136
IX
-------
List of Tables
Table 1-1. Management Practices Supported by SUSTAIN 1-4
Table 1-2. Typical Data Needs for SUSTAIN Application 1-12
Table 2-1. Simulation of Sediment Transport in SUSTAIN 2-17
Table 2-2. Transport Simulation of Other Pollutants in SUSTAIN 2-18
Table 3-1. Modules and Components in SUSTAIN 3-26
Table 3-2. Summary of Inputs, Methods, and Outputs in FM 3-27
Table 3-3. GIS Data Requirement for BMP Suitability Analysis 3-28
Table 3-4. Default Criteria for BMP Suitable Locations Used in SUSTAIN 3-29
Table 3-5. Structural BMP Options Available in the BMP Siting Tool 3-30
Table 3-6. Inputs, Methods, and Outputs of the Land Module 3-32
Table 3-7. Land Simulation Methods Used in SUSTAIN 3-33
Table 3-8. Green-Ampt Parameters 3-37
Table 3-9. List of Sediment Input Parameters for Pervious Land 3-44
Table 3-10. Range of Values for Sediment Erosion Parameters for Pervious Land 3-45
Table 3-11. Typical EMCs in Urban Runoff 3-47
Table 3-12. Summary of Inputs, Methods, and Outputs in the BMP Module 3-50
Table 3-13. Available Optional Methods for BMP Simulation Processes 3-50
Table 3-14. Representative BMPs and Recommended Simulation Methods 3-51
Table 3-15. Coefficient Cw (English units) for Rectangular Sharp-Crested Weirs 3-53
Table 3-16. Quality Ratings of Conceptual Pond Shapes Simulated by Persson et al. (1999) 3-58
Table 3-17. Recommended £' and C* Values 3-59
Table 3-18. Comparison of Aggregate vs. Distributed BMP Results 3-66
Table 3-19. Simulation Run-Time Comparison: Aggregate vs. Distributed 3-66
Table 3-20. BMPs and Associated Construction Components 3-67
Table 3-21. Components 3-70
Table 3-22. BMP Types 3-70
Table 3-23. BMP Components 3-70
Table 3-24. Component Costs 3-70
Table 3-25. Unit Types 3-70
Table 3-26. Reference Sources 3-71
Table 3-27. Structural BMPs and Major Treatment Processes 3-73
Table 3-28. Summary of Inputs, Methods, and Outputs of the Conveyance Module 3-75
Table 3-29. Typical Manning's Roughness Coefficient for Closed Pipes 3-75
Table 3-30. Typical Manning's Roughness Coefficient for Open Channels 3-76
Table 3-31. List of Sediment Input Parameters forthe Reach 3-82
Table 3-32. Summary of Inputs, Methods, and Outputs in the Optimization Module 3-83
Table 3-33. Example Control Targets for Typical Evaluation Factor Assessment in SUSTAIN 3-85
Table 3-34. Post-processor Inputs, Methods, and Outputs 3-94
Table 4-1. Watershed Characteristics 4-106
Table 4-2. Upper North Branch Oak Creek Land Use Distribution 4-107
Table 4-3. Summary of Modeled Annual Average Outflow and TSS Load in Oak Creek
Watershed 4-111
Table 4-4. BMP Parameters 4-115
Table 4-5. BMP Cost Functions for the Case Study 4-116
Table 4-6. Selected Tier-1 Solutions on the Cost-Effectiveness Curve 4-117
Table 4-7. Selected Tier-2 Best Solutions 4-120
Table 4-8. List of Input Parameters for Both the Anderson and the USGS Methods 4-124
-------
Table 4-9. Comparison of Predicted Peak Flows (ft3/s) in the Little Rocky Run Watershed 4-125
Table 4-10. Land Uses of the Case Study Area in the Little Rocky Run Watershed 4-126
Table 4-11. Major Input Parameters for Modeling of Three Subareas 4-126
Table 4-12. Subbasin SWMM Parameters 4-129
Table 4-13. Hydrologic Soil Group Distribution in Study Area Subbasins 4-129
Table 4-14. TP Buildup and Washoff Parameter Values 4-130
Table 4-15. BMP Parameters 4-131
Table 4-16. Annualized BMP Cost Function 4-132
Table 4-17. Annualized Life Cycle Cost for a Bioretention Cell with a Surface Area of 900 ft2 4-132
Table 4-18. Annualized Life Cycle Cost for a Bioswale with a Surface Area of 900 ft2 4-132
Table 4-19. 10-Yr Design Storm Peak Flows and TP Annual Load under Existing Conditions 4-134
Table 4-20. Cost-Effective Solution Details 4-136
XI
-------
Acknowledgements
The Tetra Tech project team would like to thank EPA's project officer, Dr. Fu-hsiung (Dennis) Lai, for
his active involvement in providing guidance and technical insight throughout the design and the system
development process and for his detailed review of the interim and final reports. In particular, Dr. Lai
was instrumental in incorporating in SUSTAINING use of tier analysis for integrating multiple-scale
watersheds and cost-effectiveness curves for presentation of solutions for decision planning. We
appreciate the continuous support and recognition of the significance of this project by Sally Gutierrez,
Director of EPA Office of Research and Development, National Risk Management Research Laboratory
and Anthony N. Tafuri, Chief of Urban Watershed Management Branch.
We appreciate the review and insight of the external reviewers Jim Carleton, EPA, Office of Water,
Office of Science and Technology and Dr. Arthur McGarity, Swarthmore College, Pennsylvania. We are
grateful to the team of seven optimization experts that provided additional review and excellent technical
recommendations: Dr. James P. Heaney, University of Florida, Gainsville; Dr. Manuel Laguna,
University of Colorado, Boulder; Dr. Arthur McGarity, Swarthmore College, Pennsylvania; Dr. S. Ranji
Ranjithan, North Carolina State University; Dr. Christine Shoemaker, Cornell University; Dr. Richard
Vogel, Tufts University, Boston; and Dr. Laura J. Harrell, Old Dominion University. In addition, we
would like to thank Prince George's County, Maryland, and Mow-Soung Cheng in particular, for making
the county's best management practice module and optimization routine available during the early stage
of the SUSTAIN development. We would also like to thank the Fairfax County, Virginia, Department of
Public Works and Environmental Services, Stormwater Planning Division for use of data related to the
Little Rocky Run case study. Similarly, we would like to thank the Milwaukee Metropolitan Sewerage
District and the Southeastern Wisconsin Regional Planning Commission for use of data related to the
Upper North Branch Oak Creek case study.
The project team also appreciates the feedback and suggestions from the nine beta testers: Jim Carleton,
EPA, Office of Water, Office of Science and Technology; Dr. Abigail Hathway and Dr. Simon Doncaster,
The University of Sheffield, U.K.; Dr. David Sample, Virginia Polytechnic Institute and State University;
Dr. Nguyen Khoi, Naval Facilities, Norfolk, Virginia; Scott Job, Tetra Tech; Shohreh Karimipour, New
York State Department of Environmental Conservation; Tsai You Jen, Parsons Brinckerhoff; and
Valladolid Veronica, DuPage County, Illinois. Their feedback helped us enhance the program's
functionality.
We appreciate the support of the user community and in particular the Environmental and Water
Resources Institute of the American Society of Civil Engineers in the comments and discussion of best
management practice modeling and technology over the past years. This system benefited from the
discussion and presentations during the annual conferences.
We would also like to acknowledge Dr. Guoshun Zhang and former employees of Tetra Tech, Dr. Ting
Dai and Haihong Yang, for their contribution to the successful development of SUSTAIN.
Finally, we would like to extend our appreciation to two EPA retirees, Chi-Yuan (Evan) Fan and Daniel
Sullivan, for their insight and support during the early phase of this project. Evan prepared the initial
scope of work and was the Project Officer during the project procurement in 2002 and Dan was the
Branch Chief of Urban Watershed Management Branch until 2003.
XII
-------
Acronyms and Abbreviations
BASINS Better Assessment Science Integrating Point and Nonpoint Sources
BOD Biochemical Oxygen Demand
BMP Best Management Practice
CALTRANS California Department of Transportation
COD Chemical Oxygen Demand
CSO Combined Sewer Overflow
CSTR Continuously Stirred Tank Reactor
DEM Digital Elevation Model
DEQ Department of Environmental Quality
EMC Event Mean Concentration
EPA U.S. Environmental Protection Agency
ET Evapotranspiration
GA Genetic Algorithm
GB Gigabyte
GI Green Infrastructure
GIS Geographic Information System
HSPF Hydrologic Simulation Program—FORTRAN
LID Low Impact Development
LSPC Loading Simulation Program in C++
MS4 Municipal Separate Storm Sewer System
NCDC National Climatic Data Center
NHD National Hydrography Dataset
NLCD National Land Cover Dataset
NPDES National Pollutant Discharge Elimination System
NRCS Natural Resources Conservation Service
NSGA-II Non-dominated Sorting Genetic Algorithm II
NWS National Weather Service
O&M Operation and Maintenance
PET Potential Evapotranspiration
SUSTAIN System for Urban Stormwater Treatment and Analysis INtegration
SWMM Stormwater Management Model
TKN Total Kjeldahl Nitrogen
TMDL Total Maximum Daily Load
TN Total Nitrogen
TP Total Phosphorus
TSS Total Suspended Solids
USGS U.S. Geological Survey
USLE Universal Soil Loss Equation
VFSMOD Vegetative Filter Strip Model
WINSLAMM Source Loading and Management Model for Windows
XIII
-------
Chapter 1 Introduction
Surface water degradation resulting from the effects of urbanization on hydrology, water quality, and
habitat is an issue of primary focus for multiple agencies at the federal, state, and local levels. A few
examples of critical management issues facing planners and policy makers are ensuring the protection of
source waters and the management of stormwater through peak flow mitigation, installation of sediment
and erosion control devices, or implementation of best management practices (BMPs). Many
management actions are needed throughout watersheds to achieve the desired effects on flow mitigation
and pollutant reduction; however, no single standardized solution can be effective in all locations.
Factors such as watershed size, scale, existing human activities, and natural characteristics can vary
dramatically from one place to another. The major challenge faced by decision makers is how to select
the best combination of practices to implement among the many options available that result in the most
cost-effective, achievable, and practical management strategy possible for the location of interest.
Realizing the need for improved tools to support that challenge and the opportunities presented by
emerging science and technology, the U.S. Environmental Protection Agency (EPA) initiated a research
project in 2003 to develop a fully integrated decision support framework for the selection and placement
of stormwater BMPs at strategic locations in urban or developing watersheds. Development of a software
system to meet that challenge has been conducted in a phased process. The resulting system, described in
detail in this document, is called the System for Urban Stormwater Treatment and Analysis integration
(SUSTAIN). This document and the SUSTAIN Version 1.0 system represent the culmination of work
under Phase II of development. The software, companion user manual, and periodic updates will be
available on the SUSTAIN Web site hosted by EPA, (http://www.epa.gov/ednnrmrl/models/sustain/).
This document describes the rationale for developing the framework and the uses of the framework;
explains the system's design, structure, and performance; details the underlying methods and algorithms
that provide the framework's predictive capabilities; and demonstrates the framework's capabilities
through two case studies. The initial needs analysis and model review documentation developed under
Phase I are also included in the appendices. This document, where appropriate, also examines the
limitations of the current framework and recommendations for enhancing the framework to be addressed
in future development phases.
1.1. Proj ect Rationale
A wide range of programs exist in the United States to support the protection and restoration of
waterbodies (i.e., rivers, lakes, estuaries). Most programs involve linking land-based actions to water
quantity or quality goals with the ultimate goal of reducing the impacts on receiving waters. Models of
varying scales and complexity have long been a part of developing mitigation plans, identifying
management needs, and evaluating alternatives. Examples of situations where modeling can support
decision making include source water protection plans, municipal separate storm sewer system (MS4)
permits under the National Pollutant Discharge Elimination System (NPDES) Stormwater Program
(Phase I and II), total maximum daily load (TMDL) implementation plans, and watershed-based master
plans and restoration studies.
In each case, water quality professionals need a framework to help address key stormwater management
issues, e.g., to do the following:
1-1
-------
• Evaluate and select management options to achieve a loading target set by a TMDL
• Develop cost-effective management options to implement a municipal stormwater program
• Evaluate pollutant loadings and identify appropriately protective management practices for a
source water protection study
• Determine a cost-effective mix of green infrastructure (GI) measures to meet optimal flow
reduction goals in a combined sewer overflow (CSO) control study
Over the past decade, significant progress has been made in expanding our understanding, through
detailed laboratory and field studies, of the wide array of available management techniques and their
function and impact on urban hydrology and water quality processes. Today, managers increasingly
incorporate a combination of on-site, GI technologies with more traditional structural practices as part of
comprehensive watershed restoration plans. As a result, many municipalities implement various site-
scale techniques (i.e., bioretention, rain barrels, swales, infiltration trenches) at different points throughout
a drainage area to mitigate both the flow and associated pollutant impacts of urban drainage. Practitioners
now need to evaluate both the localized site-scale benefits and the cumulative effects of implementing
hundreds or even thousands of those practices across a broad watershed landscape.
Concurrent with the evolution in management techniques, significant advances have been made in
information technology over the past decade. Previous modeling of management alternatives was limited
to highly simplified approaches for larger-scale regional studies. Next generation modeling systems now
enable more detailed simulation techniques in combination with optimization tools, resulting in the ability
to rapidly evaluate and compare multiple alternatives. Significantly faster computational speeds allow for
interactive consideration of process-based simulations of flow and water quality with optimization
searches. Software that facilitates spatial analysis, database management, and model execution is now
readily available for practical application. Integration of simulation techniques with geographic
information systems (GIS) has improved our ability to evaluate watershed management through multiple
scales and at varying levels of complexity. Improved scientific understanding and advances in
computational resources have now provided the opportunity to build more sophisticated and robust water
resources modeling tools to support decision making.
On the basis of an understanding of the needs of the user community, SUSTAIN was developed to address
the following major design objectives:
• It is intended for knowledgeable model users, including those at the local level, who are familiar
with the technical aspects of watershed modeling
• It provides users with the ability to evaluate the effects of multiple management practices and
placement strategies to support decision making
• It is specifically designed for and applicable to mixed land uses present in predominantly urban
watersheds
SUSTAIN includes hydrologic/hydraulic and water-quality modeling in watersheds and urban streams. It
has the capability to search for optimal management solutions at multiple scales to achieve desired water-
quality objectives based on cost-effectiveness.
SUSTAIN was developed by combining publicly available modeling techniques, costs of management
practices, and optimization tools in a geographically based framework to achieve the design objectives.
SUSTAIN facilitates the objective analysis of multiple water quality management alternatives while
1-2
-------
enabling consideration of interacting and competing factors such as location, scale, and cost. In
developing SUSTAIN, the most applicable algorithms for simulating urban hydrology, pollutant loading,
and treatment processes were packaged together from those in multiple, distinct models. The simulation
processes incorporated into SUSTAIN have not been known to be previously bundled in a publicly
accessible modeling framework.
1.2. Overview of SUSTAIN
SUSTAIN is a framework that facilitates a comprehensive stormwater management analysis of watersheds
at multiple scales. SUSTAIN was carefully constructed to ensure a seamless package that provides a
consistent level of technical rigor, employs the latest technology, and performs cost-effectiveness analysis
to derive practical solutions to real-world problems. SUSTAIN includes algorithms for simulating urban
hydrology, pollutant loading, and treatment processes packaged from multiple models that individually
address such processes. To provide flexibility for future updates, the system uses linked modules that
perform simulations on watershed land surfaces, in management practices, and through routing networks.
SUSTAIN uses a graphical interface to allow users to visualize the study area, select locations for
placement of management practices, and define the linkages among the various landscape features. The
analytical framework lets users apply optimization tools to explore the wide range of possible cost-
effective solutions.
Because many models are used to address watershed problems, and some regions have a long history of
model development and testing, SUSTAIN was designed to interface with external models. Through the
use of file exchanges (i.e., time series files), SUSTAIN can import externally generated watershed
modeling information and can export time series results to receiving water models for additional detailed
analysis.
J.2.J. Structure of SUSTAIN
SUSTAIN is built on a base platform interface using ArcGIS, which provides the user access to the
framework components: a BMP siting tool; a watershed runoff and routing module; a BMP simulation
module; a BMP cost database; a post-processor; and an optimization module.
Figure 1-1 shows a generalized schematic of the overall framework. The ArcGIS-based Framework
Manager (FM) is the overarching component that manages the data exchanges between the framework
components. It provides linkages between external inputs, the land simulation, the BMP simulation, the
conveyance simulation, the optimization module, and the post-processor. The FM checks for necessary
data requirements before calling for simulation and optimization components.
Each module in the framework serves a specific function and is typically applied in series. The
application usually begins with the use of the BMP siting tool, which uses the ArcGIS platform and user-
guided rules to determine site suitability for various BMP options (Table 1-1). The land simulation
module is used to generate runoff time series data to drive the BMP simulation. The conveyance module
provides routing capabilities between land segments or BMPs or both. Users also have the option to
import time series data from external watershed models (e.g., Hydrologic Simulation Program Fortran
(HSPF) or Stormwater Management Model (SWMM)) instead of performing new land simulations in
SUSTAIN.
1-3
-------
Interpretation
(Post Processor)
Figure 1-1. Overview of SUSTAIN components.
The process-based BMP module provides simulation of
management practices by using a combination of processes for
storage retention, open-channel controls, filtration, biological
purification, and mechanical structure facilitated separation. The
cost database is organized according to BMP construction
components (e.g., grading, backfilling, filter fabric) and populated
with unit costs for each component. The optimization module uses
results from other modules in the framework for evaluating and
selecting a combination of BMP options that achieve a given
pollutant-reduction goal at minimum cost. The optimization
module is designed to efficiently search for this combination of
BMPs. Finally, a post-processor presents the optimization results
in a cost-effectiveness curve.
From the GIS framework, the user first sets up a project
representing a network of drainage areas, BMPs, and routing
components. SUSTAINthen uses externally generated land use-
associated flow and water quality time series data or internally
generated data from BMP contributing areas and routes them
through BMPs to predict flow and water quality time-series data at
selected downstream locations. The user defines the assessment
locations in the watershed where results are to be analyzed or
compared (Figure 1-2).
Table 1-1. Management
Practices Supported by
SUSTAIN
Management Practice
Bioretention
Cistern
Constructed Wetland
Dry Pond
Grassed Swale
Green Roof
Infiltration Basin
Infiltration Trench
Porous Pavement
Rain Barrel
Sand Filter (non-surface)
Sand Filter (surface)
Vegetated Filterstrip
Wet Pond
1-4
-------
Figure 1-2. Watershed assessment points.
SUSTAIN's optimization capability helps users identify desired economical BMP solutions that achieve
user-defined management target(s). Another benefit of the framework is its ability to reveal the BMP
cost- and pollutant-reduction effectiveness relationship, referred to as the cost-effectiveness curve. A
sample cost-effectiveness curve is shown in Figure 1-3. Each point on the curve represents an optimal
combination of BMPs that will collectively remove the targeted amount of pollutant load at the least cost.
The BMP cost-effectiveness curve provides valuable information on the minimum costs at various
reduction goals, the maximum achievable pollutant reductions, as well as the marginal costs.
C
o
3
•o
0)
re
o
Cost
Figure 1-3. Sample cost-effectiveness curve.
1-5
-------
1.2.2. Multiple Scale Application Features
Practitioners are confronted daily with the need to evaluate management practices at multiple scales, from
an individual site to regional watershed studies. The site-scale evaluation might require a detailed
assessment of individual BMPs or combinations of BMPs (i.e., treatment trains). Larger-scale watershed
studies, typically over 100 square miles, could involve hundreds or thousands of individual management
practices to achieve a desired cumulative benefit. Simulating and performing cost comparisons for each
of these individual distributed BMP options would place a significant challenge on the accuracy and
simulation time for modeling. Two approaches were developed in SUSTAIN to address the watershed
scaling issue: aggregation and tiered or nested analysis methods. These methods facilitate the use of
SUSTAIN at multiple scales as shown in Figure 1-4.
Scale in Watershed Planning
Uses of SUSTAIN
Regional
>100
sq. mi.
Tiered watershed approach
Subdivide into relevant management
zones
Pilot studies on selected watersheds
Large watershed plan
10-100
sq. mi.
Tiered and aggregate watershed
approach
Detailed modeling of selected high
priority watersheds
Midsize watershed plan
• Tiered and aggregate BMP
1-10 sq. approach used for initial planning
mi. • Detailed modeling on all or some
watersheds
Site scale
<1 • Detailed modeling and optimization
sq. mi. of management practices
Figure 1-4. SUSTAIN^s multiple scales of application.
As an alternative to the explicit representation and routing of multiple distributed BMPs, the aggregate
BMP approach creates a virtual BMP that represents all similarly functioning treatment devices in a
watershed. This option can significantly reduce computational effort, especially when distributed BMPs
are involved in the optimization process as decision variables. The aggregated approach uses four generic
BMPs in sequence, each representing the function of many similar BMPs: on-site interception, on-site
treatment, routing/attenuation, and regional storage/treatment.
For large watersheds that require detailed analyses, SUSTAIN provides a methodology for tiered or
sequenced analysis. As illustrated in Figure 1-5, a relatively large watershed can be subdivided into
several smaller subwatersheds on which detailed analysis is performed to derive a tier-1 cost-
effectiveness curve. The tier-2 cost-effectiveness curve is derived from the three tier-1 curves by
considering all feasible optimal combinations of BMPs that produce the target load reduction at the
minimum cost.
1-6
-------
The tiered approach can be applied to large watersheds that contain several subareas and to small
watersheds that require the development of a detailed management plan, e.g., at a parcel or a street block
level.
C i
O '
T3 '*=
ro o
O 3
-J T3
0
, X
r^
cost y
^_
SUSTAIN Tier 1
„
SUSTAIN Tier 1
Target Load
Reduction
SUSTAIN Tier 2
Figure 1-5. Tiered application of SUSTAIN for developing cost-effectiveness curves.
The tiered optimization in SUSTAIN not only provides an efficient and manageable means of analysis for
large-scale applications, but also allows users flexibility in the placement of assessment points and in
evaluating explicit expectations of load reductions at upstream locations.
1.3. The Role of SUSTAIN in Watershed Applications
Various practitioners, municipalities, and watershed groups at the regional and local levels can use the
SUSTAIN framework to address a variety of management practice planning questions. Users might turn
to SUSTAIN for the following:
• Developing TMDL implementation plans
• Identifying management practices to achieve pollutant reductions in an area under an MS4
stormwater permit
• Determining optimal GI strategies for reducing volume and peak flows to CSO systems
• Evaluating the benefits of distributed GI implementation on water quantity and quality in urban
streams
1-7
-------
The SUSTAIN modeling framework can be employed at multiple phases in the watershed management
process and at varying levels of detail (Figure 1-6). When applied early in the study phase, the analysis is
typically at a low level of detail to explore the potential benefits of BMPs. After the initial assessment,
SUSTAIN can be applied in greater detail as part of implementation planning to help identify the preferred
management practices and develop the associated capital costs. Once a plan is implemented, SUSTAIN
can be used to track and assess the performance of the installation using the monitoring data collected
before and after installing BMPs. The monitored data can also be used to recalibrate and reverify
SUSTAIN for extrapolating future benefits from additional BMPs.
Watershed Planning Process Uses of SUSTAIN
Identify Problem(s) &
Set Goals
Generalized assessment of
management impacts and
load reduction potential
Develop Plan
Implement Plan
Refin(e and adapt
plan f needed \f
Track Progress
Achieve
Management Goals
Predict load reduction and cost
for multiple management
alternatives
Support selection of an optimal
implementation plan
Evaluate project phases (cost
and load reduction at each
phase)
Recalibrate SUSTAIN based
on newly collected data
Evaluate future benefits of
implementation and/or
adaptation of plan
Figure 1-6. Using SUSTAIN in the watershed planning process.
Within the context of each application, SUSTAIN provides the flexibility to support the application goals
and address the localized issues of concern. Two examples of applications as described below include
TMDL development and implementation, and the use of GI to help control CSOs.
1.3.1. TMDL Development and Implementation
TMDL projects are plans to reduce loadings to meet a defined water quality standards objective in
impaired water (USEPA 1991). Each state defines impaired waters on its Clean Water Act (CWA)
section 303(d) list. For lake TMDLs, the 303(d) listing might identify a nutrient impairment that results
in periodic algal blooms and is limiting the ability of the lake to be used for recreation and as aquatic
habitat. The TMDL development requires estimating nutrient loadings and evaluating the loading
threshold that would keep the lake within acceptable water quality conditions and meeting applicable
water quality standards. SUSTAIN uses the embedded SWMM rainfall-runoff routines to develop the
loading characteristics from the land areas. The framework is ideally suited to evaluate the load reduction
1-8
-------
potential as part of examining the reasonable assurance that the TMDL can be achieved at the prescribed
load reductions. After the TMDL is developed, SUSTAIN can be used to develop the implementation plan
that identifies the best combination of management practice type(s) and location(s), and the associated
cost load reductions.
In some areas, TMDL implementation is addressed by a municipal stormwater permit. For communities
working to comply with wasteload allocations assigned in a TMDL, SUSTAIN provides a method to
integrate stormwater permit activities with the requirements of the TMDL. Capitalizing on mapping and
data collection activities typically undertaken as part of the MS4 implementation, SUSTAIN can be used
to enumerate specific measures necessary for meeting TMDL reductions throughout the affected area.
The framework can be used to pinpoint the best locations for optimizing pollutant reductions and to
determine the mix of management practices that will achieve necessary load reductions for the least cost.
Examples of the TMDL-related investigations that SUSTAIN supports include the following:
• Optimizing the geographic focus of management activities (near the waterbody of concern or
away from it)
• Evaluating the benefits of installing rain barrels or rain gardens in a near-lake region
• Enumerating specific management practices that must be implemented to satisfy the TMDL
• Developing a funding request
• Developing a projection of reduction potential for phased installation over time
1.3.2. Evaluation of GI Practices as Part of a CSO Control Program
Even with advances in sewer technology (e.g., sewer separation and deep storage), problems still remain
with the operation of existing urban wastewater systems (NRDC 2006). Examples include impaired
performance of wastewater treatment plants resulting from the influx of stormwater (infiltration and
inflow), constraints on urban growth caused by an inadequate infrastructure, and aging combined sewer
systems, which can require costly rehabilitation (USEPA 2004). CSO control programs typically focus
on infiltrating or storing runoff to minimize peak flows to collection systems and reduce the frequency
and size of overflow events. Programs are developed and implemented to comply with mandated CSO
long-term control plans. Such challenges have led to the development of sustainable strategies for urban
stormwater and wastewater management and new alternatives to the traditional centralized sewer systems,
which comprise laterals, submains, and trunk lines all leading to a central treatment facility (Chocat et al.
2007). Although large storage structures, tunnels, and sewer separation have been used successfully to
significantly reduce CSOs in major cities, increased effectiveness and multiple benefits can be derived by
adopting new GI approaches in combination with traditional approaches.
As a means to reduce volume entering a wastewater system and reduce its peak flows, GI can be applied
through source controls and engineered BMP systems to infiltrate, evapotranspirate, or store stormwater
runoff for beneficial uses (USEPA 2007). Those approaches are to keep stormwater runoff from entering
a combined sewer system and reduce overflows (USEPA 2004). In addition, many GI approaches can be
included in adaptive management strategies designed to be resilient to such system changing factors as
population growth and climate change (USEPA 2008).
SUSTAIN is designed to support linkage to other related models such as detailed sewer system models or
receiving water models of affected rivers, lakes, and estuaries. Where existing sewer and watershed
models are available, SUSTAIN can be used to predict the most inexpensive GI practices that will result in
reduced overflow volumes and frequency.
1-9
-------
1.4. SUSTAIN Application Process
A typical SUSTAIN application scenario begins with the definition of study objectives, followed by data
collection, project/model setup, formulation of the optimization problem, and analysis of results. Figure
1-7 is a flow diagram illustrating the typical step-by-step process in SUSTAIN applications.
p-
•o
£
« |
Jl
t3 re
o> o>
51-0
o o
si
« ®
re r-
o c
= (A
Q. C
Q. O
CO '^
m W
015
Data Collection & Analysis
• Study area review
• CIS data: land use, stream, DEM, BMP sites, etc.
• Watershed and BMP information/data
• Compile monitoring data (calibration/validation)
Project Setup
• BMP representation: placement, configuration, and cost
• Land/Watershed Representation
• Routing network
• Assessment point(s)
• Test system application (externally calibrated model)
• Calibrate/validate model (internal model)
Put Optimization Processor to Work
• Select decision variables (BMP dimensions)
• Select assessment points (BMP/Outlet locations)
• Select evaluation factors, control targets (end points)
Results Analysis and Representation (Post-Processor)
• Optimum BMP dimensions
• Alternate solutions
Figure 1-7. SUSTAIN application process.
Fundamental to the setup and application of SUSTAIN is a clear definition of the study objective(s)—
What is the question that is to be answered by the analysis? For example, the objective of the study
might be to identify the set of management options (including both site- and regional-scale techniques)
that achieve a required level of pollutant load reduction (i.e., annual load in Ibs/yr). For a CSO study, the
objective might be stated as, "to reduce frequency of overflow through extensive retrofit of the drainage
area." The reduction in overflow can be measured by the magnitude of peak flows in a collection system.
The study objectives will define the scope and extent of the SUSTAIN application, which could include
the areas to be modeled, runoff and pollutant factors to be simulated, additional data collection needs, the
locations where the output will be evaluated (i.e., assessment points), and the determination of the
optimization evaluation factors and control targets (i.e., endpoints). At each control target, SUSTAIN is
capable of producing outputs in various time averaging periods and frequencies of occurrences that will
facilitate the evaluation and comparison of management alternatives. The following lists the examples of
output variations.
• Average annual flow volume percent reduction based on an existing condition
• Average annual flow volume
• High-flow rate and allowed maximum duration (user specified)
1-10
-------
• Peak-flow value and maximum exceedance frequency
• Average annual sediment/pollutant load percent reduction with respect to an existing condition
• Average annual sediment/pollutant load (load target)
• Average sediment/pollutant concentration (the maximum average concentration allowed)
• High sediment/pollutant concentration and duration (concentration threshold value and allowed
maximum duration when concentration exceeds the threshold)
• Long-term average sediment/pollutant load (daily, monthly, annual, or any user specified time
frame)
• Exceedance frequency (the threshold value and maximum exceedance frequency allowed)
SUSTAIN has been designed with inherent flexibility in the formulation and setup of the application. The
careful definition of the project objective, associated evaluation factors, and control targets will ensure the
most appropriate and useful application.
The data collection process for a SUSTAIN application is similar to most modeling projects and involves a
thorough compilation and review of information available for the study area. It generally includes
gathering applicable regional and site-scale GIS data layers, digital elevation model (DEM) data, stream
networks, locations of BMPs, land use data, critical source information, and monitoring data for
calibration and validation. A summary of typical data needs is shown in Table 1-2.
Setting up the SUSTAIN project involves using the data collected to establish a representation of the land
and pollutant sources in the watershed as well as the routing network, assessment points, and management
practices to be evaluated. For site-scale analysis of management practices, locally derived higher-
resolution site scale data will likely be required.
If the continuous time series data of flow and associated sediment/pollutants from a locally calibrated
model study is available, the data can be imported into SUSTAIN without recreating them. Most models
that operate on an hourly or shorter time step, such as HSPF, SWMM, Source Loading and Management
Model for Windows (WINSLAMM), are compatible with SUSTAIN. When importing information from
an externally generated model, the SUSTAIN application builds on the documentation, testing, calibration,
and validation of the external model.
After project setup, the optimization module synthesizes information from the BMP, land, and
conveyance modules and generates solutions that are looped back for evaluation using the same modules
again. Via this evolutionary search process, the optimizer identifies the best or most cost-effective BMP
solutions according to the user's specific conditions and objectives. Finally, the post-processor analyzes
optimization results using specific graphical and tabular reports that facilitate the classification of storm
events for analysis, viewing the time series of specific storm events, evaluating BMP performance by
storm event, and developing the cost-effectiveness curves for treatment alternatives.
1-11
-------
Table 1-2. Typical Data Needs for SUSTAIN Application
Data
Land use
Land use
lookup
External
Model
Digital
Elevation
Data (DEM)
Stream
Network
Precipitation
Other weather
data
Pipes
Stream
Geometry
Management
Practices
Flow
Water Quality
Data
Type
ESRI
Grid
Dbf Table
ASCII
Text Files
ESRI
Grid
ESRI
Shape
File
ASCII
Text File
ASCII
Text File
Data
Entry
Data
Entry
Data
Entry
ASCII
Text File
ASCII
Text File
Need
Required for defining land
use distribution
Required for assigning
land use categories and
groupings
Required for external
model linkage
Required for automatic
delineation of drainage
areas
Required for automatic
delineation of drainage
areas and for placing on-
stream management
practices
Required for internal land
simulation and for
estimating storm sizes for
the post-processor
Required if snow melt is
simulated for internal land
simulation
Required if pipe/conduit is
simulated
Required if stream routing
is simulated
Required
Required for calibration of
internal modeling of
runoff; recommended for
system testing
Required for calibration of
internal modeling of water
quality; recommended for
testing of water quality
predictions
Data Source
National Land Cover Dataset (NLCD)
(http://seamless.usss.sov/website/seamless/vie
wer.php) or locally derived
Standard National Land Cover Dataset
(NLCD) land cover code for NLCD land use
(http://landcover.usss.sov/classes.asp).
or land cover mapping code for locally derived
data
Time series generated by calibrated model; by
land use
(http://seamless.usss.sov/website/seamless/vie
wer.php) or locally derived source
National Hydrography Dataset (NHD) from
http://nhd.usss.gov/data.html
National Climatic Data Center (NCDC).
NCDC Summary of the Day (daily data) can
also be obtained from (Earthlnfo Inc.,
http://www.earthinfo.com).
NCDC (temperature, evaporation, and wind
speed)
Shape and dimensions (e.g., length, width,
diameter)
Cross-sectional geometry (shape and related
dimensions)
Characteristics of installed and proposed
management practices (e.g., size, shape,
media, design specification); dependent on
type of practice
USGS real time data
(http://waterdata.usss.gov/nwis/rt) or local
sampling
USGS surface water data
(http://waterdata.usss.gov/nwis/sw) or
EPA STORET data
(http://www.epa.sov/storet/dw home html) or
local sampling
1-12
-------
1.5. About this Report
This report provides the description and documentation to support the release of of SUSTAIN Version 1.0.
As it is developed, EPA will release additional model information on the SUSTAIN Web site. The Web
site also provides user guidance and responses to frequently asked questions regarding the operation and
use of the model.
This SUSTAIN documentation report is organized as follows:
Chapter 1 provides a general overview of the framework, its development, typical applications, and
application process.
Chapter 2 describes the structure of the framework, the roles and interactions of its major
components, and its operational characteristics.
Chapter 3 provides the detailed documentation of the analytical procedures and simulation processes,
including equations and variables, which are adopted and incorporated into various parts of SUSTAIN.
Chapter 4 presents two case studies to demonstrate how the framework is applied for selection and
placement of BMPs.
The appendices include a needs analysis for developing a comprehensive placement framework, a
review of land and BMP simulation models, and a summary of expert opinions on the current state-
of-the-art in optimization concepts and methods to support development of the optimization
component in SUSTAIN. It includes the rationale and supporting information used in formulating the
framework design and selecting land and BMP simulation techniques that appear in SUSTAIN.
1-13
-------
Chapter 2 SUSTAINDesign and Structure
SUSTAIN is a comprehensive, multiscale watershed and water quality modeling application built on an
ArcGIS platform linked to multiple simulation modules, an optimization module, and a post-processor,
which analyzes and helps interpret the results. The modular design of SUSTAIN has multiple advantages
compared to previous modeling applications including the ability to incorporate simulation of new
management practices as they are evolved, to operate independently for specific small watershed
applications, and to provide flexibility to address multiple watershed scales.
This chapter describes the system's infrastructure, its major modules, and software platforms. It also
explains how they are linked and interact.
The SUSTAIN installation requires ESRI's ArcGIS 9.3 and the Spatial Analyst extension. The application
is compatible with Microsoft Windows 98, 2000, NT, XP, and Vista operating systems and requires at
least 1GB of computer memory and 5GB of free space on the hard disk. The system also requires
Microsoft Excel 2003, which is used as a post-processor for analyzing and interpreting results.
SUSTAIN comprises the following modules:
Framework Manager—to serve as the command module of SUSTAIN, manage data for system
functions, provide linkages between the system modules, and create a simulation network to guide the
modeling and optimization activities
Land module—to generate runoff and pollutant loads from the land through internal land simulation
or importing precalibrated land simulation time series
BMP module—to perform process simulation of flow and water quality through BMPs
Conveyance module—to perform routing of flow and water quality in a pipe or a channel
Optimization module—to evaluate and identify cost-effective BMP placement and selection
strategies for a preselected list of potential sites, applicable BMP types, and ranges of BMP size
Post-Processor—to perform analysis and summarization of the simulation results for decision
making
2.1. Framework Manager
The FM performs data management, spatial analysis, and network visualization. It integrates components
from the GIS network, such as streams, conduits, and land uses, with relevant simulation modules, draws
external time series data (e.g., rainfall, runoff) as required, and checks for necessary data requirements
before calling for simulation and optimization components.
SUSTAIN is designed to interactively identify and manage the required databases, including geographic
and tabular data sets. The primary function of data management is to define the paths where data are
stored and to identify required data elements. SUSTAIN provides the option to store required geographic
2-14
-------
data on the hard disk or in a file-based geodatabase, which is a native data structure used by ArcGIS. The
geodatabase is composed of tables and queries that allow data sharing and interchange among SUSTAIN's
modules.
The FM builds on the ESRI ArcGIS (version 9.3) platform to support the placement of BMPs, delineation
of BMP tributary drainage areas and flow paths, and development of a schematized watershed simulation
network that might include land parcels, management practices, and stream reaches. The GIS component
also serves as the user interface and includes the main application window with menus, buttons, and
dialog boxes. The GIS interface allows a user to read and edit spatial and temporal data sets.
All commonly used Microsoft Office applications can be easily linked to the platform. Microsoft Excel, a
popular and powerful application for displaying and manipulating simulation time series data and
scientific graphics, was chosen as the post-processor for SUSTAIN.
Figure 2-1 shows the framework design, including system components, relationships between
components, and general flow of information.
Decision Matrix
Figure 2-1. SUSTAIN components and flow chart.
The SUSTAIN framework is designed to perform the following sequence of operations.
• From the GIS view and database, the framework first develops a simulation network that defines
the relationship between land-area units, BMPs, and stream systems on a watershed
• The FM then identifies the modules (Land, BMP, Conduit, and Reach) to be used and prepares
model input files
• The FM routes the external inputs to appropriate modules and their outputs to the Output Post-
Processor or other models
• The FM sends outputs from Output Post-Processor to the Decision Optimization Engine
2-15
-------
• The Optimization Engine evaluates the current option and selects the next preferred option from
that contained in the Feasible Option Matrix on the basis of cost and defined flow and water
quality criteria. The preferred option can be a different combination of BMP locations and types.
The Feasible Option Matrix contains types, configurations, locations, and costs of feasible BMP
options
• SUSTAIN performs numerous iterations of the sequence until the user-defined convergence
criteria are met
• The tool does not automatically select the best solution but is expected to be used as a tool to
explore and test various approaches and eventually select optimal solutions on the basis of user-
defined criteria and constraints
2.2. Simulation Modules
SUSTAINS modular simulation core consists of three standalone simulation modules: the landscape
simulation module (Land), the standalone BMP module (BMP), and a conveyance simulation module
(Conduit and Reach) as shown in the central box of Figure 2-1.
To provide a seamless and efficient operation, SUSTAIN selects and incorporates simulation routines from
commonly used watershed and receiving water models (e.g., SWMM (Huber and Dickinson 1988), HSPF
(Bicknell 2001), and Loading Simulation Program—C++ (LSPC) (Tetra Tech and USEPA 2002). The
following is a summary list of simulation routines and the model where the routines were adopted:
• Watershed/landscape models: SWMM's atmospheric, land surface, and groundwater
compartments
• Conveyance and pollutant routing: in HSPF/LSPC RCHRES and SWMM Transport compartment
• BMP simulation models: Prince George's County BMP module (Tetra Tech 2001) and selected
buffer zone simulation techniques from the VFSMOD (Munoz-Carpena and Parsons 2003)
The VFSMOD is programmed in FORTRAN computer language and compiled as a standalone dynamic
link library, whereas other simulation modules are coded in the visual C++ programming language.
The standalone simulation modules are packaged within SUSTAIN to perform the generation and
transport of sediment and other pollutants at the source (land use type), in the BMP, and in the
conveyance system. Table 2-1 shows the interaction between the Land, BMP, and Conveyance modules
to handle transport of sediment in SUSTAIN. It shows the inputs, the methods used to simulate the
sediment transport, and the resulting outputs from these simulation modules. Table 2-2 shows the
interaction between the Land, BMP, and Conveyance modules to handle transport of other pollutants in
SUSTAIN. It also shows the inputs, the methods used to simulate the water quality processes, and the
resulting outputs from the simulation modules.
2-16
-------
Table 2-1. Simulation of Sediment Transport in SUSTAIN
Land
Inputs
- Climate time series data
- Coefficients and exponents for soil detachment, washoff, and scour equations for the pervious
land
- Buildup and washoff rates of solids on impervious surfaces
Simulation Methods
- Production and removal of sediment on/from pervious land segment is computed using the
HSPF sediment algorithms
- Accumulation and removal of sediment on/from the impervious land segment is computed
using the SWMM buildup and washoff algorithms
Outputs
- Outflow time series
- Concentrations time series of total sediment
1
Split total sediment concentration into sand (0.05-2.0
mm diameter), silt (0.002-0.05 mm diameter), and clay
(< 0.002 mm diameter) concentrations.
BMP
Inputs
Inflow time series.
- Concentration time series for sand, silt,
and clay particles
- Settling velocity, critical shear stress,
and density for sand, silt, and clay
particles
Simulation Methods
- Simulation of transport, deposition, and
scouring of the sediment in BMP is
computed using the HSPF sediment
transport algorithms for a lake
-Exchange of cohesive sediments with
the bed is calculated on the basis of the
bed shear stress and user-defined
critical shear stress for deposition and
for scour
- Bed shear stress is calculated as a
function of an average flow velocity,
average water depth, and mean particle
size of the bed sediment in a BMP
Outputs
- Outflow time series
- Concentration time series for sand, silt,
and clay particles
Conveyance
Inputs
- Inflow time series
- Concentration time series for sand, silt,
and clay particles
- Settling velocity, critical shear stress,
and density for sand, silt, and clay
particles
Simulation Methods
- Simulation of transport, deposition, and
scouring of the sediment in stream is
computed using the HSPF sediment
transport algorithms
- Exchange of cohesive sediments with
the bed is calculated on the basis of the
bed shear stress and user-defined
critical shear stress for deposition and
for scour
- Bed shear stress is calculated as a
function of the slope and hydraulic
radius of the reach
Outputs
- Outflow time series
- Concentration time series for sand, silt,
and clay particles
2-17
-------
Table 2-2. Transport Simulation of Other Pollutants in SUSTAIN
Land
Inputs
- Climate time series data
- Buildup and washoff rates of the selected pollutant on the pervious and the
impervious land surface
Simulation Methods
- Accumulation of the selected pollutant on pervious and impervious land
segments is computed using the SWMM buildup functions. Power,
Exponential, and Saturation functions are provided for computing the pollutant
buildup loading on the land
- Removal of the selected pollutant on pervious and impervious land segments is
computed using the SWMM washoff algorithms. Exponential, Rating Curve,
and Even Mean Concentration methods are provided for computing the pollutant
washoff loading from the land
Outputs
- Outflow time series
- Concentrations time series for the selected pollutant
BMP
Inputs
- Inflow time series
- Concentration time series for the
selected pollutant
- 1st order decay rates for the selected
pollutants
- Background concentration (C*) for the
selected pollutants
- Percent removal rate from the
underdrain media for the selected
pollutants
Simulation Methods
- Pollutant mixing is simulated with
completely mixed tanks (CSTRs) in
series in a BMP
- Pollutant removal is calculated by first
order decay with background
concentration, i.e., k-C* method
(Kadlec and Knight 1996)
- Pollutant removal through the
underdrain media is calculated on the
basis of the user-defined percent
reduction for the selected pollutants
Outputs
- Outflow time series
- Concentration time series for the
selected pollutant
Conveyance
Inputs
- Inflow time series
- Concentration time series for the
selected pollutant
- 1st order decay rates for the selected
pollutants
Simulation Methods
- Pollutant routing is computed by
assuming a single, continuously
stirred tank reactor method in a
reach
- Pollutant removal is calculated by
1st order decay rate in a reach (no
C*)
Outputs
- Outflow time series
- Concentration time series for the
selected pollutant
2-18
-------
When operated in simulation mode, computation time can be a concern for a large watershed-scale
application because many BMP/conduits might be involved. To get realistic information about their run
time, the following test runs of SUSTAIN were performed:
• Different numbers of the same BMP unit and multiple simulation periods
• Different numbers of the same conduit unit and multiple simulation periods
• Different combinations of BMPs and conduits and multiple simulation periods
The test examples were prepared for 1,10, and 50 units (BMPs, conduits, or BMP/conduit combinations)
for simulation durations of 1, 5, and 10 years. The simulation time step of 5 minutes was used for all
simulation runs. The computer configuration used for the tests was a 1.6 GHz CPU, 768 MB RAM, and
Windows XP operating system.
The results (Figure 2-2) show that the runtime increases almost linearly with the increase of the number
of simulation units (i.e., BMP, conduit, or BMP/conduit combination). The runtime also increases
linearly with the increase in simulation period.
1600 -T
1400 -
-^ 1200 -
u
w, 1000 -
| 800-
| 600 -
EC
400 -
200 -
0
BMP- 1 yr
-BMP-5yrs
-BMP-10yrs
10 20 30 40 50
Number of BMPs
a:
10 20 30 40 50
Number of Conduits
0 10 20 30 40 50 60
Number of BMPs/Conduits
Figure 2-2. Comparison of runtime for various simulation units and periods.
The results (Figure 2-3) also reveal that conduit simulation consumes much longer run times
(approximately nine times longer) than BMP simulation, mainly because it requires solving the coupled
continuity equation and Manning's equation for conduit flow routing.
160
140 -
120 -
100 -
40 -
20 -
0
lBMP-1yr CONDUIT-1 yr BBMP/CONDUIT-1 yr
133
135
25
Number of BMPs/Conduits
Figure 2-3. Comparison of runtime for 1-yr simulation of various units.
To reduce the computational burden, it is desirable to simplify the routing simulation, particularly through
conduits, during optimization runs. Additional research is needed to develop credible methods that
balance the computational efficiency and accuracy of hydraulic routing in conduits.
2-19
-------
2.3. Optimization Module
SUSTAINS optimization module uses evolutionary optimization techniques to identify the most cost-
effective BMP selection and placement alternatives that satisfy the user-defined decision criteria. The
optimization module is interfaced with the BMP and land modules during the search process in an
iterative and evolutionary fashion to pass the performance data from simulation modules and cost
information of a viable set of BMPs to the optimization module. The optimization module then
systematically compares the cost and performance data and modifies the search path to generate a new set
of viable BMP options and repeats the process until the set criteria to end the iteration are reached.
Two optimization techniques are supported by SUSTAIN: the Scatter Search method and the Non-
dominated Sorting Genetic Algorithm II (NSGA-II) method. The Scatter Search method is a meta-
heuristic search technique that has been explored and used in optimizing complex systems (Glover et al.
2000; Laguna and Marti 2002; Zhen et al. 2004). NSGA-II is an advanced genetic algorithm based on
Pareto dominance, and uses non-domination and distribution instead of fitness value to score individuals
(Deb et al. 2000). Section 3.5.2 has more expanded discussion of the Scatter Search and NSGA-II
methods.
To validate the performance of both search techniques, they were tested against a known solution. The
objectives were twofold: (1) to evaluate the ability of Scatter Search to pick a known best solution for a
single BMP given multiple pollutant performance functions and multiple pollutant load reduction
objective criteria, and (2) to evaluate the ability of NSGA-II to generate a cost-effectiveness curve for a
known, linear solution for a single BMP.
For the known best solution, a hypothetical BMP was constructed that reduced both sediment and total
nitrogen loads. The properties of this BMP were specified to yield a sediment removal effectiveness that
was exactly double its nitrogen removal effectiveness. The cost-effectiveness curves were divided into 10
equally spaced intervals. The objective functions for the optimization tests were to minimize the cost of
achieving 20 percent, 40 percent, and 90 percent pollutant removal for both sediment and nitrogen (for a
total of six hypothetical scenarios, labeled A through F). The scenario objective functions and results are
as follows:
A. Minimize the cost to achieve a 20 per cent sediment and 10 per cent nitrogen removal. There is a
minimum cost solution that achieves both of these criteria.
B. Minimize the cost to achieve a 20 per cent sediment and 20 per cent nitrogen removal. Nitrogen is
the limiting pollutant that increases optimal cost from Scenario A.
C. Minimize the cost to achieve a 40 per cent sediment and 20 per cent nitrogen removal. There is a
minimum cost solution that achieves both of these criteria.
D. Minimize the cost to achieve a 40 per cent sediment and 40 per cent nitrogen removal. Nitrogen is
the limiting pollutant that increases optimal cost from Scenario C.
E. Minimize the cost to achieve a 90 per cent sediment and 40 per cent nitrogen removal. There is a
minimum cost solution that achieves both of these criteria.
F. Minimize the cost to achieve a 90 per cent sediment and 90 per cent nitrogen removal. There is no
possible solution that can achieve both of these criteria.
The scenario results are presented in Figure 2-4, where the green circles represent both the known
solution and the solution selected by the optimizer.
2-20
-------
Nitrigen li Limiting Constraint
10% 20% 30% 40% 50% 60% 70% 80% 90% 1
Percent Reduction
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percent Reduction
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percent Reduction
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percent Reduction
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percent Reduction
Figure 2-4. Scatter Search evaluation scenario results.
The ability of NSGA-II to find and create the linear nitrogen and sediment removal cost-effectiveness
curves was also evaluated. Figure 2-5 shows the NSGA-II solution plotted against the known linear
solutions. Both the Scatter Search and NSGA-II optimization techniques were able to find a known linear
solution with 100 percent efficiency (in terms of accuracy). In addition, both optimization techniques
were able to select an optimum solution, given multiple objectives for controlling sediment and nitrogen
simultaneously. Finally, the NSGA-II technique was able to predict a known linear cost-effectiveness
curve.
* Sediment (Known)
Sediment (NSGAII)
Nitrogen (Known)
Nitrogen (NSGAII)
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percent Reduction
Figure 2-5. NSGA-II evaluation scenario results.
2-21
-------
Operationally, SUSTAINS optimization module incorporates a tiered approach that allows for cost-
effectiveness evaluation of both individual and multiple, nested watersheds to address the needs of both
regional- and local-scale applications.
2.4. Post-Processor
The interpretation of time series results can be a daunting task, especially when multiple output locations,
scenarios (e.g., without BMPs, with BMPs, and pre-developed conditions), and parameters of interest
(e.g., inflows, outflows, pollutant loads and concentrations) must be considered. Natural precipitation-
driven process simulation also produces a highly variable set of responses, ranging in magnitude,
duration, intensity, treatment containment volume, attenuation, and pollutant removal effectiveness. This
information is stored at hourly or sub-hourly intervals, and can span several years, depending on the
length of simulation.
The primary objective of SUSTAINS post-processor is to mine the modeling results to derive the most
meaningful data to characterize the effectiveness of management strategies. The post-processor achieves
this objective through the use of graphical and tabular reports of the model output. The post-processor
uses Microsoft Excel 2003 to develop the following four analysis components of the model outputs.
Storm Event Classification—This component evaluates the precipitation data that drive the simulation
and categorizes them into a series of storm events on the basis of predefined criteria for duration and
antecedent moisture conditions. It produces a set of precipitation events over which BMP performance
will be evaluated (Figure 2-6, for details regarding interpretation of the graph see Section 3.6).
Sorted by Total Precipitation Volume
Sorted by Peak Precipitation Intensity
Selected Internal (Percentile)
• Total Rainfall
|Peak Intensity
Figure 2-6. Example storm event classification graph.
Storm Event Viewer—This component is used to visualize BMP performance hydrographs and
pollutographs for specific storm events. It provides performance measurements at an assessment point for
the specified storm events (Figure 2-7, for details regarding interpretation of the graph, see Section 3.6).
2-22
-------
_ 1.4
^ 1.2
c -i n
o
8 u-4
n n
i
i
i
i
i
•
i
i
i
i
^H
_J
50
45
40
35
30
25
20
15
10
5
0
" Post-Developed
I
I
I
I
ft
A
/
/ 1
/ 1
/ 1
/ I
/ I
Figure 2-7. Example storm event viewer graph.
Performance Summary Report—This component summarizes the performance of the modeled
management strategy for all defined storm events. It paints a picture of the types of storms that are
contained at the selected assessment point, as well as those that are bypassed and untreated. The report
can be used to identify appropriate design storms that meet a specified treatment objective for modeling
evaluation (Figure 2-8, for details regarding interpretation of the graph see Section 3.6).
Relative Contribution
-Weighted Average Reduction
Precipitation Event Reductions Pre-Developed Condition
50%
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Precipitation Event Volume (in.)
Figure 2-8. Example performance summary report graph.
Cost-Effectiveness Report—This component generates the cost-effectiveness curve at a specified
assessment point and plots other inferior solutions that were attempted during the simulation. It also
characterizes several key indicators associated with the cost-effectiveness curve, such as BMP surface
storage volume, surface area, and soil storage volume. The knowledge of how these indicators change at
various points along the cost-effectiveness curve would help develop cost-effective strategies (Figure 2-9,
for details regarding interpretation of the graph, see Section 3.6).
2-23
-------
55%
• All Solutions
• Cost-Effectiveness Curve
O Selected Simulation
25%
$0.0
$05
$1.0
$1 5
$2.0 $2 5
Cost ($ Million)
$3.0
$35
$4.0
$45
Figure 2-9. Example cost-effectiveness curve.
2.5. Summary
This chapter provided an overview of the structure of SUSTAIN, system requirements, major simulation
and optimization modules, input data processing and output results interpretation, and interoperational
linkages among the modules. The framework has been designed to maximize computational efficiency
while preserving the flexibility to represent a wide range of watershed conditions and management
practice configurations.
Chapter 3 provides a detailed description of each module and relevant algorithms, and the technical
underpinnings and assumptions associated with the framework.
2-24
-------
Chapter 3 Simulation Methods and Algorithms
This chapter describes the methods and algorithms that were built into SUSTAIN. SUSTAIN's simulation
capabilities are embedded in three modules—land simulation, BMP simulation, and conveyance
simulation—that are used in combination to support a range of watershed simulation needs and are tied
together by an overarching framework manager that performs data management, BMP site selection, the
routing network creation, and other functions. In addition to the simulation modules, SUSTAIN includes
an optimization module to evaluate and identify cost-effective BMP placement strategies and a post-
processor to facilitate analysis and interpretation of model results. Table 3-1 summarizes an overview of
the modules, components, and methods included in SUSTAIN.
The FM serves as the command module of SUSTAIN to manage data for system functions, provide
linkages between the system modules, and create the necessary routing network required for simulation
and optimization activities.
In the land simulation module, 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 2001); the particle size distribution for the eroded sediments is represented as fractional
distribution of sand, silt, and clay.
The BMP module uses a combination of process-based algorithms, including weir and orifice control
structures, flow routing and pollutant transport, infiltration, evapotranspiration, and pollutant loss/decay
simulation. A functional BMP module was incorporated by adopting the Prince George's County BMP
Module (Tetra Tech 2001). The module was further enhanced by adding continuous stir tank reactors
(CSTRs) in series and associated pollutant removal based on Kadlec and Knight's (1996) k'-C* model, a
Green-Ampt infiltration option, and dynamic simulation of evapotranspiration. For stream buffer strip
simulation, the process-based algorithm applied in the VFSMOD was adopted.
The conveyance simulation module is used to simulate movement of water and pollutants among the
physical parts of the watershed (land, BMP, conduit, reach). The module simulates flow and pollutant
routing and employs the kinematic wave and CSTR approach used in the SWMM Transport
compartment. A sediment transport component that employs the well-known algorithms from the HSPF
and LSPC models is included.
This chapter documents the function, design, inputs, and outputs of each SUSTAIN module and all
associated components.
3-25
-------
Table 3-1. Modules and Components in SUSTAIN
Module
Framework Manager
Land module
BMP module
Conveyance module
Optimization module
Post-processor
Component
Data management component
BMP site selection component
Routing network component
Weather component
Hydrology component
Water quality component
Simulation component
Buffer strip component
Aggregate BMP component
Cost database component
Routing component
Storm evaluation
Storm viewer
Performance summary report
Cost-effectiveness report
Methodology
Path identification
CIS Interfaces
ArcGIS
Site suitability criteria
Highlighted suitability areas
ArcGIS interfaces
Precipitation
Snowmelt
Evaporation
Internal simulation
Infiltration: Green- Ampt equation
Overland flow
Groundwater flow
External simulation
Unit area flows and loads
File linkages
Erosion
Pollutant buildup
Pollutant washoff
Particle size distribution
Storage routing method
Infiltration/filtration methods
Evapotranspiration method
Underdrain method
Pollutant routing and removal methods
Overland flow routing
Pollutant interception
Interception
Treatment
Storage
Unit area cost estimates
Construction components
Flow routing
Sediment
Transport pollutant routing
Problem formulation
NSGA-II
Scatter Search
Tiered analysis
3-26
-------
3.1. Framework Manager
Table 3-2 provides an overview of the required inputs, the methods used to manage and process the inputs
and the resulting outputs from the FM module of SUSTAIN. Three major components of the FM are
described in Sections 3.1.1, 3.1.2, and 3.1.3.
Table 3-2. Summary of Inputs, Methods, and Outputs in FM
Framework Manager
Inputs
- Geodatabase file with spatial and tabular data
- Cost database file
- Define GIS layers and lookup tables
- Define suitability criteria for BMP locations (optional)
- Place BMPs at the suitable locations on the map
- Define assessment point(s)
Methods
- BMP suitable locations map is created using the BMP siting tool (optional)
- BMPs are placed on the map using the BMP placement tool
- BMP drainage areas are delineated using the auto/manual delineation tools. The system allows
importing the existing drainage areas
- A routing network is created by connecting a drainage area to a BMP and a downstream BMP
through a reach or a conduit conveyance system
-LAND, BMP, REACH, and CONDUIT simulations are carried out by calling the dynamic link
libraries compiled in visual C++
Outputs
- BMP suitable locations map
- Input text file for LAND simulation module
- Input text file for BMP, REACH, and CONDUIT simulation modules
- Model simulation results
- Display simulation results at the assessment point
3.1.1. Data Management Component
The data management component compiles and organizes the data required to run SUSTAIN, including
geographic data in vector, raster, and/or tabular format. It also includes a cost database (BMPCosts.mdb)
in Microsoft Access format. SUSTAIN supports the use of either a personal geodatabase or a file-based
geodatabase as the primary repository for all geographic data sets. A file-based geodatabase is a
collection of spatial and/or temporal data sets organized into a series of indexed folders and files. Each
file-based geodatabase can store up to one terabyte of information. This option is recommended over the
use of a personal geodatabase, which is limited in size to two gigabytes and does not support the storage
of raster information. During various data processing steps, SUSTAIN creates a large number of
intermediate data sets and stores them on the hard disk specified.
The data management component is navigated through two levels using GIS interfaces. The first level of
data management identifies the path to the cost database, the file geodatabase, and the temporary
directory where all the intermediate geographic data are stored. The second level of data management
identifies the required data layers. The required data set includes land use data in raster format, a land use
lookup table, stream network, DEM (mandatory for the automatic watershed delineation option), and time
series data (mandatory for the external land simulation option).
3-27
-------
3.1.2. BMP Site Selection
The BMP siting tool was developed to assist users in selecting suitable locations for different types of low
impact development (LID) techniques or conventional BMPs. The tool is implemented using ESRI's
Arc View 9.3 and the Spatial Analyst extension. Site suitability is used as the dominant factor in
identifying potential site locations (USEPA 1999a). Using GIS analysis and up to eight base data layers,
the siting tool helps users identify suitable sites for placement of structural BMPs on the basis of
suitability criteria including elevation, slope, soil type, urban land use, roads, water table depth, stream
location, and drainage area. Table 3-3 describes these eight GIS data layers that are used as the base input
data for the tool.
Table 3-3. GIS Data Requirement for BMP Suitability Analysis
GIS Layer
DEM
NLCD Land Use
Percent
Imperviousness
Soil
Urban Land Use
Road
Stream
Groundwater Table
Depth
Format
Raster file
Raster file
Raster file
Shape file
Shape file
Shape file
Shape file
Shape file
Description
The DEM is used to calculate the drainage slope and drainage area that are
used to identify the suitable locations for BMPs.
The USGS Multi-Resolution Land Characteristics Consortium NLCD land
use grid is used to eliminate the unsuitable areas for BMPs.
The impervious grid is used to identify the suitable locations for BMPs for
the given suitability criteria.
The soil data contain the soil properties such as hydrological soil group,
which are used to identify suitable locations for BMPs.
The urban land use data contain the boundaries for the buildings and the
impervious areas needed to identify suitable locations for LIDs.
The road layer is used to identify suitable locations for some BMPs that
must be placed within a specific road buffer area.
The stream layer is used to define a buffer so that certain BMP types can be
placed outside the buffer to minimize the impact on streams.
The groundwater table depth layer is used to identify suitable locations for
the infiltration BMPs; derived from monitoring data.
Source: Lai et al. 2007
The siting tool uses a site suitability criteria matrix and is populated with default criteria that the user can
change to his or her preference or local knowledge. The default criteria in the tool as shown in Table 3-4
are derived from two EPA reports (USEPA 2004a, 2004b). Users can modify these criteria through the
interface.
The output of the BMP siting tool analysis is a spatial map that highlights the areas that meet the selected
default or user-specified site criteria for placement of the available BMPs. The system stores the data in
the BMP suitability map that can be used as a backdrop during the placement of BMPs for simulation
runs. Multiple spatial maps can be created for project areas on the basis of the various criteria selected by
the user. Users can also import additional data sets or geographic coverages to further refine the utility of
the spatial maps. For example, if BMPs can be placed only on publicly owned land, an ownership layer
can be superimposed on the siting tool results to highlight potential BMP placement locations on such
land.
3-28
-------
Table 3-4. Default Criteria for BMP Suitable Locations Used in SUSTAIN
BMP
Bioretention
Cistern
Constructed
Wetland
Dry Pond
Grassed Swale
Green Roof
Infiltration
Basin
Infiltration
Trench
Porous
Pavement
Rain Barrel
Sand Filter
(non-surface)
Sand Filter
(surface)
Vegetated
Filterstrip
Wet Pond
Site Suitability Criteria
Drainage
Area
(acre)
<2
~
>25
>10
<5
~
<10
<5
<3
~
<2
<10
~
>25
Drainage
Slope
(%)
<5
~
<15
<15
<4
~
<15
<15
<1
~
<10
<10
< 10
<15
Imperviousness
(%)
>0
~
>0
>0
>0
~
>0
>0
>0
~
>0
>0
>0
>0
Hydrological
Soil Group
A-D
~
A-D
A-D
A-D
~
A-B
A-B
A-B
~
A-D
A-D
A-D
A-D
Water
Table
Depth
(ft)
>2
~
>4
>4
>2
~
>4
>4
>2
~
>2
>2
>2
>4
Road
Buffer
(ft)
<100
~
~
~
<100
~
~
~
—
-
—
~
<100
~
Stream
Buffer
(ft)
>100
~
> 100
> 100
~
~
>100
>100
~
~
> 100
> 100
~
>100
Building
Buffer
(ft)
-
<30
~
~
~
~
~
~
~
<30
—
~
~
-
To conceptualize the physical function of BMPs with regard to their associated landscape, four categories
(or types) of BMPs are presented in the siting tool: (1) point LID, (2) point BMP, (3) linear BMP, and (4)
area BMP. Point BMPs and LID include practices that capture upstream drainage at a specific location
and can use a combination of detention, infiltration, evaporation, settling, and transformation to manage
flow and remove pollutants. Linear BMPs are narrow linear shapes adjacent to stream channels that
provide filtration of runoff; nutrient uptake; and ancillary benefits of stream shading, wildlife habitat, and
aesthetic value. Area BMPs are land-based management practices that affect impervious area, land cover,
and pollutant inputs (e.g., fertilizer, pet waste). Table 3-5 shows the structural BMP options included in
BMP siting tool.
3-29
-------
Table 3-5. Structural BMP Options Available in the BMP Siting Tool
BMP Option
Bioretention
Cistern
Constructed Wetland
Dry Pond
Grassed Swale
Green Roof
Infiltration Basin
Infiltration Trench
Porous Pavement
Rain Barrel
Sand Filter (non-surface)
Sand Filter (surface)
Vegetated Filterstrip
Wet Pond
BMP Type
Point LID
Point LID
Point BMP
Point BMP
Linear BMP
Area BMP
Point BMP
Linear BMP
Area BMP
Point LID
Linear BMP
Point BMP
Linear BMP
Point BMP
3.1.3. Routing Network
The system routing network as conceptualized in Figure 3-1 provides the connectivity among the various
simulation components (land, BMP, conduit, reach) at the watershed level. After placing BMPs on the
map and creating the drainage area for each BMP, the framework manager creates the routing network by
connecting the land segments that drain to each BMP and connecting each BMP to the downstream BMP
through a reach or conduit segment. The connections are made automatically if the DEM is used to
delineate the drainage areas. Alternatively, those connections can be made manually using the network
tools in the framework manager interface.
Figure 3-1. The routing network showing the connections
among the simulation components.
3-30
-------
3.2. Land Module
The land simulation module is used to derive runoff and pollutant loads from the land in one of two ways.
By default, the land module computes the hydrograph and pollutograph using algorithms adapted from
the SWMM (version 5) land surface compartment and sediment algorithms adapted from the HSPF
model. That is called the internal simulation option, which has been tested and verified to ensure accurate
transformation of the code from the original models. The second option is to use externally generated
time series to represent hydrology and water quality at the landscape level. The external option allows
importation of the hydrograph and pollutograph for each land use category from a pre-calibrated external
watershed model such as HSPF or LSPC.
Figure 3-2 is a schematic of the land simulation processes that produce runoff from land including time-
varying rain or snow accumulation and melting, evaporation from ponded surface, infiltration of rain or
snowmelt into unsaturated soil, percolation of infiltrated water into groundwater, and nonlinear reservoir
routing of overland flow.
Infiltration
Evapotranspiration
/\ Surface Runoff
V
LAND Simulation Processes
V
Groundwater Recharge
Figure 3-2. Schematic showing the land simulation processes.
Table 3-6 provides an overview of the required inputs, the methods used to process the inputs and
simulate the hydrologic and water quality processes occurring on the landscape, and the resulting outputs
of the land simulation module.
3-31
-------
Table 3-6. Inputs, Methods, and Outputs of the Land Module
Land Module
Inputs
Internal Option
- Define pollutants
-Define fraction of total sediments as sand, silt, and clay from each land use category
- Reclassify/group land use categories (optional)
- Define meteorological data (user-defined time step)
- Define pollutant properties
- Define land use properties
- Define rain gauge properties
- Define aquifer properties (optional)
- Define snowpack properties (optional)
- Define watershed properties
External Option
- Hourly time step
- Define pollutants
-Define fraction of total sediments as sand, silt, and clay from each land use category
- Reclassify/group land use categories (optional)
- Assign pre-calibrated land output time series for each land use group
Methods
Internal Option
- Weather data is processed to convert precipitation values to snow or rain according to the
temperature
- Snowmelt is computed using the degree-day and National Weather Service equations
- Evapotranspiration is calculated using a constant ET rate or time series values supplied by the user
- Infiltration is computed using the Green-Ampt equation
- Overland flow is computed using the Manning's equation
- Groundwater outflow is computed as a function of groundwater and surface water heads
- Production and removal of sediments on pervious land is computed using the processed-based
algorithms adopted from the HSPF model
- Buildup and washoff of sediments on impervious land is computed using the algorithms adopted
from the SWMM
- Total sediment is divided into three sediment classes (sand, silt, and clay) according to a user-
specified fraction for each class from each land use category
- Pollutants buildup and washoff rates are computed using the functions adopted from the SWMM
- Outflow, sediment, and pollutants are aggregated for all land use categories
External Option
- Total sediment is divided into three sediment classes (sand, silt, and clay) according to a user-
specified fraction for each class from each land use category
- Within each catchment area, unit-area outflow, sediment, and pollutant loads for each land unit are
multiplied by actual land area to derive aggregate land contribution
Outputs
- Hourly outflow time series
- Hourly sediment (sand, silt, and clay) concentration time series
- Hourly pollutant concentration time series
3-32
-------
Three major components compose the land simulation module: weather, hydrology, and water quality.
Multiple options are provided for representing various processes as are outlined in Table 3-7.
Table 3-7. Land Simulation Methods Used in SUSTAIN
Process
Rainfall
Snowmelt
Evaporation
Infiltration
Groundwater flow
Overland flow
Pollutant buildup
Pollutant washoff
Street cleaning
Sediment erosion
and transport
Particle size
distribution
Option 1
Weather data file
Degree-Day equation;
NWS equation
Constant value
Green-Ampt
Modified two-zone
groundwater model
Non-linear reservoir
Power function
Exponential function
User-specified pollutant
removal efficiency
Production and removal
from the pervious land;
Buildup and washoff from
the impervious land
User defined (sand, silt,
clay)
Option 2
~
~
Monthly average
value
~
~
Exponential
function
Rating curve
~
~
Option 3
~
~
User-supplied
time series
~
~
Saturation
function
Event mean
concentration
~
—
Reference
Rossman 2005
Rossman 2005
Rossman 2005
Rossman 2005
Rossman 2005
Bicknell et al.
2001
Rossman 2005
Rossman 2005
Rossman 2005
Rossman 2005
Bicknell et al.
2001
Rossman 2005
Bicknell et al.
2001
The following paragraphs explain in greater detail the methods and algorithms implemented in the
weather, hydrology, and water quality components of the land simulation module.
3.2.1. Weather Component
The weather component of the land module is adapted from the SWMM atmospheric compartment
(Rossman 2005) that uses the daily air temperature, evaporation, and wind speed data from the user-
specified climate file. The format for climate file is consistent with that used in the SWMM, where each
line in the file contains a recording station name, year, month, day, maximum temperature, minimum
temperature, and optionally, the evaporation rate and wind speed. The data must be in U.S. units:
temperature in degrees F, evaporation in in./day, and wind speed in mi/hr, all separated by one or more
spaces.
An excerpt from the climate file format might look as follows:
ST93738 2007 1 1 43 32 0.12 13.9
ST93738 2007 1 2 45 23 0.04 5.84
ST93738 2007 1 3 54 24 0.07 4.21
The precipitation data is input in a separate file where each line of the file contains the station ID, year,
month, day, hour, minute, and non-zero precipitation reading, all separated by one or more spaces.
3-33
-------
An excerpt from the precipitation file format might look as follows:
ST4489032007 1 100000.12
ST4489032007 1 1 01 000.04
ST448903 2007 1 2 16 00 0.07
Precipitation
SUSTAINS land simulation module uses the precipitation datatype in any one of these three formats: (1)
intensity, where the value is an average rate (in./hr) over the recording interval; (2) interval volume,
where the value is the volume of rain that fell in the recording interval (in.); or (3) cumulative volume,
where the value represents the cumulative rainfall that has occurred since the start of the last series of
non-zero values (in.). The precipitation values are converted to snow amount according to the user-
specified temperature below which precipitation falls as snow instead of rain (Rossman 2005).
Snowmelt
Snowmelt is computed at each step using a degree-day equation when it is dry and the National Weather
Service (NWS) River Forecast System - Snow Accumulation and Ablation Model (Anderson 1973)
during rainfall periods (Huber and Dickinson 1988). The two equations are presented below.
Degree-Day Equation
During periods of no rainfall, snowmelt is computed by the Degree-Day equation:
Smelt = DHMx(Ta-Tbase) (3.1}
where
Smelt = snowmelt rate (water equivalent in./hr),
DHM= melt coefficient (water equivalent in./hr-°F),
Ta = air temperature (°F), and
Tbase = snowmelt base temperature (°F).
NWS Equation
During periods of rain, snowmelt is computed using Anderson's (1973) NWS equation. Anderson
combines the appropriate terms for each heat budget component into one equation for the melt rate:
Smelt = (Ta - 32] x (o. 00167 + Syx U ad] + 0.007x Prec)+ 8.5 x U ad] x(EA-0.18] (3-2)
where
Smelt = snowmelt rate (water equivalent in./hr),
Ta = air temperature (°F),
5y=7.J/(in.-Hg/°F),
y= psychometric constant (in.-Hg/°F),
Uadj = wind speed function (in./in.-Hg-hr),
Prec = rainfall intensity (in./hr), and
EA = saturation vapor pressure at air temperature (in.-Hg).
The psychometric constant y is calculated as
y = 0.000359x PA (3.3)
where
y= psychometric constant (in.-Hg/ F) and
3-34
-------
PA = atmospheric pressure (in.-Hg).
Average atmospheric pressure is calculated as a function of elevation, z:
PA = 29.9-1.02
1000
+ 0.0032
1000
(3-4)
where
z = average elevation (ft).
The wind speed function, Uadj, accounts for turbulent transport of sensible heat and water vapor.
Anderson (1973) gives the following equation:
where
Uadj= 0.006
Uadj = wind speed function (in./in.-Hg-hr) and
M = average wind speed (mi/hr).
(3-5)
The saturation vapor pressure, EA, is given by the following exponential approximation:
EA = 8.1175 x 106 xexp
-7701.544
(Ta+405.0265}
(3-6)
where
EA = saturation vapor pressure at air temperature (in.-Hg) and
Ta = air temperature (°F).
Evaporation
Evaporation is calculated for standing water on land surfaces, subsurface water in groundwater aquifers,
and water held in storage units. On the basis of the approach used in SWMM, evaporation is subtracted
from the rainfall or water storage area prior to calculating infiltration. Evaporation rates can be stated as
one of these three forms: a single constant value, a set of monthly average values, or a user-supplied time
series input in the climate data file. If a climate file is used, the user-specified monthly pan coefficients
are used to convert the pan evaporation data to free water-surface values (Rossman 2005).
3.2.2. Hydrology Component
The hydrology component simulates the rainfall runoff processes and provides the linkage between the
meteorological information and movement of water into and across the land surface. The methods
selected provide time variable response to meteorological inputs while using well-established methods for
simulation. By building on methods that are established in the literature, users can rely on literature
values and industry practice to develop input parameters for initial application and to use as a starting
point for calibration.
The hydrology component of the land simulation module is adapted from the SWMM land surface and
groundwater compartments (Rossman 2005). SUSTAINuses the Green-Ampt method to compute the
amount of infiltration of rainfall on the pervious land area into the unsaturated upper soil zone. The
surface runoff is computed using Manning's equation.
3-35
-------
Infiltration Using Green-Ampt Equation
The Green-Ampt infiltration method assumes that a sharp wetting front exists in the soil column which
separates the unwetted 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).
If/ > Ks, then/= /, until F = F, =— —
1 - I/KS
Following surface saturation,
„ dF T
F
(3-7)
For / > K,, and/= / for / < K,
where
/ = inflow rate (in./hr),
F= amount of infiltration (in.),
Fs = amount of infiltration up to surface saturation (in.),
F = infiltration rate (in./hr),
Ks = saturated hydraulic conductivity (in./hr),
6S = saturated moisture content,
9i = initial moisture content, and
ytf = average wetting front suction head (in. of water).
This differential equation is solved iteratively to determine/at each time step by using Newton-Raphson
method. The infiltration volume during the time interval is equal to the inflow volume if the surface does
not saturate. If saturation occurs during the time interval, the infiltration volumes over each stage of the
process within the time steps are calculated and summed. When there is no inflow, any water ponded on
the surface is allowed to infiltrate and added to the cumulative infiltration volume. In using the Green-
Ampt method, a complication occurs when the inflow rate starts at a value above, drops below, and then
rises above Ks again during the infiltration computation. In such a case, the moisture content needs to be
redistributed as the assumption of saturation from the surface down to the wetting front does not hold. A
major advantage of the Green-Ampt method is that the input parameters (i.e., Ks, y/f,6s, 9t) can be
determined from physical measurements. As shown in Table 3-8, Rawls et al. (1983) provide typical
values for the parameters.
Overland Flow
The conceptual view of the surface runoff calculation in SUSTAIN is illustrated in Figure 3-3, which is
adapted from the SWMM5 user's manual (Rossman 2005). The surface of each subwatershed is treated
as a nonlinear reservoir. Inflow comes from precipitation and upstream subwatersheds. The outflows are
infiltration, evaporation, and surface runoff to downstream areas. The maximum surface storage capacity
is composed of ponding volume, surface wetting volume, and interception volume, normalized by surface
area and is represented as depth. Surface runoff per unit area, Q, occurs only when the surface water
3-36
-------
depth exceeds the maximum surface storage depth, dp, in which case the outflow is given by Manning's
equation:
(3-8)
n
where
Q = outflow rate (cfs),
W= subwatershed width (ft),
n = Manning's roughness coefficient,
d = water depth (ft),
dp= depth of depression storage (ft), and
S = subwatershed slope (ft/ft).
Table 3-8. Green-Ampt Parameters
Soil Texture
Class
Sand
Loamy Sand
Sandy Loam
Loam
Silt Loam
Sandy Clay Loam
Clay Loam
Silty Clay Loam
Sandy Clay
Silty Clay
Clay
Saturated
Hydraulic
Conductivity
(in./hr)
4.74
1.18
0.43
0.13
0.26
0.06
0.04
0.04
0.02
0.02
0.01
Suction Head
(in.)
1.93
2.40
4.33
3.50
6.69
8.66
8.27
10.63
9.45
11.42
12.60
Porosity
(Fraction)
0.437
0.437
0.453
0.463
0.501
0.398
0.464
0.471
0.430
0.479
0.475
Field
Capacity
(Fraction)
0.062
0.105
0.190
0.232
0.284
0.244
0.310
0.342
0.321
0.371
0.378
Wilting Point
(Fraction)
0.024
0.047
0.085
0.116
0.135
0.136
0.187
0.210
0.221
0.251
0.265
Source: Rawlsetal. 1983
Evaporation Rain/Snowmelt
Infiltration
Source: Rossman 2005
Figure 3-3. Conceptual view of surface runoff.
Subwatershed width (W) can be estimated by dividing the subwatershed area by the length of the
representative flow path. The depth of water over the subwatershed is continuously updated with time by
solving a water balance Equation (3-9) for the subwatershed.
3-37
-------
dd L4W(d f3 sia = + WCQN(d f3 (3.9)
dt A-n
49 -W -S
'4
1'2
.
A-n
where
WCON = parameter for overland flow routing,
d = water depth (ft),
t = time (sec),
W= subwatershed width (ft),
A = surface area of subwatershed (ft2),
n = Manning's roughness coefficient,
ie = rainfall excess (ft/s),
dp= depth of depression storage (ft), and
S= subwatershed slope (ft/ft).
Groundwater Flow
Accounting for groundwater inputs is of greater significance with larger watersheds for the purpose of
accounting for baseflow in streams. Groundwater simulation provides an important link between surface
and subsurface flow. In SUSTAIN, groundwater flow is simulated using the SWMM formulation
(Rossman 2005). It also employs some modifications based on HSPF techniques for simulating the
interaction between saturated soil water and unsaturated soil water when the water table approaches or
rises above the ground (Bicknell et al. 2001). Those modifications were made to smooth out the
groundwater outflow response to account for the interaction of rising groundwater storage with the
unsaturated zone storages, cohesive water storage, and gravity water storage. Without those
modifications, the groundwater level is a function of gravity storage only. The water in cohesive water
storage is not available for groundwater outflow but is subject to evapotranspiration. It is assumed in
SUSTAIN 'that there is no interaction between land groundwater and BMP deep percolation. Deep
percolation water from BMPs is lost from the system.
Two-zone Groundwater Model from SWMM
In this formulation, groundwater flow is a function of groundwater and surface water heads in the
discharge channel, as shown in Equation (3-11).
Qg, = 4 (Hgv - E)B> - A2 (Hsw - E)B> + A3 Hgw Hsw (3-1 1
where
Qgw = groundwater flow (cfs),
Hgw = elevation of groundwater table (ft),
HSW = elevation of surface water at receiving node (ft),
E = elevation of node invert (ft),
A! = groundwater flow coefficient,
BI = groundwater flow exponent,
A2 = surface water flow coefficient,
B2 = surface water flow exponent, and
A3 = surface -groundwater interaction coefficient.
3-38
-------
The SWMM's two-zone groundwater model is shown in Figure 3-4. The upper zone is unsaturated at
varying moisture content, which is updated at each time step of infiltration and regeneration of infiltration
capacity simulation. The lower zone is saturated, and hence its moisture content is fixed at the soil
porosity.
•*TOT
Source: Rossman 2005
Figure 3-4. Two-zone groundwater model adapted from SWMM.
The fluxes shown in Figure 3-4 are expressed as volume per unit area per unit time and consist of the
following:
fi= infiltration from the surface;
/EU = evapotranspiration from the upper zone, which is a fixed fraction of the unused surface
evaporation;
fu = percolation from the upper to lower zone, which depends on the upper zone moisture content 6
and depth dv;
fsL = evapotranspiration from the lower zone, which is a function of the depth of the upper zone dv;
fL = percolation from the lower zone to deep groundwater, which depends on the lower zone depth
dL; and
fo = lateral groundwater interflow to the conveyance network, which depends on the lower zone
depth dL as well as depths in the receiving channel.
Two-zone Groundwater Interaction from HSPF
The effects of a rising water table differ under low water table and high water table conditions. A
modified, two-zone groundwater representation in Figures 3-5 and 3-6 provides a higher-resolution option
for characterizing subsurface conditions and improvements to runoff computation. Figure 3-5 shows the
schematic of a two-zone soil moisture storage layer under low water table conditions, while Figure 3-6
illustrates that under high water table conditions.
The saturated and unsaturated zone interactions are a function of water transfer rates, existing saturation
levels, and physical characteristics of the soils such as porosity. For modeling purposes, the total porosity
is divided into two parts: porosity in micropores (rjmi, cohesion water) and porosity in macropores (rjma,
gravitational water). Cohesion water is bonded in soil by capillary forces, and it is roughly equal to the
difference between the wilting point and field capacity. Gravitational water drains from soils in the
unsaturated zone by gravity forces.
The groundwater level is the elevation of the saturated zone above an arbitrary datum such as mean sea
level. The active groundwater storage is gravity water stored above the water elevation of a channel that
3-39
-------
is within or adjacent to the land. A lower elevation is the maximum depth where soil moisture varies
seasonally due to evapotranspiration.
When the groundwater elevation is below the lower elevation (within the saturated Zone 1 as shown in
Figure 3-5), there is no interaction between the saturated and the unsaturated zones. Groundwater
elevation in this zone is computed as a function of the groundwater storage and the total porosity (the sum
of macropores and micropores).
n . n
Land Surface
Lower
Evapotranspiration
Depth
Active Groundwater
Storage
^B Datum
Source: Bicknell et al. 2001
Figure 3-5. Two-zone soil moisture storage under low water table condition.
The groundwater elevation in this zone is calculated as:
(3-12)
where
Hgw = groundwater elevation (ft),
Sgw = total groundwater storage (ft),
rjni = soil porosity in micropores (large pores for cohesive water), and
?7ma = soil porosity in macropores (large pores for gravitational water).
When the groundwater elevation reaches the higher elevation (within the unsaturated Zone 2 as shown in
Figure 3-6), the groundwater storage starts interacting with the upper zone storages. Rising groundwater
that occupies micropores is reassigned to the upper zone cohesive water storage. Groundwater storage
shares macropores with the upper zone gravity water storage and is subject to evapotranspiration.
Changes in groundwater storage are distributed between upper zone storages and groundwater storage,
according to their relative saturation levels. Groundwater elevation in this zone is a function of upper
zone water in macropores and groundwater storages and is calculated as:
Hew — •
(3-13)
where
Het = elevation at the maximum depth due to seasonal evapotranspiration (ft),
Sgw = total groundwater storage (ft), including all water in the lower saturated zone and in
macropores of the upper zone saturated soil,
3-40
-------
Suz = upper zone water storage in macropores of the unsaturated soil (ft), and
Slz = groundwater storage below Het (ft), which is equal to Het x (rjmi + rjma).
Land Surface
Lower
Evapotranspi ration
Depth
Active
Groundwater
Storage
Datum
Source: Bicknell et al. 2001
Figure 3-6. Two-zone soil moisture storage under high water table conditions.
When the groundwater elevation is below Hch, the channel water elevation, there will be no outflow from
the groundwater storage. When the groundwater elevation is above Hch, the groundwater outflow is
computed as a function of the active groundwater storage, i.e., the gravity water storage above the channel
water level.
3.2.3. Water Quality Component
The water quality component performs the transport of pollutants on the basis of total flow (runoff and/or
infiltrated groundwater outflow) computed in the hydrology component. The simulation methods
included for routing of sediments and pollutants are adapted from the SWMM land surface compartment
(Rossman 2005) and from HSPF for sediment production and removal from pervious lands (Bicknell et
al. 2001). SUSTAIN can simulate the generation and transport of any number of user-defined pollutants
and divides them into two major groups: sediment and non-sediment pollutants. To facilitate sediment
routing, the total sediment load is divided into three sediment classes—sand, silt, and clay—and the
model allows users to define their distribution fractions. For pollutants that are associated with sediment,
co-fractions are used to quantify the mass of pollutant as a direct proportion of sediment mass. The total
sediment/non-sediment load is simulated for each defined land use category, and then the total is summed
within each subwatershed or BMP drainage area for routing to a BMP or conduit component.
Pervious Land Segment
SUSTAIN computes the sediment load using the HSPF sediment algorithms (SEDMNT) and all non-
sediment pollutant loadings using the SWMM buildup and washoff algorithms for pervious land segments
(Lai etal. 2007).
Production and Removal of Sediment
HSPF simulates sediment production as a function of detachment/washoff or direct scour from a soil
matrix. It assumes that the soil matrix contains an unlimited supply of sediment. User-specified,
physically based model parameters are used to determine the specific rates and modes of how sediment is
made available for transport with runoff. For example, the supporting management practice factor in the
soil detachment by rainfall equation was based on the P factor in the Universal Soil Loss equation
(USLE) (Wischmeier and Smith 1965). It is introduced to better evaluate agricultural conservation
3-41
-------
practices on reducing erosion potential. Figure 3-7 represents the storages and fluxes used to simulate the
detachment, attachment, and removal involved in the erosion processes on the pervious land surface.
SEDscr
Scour of
Soil Matrix
By Water
Source: Bicknell et al. 2001
Figure 3-7. Schematic of sediment production and removal processes.
Removal of sediment by water is simulated as washoff of detached sediment in storage (SEDWO) and scour
of matrix soil (SEDSCI). The washoff process involves two parts: the detachment/attachment of sediment
from/to the soil matrix and the transport of this sediment. Detachment (SEDdet) occurs by rainfall.
Attachment occurs only on days without rainfall; the rate of attachment is specified by parameter Cafflx.
Transport of detached sediment is by overland flow. The scouring of the matrix soil is simplified into one
process by combining both pickup and transport by overland flow.
Sediment Detachment by Rainfall
Kinetic energy from rain falling on the sediment detaches particles which are then available to be
transported by overland flow. The equation that simulates detachment is:
r {At)
where
SEDdet = sediment detached from the soil matrix by rainfall (tons/acre/interval),
At = number of hours/interval,
Cr = fraction of the land covered by snow and vegetation,
P = supporting management practice factor,
K, = coefficient for detachment of soil by rainfall,
PCp = rainfall (in./interval), and
Jr = exponent for detachment of soil by rainfall.
Sediment Removal by Overland Flow
When simulating the washoff of detached sediment, the transport capacity of the overland flow is
estimated and compared to the amount of detached sediment available. The transport capacity is
calculated by the equation:
(3-14)
3-42
-------
where
SEDcap = transport capacity of detached sediment in overland flow (tons/acre/interval),
At = number of hours/interval,
Ks = coefficient for detached sediment by overland flow,
qs = overland flow (in./interval), and
Js = exponent for detached sediment by overland flow.
When SEDcap is more than the amount of detached sediment in storage (Sdet), the flow washes off all the
detached sediment storage, and SEDWO becomes equal to Sdet. However, when SEDcap is less than Sdet, the
situation is transport limiting, so SEDWO is equal to SEDcap.
Direct detachment and transport of the soil matrix by scouring (e.g., gullying) is simulated with the
equation:
where
SEDscr = scour of matrix soil (tons/acre/interval),
At = number of hours/interval,
Kg = coefficient for scour of the matrix soil,
qs = surface flow (in./interval), and
Jg = exponent for scour of the matrix soil.
The sum of the two fluxes, SEDm and SEDscr, represents the total sediment outflow (SEDSO) from the land
segment.
Re-attachment of Detached Sediment
Sediment attachment to the soil matrix is simulated by changes in SEDdet. Because the soil matrix is
considered to be unlimited, no addition to the soil matrix is necessary when this occurs. Sdet is diminished
at the start of each day following a day without precipitation. This decrease is calculated by multiplying
Sdet by (1.0 - Caflx), where Caflx is the fraction by which detached sediment storage decreases each day as a
result of soil compaction. This fraction is a calibration parameter.
Pervious Land Sediment Input Parameters
Table 3-9 shows the recommended ranges of input parameters for simulating the pervious land segment.
The minimum and maximum ranges given in Table 3-9 are for the numerical stability of the model. The
actual values fall within those ranges and are typically defined by calibration and user experience.
Additional HSPF parameterization guidance is available as part of the Better Assessment Science
Integrating Point and Nonpoint Sources (BASINS) technical note series at
http://www.epa.gov/waterscience/basins/docs/tecnote8.pdf (USEPA 2006).
3-43
-------
Table 3-9. List of Sediment Input Parameters for Pervious Land
Parameters
P
Kr
Jr
Caffix
cr
Ks
Js
Kg
JS
Default Value
1.0
0.0
none
0.0
0.0
0.0
none
0.0
none
Min. Value
0.001
0.0
none
0.0
0.0
0.0
none
0.0
none
Max. Value
1.0
none
none
1.0
1.0
none
none
none
none
Units
none
-
-
per day
none
-
-
-
-
Source: Bicknell etal. 2001
Pervious Land Sediment Erosion Calibration
The erosion process on pervious land areas is represented as the net result of detachment of soil particles
by raindrop impact on the land surface and subsequent transport of the fine particles by overland flow.
The primary sediment erosion calibration parameters are as follows:
Kr = coefficient in soil detachment equation (pervious area)
Ks = coefficient in sediment washoff equation (pervious area)
Although a number of additional parameters are involved in sediment erosion calibration, such as those
related to vegetation cover, agricultural practices, rainfall, and overland flow intensity, Kr and Ks are the
primary parameters controlling sediment loading rates. Kr is usually estimated as equal to the erodibility
factor, K, in the USLE (Wischmeier and Smith 1965) and then is adjusted in calibration. Ks is primarily
evaluated through calibration and past experience.
While Table 3-9 presents possible parameter ranges to ensure model stability, Table 3-10 lists the
sediment parameters along with typical and possible minimum and maximum ranges of values based on
application experience over the past 20 years. In addition, the HSPFParm database (USEPA 1999)
provides calibrated parameter values for numerous watersheds across the United States. Additional
guidance in sediment erosion calibration is provided in the HSPF Application Guide (Donigian et al.
1984).
Pollutant Buildup
In SUSTAIN, the amount of pollutant buildup over land is computed using one of the following functions
available in SWMM, as a function of the number of preceding dry-weather days (Rossman 2005).
Power Function
Pollutant buildup, B, is accumulated proportional to time, t, raised to defined power C3 until a maximum
limit is achieved:
where
(3-17)
B = pollutant buildup (mass per unit area, e.g., Ibs/acre),
Ci = maximum buildup possible (mass per unit area, e.g., Ibs/acre),
C2 = buildup rate constant (mass per unit area per unit time, e.g., Ibs/acre/day),
C3 = time exponent, and
3-44
-------
At = time, e.g., number of days.
Table 3-10. Range of Values for Sediment Erosion Parameters for Pervious Land
Name
P
Kr
Jr
Caffh
Cv
K,
J,
Kg
Jg
Definition
Management
practice factor
fromUSLE
Coefficient in
the soil
detachment
equation
Exponent in
the soil
detachment
equation
Daily reduction
in detached
sediment
Fraction land
surface
protected from
rainfall
Coefficient in
the sediment
washoff
equation
Exponent in
the sediment
washoff
equation
Coefficient in
soil matrix
scour equation
Exponent in
soil matrix
scour equation
Units
None
None
Per
day
None
None
None
Range of Values
Typical
Min
0.0
0.15
1.5
0.03
0.0
0.5
1.5
0.0
1.0
Max
1.0
0.45
2.5
0.1
0.9
5.0
2.5
0.5
3.0
Possible
Min
0.0
0.05
1.0
0.01
0.0
0.1
1.0
0.0
1.0
Max
1.0
0.75
3.0
0.5
0.98
10.0
3.0
10.0
5.0
Function of...
Land use,
agricultural
practices
Soils
Soils, climate
Soils,
compaction,
agricultural
operations
Vegetal cover,
land use
Soils, surface
conditions
Soils, surface
conditions
Soils, evidence
of gullies
Soils, evidence
of gullies
Comments
Use the P factor
from USLE
Estimate from the
soil credibility
factor (K) in USLE
Usually start with
value of 2.0
Reduces fine
sediments
following tillage
Seasonal/ monthly
values are often
used
Primary sediment
calibration
parameter
Usually use value
of about 2.0
Calibration, used
only if there is
evidence of gullies
Usually use value
of about 2. 5
Source: Donigian and Love 2003
Exponential Function
Pollutant buildup, B, follows an exponential growth curve that approaches a maximum limit
asymptotically:
(3-18)
where
B = pollutant buildup (mass per unit area, e.g., Ibs/acre),
Ci = maximum buildup possible (mass per unit area, e.g., Ibs/acre),
C2 = buildup rate constant (per time, e.g., per day), and
3-45
-------
At = time, e.g., number of days.
Saturation Function
Pollutant buildup, B, begins at a linear rate then slows down over time until a saturation value is reached:
D CJ
8 = ^ (3-19)
where
B = pollutant buildup (mass per unit area, e.g., Ibs/acre),
Ci = maximum buildup possible (mass per unit area, e.g., Ibs/acre),
C2 = half-saturation constant (days to reach half of the maximum buildup), and
t = time, e.g., number of days.
Pollutant Washoff
The accumulated pollutants on a pervious land surface are washed off during runoff periods from a choice
of available SWMM functions (Rossman 2005), which are all supported by SUSTAIN. These functions
include the exponential washoff, rating curve washoff, and an event mean concentration (EMC).
Exponential Washoff
The washoff load, W, is proportional to the product of runoff raised to the defined power C2 and to the
amount of pollutant buildup remaining at each simulation timestep:
W = C1xqsC2xB (3-20)
where
W= pollutant washoff load (mass per unit area per time, e.g., Ibs/acre/hr),
Ci = washoff coefficient,
C2 = washoff exponent,
qs = runoff rate per unit area (e.g., in./hr), and
B = pollutant buildup (mass per unit area, e.g., Ibs/acre).
Rating Curve Washoff
The rate of pollutant washoff, W, is proportional to the runoff rate raised to the defined power C2:
where
W= pollutant washoff load (mass per unit area per time, e.g., Ibs/acre/hr),
Ci = washoff coefficient,
C2 = washoff exponent, and
qs = runoff rate (user-specified flow units, e.g., in./hr).
Event Mean Concentration (EMC)
This is a special case of Rating Curve Washoff where the exponent is 1.0 and the coefficient Ci represents
the washoff pollutant concentration in mass per volume. The typical EMCs for selected pollutants in
urban runoff are shown in Table 3-11.
3-46
-------
Table 3-11. Typical EMCs in Urban Runoff
Pollutant
TSS (mg/L)
BOD (mg/L)
COD (mg/L)
Total P (mg/L)
TKN (mg/L)
NO2/NO3-N (mg/L)
Total Cu (ng/L)
Total Pb (ng/L)
Total Zn (ng/L)
EMC
180-548
12-19
82-178
0.42-0.88
1.90-4.18
0.86-2.2
43-118
182-443
202-633
Source: USEPA1983
Impervious Land Segment
SUSTAIN computes sediment and all other pollutant loadings from the impervious land segment using the
pollutant buildup and washoff algorithms as defined in the previously described Pervious Land Segment
section. For impervious land segments, SUSTAIN supports a street-sweeping algorithm adopted from
SWMM (Rossman 2005). The user can specify days between sweeping, days since the last sweeping at
the start of the simulation, the fraction of buildup of all sediment types (sand, silt, and clay) available for
removal by sweeping, and the fraction of available buildup for each sediment type removed by sweeping.
The parameters can differ by type of land use.
3.2.4. Important Considerations and Limitations: Land Module
Rainfall-runoff time series data are the drivers for BMP simulation and network routing in SUSTAIN.
The relative modeled response from one land use to another is influenced by the physical characteristics
of the land as defined by the model setup. In any modeling application, many important factors must be
considered. A few of these considerations are highlighted for guidance on configuration and
interpretation of results when applying the land module. They include model testing considerations and
land segmentation.
Model Testing: Calibration and Validation
Calibration and validation is a process during which monitoring data are split into two independent
periods: calibration and validation. Ideally, those are two typical periods (not extreme conditions) within
a typical range of flow conditions. During the calibration period, key parameters are adjusted within
reasonable ranges until the best fit with the observed data is determined. The performance of the
calibrated model is then tested with data from a separate validation period.
The SWMM-based method available in SUSTAIN, as well as similar rainfall-runoff models and methods
used for externally generating time series, though physically based, are empirical in how they are applied.
The models require calibration and validation of estimated model results with observed data. Observed
data that are used for calibration are often collected at locations that drain multiple land use types.
Because individual modeled land use time series are the fundamental units for runoff generation in
3-47
-------
SUSTAIN, there is a need to ensure with reasonable confidence that modeled results are meaningful and
applicable.
An ideal modeling data set for calibration includes monitoring several smaller watersheds, each with
relatively homogeneous land uses (i.e., low-density urban, high-density urban, forest) for a range of storm
conditions. For each monitored storm event, the recommended data includes high-resolution
precipitation, flow hydrograph, and discrete or composite water quality sampling for pollutants of
concern. Using the observed precipitation, flow, and water quality data, the model is developed and
tested for each land use type on the basis of physical characteristics of the drainage area. Standard
calibration techniques are typically applied to assess how well the observed and the modeled time series
data match.
A limitation of any modeling effort is the inability to represent 100 percent of nature's heterogeneity; the
recognition of that fact influences the interpretation and application of calibrated model results. The setup
of SUSTAIN is based on building representative land use times series data that allow users to make
comparisons and extrapolate responses for current and potential future land use and management
conditions. Testing each land use and the combined behavior of mixed land use watersheds against
observed data is used to adjust model parameters for the best fit and to confirm model results. One
common use of the calibrated data set is the extrapolation of results at one location to represent a response
at other similar but ungaged locations. SUSTAIN includes land use reclassification to facilitate the
application of models to various future or managed conditions and can be used to develop locally
homogenized responses of similar land units. The primary objective of the model is to capture the unique
essence of one type of land-use response relative to other types. When the available calibration data are
limited, that objective becomes even more important for either extrapolating time series response from
one calibrated area to another, or adopting loading estimates for various land uses from literature.
Particle Size Distribution
Most land-based erosion and sediment simulation techniques compute total eroded sediment load.
Because the simulation land segments are discretized by the land use category, erosion is simulated
uniquely and is characteristic of that land use category. For sediment produced by each land use, particle
size distribution is represented as fractional distribution of the total sediment. That means that the total
land-based sediment load, multiplied by the corresponding size fraction, gives the actual amount of
sediment computed for each sediment size class. The user can allocate sediment into three classes (sand,
silt, and clay). The three sediment classes are used to represent sediment response behavior during
transport in subsequent BMP or conduit modules or both.
Particle size distribution and associated pollutant concentrations provide an important linkage to BMP
simulation algorithms. For example, BMPs that remove pollutant through trapping and settling of
sediments will be especially sensitive to the user-specified particle size distributions. If residence time
within the BMP is short, only larger particles might be removed effectively. For externally generated
time series data, SUSTAIN supports the option to specify each sediment class as an independent time
series rather than apply a size distribution to a bulk sediment mass. This approach might be desirable in
cases where particle size distribution changes dramatically during the course of a single storm event and
detailed sediment monitoring data are available to justify the modeling approach.
Land Segmentation
Finally, depending on the size of the watershed being modeled, regional considerations can influence the
robustness and utility of the modeled land units. In fact, the way land units are classified from the onset
can have a strong bearing on how representative or portable the modeled land unit response will be. For
example, in some places, it might be sufficient to model land units on land use alone; whereas, in others,
the use of a hydrologic response units (HRUs) approach might provide a better representation. An HRU
3-48
-------
is a combination of multiple physiographic characteristics, such as land use, soils, or slope. In flat regions
where soils are relatively homogeneous, it might be sufficient to use a purely land-use-based
classification approach. However, if soils or slopes are heterogeneous across multiple land units, the
HRU approach might be a better way of organizing land units.
3.3. BMP Module
The BMP module is designed to provide a process-based simulation of flow and pollutant transport
routing for a wide range of structural BMPs. It is designed so that new BMPs and alternative solution
techniques can be added over time. The BMP module performs the following hydrologic processes to
reduce land runoff volume and attenuate peak flows: evaporation of standing surface water, infiltration of
ponded water into the soil media, deep percolation of infiltrated water into groundwater, and outflow
through weir or orifice control structures. Figure 3-8 shows a schematic of the BMP simulation
processes.
BMP
Simulation
Processes
Figure 3-8. A schematic showing the BMP simulation processes modeled in SUSTAIN.
Table 3-12 provides an overview of the required inputs, the methods used to manage and process the
inputs, and the resulting outputs of the BMP module.
Table 3-13 provides a summary of the key BMP simulation processes included in SUSTAIN. Option 1 is
the default option. Option 2 provides a more data intensive alternative that simulates additional physical
processes in the BMP. With regard to pollutant removal, Option 1 differs from Option 2 in that it does
not include background concentration C*. Users can select processes from either option depending on the
available data and level of detail required.
Table 3-14 lists the BMP types and the associated applicable simulation methods. The BMP simulation
techniques are chosen and implemented to provide a reasonable representation of the physical processes
associated with detention, retention, and infiltration.
3-49
-------
Table 3-12. Summary of Inputs, Methods, and Outputs in the BMP Module
BMP Module
Inputs
- Define BMP dimensions
- Define substrate (soil and underdrain media) properties
- Define sediment settling and transport parameters
- Define pollutant removal and routing parameters
- Define cost for each functional component of a BMP
- Hourly inflow time series
- Hourly sediment (sand, silt, and clay) concentration time series
- Hourly pollutant concentration time series
Methods
- Evapotranspiration is calculated (user-selected constant, monthly, or daily values;
derived from daily temperature using Hamon method)
- Infiltration is computed using the Green-Ampt or Holtan methods
- Deep percolation is calculated according to user-specified background infiltration
rate
- Surface outflow is computed using weir or orifice equations
- Underdrain outflow is computed using orifice equation
- Sediment (sand, silt, and clay) settling and routing is computed using the
processed based algorithms adopted from the HSPF model
- Pollutant removal is calculated using 1st order decay or k-C* method
- Pollutant routing is computed by using completely mixed or CSTR in series
method
Outputs
- Sub-hourly outflow time series
- Sub-hourly sediment (sand, silt, and clay) concentration time series
- Sub-hourly pollutant concentration time series
Table 3-13. Available Optional Methods for BMP Simulation Processes
Processes
Flow Routing
Infiltration
Evapotranspiration
Pollutant Routing
Pollutant Removal
Buffer Strip (Sheet Flow)
Flow Routing
Buffer Strip Sediment
Trapping
Buffer Strip (Sheet Flow)
Pollutant Removal
Option 1
Stage-outflow storage routing using weir or
orifice equations
Green-Ampt method
Constant ET rate or monthly average value,
or daily values
Completely mixed, single CSTR
1st order decay
Kinematic wave overland flow routing
Process-based Univ. of Kentucky sediment
interception simulation method as applied
in VFSMOD
1st order decay
Option 2
For swale: kinematic routing by
solving the coupled continuity
equation and Manning's equation
Holtan-Lopez equation
Calculate potential ET using
Hamon' s method
CSTRs in series
k'-C* method
~
~
~
3-50
-------
Table 3-14. Representative BMPs and Recommended Simulation Methods
BMP
Recommended Simulation Methods
Detention pond
Constant ET rate or monthly average value, or daily values
Calculate potential ET using Hamon's method
Stage-outflow storage routing using weir or orifice equations
Completely mixed pollutant routing
CSTR in series pollutant routing
First order decay (k'-C* method)
Sediment settling and transport
Constructed wetland
Green-Ampt method
Holtan-Lopez equation
Constant ET rate or monthly average value, or daily values
Calculate potential ET using Hamon's method
Stage-outflow storage routing using weir or orifice equations
Completely mixed pollutant routing
CSTR in series pollutant routing
First order decay (k'-C* method)
Sediment settling and transport
Bioretention
Green-Ampt method
Holtan-Lopez equation
Constant ET rate or monthly average value, or daily values
Calculate potential ET using Hamon's method
Stage-outflow storage routing using weir or orifice equations
Completely mixed pollutant routing, single CSTR
1st order decay, no C*
Underdrain percent reduction (user defined)
Infiltration trench
Green-Ampt method
Holtan-Lopez equation
Constant ET rate or monthly average value, or daily values
Calculate potential ET using Hamon's method
Stage-outflow storage routing using weir or orifice equations
Completely mixed pollutant routing, single CSTR
1st order decay, no C*
Hydrodynamic storage
device
Stage-outflow storage routing using weir or orifice equations
Completely mixed pollutant routing
1st order decay, no C*
Sedimentation
Grassed swale
Kinematic flow routing by solving the coupled continuity equation and
Manning's equation
Completely mixed pollutant routing, single CSTR
1st order decay, no C*
Sediment settling and transport using user defined settling velocity and
critical shear stress
Vegetated filterstrip
Kinematic wave overland flow routing
Process-based sediment interception simulation method (VFSMOD)
1st order decay pollutant removal, no C*
The following describes in more detail the methods and algorithms implemented in the BMP simulation
module. Section 3.3.1 describes the BMP simulation component, Section 3.3.2 the buffer strip
component, Section 3.3.3 the aggregate BMP component, and Section 3.3.4 the cost database component.
3-51
-------
3.3.1. BMP Simulation Component
BMPs in SUSTAIN are simulated using a combination of fundamental algorithms to represent the
processes of storage, routing, infiltration, evapotranspiration, underdrain infiltration, and pollutant routing
and removal. The fundamental algorithms associated with each method are shown below.
Storage Routing Method
Water balance storage routing is a commonly used method for flow routing in ponds and impoundments.
where
AV/At = I-O (3-22)
AV = change in storage (volume),
At = time interval (time),
/ = inflow (volume per unit time), and
O = outflow (volume per unit time).
Stage-outflow relationships are widely used for flow routing through an orifice or over a weir as shown in
Figure 3-9.
Reservoir level
H
Sharp crested weir
Orifice diameter
Figure 3-9. Wetland/lake/reservoir weir and orifice outflow.
Weir Outflow
Three commonly used weir types (i.e., sharp-crested rectangular weir, sharp-crested triangular weir, and
broad-crested rectangular weir) are supported in SUSTAIN.
The equation for the rectangular, sharp-crested weir overflow is (Linsley et al. 1992):
Qw=CwLwh3/2
where
Qw = outflow over sharp-crested weir (ft3/s),
Cw = coefficient of discharge,
Lw = length of weir crest (ft), and
h = depth of the water above weir crest (ft).
Values of Cw (English units) for sharp-crested rectangular weirs are given in Table 3-15.
(3-23)
3-52
-------
Table 3-15. Coefficient Cw (English units) for Rectangular Sharp-Crested Weirs
H/h
0.5
1.0
2.0
10
co
Head h on Weir, ft
0.2
4.18
3.75
3.53
3.36
3.32
0.4
4.13
3.71
3.49
3.32
3.28
0.6
4.12
3.69
3.48
3.30
3.26
0.8
4.11
3.68
3.47
3.30
3.26
1.0
4.11
3.68
3.46
3.29
3.25
2.0
4.10
3.67
3.46
3.29
3.25
5.0
4.10
3.67
3.45
3.28
3.24
Source: Linsleyetal. 1992
Hd= Height of the weir
The equation for the triangular (V-notch) sharp-crested weir overflow is (Linsley et al. 1992):
(3.24)
where
Qw = outflow over sharp-crested weir (ft3/s),
Cw = coefficient of discharge, default value is 0.58 for English units,
h = depth of the water above weir crest (ft),
6= vertex angle of the V-notch, and
g = acceleration of gravity (32.2 ft/s2).
True broad-crested weir flow occurs when the upstream head above the crest is between about 1/20 and
1/2 the crest length in the direction of flow (USER 2001). Equation (3-25) is applicable to broad-crested
weirs, and it is recommended that weir coefficient Cw be determined by measuring the flow at various
flow rates (Linsley et al. 1992). The value of the weir coefficient varies with h/Hd. One way of
estimating Cw is to use the equation derived by Fox (University of British Columbia Department of
Mechanics (No date) Fluid Dynamics Course Notes):
0.65 2 r—
~-^2s
c =
where
h = depth of the water above the weir crest (ft),
Hd = height of the weir (ft), and
g = acceleration of gravity (32.2 ft/s2).
Orifice outflow
The equation for the orifice flow is:
(3-25)
(3-26)
where
Qo = outflow through orifice (ft3/s),
C0 = orifice coefficient of discharge,
A0 = orifice cross sectional area (ft2),
3-53
-------
g = acceleration due to gravity (ft/s2), and
H = depth of the water level above the orifice (ft).
Infiltration/Filtration
SUSTAIN supports two options for the simulation of infiltration in BMPs: (1) the Holtan-Lopez equation
adopted from the Prince George's County BMP module (Tetra Tech 2001) and (2) the Green-Ampt
equation (for details, see Section 3.2.2) as is applied in the SWMM (Rossman 2005).
Holtan-Lopez Empirical Model
The Holtan-Lopez empirical model computes the infiltration rate as a function of the actual available soil
water storage, Sa, of the surface soil layer, as shown below (Maidment 1993):
f = GRIxAxSaL4+fc (3-27)
where
/= infiltration rate (in./hr),
GRI = growth index of vegetation in percent maturity, varying from 0.1 to 1.0,
A = infiltration capacity (an index representing surface-connected porosity and density of plant
roots),
Sa = available storage in the surface layer (in.), and
fc = constant final infiltration rate (in./hr).
In Equation (3-27), A is the vegetative parameter that characterizes surface-connected porosity and the
density of plant roots, which affect infiltration (a value of 0.8 is a typical number for sod or vegetation
that would be found in a BMP). fc is the final constant infiltration rate (in./hr), which is a function of the
hydrologic soil group. The value offc ranges from 0.3 in./hr for group-A soils to between 0.0 and 0.05
in./hr for group-D soils (Maidment 1993). In a continuous calculation, the available soil storage (Sa) and
infiltration rate (/) are computed at each simulation time step. Available soil storage is updated each time
increment and the infiltration is calculated.
This method was developed using the premise that soil moisture storage, surface-connected porosity, and
the effect of root density of the control soil layer are the dominant factors influencing the infiltration
process.
A difficulty with using this method is estimating the control soil layer depth. For simulating the
infiltration process, it is assumed that the soil column depth is the control depth because BMP devices
normally have a confined soil/substrate layer.
Green-Ampt Infiltration Equation
This method is discussed in Section 3.2.2 in the hydrology component of land simulation module. The
Green-Ampt equation can be applied to both surface runoff and BMP simulation.
When performing BMP infiltration simulation, the impact of the underdrain layer, the impermeable
bottom layer, or both, on the infiltration process needs to be considered in the simulation. Because the
Green-Ampt method can be applied to a layered soil column, the underdrain layer can be represented as a
separate layer under the soil column. In cases where an impermeable layer is present at the bottom of the
soil column, the infiltration rate ceases when the soil storage capacity is reached.
A drawback of the Green-Ampt method is that it does not include a parameter to explicitly reflect the
effect of the vegetation root zone on the infiltration rate.
3-54
-------
Evapotranspiration
Potential evapotranspiration (PET) time series can be estimated on the basis of the U.S. Weather Bureau
Class A pan records with adjustment. For instance, the HSPF WDM utility estimates PET using the pan
records, then the HSPF simulation model is used to adjust the time series for snow accumulation and melt
(Bicknell et al. 1997). When snow conditions are absent, only PET and precipitation are required.
However, when snow conditions are considered, air temperature, rainfall, snow cover, water yield, and ice
content of the snowpack are also required, and the evaporation data are adjusted. The input evaporation
values are reduced to account for the fraction of the land segment covered by the snowpack.
Several methods are available to estimate PET. The Penman-Monteith method (Maidment 1993) requires
values for solar radiation, air temperature, relative humidity, and wind speed. The Priestley-Taylor
method (Maidment 1993) requires solar radiation, air temperature, and relative humidity. The third,
Hargreaves method (Maidment 1993) requires air temperature only.
SUSTAIN provides three options to estimate PET: (1) rely on the user-supplied monthly PET rate (2)
calculate PET from the user-supplied pan evaporation time series input and monthly pan coefficients, and
(3) calculate the PET rate using Hamon's method (1961).
Hamon's (1961) method generates daily PET using air temperature, a monthly variable coefficient, the
number of hours of sunshine (computed from latitude), and absolute humidity (computed from air
temperature).
pfTT - r< x r> 2 v n (3-28)
rn,l -^TS^-^hr X Pws ^ '
where
PET = daily PET (in.),
CTS = monthly variable coefficient, and
Dhr = possible hours of sunshine computed as a function of latitude and time of year.
= 2J6.7xpws
Pws Tav+273.3
where
pws = saturated water vapor density (absolute humidity) at daily mean air temperature (g/cm3)
and
Tm = mean daily air temperature ( C).
™ =6.J08xexp
17.26939 xT
(3-30)
, m+273.3
where
pws = saturated vapor pressure at the air temperature.
Hamon (1961) suggests a constant value of 0.0055 for CTS. However, monthly values can be specified to
avoid underestimating PET in some areas, especially for the winter months.
3-55
-------
Calculate Actual Evapotranspiration
Once PET is determined, the actual evapotranspiration (ET) is calculated as a function of PET and soil
moisture storages. While PET represents the maximum possible achievable ET on the basis of
atmospheric conditions alone, actual ET is determined using an accounting of the status of the various
components of the hydrologic budget. The actual ET is equal to PET when the soil moisture is greater
than or equal to the moisture at the field capacity and there is no actual ET if the moisture content is less
than or equal to the moisture at the wilting point.
Underdrain Method
Underdrain Outflow
The underdrain outflow in a BMP is modeled using a simple water balance concept. The available
underdrain storage is represented as the total of void spaces beneath the upper soil layer. Inflow into
underdrain storage is limited by the final infiltration rate of the upper soil layer. Because the primary
function of the underdrain is to provide additional water storage and to delay outflow, the outflow pipe
draining the underdrain layer is placed at the interface between the upper soil layer and the underdrain
layer. Figure 3-10 illustrates the function of underdrain together with other substrate model components.
Evap otrans pirati on
Soil Layer
Infiltration
Underdrain
Outflow
1 £-"""•'•'"' '-> ..-•'''•1':--V'V-.
Soil Media
Storage and
Filtration
Underdrain
Storage
Background
Infiltration
Figure 3-10. Processes considered in an underdrain structure.
Outflow from the underdrain layer is assumed to be unrestricted; therefore, no pipe outflow is required.
Underdrain outflow is part of the modeled BMP effluent and occurs when all available underdrain storage
is used up, when the water level meets or exceeds the underdrain level, or when both occur. Each
infiltration management practice can be modeled with or without underdrain outflow. If underdrain
outflow is enabled, the user must specify the thickness of the underdrain storage layer, the media void
fraction, and the background infiltration rate (Figure 3-10). Water and pollutants are removed from the
system entirely through background infiltration.
Underdrain Filtration of Pollutant
If underdrain is specified in the soil properties of a BMP, additional reduction in pollutant concentration
from underdrain routing is simulated in the module using the underdrain percent removal, which is a
user-supplied parameter. This option allows users the flexibility in estimating pollutant removals through
the soil media of a BMP.
3-56
-------
Pollutant Routing and Removal Methods
The methods of pollutant routing to achieve pollutant reduction are described in this subsection for a
completely mixed system and a multiple impoundments in series. The flow through a plug flow reactor
(PFR), as a series of infinitely thin coherent plugs, each with a uniform composition, is perfectly mixed in
the cross direction but not in the longitudinal direction (direction of flow). Each plug of differential
volume is considered as a separate entity, with an infinite simally small volume and requires a very small
time steps (in seconds). SUSTAIN uses one minute to hourly time step to simulate flow and pollutant
routing and does not support plug flow option in the current version. However, it can be seen that an
infinite number of small continuously stirred tank reactors (CSTRs) operating in series would be
equivalent to a PFR.
First-Order Decay with Complete Mixing
This method is commonly used and suitable for small ponds when complete mixing is likely.
= IftJC, ft) - 0(t)C(t) - KCft) V(t) (3-31)
at
where
V= reservoir volume (ft3),
Cj = influent pollutant concentration (mg/L),
C = effluent and reservoir pollutant concentration (mg/L),
/ = inflow rate (ft3/s),
O = outflow rate (ft3/s),
t = time (sec), and
K = decay coefficient (1/s).
Continuously Stirred Tank Reactors in Series and Kadlec and Knight 's Model
CSTRs in series are used to represent a hydraulic condition intermediate between completely mixed and
plug flow (Wong et al. 2001, 2002). That method is applied for simulating first-order pollutant removal
processes (e.g., settling, decay) that occur in ponds, wetlands, and other similar BMPs. The calculation
begins by estimating the number of reactors in series to be selected to represent the shape of the BMP,
followed by applying first order kinetics with nonreactive background concentration (the k-C* model;
Kadlec and Knight 1996).
Step 1: Estimate N, the number of CSTRs in series.
N, the number of CSTRs in series, can be approximated on the basis of BMP shape (Persson et al. 1999;
Wong et al. 2001, 2002). Values of TV for the various pond shapes, shown in Figure 3-1 1, are presented in
Table 3-16. Highest TV values are for ponds with a distributed inflow (pond E), baffles (pond G), and very
elongated flow or high length to width ratio (pond J).
3-57
-------
Figure 3-11. Conceptual pond shapes simulated by Persson et al. (1999).
Table 3-16. Quality Ratings of Conceptual Pond Shapes Simulated by Persson et al. (1999)
Pond
J
G
E
P
Q
I
K
A
B
O
D
H
C
N*1/(1-X)
10.0
4.2
4.1
2.6
2.5
.7
.6
.4
.4
.3
.2
.1
.1
Qualitative
Rating
Good
Satisfactory
Poor
Step 2: Apply first-order decay (k' - C* model) to each CSTR.
After selecting the number of reactors, pollutants are modeled for each tank at each time step using the
first-order kinetic model, described in Equation (3-32).
where
(Cout-C )/(Cm-C ) = e
C* = background concentration (mg/L),
Cin = input concentration (mg/L),
Cout = output concentration (mg/L),
q = hydraulic loading or overflow rate (m/yr),
k' = k-h = rate constant (m/yr),
k = first order decay rate (1/yr), and
h = pond depth (m).
(3-32)
3-58
-------
This equation is computed separately for each time step at each CSTR. The main difference between this
equation and ordinary first-order decay modeling for a CSTR is the inclusion of C*, the background
concentration, below which the effluent cannot fall. Another advantage of this method is that using an
areal rate constant (units of depth/time) instead of' ^volumetric one (units of inverse time) helps avoid
having to specify an average depth or the volume for odd natural configurations; instead, only the pond
surface area is required to compute the hydraulic loading rate q.
Wong et al. (2002) recommend some k' and C* values as shown in Table 3-17, on the basis of limited
model calibration for total suspended solids (TSS), total phosphorus (TP), and total nitrogen (TN) in
urban areas near Melbourne. Those values should be used with caution. However, they could be used as
a starting point in the absence of local data.
Table 3-17. Recommended k' and C* Values
Treatment Measures
Sedimentation Basins
Ponds
Vegetated Swales
Wetlands
k'
(m/yr)
TSS
15,000
1,000
15,000
5,000
TP
12,000
500
12,000
2,800
TN
1,000
50
1,000
500
C*
(mg/L)
TSS
30
12
30
6
TP
0.18
0.13
0.18
0.09
TN
1.7
1.3
1.7
1.3
Localized calibration can be performed to customize the simulation technique for specific areas. The C*
and k' values can be determined or calibrated using monitoring data (particle size distribution in
particular) and treatment measure design specifications.
3.3.2. Overland Flow Routing and Pollutant Interception
The following algorithms for overland flow routing and pollutant interception simulation are employed in
SUSTAIN for buffer strip simulation.
Kinematic Wave Overland Flow Routing Method for Filter Strip Simulation
Overland flow through filter strips can be simulated using a kinematic wave method and solving the
coupled continuity and Manning's equations.
Mathematical Model
Continuity equation
dt dx
Manning's equation
(3-33)
(3-34)
n
where
h = overland flow depth (ft),
3-59
-------
q = overland flow per unit width of the subcatchment (ft2/s),
n = Manning's roughness coefficient,
ie = rainfall excess depth (ft/s),
t = time (s), and
S0 = subcatchment slope.
Initial condition
h = 0; Q0
where, h0 can be 0, a constant, or a time-dependent function, such as the incoming hydrograph from the
adjacent subcatchment. The rainfall excess, ie, can be calculated from the hyetograph and Green-Ampt
infiltration method at each time step.
Numerical Solution
The coupled continuity equation and Manning's equation are solved using Petrov-Galerkin (PG)
formulation to compute the flow rate (q), velocity (v), and depth (h) throughout the plane for each time
step. Kinematic shocks (oscillations in the solution) are introduced when a sudden change in conditions
(e.g., slope, roughness, and inflow) occurs. For filter strip simulation, as the soil surface conditions are
updated for each time step, the potential for kinematic shocks is further increased. The PG finite element
method was found to reduce the amplitude and frequency of oscillations compared to a conventional
Bubnov-Galerkin finite element solution, thus improving the model stability for situations that are subject
to kinematic shocks (Munoz-Carpena et al. 1993).
VFSMOD Algorithms for Sediment Interception
The sediment interception algorithm used in VFSMOD considers that when runoff reaches the upstream
edge of the filter, the vegetation provides a sudden increase in hydraulic resistance, which slows the flow,
lowers its transport capacity, and causes deposition of the coarse material (particle diameter dp > 0.0037
cm), which is carried mostly as bed-load transport. The sediment trapped in the first section of the filter
forms a geometrical shape (the wedge zone), which is either triangular when Y (f)
-------
ENTRY
Field
Wedge Zone
9si> qln : (1
Suspended Load Zone
EXIT
I
Flow needed at points
Figure 3-12. Filter description for the sediment transport algorithm.
Einstein's bed-load transport Equation (3-35) is solved using the method proposed by Barfield (Muiioz-
Carpena 1993) to compute the sediment transport capacity.
. -0.25
7 Rsksk
= 1.08
gsk
where
y= water density (g/cm3),
YS = sediment density (g/cm3),
dp = particle diameter (cm),
gsk = sediment load (g/cm-s) at point k (k = 1,2),
Sk = slope at point k,
g = gravitational constant (cm/s2),
S.dt
(3-35)
Rsk = spacing hydraulic radius at point k, defined as ,
2dft+ss
Ss = grass spacing (cm),
dfk = modified flow depth (cm) at point k, defined as— * ,— , and
T) 2/3 I o
**-sk V k
qk = unit width flow rate (cm2/s) at point k.
It was assumed that only sediment at the fine sand/silt threshold (diameter > 0.0037 cm) is considered in
the wedge zone sediment routing and that fine sediment (diameter < 0.0037 cm) runs through to the
suspended zone (Munoz-Carpena 1993). Therefore, the user is required to input the percentage of coarse
particles from incoming sediment that will be routed through the wedge. The calculated sediment
transport capacity is compared with incoming sediment concentration. If the incoming sediment
concentration is higher, deposition at the wedge will occur; if lower (meaning that there is enough energy
3-61
-------
to transport sediment through the wedge and no deposition occurs), all sediment is transported to the
suspended sediment zone (zones C(t) and D(t)).
After the downside of the wedge, two zones, C and D, form the suspended load zone. It was assumed that
on zone C, sediment has covered the indentations of the surface so that bed-load transport and deposition
occur, but the soil slope is not significantly changed. All bed-load transported sediment is captured
before reaching zone D so only suspended sediment is transported and deposited in this zone until the
flow reaches the end of the filter. Flow values at point 3 and the exit point are needed for the calculation
and are provided by the flow module. The trapping capacity (Tr) Equation (3-36) developed by Tollner et
al. (1976) is used to simulate the sediment trapping for the suspended load zone:
T ^
where
(3-36)
gS2 = sediment load at point 2 (g/cm-s),
gso = sediment load at the output point (g/cm-s),
V3 = mean velocity at point 3 (cm/s),
Vf= fall velocity (cm/s),
v = kinematic viscosity of water (cm2/s), and
L = effective filter length (cm).
Particle Deposition and Sediment Transport
On the basis of the laboratory study by Deletic (2001) the particle deposition or trapping efficiency for
sediment fraction s (particle with diameter of ds) can be estimated as a function of the particle fall
number, Nfs, which is calculated as:
Nfs= (3-37)
f's hV
where / is the flow length, Vsthe Stokes' settling velocity of particle size ds, and Vis the average mean
flow velocity between grass blades. The sediment trapping efficiency 7^ is expressed as:
N0.69
T = _ £ _ (3-38)
r's N°f69+4.95
J's
The suspended sediment transport equation is expressed as:
| dq^ = ^d2 (hqss/q) ^ (3.39)
dt dx dx2
where qS:S is the sediment loading rate of fraction s per unit width, Dis is the dispersion coefficient, and As
is the trapping efficiency for fraction s per unit length calculated as As=TriS/l.
3-62
-------
Terrain Surface Level and Slope Changes
The rise in the surface level, z, is modeled as the integral of trapped particles of all fractions of particle
sizes and is expressed as:
(3-40)
where/? is the porosity of deposited sediment.
The water quality for the pollutant (other than suspended sediment) is simulated by applying the most
commonly used first order decay model. A first order decay is equivalent to an exponential decay,
represented by the Equation (3-41).
Ct=C0e-kt (3-41)
where
Ct = concentration at time t (mass per volume),
C0 = initial concentration at time zero (mass per volume), and
k = reaction rate (per timestep).
For the current phase of SUSTAIN development, the decision was made to include this complex, process-
based simulation for filter strip simulations. This model is CPU intensive and, thus, is not used during the
optimization process. Its purpose in SUSTAIN is for evaluation of filter strip performance only (see
Section 3.3.6). Future phases will include a simplified method that can be incorporated into the
optimization module.
3.3.3. Aggregate BMP Component
The aggregate BMP component provides an optional method for assessing the combined impact of
multiple BMPs on the watershed runoff and pollutant load. The formulation was developed to represent
the aggregate characteristics of distributed BMP, while reducing the user effort for model setup and
computation time needed for simulation and optimization. While the BMP module in SUSTAIN performs
explicit simulation of individual BMP practices for a defined management area, the aggregate BMP
component evaluates storage and infiltration characteristics for multiple BMPs simultaneously without
explicit recognition of their spatial distribution and routing characteristics in the selected watershed.
As illustrated in Figure 3-13, an aggregate BMP consists of a series of process-based components,
including on-site interception, on-site treatment, routing attenuation, and regional storage/treatment. For
the aggregate BMPs, users are asked to input design drainage area and number of units for each aggregate
BMP component. Each component can be used to represent one of a number of individual BMPs. For
example, on-site interception can be represented as a cistern, rain barrel, or green roof, while on-site
treatment can be bioretention, porous pavement, or an infiltration trench. Each individual component can
be enabled or disabled according to the desired functionality and sized and parameterized using the BMP
templates that are identical to that of the individual BMPs.
3-63
-------
On-Site
Interception
Routing
Attenuation
Outlet
On-Site
Treatment
Regional
Sto rage/Treatm e n t
Figure 3-13. Generic aggregate BMP schematic.
Aggregate BMPs can be applied to a user-defined drainage area. The land use distribution of the drainage
area is automatically calculated on the basis of the land use map and populated into the land use
distribution/assignment table as shown in Figure 3-14. Users then assign the percentage of each land use
that contributes to each of the aggregate BMP components (Figure 3-14). Runoff and pollutant loads
from the total drainage area of each component are lumped and routed through the respective component.
The relative scales and sizes of individual BMPs are preserved in the aggregate representation.
v Aggregate BMP Landuse Distribution
Select Subwatershed h ^ ,-.
• «M* (@)
Land Use Distribution
>
Landuse Group/Info Type
BMPID
Category
BMPTvpe
Hiqh-Density^Residential Impervious
Hiqh-DensiVResidential F'ervious
Downstream ID
Area (ac.)
8.95
224
RainBarrell m BioRetentionBasin! ml Outlet \%\
5 |e |o
On-Site Interception
Rain Barrel
25
0
6
On-Site Treatment
BioRetentionBasin
75
0
0
Outlet
Outlet
0
100
0
Figure 3-14. Illustration of land use area assignment to aggregate BMP components.
To investigate the applicability of the aggregate BMP approach, a test case was conducted to compare the
results of a SUSTAIN simulation using the aggregate approach to one using the distributed approach
representing the same BMP scenarios. The distributed approach represents a fully articulated BMP and
routing network, whereas the aggregate approach represents the component responses. Because of the
lumped representation, the aggregate approach does not consider detailed routing between components. It
is assumed that for small basins, the associated short time of concentration means that a fully articulated
routing simulation is not necessary. Similarly, there is presumably some watershed size threshold, above
which the lumped routing assumption might no longer be appropriate.
That presumption was evaluated by developing five test simulation drainage areas of different sizes (1.3
acres, 7.8 acres, 31.2 acres, 128 acres, and 256 acres). For each size, three simulation scenarios were
applied: (1) an aggregate BMP representation, (2) a fully articulated network with conduit routing, and
(3) a fully articulated network without conduit routing (conduit dimensions were set equal to zero). In
this way, the test was designed to highlight the relative importance of the routing component to overall
simulation results. Figure 3-15 illustrates the testing concept, showing the relative complexity of the
distributed routing network scenarios as they increase in size vs. that of the aggregate representation.
One-year simulations were conducted for each scenario with each drainage area size. Three factors were
3-64
-------
computed and summarized for each run: total annual flow volume, peak flow rate, and total annual TSS
load.
Table 3-18 summarizes the results of the comparison. Among the three factors examined, peak flow rate
is most sensitive to drainage area size. For example, at 128 acres, the peak flow rate values of the
aggregate representation are significantly different from the distributed representations even though total
flow volume and total TSS load values remain relatively constant.
At the drainage area size of 256 acres, the total TSS load of the aggregate representations were drastically
different from the distributed ones; however, the differences are much smaller for drainage areas less than
256 acres (less than 2 percent difference). Simulation run times for various scenarios are listed in Table
3-19, showing that on the basis of this initial testing and as expected, the distributed representation
requires a much longer run time than the aggregate representation.
1.3 acres 7.8 acres 31.2 acres 128 acres 256 ac
Rooftop
Agg. Bio retention
Pavement
Figure 3-15. Aggregate BMP testing configuration.
3.3.4. BMP Cost Database Component
The BMP module costing component provides the underlying cost database used by the optimization
component in evaluating BMP scenarios. The compilation of cost information was predicated on
obtaining cost data in a format that could be input into a uniform database. This compilation was
constrained by the non-uniformity with which available cost information for BMP construction is
reported. As a result, it was determined that the best approach for building SUSTAIN's cost database was
to determine unit costs (i.e., cost per square foot) for individual construction components of the overall
BMP. Construction components include excavation, grading, filter fabric, and so forth. Basing the cost
estimation routines on basic construction components rather than the whole BMP installation is aimed to
minimize differences encountered because of site or locality factors. Users have the ability to override
the data with their own locally derived information.
3-65
-------
The cost database was developed by identifying individual construction components for each BMP
technique simulated by SUSTAIN. Table 3-20 outlines the construction components for which data were
compiled, the assumptions governing the general characteristics of the construction component (e.g.,
hardwood mulch versus pine straw mulch), and the associated BMP techniques. Unit costs for each
component were compiled from retail sources and from reference documents and programs involved in
BMP implementation at the local, state, and federal levels.
Table 3-18. Comparison of Aggregate vs. Distributed BMP Results
Scenario
Aggregate
Representation
Distributed
Representation
w/o
Routing
Routing
% Difference vs. Aggregate
Representation
Distributed
w/o Routing
Distributed
with Routing
Total flow volume (ftVyr)
1.3 acre with BMPs
7.8acrewithBMPs
3 1.2 acre with BMPs
128 acre with BMPs
256 acre with BMPs
32,845
197,038
787,793
3,150,904
6,301,805
33,401
200,408
808,312
3,233,250
6,466,500
33,924
203,956
822,903
3,280,110
6,563,871
1.69
1.71
2.60
2.61
2.61
3.29
3.51
4.46
4.10
4.16
Peak Flow Rate (cfs)
1.3 acre with BMPs
7.8 acre with BMPs
3 1.2 acre with BMPs
128 acre with BMPs
256 acre with BMPs
0.7
4.0
15.2
58.3
120.8
0.7
4.1
16.3
65.1
175.7
0.7
3.9
15.7
60.4
91.7
2.00
1.75
6.83
11.58
45.41
-0.59
-1.75
3.09
3.52
-24.09
TSS load (Ib/yr)
1.3 acre with BMPs
7. 8 acre with BMPs
3 1.2 acre with BMPs
128 acre with BMPs
256 acre with BMPs
280
1,679
6,727
26,907
53,814
282
1,690
6,817
27,269
68,794
281
1,669
6,725
26,925
67,685
0.57
0.63
1.34
1.35
27.84
0.20
-0.60
-0.03
0.07
25.78
Table 3-19. Simulation Run-Time Comparison: Aggregate vs. Distributed
Scenario
1.3 acre with BMPs
7. 8 acre with BMPs
3 1.2 acre with BMPs
128 acre with BMPs
256 acre with BMPs
Aggregate
Representation
< 0.3 sec
< 0.3 sec
< 0.3 sec
< 0.3 sec
< 0.3 sec
Distributed
w/o Routing
1 sec.
12 sec.
60 sec
7 min
12 min
Distributed
w/ Routing
14 sec
90 sec.
7 min
30 min
50 min
Notes
19 BMPs, 5 conduits
114 BMPs, 30 conduits
456 BMPs, 120 conduits
1,824 BMPs, 480 conduits
3,648 BMPs, 961 conduits
3-66
-------
Table 3-20. BMPs and Associated Construction Components
Construction
Component
Excavation
Grading/finishing
Backfilling
Soil/Planting
Media
Filter Fabric
Gravel 1
Gravel 2
Gravel 3
Underdrain Pipe
Mulch
Rain Barrel
Green Roof
System
Grass
Perennials
Small Trees
Woody Shrubs
Inlet Structure
Outlet Structure
Observation Well
Seal
Porous Paving
Material
Description
Using light equipment
Generally using light equipment or hand tools to
minimize compaction except in the case of the
creation of detention ponds where compaction is
necessary
Replacing excavated area with soil/planting media;
required when using amended growing media
A typical mix is 50% sand, 30% planting soil (low
clay content), and 20% shredded hardwood mulch
Often placed between gravel reservoir and
underlying/overlying soil to reduce clogging of the
reservoir void spaces
Porous pavement filter course, smaller particle sizes
Reservoirs; slightly larger particle sizes, no fines
Erosion control (rocks/riprap)
4-in. perforated PVC
Shredded hardwood
Framework assumes use of sod
Assumes planting density of 1 ft o.c. for 1-gal plants
Assumes planting densities of 15 ft o.c.
Assumes planting densities of 3 ft o.c.
4-in. PVC pipe
Bentonite (as opposed to geotextile)
Applicable BMP Techniques
Bioretention Basin; Vegetated
Swale; Porous Pavement; Wet
Pond; Dry Pond; Wetland;
Infiltration Basin
Bioretention Basin; Vegetated
Swale; Porous Pavement; Wet
Pond; Dry Pond; Wetland;
Infiltration Basin; Buffer Strip
Bioretention Basin; Wetland;
Infiltration Basin
Bioretention Basin; Wetland;
Infiltration Basin; Green Roof
Bioretention Basin; Porous
Pavement; Infiltration Basin
Porous Pavement
Bioretention Basin; Porous
Pavement; Infiltration Basin
Wet Pond; Dry Pond; Wetland
Bioretention Basin; Porous
Pavement
Bioretention Basin; Wetland;
Infiltration Basin
Rain Barrel
Green Roof
Bioretention Basin; Vegetated
Swale; Wet Pond; Dry Pond;
Infiltration Basin; Buffer Strip;
Wetland
Bioretention Basin; Infiltration
Basin; Wetland
Bioretention Basin
Bioretention Basin
Wet Pond; Dry Pond; Wetland
Wet Pond; Dry Pond; Wetland
Infiltration Basin
Wet Pond
Porous Pavement
3-67
-------
Unit Cost Data Sources
Costs for BMP construction components were obtained from the following sources.
Wholesale/Retail Bulk Material Pricing
Several wholesale or retail companies were inventoried to provide bulk material pricing. Retailers were
contacted to provide unit pricing for various raw materials such as mulch, sand, stone, and other
commercial landscape materials. Retailers were also contacted to provide pricing for rain barrels. Unit
cost data were compiled in 2007.
EPA Stormwater Technology Fact Sheets
EPA's Office of Water Municipal Technology Assessment Program supports innovative and alternative
technology development through a number of efforts and partners. In 1999 the program produced a series
of Stormwater Technology Fact Sheets (832-F-99-001 to 048), which are at
http://www.epa.gov/owmitnet/mtb/mtbfact.htm. The series provides information on advantages and
disadvantages, design criteria, operations and maintenance, performance, and cost estimates regarding a
range of management technologies including bioretention, catch basins, flow diversion, infiltration
trenches, modular systems, porous pavement, and others.
California Department of Transportation (CALTRANS)
In the late 1990s, CALTRANS began a study to evaluate structural BMPs for Stormwater treatment.
Among other things, the study evaluated removal efficiencies, capital costs, and annual operation and
maintenance (O&M) costs of individual applications.
Fairfax County BMP Fact Sheets
Fairfax County, Virginia, developed 25 fact sheets for inclusion in the county's Public Facilities Manual
that present an overview of the management strategies and technologies for various current or potential
BMP/LID techniques used in the county. The fact sheets address seven different BMP categories
including: (1) bioretention systems, (2) filtering technologies, (3) permeable pavements, (4) site design
strategies, (5) soil amendments, (6) vegetative systems, and (7) water conservation/reuse. The
information in each fact sheet is consistent so that relative comparisons can be made on the critical
design, construction, and maintenance issues. Information includes a general description of the BMP,
water quality and quantity controls, location; design construction and materials, cost, maintenance,
performance and inspection, and potential LEED (Leadership in Energy and Environmental Design)
credits. The fact sheets are at http://www.lowimpactdevelopment.org/fairfax.htmtfffx factsheet.
Natural Resources Conservation Service (NRCS) Cost Share Data
The U.S. Department of Agriculture's (USDA's) NRCS program Web site provides cost lists and tools
developed by NRCS field office staff to support BMP cost-sharing programs under the department's
Environmental Quality Incentives Program and other programs. Documents containing unit cost data for
BMP components are available for various years for 34 states.
Michigan Department of Environmental Quality's (DEQ 's) 319 BMP Cost Database
The Michigan DEQ's Nonpoint Source Program administers the CWA section 319 grant program for the
state. It requires grantees to submit BMP cost share information for purposes of tracking the cost and
location of BMPs installed with DEQ Nonpoint Source grant funding and documenting expenses for cost-
share practices. DEQ is required to produce annual reports for the EPA and the Michigan legislature to
detail the use of state and federal grant funds. In addition, the data are also used to share information
about specific practices with current and potential grant recipients. The SUSTAIN cost database adapted
3-68
-------
cost data for certain practices such as green roofs and porous pavement because they were reported in the
same units as the cost module database. Costs for other relevant BMPs are reported in the DEQ BMP
Cost Share database; however, component costs are not broken down. As a result, these data were not
used.
Green Roofs
EPA's Heat Island Web site and the Great Lakes Water Institute's Web site provide installation and
maintenance costs per unit for green roofs in urban areas. The Web sites are
http://www.epa.gov/heatisland/mitigation/greenroofs.htm
http://www.glwi.uwm.edu/research/genomics/ecoli/greenroof/roofinstall.phptfcosts.
Minnesota Stormwater Manual (version 1.1)
The Minnesota Stormwater Manual (Minnesota Pollution Control Agency 2005) provides unit costs for
several BMPs. The Stormwater Manual discusses a variety of BMP approaches designed to lessen the
impacts of urban development. The Manual explores an array of BMPs that can be implemented to
control sediment and reduce runoff in a practical and flexible manner.
Cost Database
The cost database in the BMP module is a Microsoft Access database containing records related to unit
costs of BMP construction components relevant to the techniques simulated by SUSTAIN. In addition to
the unit cost per component, each record contains information related to the source of the data and, to the
extent possible, the year or general time period from which the cost data were recorded. In addition,
O&M cost estimates were available for a limited number of records. All unit cost data were entered into
the unit cost table using the unit in the original source; conversions to a consistent unit of measure take
place in the system using appropriate conversion factors.
Using the Cost Component
The Cost Component provides the flexibility of choosing the construction components and the associated
costs for estimating the total cost of the BMP.
User Data
Users also have the option to enter their own cost data as well as operations and maintenance costs.
Selection of Data
Cost estimates in the database are taken from sources across the country; however, not all states and
regions are represented. Users can choose to base cost estimates on an average of all data in the database
or they can choose to use only select sources of data if they deem specific sources to be adequately
representative. In addition, the interface allows users the option of not including specific components
when calculating the unit cost of a BMP. For example, in an area where it is not necessary to include an
impermeable seal on the bottom of a wet detention pond, the cost component can be turned off. O&M
data were limited and, thus, are not included in the cost database.
Table 3-21 through Table 3-26 provide additional information related to fields and tables that compose
the database.
3-69
-------
Table 3-21. Components
Column
Components_ID
Components_txt
Components_Desc
Details
Component ID number
Construction component
General description of the component
Type
Numeric
Text
Text
Table 3-22 BMP Types
Column
BMPTypeJD
BMPType_Code
BMPType_Desc
Details
BMP Type ID number
Lookup code (no spaces) (e.g., Bioretention Basin)
Description of the BMP Type (e.g., Bioretention Basin)
Type
Numeric
Text
Text
Table 3-23. BMP Components
Column
BMP_Component_ID
Component_ID
BMP TypeJD
Details
Unique record identifier
Component ID number as in components table
BMP Type ID number as in BMP Types table
Type
Numeric
Numeric
Numeric
Table 3-24. Component Costs
Column
Component_Cost_ID
Component_ID
Unit
Cost
Year
Source_ID
Locale
UseFlag
Note
OrigJJnit
Orig_Cost
Details
Unique record identifier
Construction Component
Unit on which cost estimate is based (e.g.,, ft, ft2, ft3, etc.)
Cost in dollars
Year in which the construction cost estimate was developed for conversion
and baseline comparisons
Code for the reference from which unit cost data were taken
Geographic area for which cost estimate is applicable (e.g., national, state,
city specific)
Flag to differentiate useable records
Notes regarding the data
Original unit on which cost estimate is based, before conversion to standard
units
Original cost in dollars, before standardizing units
Type
Numeric
Numeric
Text
Numeric
Numeric
Text
Text
Numeric
Text
Text
Numeric
Table 3-25. Unit Types
Column
UnitTypeJD
UnitType_Code
UnitType_Desc
Details
Unit Type ID number
Code for the unit of cost data used for specific BMP component costs (e.g.,
ft2)
Text description of unit
feet)
of cost data for BMP component cost (e.g., square
Type
Numeric
Text
Text
3-70
-------
Table 3-26. Reference Sources
Column
Reference_Source_ID
Source_Type
Title
Year
Author
Publication_Street
Publication_City
Publication_State
Publication_ZIP
Reference_Number
Prepared_by
Note
Details
Unique record identifier
Assigned code for the reference/source from which unit cost data were
taken
Title of reference
Year reference was published
Agency under whose name the reference was developed and distributed
Mailing street address, if listed
City listed for publishing agency /author; NOT city for which data are
derived
State listed for publishing agency /author; NOT state for which data are
derived
ZIP Code listed for publishing agency/author; NOT the ZIP Code for which
data are derived
Any code listed on the report title page used by author/issuing agency to
identify and track official publications
Developer of the reference (if work was completed on behalf of public
agency by contractor)
Notes regarding the source and/or data
Type
Numeric
Text
Text
Numeric
Text
Text
Text
Text
Numeric
Text
Text
Text
Factors Affecting Development of Cost Database and Implications for Use
Format of available cost data
All data used to populate the cost database are from nonproprietary sources. It is important to note that
the nature of database construction requires that information be consistent and uniform. Unfortunately, a
wide variety of formats are used in which cost data for BMP and construction components are reported.
That presents a serious limitation to the amount of cost data available for the current phase of SUSTAIN
development. To maximize the utility of the optimization functions of SUSTAIN, future development
phases must include development of additional unit cost data.
A limited number of estimates were available for O&M costs. If a range of costs was given in the
original source, such as 5 through 7 percent of construction costs, the average was used in the database
(e.g., 6 percent).
NRCSdata
The NRCS unit cost data represents a significant portion of the cost records in the SUSTAIN cost
database. The information available from the NRCS includes cost lists and various tools (such as
spreadsheets) developed by NRCS field office staff to support BMP cost-sharing programs under the
USDA's Environmental Quality Incentives Program and other programs. In general, the BMPs and cost
data represented in the NRCS data are focused on rural, agricultural applications; however, several
projects have data available in counties with both urban and rural areas. Project and construction
component costs are obviously reflective of local and regional economic factors.
Note, too, that different states have adopted different approaches to developing this information; while
some state NRCS field offices responsible for developing this information employ economists, some do
not. In Virginia, for example, costs are based on either real project numbers or on bids submitted by local
3-71
-------
contractors for typical applications. Atypical (expensive, technically complicated, or unusual) projects
are generally excluded from the universe of estimates used to generate average cost lists.
The cost share data for 28 states were accessed and reviewed during development of the database, and
appropriate records were selected for inclusion on the basis of consistency with the BMPs and
components represented in SUSTAIN, as well as the applicability of units for which costs were reported.
The potential exists that cost estimates derived from the NRCS data are low, relative to costs for projects
in more urban areas, which might be expected to have higher unit costs because of higher operating,
equipment rental, land, and other costs. Efforts were made to select data for inclusion in the database to
address this potential for underestimation. For example, the Alabama cost estimator spreadsheet includes
four records related to unit costs for the SUSTAIN component backfill. The analogous component in the
Alabama Cost Estimator spreadsheet is called Earth/ill. Cost estimates are available for < 3,000 yd3;
3,000-10,000 yd3; 10,000-30,000 yd3; and 30,000+ yd3. Unit costs for the smallest volume (< 3,000 yd3)
are highest; therefore, it was the estimate selected for inclusion in the SUSTAIN cost estimate database.
Where the NRCS data provide multiple unit costs for a component, the highest cost is used.
Because the NRCS data represent such a significant proportion of the publicly available unit cost data
related to implementation and construction of BMPs, they were included in the cost database for
SUSTAIN. However, in recognition of the potential limitations associated with applying those data to
urban areas, users may opt to exclude all NRCS data from cost calculations. Note that excluding NRCS
data will result in a more restricted data set from which the model's cost estimates will be based.
3.3.5. Summary of Management Practices and Treatment Processes in SUSTAIN
SUSTAIN provides multiple ways for simulating management practices that are widely used to treat
runoff and mitigate flow volumes. Treatment processes may be grouped into the following categories:
• Storage/detention or flow attenuation (i.e., those detaining and/or attenuating water)
• Infiltration (i.e., those infiltrating water to the ground)
• Filtration (i.e., those passing water through a porous medium)
• Evapotranspiration (i.e., those losing water from surface and/or soil column)
• Water quality (i.e., those performing pollutant removal)
Table 3-27 lists the structural BMP types handled by SUSTAIN and the associated treatment processes.
3.3.6. Important Considerations and Limitations of the BMP Module
Selecting an Appropriate Simulation Method
The BMP module provides an array of simulation options with varying degrees of sophistication in the
required input and computational rigor. Such options provide flexibility for users to customize their
problem formulations to suit the project needs. The user should carefully select the method or approach
considering the overall problem formulation, availability of supporting data, and expected outcomes. For
example, users can choose between the two infiltration options provided: the simple Holtan infiltration
method or the iterative Green-Ampt infiltration method. To optimize BMP selection for a large study
area having several infiltration BMPs, it might be advantageous to select the Holtan method over the
Green-Ampt method because it might result in shorter computation time (e.g., reduced runtime for a
single run by several seconds). Because it might be required to perform hundreds or thousands of BMP
simulations during an optimization analysis, the accumulated run time savings of a few seconds per run
3-72
-------
could be significant. On the other hand, users should also consider simulation accuracy and localized
investment in model simulation in determining the appropriate technique. For example, if the Green-
Ampt method has already been parameterized and successfully applied for land simulation (in SWMM), it
would be wise to use the same method for simulating infiltration in the Land and BMP modules.
Table 3-27. Structural BMPs and Major Treatment Processes
Structural BMP Types
Bioretention
Cistern
Constructed Wetland
Dry Pond
Grassed Swale
Green Roof
Infiltration Basin
Infiltration Trench
Porous Pavement
Rain Barrel
Sand Filter (non-
surface)
Sand Filter (surface)
Vegetated Filterstrip
Wet Pond
Storage/Detention
or Flow
Attenuation
0
+
+
+
0
(o)
(o)
(o)
-
+
(o)
(o)
(o)
+
Infiltration
+
-
(o)
(o)
+
-
+
+
+
-
0
0
0
(o)
Filtration
0
-
-
-
(o)
(o)
0
0
(o)
-
+
+
+
-
Evapotranspiration
0
-
+
+
-
+
-
-
-
-
_
-
0
0
Water
Quality
+
-
+
0
0
-
0
0
0
-
+
+
+
0
Note: () optional function; + major function; o secondary function; - insignificant function
Nonstructural BMP Representation
The previous discussions focused on the simulation methods for structural BMPs, but viable nonstructural
BMPs are potentially effective. For example, street sweeping is a nonstructural BMP; however, it is
handled as part of the land simulation module because it directly reduces pollutant sources upstream of
structural BMPs. Other pollutant source control practices, like fertilizer or pet waste management, are
BMPs; but for a similar reason, they are accounted for through the land module. With the exception of
street sweeping, SUSTAIN does not provide explicit representation for source control actions. Using
standard modeling practices, land use characteristics in the land module or in an external model can be
modified to represent changes in land management. The user can create alternative land use categories
that represent areas with and without nonstructural BMPs. Note that SUSTAIN does not support the use
of land uses as a decision variable in cost-optimization.
Using VFSMOD in SUSTAIN
The VFSMOD component for the simulation of vegetated filter strips has limitations in terms of how it is
integrated within SUSTAIN. While VFSMOD has a detailed representation of sediment transport through
a filter strip, it uses a simple first-order decay representation for water quality constituents and does not
address sediment-associated pollutants. In addition, SUSTAIN acts as a pre-processor for generating an
input file for VFSMOD, and data flow to VFSMOD is one-directional. For that reason, a filter strip
modeled in VFSMOD cannot be integrated with other BMPs or be included in an optimization
formulation. Consequently, VFSMOD is primarily used as a BMP evaluation component for assessing
filter strip performance.
3-73
-------
Using the Aggregate BMP
The behavior of an aggregated BMP type in a treatment train conceptually represents the collected
behavior of all BMPs of the same type in the watershed. No attenuation of flow and pollutants, such as
from routing through a reach or a conduit, occurs from one upstream aggregated BMP to the one
downstream. For small drainage areas, that assumption usually has less impact because the routing effect
is small and the travel time or time of concentration is short. As drainage area increases, so does the
compounded impact of routing through a large conveyance network. The aggregate BMP simulation
results will begin to diverge from the detailed distributed simulation. In summary, the aggregate BMP
option tends to work best in watersheds that have a time of concentration similar to the simulation time
step. Given a one-hour time step simulation, a good size for watershed with a low- to moderate slope is
between 50 and 150 acres. The aggregated BMP is probably best used for screening-level analysis to
determine the treatment potential of a watershed. Once treatment targets have been established, further
analysis can be perform on a more fully articulated and detailed BMP network.
Operation and Maintenance Assumptions and Costs
The BMP cost database does not include O&M costs, and provisions are made to allow users to enter
values in terms of a fraction of construction costs or as an added fixed cost. Because of the limited and
highly variable nature of the O&M cost information, the cost is assumed to be evenly distributed over the
entire life cycle of the BMP. In addition, the BMPs are assumed to be maintained to perform as designed
and that performance does not degrade over time.
3.4. Conveyance Module
The conveyance module performs routing of flow and water quality through a conduit. In SUSTAIN,
conduits are pipes or one-dimensional open channels that move water from one node to another in a
watershed routing network. The cross-sectional shapes of a conduit can be selected from a variety of
standard open and closed geometries. Irregular, natural, cross-section shapes are supported, as are user-
defined, closed shapes. Flow and pollutant routing are simulated using algorithms from SWMM (version
5) transport compartment (Rossman 2005). Sediment routing is simulated using reach sediment transport
algorithms from HSPF (Bicknell et al. 2001). Table 3-28 provides an overview of the required inputs, the
methods used to manage and process the inputs, and the resulting outputs of the conveyance module.
3.4.1. Methodology
This section discusses the methodologies applied by the conveyance module to handle flow routing,
sediment settling and routing, pollutant removal, and pollutant routing.
Flow Routing
SUSTAIN uses the kinematic wave method to perform flow routing simulation. This method solves the
continuity equation along with a simplified form of the momentum equation. Kinematic wave routing
allows flow to vary both spatially and temporally in a conduit and results in attenuated and delayed
outflow hydrographs as inflow is routed (Rossman 2005). The maximum flow that can be conveyed
through a conduit is the full-flow Manning equation value. Any flow in excess of this conduit capacity is
bypassed to the downstream node as an untreated overflow. The typical Manning's roughness
coefficients for closed pipes and open channels are shown in Table 3-29 and Table 3-30, respectively.
3-74
-------
Table 3-28. Summary of Inputs, Methods, and Outputs of the Conveyance Module
Inputs
- Define conduit dimensions
- Define conduit initial condition parameters
- Define reach cross-section for irregular shape
- Define sediment settling and transport parameters
- Define pollutant removal and routing parameters
- Hourly inflow time series
- Hourly sediment (sand, silt, and clay) concentration time series
- Hourly pollutant concentration time series
Methods
- Flow routing is computed using kinematic wave method
- Sediment (sand, silt, and clay) settling and routing is computed using the
process-based algorithms adopted from the HSPF model
-Pollutant removal is calculated using the 1st order decay method
- Pollutant routing is computed using the CSTR method
Outputs
- Sub-hourly outflow time series
- Sub-hourly sediment (sand, silt, and clay) concentration time series
- Sub-hourly pollutant concentration time series
Table 3-29. Typical Manning's Roughness Coefficient for Closed Pipes
Conduit Material
Asbestos-cement pipe
Brick
Cast iron pipe
Cement-lined and seal coated
Concrete (monolithic)
Smooth forms
Rough forms
Concrete pipe
Corrugated-metal pipe
( 1/2 -in. x 2 -2/3 -in. corrugations)
Plain
Paved invert
Spun asphalt lined
Plastic pipe (smooth)
Vitrified clay
Pipes
Liner plates
Manning's Roughness
0.011-0.015
0.013-0.017
0.011-0.015
0.012-0.014
0.015-0.017
0.011-0.015
0.022-0.026
0.018-0.022
0.011-0.015
0.011-0.015
0.01-0.015
0.013-0.017
Source: ASCE 1982
3-75
-------
Table 3-30. Typical Manning's Roughness Coefficient for Open Channels
Conduit Material
Lined Channels
Asphalt
Brick
Concrete
Rubble or riprap
Vegetal
Excavated or dredged
Earth, straight and uniform
Earth, winding, fairly uniform
Rock
Unmaintained
Natural channels (minor streams, top
width at flood stage < 100 ft)
Fairly regular section
Irregular section with pools
Manning's Roughness
0.013-0.017
0.012-0.018
0.011-0.020
0.020-0.035
0.030-0.40
0.020-0.030
0.025-0.040
0.030-0.045
0.050-0.140
0.030-0.070
0.040-0.100
Source: ASCE 1982
Sediment Transport
This section describes the transport, deposition, and scour of inorganic sediment in free-flowing reaches
using the HSPF algorithms (Bicknell et al. 2001). Figure 3-16 shows the principal state variables and
fluxes involved in the sediment transport processes.
Total
Sediment
Outflow
_
Suspended
Storage
SEDdsJ
Deposition
Scour
°bed_i
Bed
Storage
for Sand
for Silt
for Clay
Figure 3-16. Schematic of sediment transport, deposition, and scour in conduits.
Both the migration characteristics and the adsorptive capacities of sediment vary significantly with
particle size. To facilitate analyses to account for the effects of particle sizes, SUSTAIN divides the
3-76
-------
inorganic sediment load into three components (sand, silt, and clay), each with its own properties. Sand
has a particle size ranging from 0.05 millimeter (mm) to 2.0 mm in diameter, silt from 0.002 mm to 0.05
mm in diameter, and clay smaller than 0.002 mm.
The system assumes that scour or deposition of inorganic sediment does not affect the hydraulic
properties of the conduit. Furthermore, it is assumed that sand, silt, and clay deposit in different areas of
the conduit bed so that the deposition or scour of one material is not linked to the changes of others.
Longitudinal movement of bed sediments by flow shear stress is not modeled.
First, the volume occupied by each component of bed sediment is calculated as shown in Equation (3-42).
Vbd.=?^L (3-42)
e ~ P_t
where
Vbedj = volume occupied by component / of bed sediment (ft3),
Sbedj = bed storage of component /' of sediment (Ib), and
Pj = particle density of component /' (lb/ft3).
The volumes of the three components of bed sediment are summed, and the total bed volume is adjusted
to account for voids in the sediment (i.e., the porosity):
i=3
Vbed_r (3.43)
vh=—
O j
l-r\
where
Vh = volume of bed (ft3),
Vbedj = volume of sediment contained in the bed (sand, silt, and clay) (ft3), and
77 = porosity of bed sediment (ratio of pore volume to total volume).
Finally, the depth of bed sediment is calculated as:
db=T^Wb <3-44)
where
db = depth of bed (ft),
Vb = volume of bed (ft3),
Lr = length of conduit (ft), and
Wb = effective width of bed (ft).
Cohesive sediments
Two steps are used to model the deposition, scour, and transport processes of cohesive sediments (silt and
clay). The first step computes the advective transport and the second step calculates the amount of
deposition or scouring on the basis of the bed shear stress.
3-77
-------
Advective Transport of Constituent
This section computes the concentration of material in a conduit and the quantities of material that leave
the conduit due to longitudinal advection. Two assumptions are made in the solution technique for
normal advection: each constituent is uniformly dispersed throughout the waters of the conduit and is
completely entrained by the flow—that is, the material moves at the same horizontal velocity as the water.
The equation of continuity can be written as:
SEDm - SEDro = (C x V) - (C. x V.) (3-45)
where
SEDin = total inflow of sediment over the interval (Ib),
SEDm = total outflow of sediment over the interval (Ib),
Cs = sediment concentration at the start of the interval (lb/ft3),
C = sediment concentration at the end of the interval (lb/ft3),
Vs = volume of water stored at the start of the interval (ft3), and
V= volume of water stored at the end of the interval (ft3).
The other basic equation states that the total outflow of material over the time interval is a weighted mean
of two estimates; one based on conditions at the start of the interval, the other on ending conditions:
SEDm = (C, xQsx js) + (CxQxcojs) (3-46)
where
Qs = outflow rate at the start of the interval (ft3/time interval),
Q = outflow rate at the end of the interval (ft3/time interval),
js = weighting factor, and
cojs = 1 -js.
By combining Equations (3-45) and (3-46) we can solve for the concentration C:
c = SEDin+Csx(Vs-Q,xjs) (3_47)
V + Q x cojs
The total amount of material leaving the conduit during the interval is calculated using Equation (3-46).
If the conduit goes dry during the interval, the total amount of material leaving the conduit is the sum of
the material coming in and the material leaving based on the concentration at the start of the interval:
SEDm=SEDtn+(C.xQ.xjs) (3-48)
Deposition and Scouring
Exchange of cohesive sediments with the bed is dependent upon the shear stress exerted on the bed
surface. When the shear stress (r) in the conduit is less than the user-supplied, critical, shear stress for
deposition (rcd), sediment deposition occurs. On the other hand, when the shear stress is greater than the
user-supplied, critical, shear stress for scour (TCS), scouring of cohesive bed sediments takes place. The
rate of deposition for a particular fraction of cohesive sediment is based on a simplification of Krone's
(1962) equation in the following form:
3-78
-------
D = coxCx \l-—\ (3-49)
I TCJ
where
D = rate at which sediment settles out of suspension (lb/ft2 interval),
co = settling velocity for cohesive sediment (ft/interval),
C = concentration of suspended sediment (lb/ft3),
T = shear stress (lb/ft2), and
rcd = critical shear stress for deposition (lb/ft2).
The rate of change of suspended sediment concentration in the conduit due to deposition can be expressed
as:
-- (3.50)
dt d,
where
dav = average depth of water in the conduit (ft).
By substituting the expression for deposition rate (D) from Equation (3-49), and integrating and
rearranging Equation (3-50), a solution can be obtained for the concentration of suspended sediment lost
to deposition during a simulation interval (Cdep):
1-exp -— x \1-— \\ (3-51)
L II = settling velocity for sediment fraction (ft/interval),
dm = average depth of water in conduit (ft),
r = shear stress (lb/ft2), and
rcd = critical shear stress for deposition (lb/ft2).
The user must supply values for settling velocity (ca) and critical shear stress for deposition (rcd) for silt
and clay fractions in cohesive sediment.
The amount of sediment in suspension (Ssus) is updated by subtracting the amount settled. Likewise, the
amount of sediment in bed (Sbed) is updated by adding the amount settled on it.
The rate of resuspension, or scour, of cohesive sediments from the bed is derived from a modified form of
Partheniades' (1962) equation:
S = ^x—-J (3-52)
1T« )
where
S = rate at which sediment is scoured from the bed (lb/ft2 interval),
3-79
-------
// = erodibility coefficient for the sediment fraction (lb/ft2 interval), and
TCS = critical shear stress for scour (lb/ft2).
The rate of change of suspended sediment fraction concentration in the conduit due to scour can be
expressed as:
(3.53)
By substituting the expression for scour rate (S) from Equation (3-52) and integrating and rearranging
Equation (3-53), a solution can be obtained for the concentration of suspended sediment added to
suspension by scour during a simulation interval (Csa):
T
T_
(3-54)
d.
where
ju = erodibility coefficient (lb/ft2 interval), and
dav = average depth of water (ft).
The user is required to supply values for the erodibility coefficient (/LI) and critical shear stress for scour
(TCS) for each fraction of cohesive sediment (silt and clay) that is modeled.
The amount of sediment in suspension (Ssus) is updated by adding the scoured mass, as is the amount of
sediment in bed (Sbecj) by subtracting the scoured mass.
If the amount of scoured sediment is greater than the original sediment in the bed, all sediment in the bed
will be resuspended and the amount of sediment in the bed is set to zero.
Non-cohesive Sediment
Erosion and deposition of sand, or non-cohesive sediment, is affected by the amount of sediment that the
flow is capable of carrying. If the amount of sand being transported is less than the flow can carry for the
hydrodynamic conditions of the conduit, sand is scoured from the bed. This occurs until the actual sand
transport rate becomes equal to the carrying capacity of the flow or until the available bed sand is all
scoured. Conversely, deposition occurs if the sand transport rate exceeds the flow's carrying capacity.
The sand transport capacity for a conduit is calculated by using an input power function of the velocity.
The potential sand concentration (Cp) is determined by the following conversion:
Cp=kxvJ (3-55)
where
Cp = potential sand concentration (lb/ft3),
k = coefficient in the sandload suspension equation (input parameter),
j = exponent in sandload suspension equation (input parameter), and
vav = average velocity (ft/s).
3-80
-------
The potential outflow of sand (SEDprc) is calculated as:
SEDpm = (C, x05 x js)+(Cp xQxcojs) (3-56)
where Cs, Qs,js, Q, and cojs are as previously defined for Equations (3-45) and (3-46).
The potential scour from, or deposition to, the bed storage is found using the continuity equation:
SEDpds=(VxCp)-(VsxCs)+SEDpro-SEDm (3-57)
where
SEDpds = potential scour (+) or deposition (-) (Ib),
Cp = potential sand concentration at the end of the interval (lb/ft3),
Cs = sand concentration at the start of the interval (lb/ft3),
SEDpro = potential outflow of sand over the interval (Ib), and
SEDin = inflow of sand during the interval (Ib).
The potential scour is compared to the amount of available sand for resuspension. If scouring potential is
less than the available sands, the demand is satisfied in full and the bed storage is adjusted accordingly. If
the potential scour cannot be satisfied by bed storage, all the available bed sand is suspended, and the bed
storage is exhausted. The concentration of suspended sand (Q is calculated as:
c = SEDm+SEDds+Csx(Vs -QsxjS) (3_5g)
V + Q x cojs
where
C = concentration of sand at end of interval (lb/ft3),
Cs = concentration of sand at start of interval (lb/ft3),
SEDin = inflow of sand during the interval (Ib), and
SEDds = sand scoured from, or deposited to, the bottom (Ib).
The total amount of sand leaving the conduit during the interval is calculated using Equation (3-58). If a
conduit goes dry during an interval, or if there is no outflow from the conduit, all the sand in suspension
at the beginning of the interval is assumed to settle out, and the bed storage is correspondingly increased.
Sediment Transport Input Parameters
Parametric information required for silt and clay includes particle diameter (0), particle settling velocity
in still water (a>), particle density (p), critical shear stress for deposition (rcd), critical shear stress for scour
(TCS), and erodibility coefficient (ju). Parameter values required for sand include median bed sediment
diameter (fc0) and particle settling velocity (CD). Table 3-31 shows the range of input parameters that are
recommended.
Sediment Transport Calibration
Sediment transport parameters are typically derived by calibration. The calibration process involves
establishing initial parameter values and a subsequent adjustment process. The eroded material from each
land use category is fractionated into sand, silt, and clay before entering a conduit using available soils
information; typically, a single fractionation scheme is used for all conduits. The fraction should reflect
the relative percent of the surface material (i.e., sand, silt, clay) available for input to the conduit.
3-81
-------
Investigation of the bed material composition will also help provide insight into appropriate fractionation
values.
The initial sediment parameters—such as particle diameter, particle density, settling velocity, and bed
depth and composition—and beginning calibration parameters can be evaluated from sources such as
local/regional data, past experience, and handbooks or literature. The parameter values can then be
adjusted on the basis of available site-specific data and calibration.
Table 3-31. List of Sediment Input Parameters for the Reach
Parameters
50
(O
V
P
Ted
TCS
k
j
Default Value
0.0
0.01
none
0.0
2.65
1.0E10
1.0E10
0.0
0.0
Min. Value
0.0
0.0001
0.02
0.0
1.0
l.OE-10
l.OE-10
0.0
0.0
Max. Value
0.003
100.0
500.0
none
4.0
none
none
none
none
Units
in.
in.
in./s
Ib/ft2/day
lb/ft3
lb/ft2
lb/ft2
~
~
Source: Bicknell etal. 2001
Pollutant Transport
The conveyance module simulates pollutant routing assuming that the conduit behaves as a CSTR. The
algorithms are adopted from SWMM (version 5) transport compartment (Rossman 2005). The pollutant
concentration exiting the conduit is determined by integrating the conservation of mass equation, using
average values for quantities that might change over the time step such as flow rate and volume.
SUSTAIN simulates sediment as a primary pollutant and assumes that all other pollutants follow the 1st
order decay in a conduit as shown in Equation (3-41).
3.4.2. Important Considerations and Limitations of the Conveyance Module
Use of Kinematic Wave Routing Method
Several issues are important to consider when applying the conveyance module to perform stream
routing. The desire to maintain a high resolution of spatial detail must be balanced with the need to
preserve time of concentration along the in-stream flow network in the watershed. In-stream travel time
can have a significant influence on model stability and accuracy. SUSTAIN incorporates two types of
routing algorithms: storage routing and kinematic wave. Those algorithms are most accurate when the
flow time of the flood wave through individual reaches approximates the simulation time step. This is
achieved during model configuration by either: (1) selecting a simulation time step that is as small as the
travel time through the smallest transport segment in the network, or (2) sizing the transport segments in
the network to have travel times that are at least greater than or equal to the simulation time step.
In cases most likely encountered in the context of SUSTAIN applications, a kinematic wave routing
method can usually maintain numerical stability with time steps on the order of 5 to 15 minutes. If those
effects are not expected to be significant, kinematic wave routing can be an accurate and efficient method,
especially for long-term simulations. Finally, it is important to recognize that kinematic wave routing
does not account for backwater effects, entrance/exit losses, flow reversal, or pressurized flow and is
restricted to dendritic network layouts.
3-82
-------
Simulation Time and Number of Conveyance Elements
Another aspect to consider in setting up a conveyance network for simulation is to determine the number
of conveyance elements. The conveyance system is found to be the most computationally intensive
component of the network. It often needs an order of magnitude more time to run water through a
conveyance than through a BMP. For this reason, the strategic selection and sizing of conveyances (i.e.,
using one representative pipe in place of several actual pipes in a section of the network) can have
significant benefit in terms of computation time savings. If a conveyance network is a part of an
optimization formulation, a saving of several seconds per simulation run will translate into a significant
reduction in computational time because hundreds or even thousands of optimization runs are often
required to arrive at optimal solutions.
3.5. Optimization Module
SUSTAIN includes an optimization module to develop cost-effective BMP placement and selection
strategies on the basis of a pre-selected list of potential sites and applicable BMP types and size ranges.
The module uses evolutionary optimization techniques to perform the searches for optimal combinations
of BMPs that meet the user-defined decision criteria. Table 3-32 summarizes the required inputs,
methods used, and outputs and Figure 3-17 presents a conceptual overview of the module.
The optimization module works hand-in-hand with the BMP, land, and conveyance modules during the
search process in an iterative and evolutionary fashion. The simulation modules evaluate the BMP
performance, as defined via evaluation factors, and cost data of a set of chosen BMP options and pass that
information to the optimization engine. The optimization engine synthesizes the information, modifies
the search path, and generates new solutions that are repeatedly evaluated using the simulation modules.
Through this evolutionary search process, the module will progressively march toward the identification
of the best or most cost-effective BMP solutions that meet the user's specific conditions and objectives.
Table 3-32. Summary of Inputs, Methods, and Outputs in the Optimization Module
Inputs
-Define decision variables (the size ranges of potential BMPs )
- Define assessment point(s) and evaluation factor(s)
- Define management targets (for the minimize cost option)
- Define BMP cost functions
Methods
- For the minimize cost option, optimization search is performed using Scatter Search technique
- For the generate cost-effectiveness curve option, optimization search is performed using NSGAII technique
Outputs
- For the minimize cost option, the optimization process outputs optimal solutions that meet the specified
treatment targets
- For the cost-effectiveness curve option, the optimization process outputs the optimal solutions along the cost-
effectiveness curve
3.5.1. Problem Formulation
The objective of the optimization module is to determine BMP locations, types, and design configurations
that minimize the total cost of management while satisfying water quality and quantity constraints. To
formulate an optimization problem, SUSTAIN requires the user to specify three sets of information:
decision variables, assessment points and evaluation factors, and management targets.
3-83
-------
ArcGIS Map Interface
Generate BMP
Model Input File
Changeable Variables
(#, min, max, increment)
Control Target
SOLUTION
Figure 3-17. Conceptual overview of the optimization module.
Decision Variables
Placing BMPs at different spatial levels or locations (or both) affects the overall cost-effectiveness of the
stormwater control system (Zhen 2004). Therefore, BMP location represents one important decision
variable for optimization. The possible BMP locations are typically pre-selected on the basis of multiple
factors, including availability of space site characteristics (slope, soil infiltration rates, and water table
elevation) and other logistical considerations. Another important decision variable involves BMP
configuration. At a given feasible location of a BMP type, the configuration parameters can be treated as
decision variables with the specified minimum, maximum, and discreet search interval values.
Assessment Point(s) and Evaluation Factor(s)
An assessment point is a location where the water quality or quantity parameters or both are evaluated.
Figure 3-18 shows an example of assessment points that can be at the watershed outlet, key tributary
outlets, and the downstream node of a stream segment.
Figure 3-18. Illustration of assessment points.
3-84
-------
SUSTAIN provides the user an evaluation factor selection menu when defining the optimization problem.
The framework allows for various averaging periods and frequencies consistent with that for typical water
quality criteria and TMDL related allocations. Table 3-33 lists the evaluation factor options in SUSTAIN.
Table 3-33. Example Control Targets for Typical Evaluation Factor Assessment in SUSTAIN
Control Target
Target Value
Note
Flow
Peak Discharge
Annual Average
Volume
Exceeding
Frequency
cubic feet per sec
percent reduction of the existing condition
fraction between existing and pre-developed conditions
cubic feet per year
percent reduction of the existing condition
fraction between existing and pre-developed conditions
times per year of a given threshold flow rate (cfs)
percent reduction of the existing condition
fraction between existing and pre-developed conditions
~
0-100
0-1
-
0-100
0-1
~
0-100
0-1
Parameters related
to increased runoff
from urbanization
Sediment
Annual Average
Load
Annual Average
Concentration
Maximum Days
Average
Concentration
pounds per year value
percent reduction of the existing condition
fraction between existing and pre-developed conditions
milligram per liter value
percent reduction of the existing condition
fraction between existing and pre-developed conditions
milligram per liter value of given days
percent reduction of the existing condition
fraction between existing and pre-developed conditions
-
0-100
0-1
~
0-100
0-1
~
0-100
0-1
Parameters to meet
the water quality
standards or
biologically
derived parameters
to meet designated
uses in waterbody
of concern
Pollutants (TN, TP, or User Defined)
Annual Average
Load
Annual Average
Concentration
Maximum Days
Average
Concentration
pounds per year value
percent reduction of the existing condition
fraction between existing and pre-developed conditions
milligram per liter value
percent reduction of the existing condition
fraction between existing and pre-developed conditions
milligram per liter value of given days
percent reduction of the existing condition
fraction between existing and pre-developed conditions
~
0-100
0-1
~
0-100
0-1
~
0-100
0-1
Parameters to meet
the pollutant
criteria (numeric
concentration or
frequency of
exceedance) or
TMDL (load
allocation) or other
locally defined
water-protection
goals
Management Targets
Management targets can be related to either water quality or quantity. The user specifies the water quality
or water quantity target value or range for each assessment point.
3-85
-------
3.5.2. Optimization Algorithms
Evolutionary search techniques have shown great promise in their ability to solve nonlinear,
multiobjective, complex optimization problems such as the one above (Zhen and Yu 2004; Dorn and
Ranjithan 2003; Harrell 2001; Perez-Pedini et al. 2005). Though such techniques demand extensive
computing time, rapid advances in computing power and speed make the techniques more practical and
applicable than ever before.
Numerous evolutionary algorithm-based multiobjective optimization procedures are available. Generally,
the search techniques can be classified into two categories, i.e., (1) constraint method-based evolutionary
algorithms, which use single-objective evolutionary algorithms to solve multiobjective optimization
problems by transforming the multiobjective problem to a single-objective problem via the constraint
method, and then solving it iteratively; (2) multiobjective evolutionary algorithms, which solve the
problem in a single pass, where the population represents the set of nondominated solutions. In the
SUSTAIN framework, two search techniques are implemented: Scatter Search and Nondominated Sorting
Genetic Algorithm-II (NSGA-II)). Scatter Search is a single-ojective evolutionary algorithm, and NSGA-
II is a multiobjective evolutionary algorithm.
Scatter Search is introduced by Glover (1977) as a heuristic for integer programming that expanded on the
concept of surrogate constraints. The Scatter Search method is an evolutionary search technique that has
been explored and used in optimizing complex systems (Glover et al. 2000). Scatter Search shares some
commonalties with the widely applied single-objective genetic algorithms (GAs) because both techniques
are population-based approaches. However, Scatter Search and GA have a number of distinct features of
their own. GA approaches are predicated on the idea of choosing parents randomly to produce offspring
and then introduce randomization to determine which components of the parents should be combined. By
contrast, the Scatter Search approach does not emphasize randomization or being indifferent to choices
among alternatives. Instead, it is designed to incorporate strategic responses, both deterministic and
probabilistic, that take account of evaluation history. Scatter Search focuses on generating relevant
outcomes without losing the ability to produce diverse solutions (Laguna and Marti 2002). Because of
that feature, it serves as a better optimization technique for identifying the near-optimal solution with a
specific target value.
NSGA-II is one of the most efficient, multiobjective, evolutionary algorithms using the elitist approach
(Deb et al. 2002). In NSGA-II, solutions are sorted on the basis of the degree of dominance within the
population (i.e., if a given solution is not dominated by any other solution, that solution has the highest
possible fitness). In addition, the algorithm seeks to preserve diversity along the first non-dominated
front so that the entire Pareto-optimal region is found. NSGA-II has gained popularity in recent years and
showed superiority over other multiobjective evolutionary algorithms, e.g. Pareto-Archived Evolution
Strategy and Strength-Pareto EA, when applied to solve optimization problems associated with watershed
management (Dorn and Ranjithan 2003).
Comparison of Scatter Search and NSGA-II
Both Scatter Search and NSGA-II are population-based evolutionary optimization techniques; however,
they employ different search strategies. Scatter Search refines a scatter pattern around the targeted
objectives by replacing members of a reference population, while NSGA-II defines a population as
individual solutions along a cost-effectiveness curve and refines the entire population with better
solutions until the final solution approaches the true Pareto frontier. Pareto optimality is a concept
commonly applied in economics and engineering. At any cost or reduction level, a solution is said to be
Pareto optimal when no further improvements can be made. Figure 3-19 is a conceptual representation of
the two search routines. In Figure 3-19, the Pareto optimum frontier is represented in two-dimensional
space as the solid cost-benefit arc in both graphs. A new dimension would be added for each additional
3-86
-------
pollutant reduction target in the objective function. The concentric circles in the Scatter Search graph
illustrate progressively better (narrower) reference populations, until the final set of best solutions are
found clustered around the point along the Pareto frontier. The dashed lines in the NSGA-II graph
illustrate progressively improving cost-benefit relationships with each new generation of solutions until
the true Pareto frontier is realized in the last generation. Scatter Search clusters solutions around the
defined objective on the Pareto frontier, while NSGA-II distributes the solutions along the entire trade-off
frontier. Increasing the resolution for NSGA-II means increasing the number of individuals in the
population, which increases the number of generations needed to find the Pareto optimal frontier.
$
Scatter Search
$
NSGAII (Non-dominated
Sorting Genetic Algorithm II)
Figure 3-19. Comparison of Scatter Search and NSGA II optimization techniques.
For the minimize cost option, Scatter Search is more efficient (finding the best solutions with fewer runs
of the simulation module) because the search is more focused around the target. For the cost-effectiveness
curve option, NSGA-II (genetic algorithm) is more efficient because it applies the non-dominated sorting
technique and the search proceeds in a manner of fronts.
Scatter Search
The major operation steps of Scatter Search are described below.
Generating a starting set of diverse points
Generating a starting set of diverse points is accomplished by dividing the range of each variable into four
sub-ranges of equal size. Next, a solution is constructed in two steps: a sub-range is first randomly
selected and then a value is randomly chosen from the selected sub-range. The starting set of solution
points also includes all variables at their lower bound, all variables at their upper bound, all variables at
their midpoints, and other solution points suggested by the user.
Choosing a subset of diverse points as the reference set
The reference set (RefSet), is a collection of both high-quality solutions and diverse solutions that are
used to generate new solutions. Specifically, the RefSet consists of the union of two subsets, RefSetl and
RefSet2, of size bt and b2, respectively. That is, \RefSet\ =b = b1 + b2. The construction of the initial
reference set starts with the selection of the best bi solutions from the starting set of diverse points (/*).
The notion of best in this step is a measure given by the evaluation of the objective function. These
3-87
-------
solutions are added to RefSet and deleted from P. For each improved solution in P - RefSet, the
minimum of the Euclidean distances to the solutions in RefSet is computed. Euclidean distance is the
straight line distance between two points. For example, in a two-dimensional plane, the Euclidean
distance is the straight line between point 1 at (xh yi) and point 2 at (x2, y2) and is equal
Then, the solution with the maximum of these minimum distances is selected. This solution is added to
RefSet and deleted from P and the minimum distances are updated. This process is repeated b2 times.
The resulting reference set has bt high-quality solutions and b2 diverse solutions.
Starting the search for the optimal solution by using a linear combination method to construct new
solution points from the reference solution points
The linear combination is based on the three types of formulation, in which x' and x" are reference
solution points, and x;_5 is the newly generated solution points:
x3=x' + d
x — x
where d = r - and r is a random number in the range of (0, 1).
Updating the RefSet
In the course of searching for a global optimum, the RefSet is continuously updated. The solutions
having better quality, or ones that can improve the diversity of the reference set, replace the old points in
the set.
Stop the search if the stopping criteria are met
The stopping criteria can be defined, at the user's option, either as the maximum number of iteration runs,
or the minimum improvement between updates of the reference set, or both, in which case, the search
process will be stopped when either of the criteria is met.
NSGA-II
The major operation steps of NSGA-II are described below.
Creation of first generation
When applying the NSGA-II, a random parent population (P0) consisting of TV solutions is first created.
The population is then sorted by the non-dominant level. A solution x(1> is non-dominant to another
solution when x(1> performs no worse than the other solution in all objectives, and x(1> performs better than
the other solution in at least one objective. At the end of the sorting, each solution is assigned a fitness
(or rank) equal to its non-dominant level, with a smaller value indicating that the solution is dominated by
fewer other solutions.
The processes of tournament selection, crossover, and mutation are used to create a child population (Q0),
which has a same size ofP0 with N solutions.
3-88
-------
Main loop
In the first step of the main loop, the parent population and the child population are combined (R0). The
population of R0 will have 27V solutions. The 27V solutions in R0 are then sorted according to non-
domination. Elitism is ensured in this step because both the parent and the child population are used in
the sorting. The sorted 27V solutions will form various best non-dominated subsets (in which all the
solutions are non-dominant to each other, but overall they dominate other subsets). The first TV solutions
from the ranked best non-dominant subsets, F}, ..., Ft, are then selected to form a new parent population
(Pi). The new parent population is used to create a new child population (O;), and the process continues
until the stopping criteria are met.
The NSGA-II uses the crowding distance (the size of the largest cuboid enclosing solution x(1) without
including any other solution in the population) concept to maintain solution diversity. That is, in cases
where two solutions have the same non-domination rank, the solution with larger crowding distance is
always preferred.
Stopping criteria
The user can make the NSGA-II stop when the new parent population does not change for two
consecutive loops. The stopping criterion can also be that the fitness function does not improve after a
certain number of iterations.
Optimization Options
SUSTAIN provides two optimization options: (1) cost minimization, and (2) cost-effectiveness curve. In
the cost-minimization option, the optimization search process identifies the near-optimal solutions that
meet the user-specified management targets. With the cost-effectiveness curve option, the optimization
process reveals all the cost-effective solutions within the user-specified management target range.
Cost-Minimization Option
With the objective of minimizing cost subject to desired water quality or water quantity objectives (or
both) at a specified location (assessment point), the optimization problem formulation can be
mathematically expressed as below. In the formulation, a group of BMPj (i = !,...,«) forms the decision
matrix, which defines the optimization engine's search domain. For each potential location, the user
defines the feasible range of BMP type and configuration parameters.
The objective is to:
n
Minimize ^Cost(EMPt)
i=l
subject to:
< maX and
Lk < Lmaxk
where
BMPi = a set of BMP configuration decision variables associated with location /',
QJ = the computed amount of water quantity factor at the assessment pointy,
QmaXj =the maximum value of the water quantity factor targeted at the assessment point j,
Zk = the computed amount of water quality loading factor at the assessment point k, and
Lmaxk = the maximum value of the water quality loading targeted at the assessment point k.
3-89
-------
Cost-Effectiveness Curve Option
Under the cost-effectiveness curve option, the search aims at identifying the cost-effective solutions
within the specified management target range. The multiobjective problem can be expressed as follows:
n
Minimize '^Cost(BMPJ) and
i=l
Minimize EF
where
BMPj = a set of BMP configuration decision variables associated with location /' and
EF = the management evaluation factor (EF) at one given assessment point, and the EF can be
any of the options listed in Table 3-33.
3.5.3. Regional Application
SUSTAIN is able to evaluate management practices at multiple scales, ranging from local to watershed
applications. Placement of BMPs at different spatial levels (i.e., on-site, subregional, and regional)
affects the overall cost-effectiveness of the storm water control system (Zhen and Yu 2004). Management
plans often need to evaluate the cumulative benefit of management practices at multiple- scale watersheds
on downstream water quality in rivers, lakes, or estuaries. The site or local-scale evaluation involves
simulation and analyses of individual BMPs and various combinations of practices and treatment trains to
derive local runoff quantity and quality. For a larger-scale watershed, there could be hundreds or
thousands of individual management practices that are implemented to achieve a desired cumulative
benefit. SUSTAIN incorporates an innovative, tiered approach that allows for cost-effectiveness
evaluation of both individual and multiple nested watersheds to address the needs of both regional and
local-scale applications (Figure 3-20). This section describes the procedures of the tiered optimization
analysis approaches in SUSTAIN for sequentially identifying cost-effective BMPs on a regional scale.
A relatively large watershed can usually be subdivided into several smaller subwatersheds as shown in
Figure 3-20. Users need to select, with, say, the use of the siting tool in SUSTAIN, an appropriate suite of
feasible BMP options (types, configurations, and costs) at strategic locations for each subwatershed.
SUSTAINihen generates the time series rainfall-runoff data from BMP drainage areas and routes them
through BMPs, in parallel or in series, to produce the quantity and quality data at downstream assessment
points. SUSTAIN uses the cost and effectiveness data to derive the cost-effectiveness curve that relates
flow or pollutant-load reductions with costs. Each point on the cost-effectiveness curve represents an
optimal combination of BMPs that will collectively remove the targeted amount of pollutant load at the
least cost.
The tiered optimization procedures implemented in SUSTAIN provides an efficient and manageable
means for large-scale applications and allows users to evaluate and optimize on the basis of the
hydrologic and water quality characteristics at the specified assessment points. Tier-1 performs the
optimization search to develop cost-effectiveness curves for each tier-1 subwatershed. In atier-2
analysis, the tier-1 solutions are used to construct a new optimization search domain and run the transport
module, if needed, with solutions from all the tier-1 subwatersheds to develop the combined cost-
effectiveness curve for the entire watershed.
3-90
-------
Target Load., __
Reduction j
SUSTAIN Tier 2
Figure 3-20. Tiered application of SUSTAIN for developing cost-effectiveness curves.
Figure 3-21 illustrates the tiered application process in more detail. At the first step (tier-1) of the tiered
optimization analysis, the cost-effectiveness curve for each subwatershed is generated by performing
continuous multiple optimization runs at incremental flow/pollutant reduction targets. In the second step
(tier-2), the search domain is constructed using the tier-1 results. As shown, the search domain for tier-2
contains the discrete solutions on the tier-1 cost-effectiveness curves at assessment points /' and/. The
third step is to perform the tier-2 optimization for the search domain constructed. The optimization
engine strategically samples the discrete options in the search domain. The cost-effectiveness of each
sample is measured, stored, and analyzed to guide the next search direction.
Tier-1 Cost Effectiveness Curve
for Assessment Point i
interval
Cost
Tier-1 Cost-Effectiveness Curve
for Assessment Point]
I"
Cost
: Tier-2 Search
Donjiajn
Tier-1 Solutions for
Assessment Point]
Figure 3-21. Construction of the tier-2 search domain using tier-1 results.
3-91
-------
Figure 3-22 illustrates the simulation process used to generate the results for measuring the cost-
effectiveness of each iteration in the tier analysis. The simulated time series outputs for all discrete points
on the tier-1 cost-effectiveness curve are stored and used when a point, hence the BMP options associated
with it, is chosen in the tier-2 analysis. Similarly, the time series runoff data of the watershed area that is
not part of the tributary areas of the tier-1 assessment points is generated and stored before the tiered
analysis. This data, however, might also be generated during the tier-2 search process. The transport
module is often required to perform routing of the time series data from the upstream tier-1 subwatershed
to merge with that for the downstream tier-1 subwatershed. In such a manner, the tiered approach is
applied to a large watershed which contains subwatersheds or to a small watershed that requires the
development of a detailed management plan at a parcel- or a street-block-level.
Simulation process for each iteration run:
Time Series:
®
|
•*
Transport
Module:
1 Channel 1
Tier-1 solution time series at assessment points
Time series for the areas/catchment that are not covered
by tier-1 assessment points
Figure 3-22. Simulation process for each iteration run.
3.5.4. Important Considerations and Limitations: Optimization Module
Importance of Calibration
The optimization engine performs iterative searches to identify cost-effective solutions. The search
process is dependent on the cost and BMP treatment effectiveness values of each BMP or a combination
of BMPs evaluated. Therefore, it is crucial to have calibrated watershed and BMP simulation modules, as
well as good BMP cost data, to ensure meaningful results. The cost-effective solutions from evolutionary
search techniques should be considered only near-optimal solutions, meaning that the solutions are not
guaranteed to be the absolute best but are believed to be close to it.
Minimizing Run-time and Computational Effort
The approximate number of iterative runs to reach the near-optimal solutions is dependent on the number
of decision variables, the number of discrete options of a decision variable, and the complexity of the
problem. In general, the number of runs needed is reduced with a reduced number of decision variables,
number of discrete options for a decision variable, and complexity of the problem. Experience gained
from performing experiments is usually useful for estimating the number of runs needed for reaching
near-optimal solutions.
The optimization process based on long-term continuous simulation can require a large amount of
computation time especially for a large watershed that has many feasible BMP sites and reaches/conduits.
To achieve computational efficiency, it is advantageous to minimize the number of non-dummy conduits,
decision variables, and search steps.
3-92
-------
When using the cost-minimization option, the user is asked to set a treatment (e.g., pollutant reduction)
target. To avoid setting an unreachable target, one can perform a simulation that maximizes treatment
scenario by setting the BMP sizes to maximum for a site for estimating the maximum treatment
effectiveness.
Tiered Analysis Interface
Although the SUSTAIN engine is capable of performing tiered optimization, the related interface is yet to
be developed. Users must manually create all the input files to perform the tiered optimization analysis.
A more robotic and automated interface will be developed to guide the users through the tiered analysis
processes.
3.6. Post-Processor for Results Interpretation
The SUSTAIN post-processing module provides a centralized location for analyzing and interpreting
simulation outputs at multiple locations, and for scenarios (e.g., existing development with and without
BMPs, and pre-development conditions) and parameters of interest (e.g., inflows, outflows, and pollutant
loads and concentrations). The framework allows users to evaluate simulation results that are highly
variable in magnitude, duration, intensity, treatment containment volume, attenuation, and pollutant-
removal effectiveness. The simulation outputs contain hourly or subhourly data, and can span several
years, depending on the length of simulation.
The primary objective of the post-processor is to mine the model results to derive meaningful information
that best characterizes the effectiveness of the modeled management strategies. This is achieved by the
use of specific graphical and tabular reports. Four components are in the post-processor: storm event
classification, storm event viewer, storm performance summary, and cost-effectiveness curve. Table 3-34
provides an overview of the required inputs, the methods used to manage and process the inputs, and the
resulting outputs from SUSTAIN"?, post-processor.
3.6.1. Storm Event Classification
Storms with identical hyetographs (i.e., precipitation pattern) can produce very different runoff responses
depending on prior moisture conditions, seasonality, and geographic location. For example, if one storm
follows a long, dry period, and another follows shortly after a relatively wet period, the runoff responses
would be different. BMP performance is similarly affected by antecedent conditions. As a result,
accurate interpretation of modeled BMP performance requires an understanding of prior precipitation and
soil moisture. SUSTAIN's post-processor facilitates this through its storm event classification method,
whereby the time interval between storms (storm interval) is specified in order to define a storm event
from a precipitation time series.
The storm interval is the number of dry days between storms necessary to restore near-equilibrium soil
moisture conditions. The interval varies by geographic region and can be selected using statistical
evaluation of precipitation time series. For example, in regions such as Seattle, Washington, where
rainfalls are frequent but of low intensity, a shorter storm event interval is more appropriate. Longer dry
intervals, say, 3 days, are typically used in drier regions that rarely get rain, such as Southern California.
Figure 3-23 shows how the derived set of storm events changes with the number of dry hours between
storms. For each storm event, the graphs in Figure 3-23 show the total precipitation and the peak
precipitation intensity during an interval, relative to other intervals. As expected, the number of events
decreases as the dry interval (number of dry hours between storms) increases.
3-93
-------
Table 3-34. Post-processor Inputs, Methods, and Outputs
Post-Processor
Inputs
- Storm event classification
- Rainfall time series file
- Number of dry days between storm events
- Evaluation period (start and end dates)
- Time series files for storm viewer and performance summary
- Development conditions: pre-developed, with, and without BMPs
- Cost-effectiveness curve
- Best solutions and all solutions
- Associated model input file
Methods
- Storm event classification: divide time series into discrete storm events, separated by a user-defined
minimum number of days for the entire evaluation period
- Storm event viewer: browse and display modeled time series comparisons for three scenarios (with and
without BMPs and pre-developed condition) for a user-selected storm event
- Storm performance summary: display summary comparisons for scenarios (with BMPs and pre-developed
condition, relative to the scenario without BMPs) for all storm events
- Cost-effectiveness curve: superimpose cost-effective solutions on all solutions
- Upon request, generate model results for a selected cost-effective solution
Outputs
- Storm event classification: browsable and sortable list of storm events
- Storm event viewer: time series comparison graphs
- Storm performance summary: load comparison graphs for all storm events and EMC graphs for all storms
- Cost-effectiveness curve: cost distribution by BMP type and storage distribution by cost interval
24 hours -» 78 events
Selected Interval •Total Rainfall HPeak Intensity
Figure 3-23. Number of storm events and total precipitation and peak intensity
as a function of dry hours between storms.
3-94
-------
In the SUSTAIN post-processor, once the storm events are defined, the user has the option of sorting the
events by total precipitation volume or by peak intensity, as shown in Figure 3-24. Note that the same
event (August 9, 2001, 6:00 p.m.) has been highlighted in both Figure 3-23 and Figure 3-24. The event
sorting provides additional insight into the details of the selected event relative to other events. For
example, in terms of total rainfall volume, the selected event is in the 53rd percentile; however, in terms of
peak intensity, it is in the 82nd percentile for the selected year. The derived storm event information is
further explored by the storm event viewer and storm performance summary analysis as described in the
next two sections.
Sorted by Total Precipitation Volume
Sorted by Peak Precipitation Intensity
Selected Interval (Percentile)
• Total Rainfall
|Peak Intensity
Figure 3-24. Precipitation events sorted by total precipitation volume and peak intensity.
3.6.2. Storm Event Viewer
Once the storm events are defined, the user can select a specific storm event to visualize its detailed time
series responses. The storm event viewer plots the time series data for comparison under the following
three conditions: (1) existing development without BMPs, (2) existing development with BMPs, and (3)
before development. The viewer illustrates how BMP performance changes with increasing or decreasing
storm size. Four example storms are shown below for illustration the use of the viewer. Those storm
events were derived using 24 dry hours as the storm separation criterion. Storm event 1 (Figure 3-25) is
the selected interval highlighted previously in the storm classification figures.
In each of the storm viewer graphs, the blue shaded area represents the developed condition without
BMPs (post-developed), the brown line represents the BMP scenario, and the green line represents the
pre-developed condition. Storm event 1 in Figure 3-25 shows excellent BMP performance because the
outflow from the BMP scenario (brown line) is significantly reduced from the post-developed condition,
indicating that the storm runoff was well controlled by the BMP installed. Figure 3-26 shows a slightly
different response. Although the peak flow for this event is similar to that of the first one (slightly lower),
the storm interval 2 generates significantly more outflow than storm event 1. This is because the peak
precipitation came after 2 hours of steady rainfall causing the ground to be saturated and thus more
bypass flow through the BMPs. After the storm has ended, outflow persists for a number of hours
because of the attenuation by BMPs in the drainage area.
3-95
-------
Post-Developed
•The BMP Scenario
•Pre-Developed Condition
Figure 3-25. Storm event 1: August 9, 2001 6:00 p.m. to August 9, 2001 8:00 p.m.
(0.66 in. to 2 wet hours, peak: 0.48 in.).
d -i n
£ n«
o
0)
n n
i
i
i
i
i
•^^
^^^H
120
100
•g 80
Post-Developed
The BMP Scenario
Pre-Developed Condition
Figure 3-26. Storm event 2: May 14, 2001 11:00 a.m. to May 14, 2001 1:00 p.m.
(0.95 in. to 4 wet hours, peak: 0.38 in.).
The dry-interval criterion is applied after the last precipitation. For a 24-hour dry interval criterion, a
storm event begins with the first non-zero precipitation and ends at a minimum of 24 hours after the end
of the last precipitation. The storm duration as is defined allows the runoff to be rightly associated with
the storm that caused it. Storm event 3 in Figure 3-27 consists of two storms because the second storm
began less than 24 hours after the end of the first storm. By definition, the second storm cannot be
evaluated independently of the first storm because antecedent moisture conditions were not stabilized
following the first storm. This approach of defining a storm event will reduce the noise associated with a
3-96
-------
time series response. Had those two storms been considered independently, the second storm in the
interval—which has a slightly lower total precipitation, longer duration, and lower peak precipitation—
would have appeared much worse than storm event 1 because of the antecedent conditions.
It is also interesting to note that only the second storm generated measurable surface runoff for the pre-
developed condition. This further illustrates the impact of antecedent moisture conditions on BMP
performance.
_ 1.4
•S 1.2
•d 1 n
S ns
Q_
0)
0.0
i
i
i
i
^^
i
i
i
" 1
50
45 -I
40
35
30 --
25
20
15
10
5
0
Post-Developed
•The BMP Scenario
•Pre-Developed Condition
A
Figure 3-27. Storm event 3: April 20, 2001 2:00 a.m. to April 21, 2001 5:00 a.m.
(1.03 in. to 12 wet hours, peak: 0.30 in.).
Finally, storm event 4 in Figure 3-28 is the largest total precipitation volume of storms considered. It
consists of a series of intermittent precipitation occurrences scattered over a 2-day period. By grouping
the events into the same storm event, their effects are evaluated together because they are not independent
hydrologically.
3.6.3. Storm Performance Summary
The derived storm events are used to generate a BMP performance summary for the simulation period.
The graph (as shown in Figure 3-29) displays summary comparisons of three scenarios (with and without
BMPs and pre-development condition). The graph shows three series. The first series (circles) shows
percent reduction by storm interval for the BMP scenario relative to the post-developed (with no BMPs)
scenario. The second series (horizontal hashes) shows percent reduction for the pre-developed condition
relative to the post-developed (with no BMPs). The series is plotted for reference purposes. If the BMP
scenario (circle) is at or below the series (hash line), it indicates that BMP performance for that storm
event is not performed at the level of pre-developed conditions. The final series is a bar graph, which
follows the right axis. For each output constituent, the series computes the pollutant contribution (in
percent) of each storm event with respect to the total rainfall series. Figure 3-29 shows a sediment
removal summary for 34 storm events sorted by sediment load on the y-axis and storm event volume on
the x-axis.
3-97
-------
350
300
I Post-Developed
11 A
/\
Figure 3-28. Storm event 4: February 8, 2001 12:00 a.m. to February 9, 2001 4:00 p.m.
(2.34 in. to 25 wet hours, peak: 0.61 in.).
Relative Contribution
-Weighted Average Reduction
Precipitation Event Reductions
100%
Pre-Developed Condition
- 50%
0%
i Precipitation Event Volume (in.)
FT
| Figure 3-29. Sediment removal summary for 34 storm events arranged
by baseline sediment load.
As shown in Figure 3-29, sediment load increases generally with increase in precipitation volume and
smaller storms can be better managed by BMPs than the larger storms. In the example, the weighted
average percent reduction is around 50 percent, which illustrates the influence of the larger load-
producing storms on the long-term average reduction. While the pre-development condition (horizontal
hashes series) would contain close to 100 percent of the current baseline flow and load under the existing
development condition, some large storms produce runoff and loads. For example, the largest load
producing storm (2.34 in., which was also previously featured in Figure 3-28) shows the poorest BMP
performance at 13 percent removal. This storm generated measurable sediment load even under the pre-
developed condition.
3-98
-------
Figure 3-30 shows the summary results of peak flow-reduction performance for this same storm series.
The overall results are similar to that for sediment load reduction. For both sediment load and peak flow
reductions, the 4.41-in. storm event performs among the highest because of its relatively low intensity.
This storm event consists of four individual small storms with relatively high peak flow rates over an 11-
day period, and hence the runoff is largely contained by the BMPs.
Relative Contribution Weighted Average Reduction ^^^—Precipitation Event Reductions Pre-Developed Condition
Precipitation Event Volume (in.)
Figure 3-30. Peak flow reduction summary for 34 storm events arranged
by peak flows.
Additionally, the post-processor is designed to produce a different graph to show a summary comparison
of peak flows or water quality constituents for the same three scenarios. For water quality constituents, it
uses the EMC, which is computed as the total pollutant load divided by the total outflow volume of the
storm. Figure 3-31 shows a peak flow comparison of three scenarios by storm event arranged by
increasing peak flows for the existing development (baseline). Similarly, Figure 3-32 shows the EMCs of
sediment for the corresponding storm events in Figure 3-30. It is interesting to note from Figure 3-32 that
sediment EMC in pre-developed condition can be larger than the other two scenarios because of low
runoff volumes when the watersheds are not yet developed. Depending on the study nature, attaining a
specified EMC might be an important goal to achieve.
Post-Developed
The BMP Scenario
Pre-Developed Condition
OOOOOOOO^OOOOOOOOOOOCM^CM^-
Precipitation Event Volume (in.)
Figure 3-31. Peak flow comparison by storm event arranged
by post-developed peak flows.
3-99
-------
25
20 --
• Post-Developed
•The BMP Scenario
• Pre-Developed Condition
Precipitation Event Volume (in.)
Figure 3-32. EMC of sediment for the corresponding storms.
3.6.4. Cost-effectiveness Curve
The final analysis component of the post-processor is creating a cost-effectiveness curve to facilitate
decision making. The post-processor can generate and display this curve directly from the output only
when the NSGA-II search method is used in the optimization module. It displays the curve one pollutant
constituent at a time; however, the post-processor still allows a user to evaluate the resulting benefit to
other constituents gained from optimizing performance for a single constituent. For example, one can
evaluate how optimizing flow reductions impacts sediment reductions. Each optimization run generates
two files from which the post-processor derives the cost-effectiveness curve: (l)AHSolutions.out, which
contains cost-benefit summaries for each intermediate optimization run, and (2) BestSolutions.out, which
contains cost-benefit summaries for the final population of points that constitutes the optimum frontier.
Figure 3-33 shows an example of a cost-effectiveness curve for sediment load reduction.
55%
50%
= 45%
40%
ig 35%
30%
25%
All Solutions
Cost-Effectiveness Curve
Selected Simulation
$0.0
$05
$1.0
$1 5
$2.0 $2 5
Cost ($ Million)
$3.0
$35
$4.0
$45
Figure 3-33. Example cost-effectiveness curve for sediment load reduction.
3-100
-------
In Figure 3-33, cost (dollars in millions) is plotted on the x-axis, while effectiveness (% sediment load
reduction) is plotted on the y-axis. All the intermediate solutions are plotted as smaller circles, while the
optimum cost-effectiveness curve solutions that form the left- and upper-most boundaries of the search
domain use more pronounced circles.
The data in a cost-effectiveness curve is related to the data generated in all the previously described post-
processor components described thus far. In Figure 3-33, as an example, the most cost-effective solution
(at a cost of around $340,000) for 50% load reduction is highlighted. The cost, performance, and time
series data associated with the BMPs that collectively provide 50% sediment load reduction can be
retrieved from the information presented in Figure 3-29, in the rest of the storm performance graphs
shown in Section 3.6.3, and the storm viewer graphs shown in Section 3.6.2. Using the post-processor,
the user can navigate along points on the cost-effectiveness curve, generate individual runs for individual
solutions on the fly, and toggle between the storm summary/individual storm viewers and the cost-
effectiveness curve to see BMP performance at multiple temporal levels of resolution.
The post-processor can produce two additional types of information from the BestSolutions.out file: cost
distribution by BMP type for a given point on the cost-effectiveness curve and BMP storage distribution
by cost interval. Figure 3-34 shows the data of BMP cost distribution versus sediment reduction
effectiveness on the cost-effectiveness curve and the 50% reduction point shown on Figure 3-32 is
highlighted. For each effectiveness curve, the figure also shows the costs associated with the selected
BMP types. The inset pie chart shows the distribution of cost by BMP type associated with 50 percent
effectiveness. Note that the porous pavement option is not in the pie chart because it is not an optimal
option to achieve the 50% reduction line to the right of the selected point in Figure 3-33. That means that
porous pavement option is likely feasible but not cost-effective for that level of load reduction. However,
it would be a cost-effective choice to achieve 52% or more reduction of sediment load as shown in Figure
3-33.
$3.5
$3.0
$2.5
^ $2.0
-POROUSPAVEMENT
• RAINBARREL
BBIORETENTION
• DRYPOND
^Selected Simulation
$1.5-
$1.0-
$0.5
$0.0
ssssssssssssssssssssssssss
t-a>r^t-cor^coinrsia>cD-'-a>
o> o o
^r in in
Effectiveness (% Reduction)
Figure 3-34. BMP cost distribution by effectiveness for sediment load reduction on the cost-
effectiveness curve.
In the optimization process, the feasible BMP types are compared and screened based on simulation
modeling, cost, and optimization input specification, to develop an order of cost-effective options for
3-101
-------
maximizing their use. Figure 3-34 implies that to achieve sediment removal from runoff, dry ponds are
the most cost-effective option, while porous pavement is the least. In some places, rain barrels are used in
conjunction with bioretention, and the use of those two BMP types in the same space results in some
variability that exhibits itself as noise in the graphs. In this example, there is no clear choice for
maximizing one technique over the other because both rain barrels and bioretention were selected at all
levels along the cost-effectiveness curve. Figure 3-35 provides detail about a point along the curve where
BMP selection is fairly well distributed among the four broad categories of BMPs. However, the point
represents a $2 million investment, which is well above the point of diminishing returns; similar
performance can be achieved for less than $0.5 million.
The cost distribution by BMP types as described represents the total cost for a given type in the drainage
area that meets the goals set at the assessment point. While that cost distribution does not provide
specific information about the spatial locations of actual BMPs, knowing the types of practices associated
with each point along the cost-effectiveness curve provides insight into the reasoning and order of
selecting individual practices.
55%
• All Solutions
O Cost-Effectiveness Cure
O Selected Simulation
25%
$0.0
$05 $1.0 $15 $2.0 $25
Cost ($ Million)
$3.0 $35 $4.0 $45
= $2.0
POROUSPAVEMENT
RAINBARREL
BIORETENTION
DRYPOND
Selected Simulation
cnmococMCOi^cocoooooooot-t-t-CM
CNCNCNCNCNCNCNCNCN
$0.0
CO CO CO
Effectiveness (% Reduction)
Figure 3-35. Example cost-effectiveness and cost-distribution pair.
3-102
-------
SUSTAIN provides another way to look at the preference of BMP types across a cost-effectiveness curve.
Figure 3-36 shows BMP storage distribution by cost and effectiveness interval. The horizontal axis is
cost, while the left vertical axis and line graph are effectiveness in terms of the percent of sediment
reduction. The right vertical scale is the amount of storage (in acre-feet), associated with surface, soil,
and underdrain storages. The graph shows which of the physical treatment processes is responsible for
providing the effectiveness. For instance, the contribution from soil/underdrain storage is relatively small
until around the $0.3 million interval (or 47% removal efficiency). At this point on the cost-effectiveness
curve shown in Figure 3-33, the primary mode of pollutant removal is the use of dry ponds. Subsurface
storage increases as bioretention plays a larger role and when porous pavement is used.
TotalSurfStorVol(acre-ft)
Sff^
(A
Cost ($ Million)
Figure 3-36. BMP storage distribution by cost-effectiveness interval.
3.6.5. Important Considerations and Limitations: Post-Processor
Microsoft Excel Macros Security Setting
The post-processor uses Microsoft Excel 2003 Visual Basic Applications (VBA) to perform summary and
analysis. VBA requires that the user's Microsoft Excel security setting be set to at least medium. The
default setting is high, which will disable all macros without notifying the user. The medium setting will
prompt the user when a spreadsheet is opened that has macros and requires a response of enable or disable
from the user to proceed. The low setting always enables macros without prompting the user for a
decision. The medium setting is recommended because the user has the option of disabling the macros if
the spreadsheet is not from a trusted source.
Functional Limitations
The post-processor is designed to perform analysis in conjunction with cost-effectiveness curves. The
curves can be produced using only the NSGA-II optimization method. While it is possible to use the
post-processor to visualize individual storms and time series data generated by the Scatter Search method,
cost-effectiveness curve evaluation uses the full functionality. Another limitation of the post-processor
with regard to time series evaluation is that it shows only total inflow versus total outflow through a given
node in the network. If an assessment point is also a BMP site, the post-processor does not have the
ability to summarize the complete history of surface and subsurface interactions (e.g., infiltration capacity
exceedence, underdrain outflow, weir and orifice outflow). In addition, the post-processor has the ability
to select and run only those solutions that appear along the cost-effectiveness curve. It cannot be used to
directly select, run, and visualize information associated with a specific point that falls below the cost-
effectiveness curve.
3-103
-------
Chapter 4 Case Studies
To best demonstrate the functionality and help users visualize how they can apply SUSTAIN, two case
studies were developed for metropolitan areas that represent typical settings for applying the framework.
Case studies provide an excellent opportunity to explore the capabilities of SUSTAIN as described in this
document in the context of area! life scenario and demonstrate the application process beginning with
data collection, problem setup, optimization, and results interpretation. The examples were specifically
developed to highlight the core functions of BMP placement and selection and the associated
development of the best solution for the user-defined problem. The case studies also demonstrate how to
apply key functions such as the multi-scale, tiered analysis, and the use of multiple control targets for
optimization.
Ideal case studies for testing SUSTAIN should be consistent with the design requirements and the
placement and selection of BMPs in urban areas. For effective demonstration, the case studies also need
to build on a history of monitoring data collection and analysis. The ideal case studies also have recent
watershed-based studies that have resulted in calibrated and validated models that can be used for setup
and comparison. The case studies were selected using the following criteria:
• Dominantly urban land use
• Water quality management needs
• History of data collection
• Calibrated/validated model application
Two locations fit the criteria and were available for use—the Milwaukee Metropolitan Sewer District
(MMSD) in Milwaukee, Wisconsin (the Oak Creek watershed), and Fairfax County, Virginia (the Little
Rocky Run watershed). For each case study, locally derived data were used to develop the project setup
and analysis. Next, specific problem objectives were identified that highlighted some important functions
of SUSTAIN. For the Oak Creek watershed, the case study focuses on placement and selection of BMPs,
using a single pollutant with a tiered approach, for effectiveness evaluation at several targeted pollutant
reduction goals. It also demonstrates how to integrate external model time series from an existing
watershed model. For the Little Rocky Run watershed, the case study focused on placement and selection
of BMPs on the basis of the evaluation of two concurrent control targets of peak flow and TSS reduction.
It also demonstrates using the internal SWMM model for generating runoff and pollutant load time series.
This chapter describes the project setup, analysis process, and results interpretation for each case study.
Through the demonstrations, the flexibility and potential of the framework is shown. Future
demonstrations and applications throughout the user community will provide valuable experiences and
insights on both the full potential of the existing framework and recommendations for the continued
improvement of SUSTAIN.
4.1. Upper North Branch Oak Creek Watershed
Milwaukee is on the southwestern shore of Lake Michigan and is the largest city in Wisconsin and 23rd
largest (by population) in the nation. Milwaukee is the main cultural and economic center of the seven-
county Greater Milwaukee Area, with an estimated population of 2,014,032 as of 2008. Four major river
4-104
-------
systems (Menomonee River, Kinnickinnic River, Oak Creek, and Root River) drain an area of more than
1,100 square miles that ultimately discharges through the Milwaukee River, in downtown Milwaukee, to
the harbor and Lake Michigan. Upstream areas are predominantly rural and agricultural including dairy
farms and crops such as corn, soybeans, and alfalfa. Nearer to the metropolitan area are suburban areas
and commuter communities. The downtown region is densely urbanized and drained by a combined
sewer system. Over the past 20 years the city and the Milwaukee Metropolitan Sewerage District
(MMSD) have taken significant steps to address CSOs by constructing a major storage tunnel. The 7.1-
mile long Northwest Side Relief Sewer is a deep tunnel that can hold up to 89 million gallons of
wastewater. The region has also led the way with various green programs such as the Greenseams
program and been an advocate of innovative stormwater management techniques such as rain gardens,
rain barrels, and downspout disconnections. Major environmental issues in the region include increased
loadings of sediment, nutrients, and pathogens associated with urbanization and agriculture, and in the
urban center CSOs.
Regional water quality and environmental protection continue to be a priority in the area. Recognizing
this need, the MMSD has led a long-range planning effort to identify improvements needed for its
facilities to accommodate growth and protect water quality through the year 2020. This effort is known
as the MMSD 2020 Facility Plan. A related planning effort, known as the Regional Water Quality
Management Plan Update (RWQMPU), was also conducted in coordination with the MMSD by the
Southeastern Wisconsin Regional Planning Commission (SEWRPC) to update the regional water quality
management plan for all the major watersheds draining through the greater Milwaukee area (SEWRPC
2007).
As part of the planning process, a comprehensive suite of models was applied to the four major tributaries
to assess current conditions and evaluate a range of management scenarios. Those models allowed
planners to evaluate the potential water quality benefits of a range of implementation measures, including
facility improvements and urban, suburban, and rural stormwater BMPs. The watershed modeling
component was developed using LSPC (Tetra Tech and USEPA 2002). Hydrology and water quality
models were developed and calibrated to provide a basis for modeling current conditions versus potential
implementation scenarios. The modeling application was developed and tested using an extensive record
of precipitation, flow, and water quality sampling. Calibration of the watershed models followed a
sequential, hierarchical process that began with hydrology, followed by sediment erosion and transport,
and, finally, calibration of chemical water quality.
The Kinnickinnic River, Menomonee River, Milwaukee River, and Oak Creek watershed models were
linked to a model of the Lake Michigan estuary so that the impact of upstream water quantity and quality
could be simulated. During the RWQMPU study, several year 2020-projected management scenarios to
reduce nutrient, bacteria, and sediment loading were evaluated (SEWRPC 2007)
The history of environmental management, watershed planning, technical analyses, monitoring, and the
availability of calibrated models made this location ideal for a case study application of SUSTAIN. A
representative subwatershed, Oak Creek, was selected for examination for the case study, and TSS was
selected as the pollutant for optimization and analysis. This case study examined the use of SUSTAIN for
tiered analysis and BMP optimization for a single parameter.
4.1.1. Project Setting
The Oak Creek watershed covers parts of the cities of Milwaukee, South Milwaukee, Cudahy, Franklin,
Greenfield, and Oak Creek and encompasses approximately 27 square miles. Table 4-1 provides a listing
of the main characteristics of the watershed.
4-105
-------
• USGS Flow Gages
Roads
/\/ Streams
Major Roads
Lakes
Oak Creek Watershed
Figure 4-1. Oak Creek Watershed and Upper North Branch Oak Creek.
Table 4-1. Watershed Characteristics
Watershed drainage area (square miles)
Miles of streams
Miles of streams listed as outstanding or exceptional
resource waters
Miles of streams on impaired waters list
General threats to stream water quality
Number of named lakes
Number of dams
Threats to lake water quality
26.2
21.2
0
13
Urban runoff
Toxics
Hydrological modification
Stream bank erosion
1
1
Nutrient enrichment
Sedimentation
4-106
-------
The watershed has three major streams—Mitchell Field Drainage Ditch, North Branch of Oak Creek, and
Oak Creek. The longest stream of the three, Oak Creek, has a perennial length of approximately 13.1
miles. North Branch of Oak Creek and Mitchell Field Drainage Ditch, which are tributaries to Oak
Creek, have perennial lengths of approximately 5.8 and 2.4 miles, respectively. There is one reservoir in
the watershed with a history of siltation and algal blooms.
Water quality in Oak Creek is degraded in part because of elevated sediment, sediment-associated total
phosphorus, and fecal coliform loads. Because sediment data were closely correlated with nutrients
specifically and peak discharges in general, TSS was selected as the pollutant on which to focus the case
study analysis.
Pollution sources in the watershed include the following:
• Stormwater runoff from impervious urban land
• Runoff from agricultural lands
• Eroding streambanks and sedimentation
• Wildlife, pets, and residential lawns
• Erosion from construction sites
• Sanitary sewer overflows and industrial discharges
• Leaking underground storage tanks, landfills, runoff from salvage yards
The Upper North Branch Oak Creek (UNB) area, shown in the upper-left corner of Figure 4-1, was
selected for further evaluation in the case study. The UNB is a headwater area of concern because of poor
habitat, elevated sediment load, and sediment-associated total phosphorus and fecal coliform loads. The
UNB has a drainage area of approximately 4.2 square miles with mixed land uses. The dominant
residential and commercial land uses compose 68 percent of the total area. Table 4-2 lists the land use
distribution of the study area. The Oak Creek watershed has discharges from 17 industrial facilities as
well as sanitary sewer overflows.
Table 4-2. Upper North Branch Oak Creek Land Use Distribution
Land Use Type
Water/wetland
Forest
Pasture/hay
Crop
Developed open space
Commercial
High density residential
Low density residential
Total
Area
(acre)
41.0
323.7
88.5
168.4
263.4
342.0
684.6
766.3
2,677.7
Area Percentage
(%)
2
12
3
6
10
13
26
29
100
The UNB watershed provides an excellent setting for demonstrating how SUSTAIN can be applied in the
implementation and tracking aspects of a watershed planning process. For this application, a single
objective of TSS load reduction was selected. In terms of SUSTAIN functionality, this case study
demonstrates the following:
4-107
-------
• An external linkage to a previously modeled set of land use time series outputs
• The use of aggregated BMPs
• The tiered optimization approach
4.1.2. Data Collection and Analysis
The SUSTAIN case study application began with a review of watershed characteristics and compilation of
related spatial and temporal model outputs. The next steps involved processing watershed model output
data into the required input data format and conducting model setup. Model setup included land use
reclassification and time series mapping, delineating potential BMP drainage areas, and selecting
assessment points. The established model configuration was then used to evaluate different treatment
scenarios to measure relative impacts, perform optimization analysis, and, finally, interpret and present
results. Figure 4-2 is a roadmap of objectives and problem formulation for the case study.
Case Study Objectives
Question to be
answered:
What are cost-effective
strategies for reducing
nonpoint source loads
from a mid-sized urban
watershed using
centralized BMPs,
distributed BMPs, or a
combination of both?
Control Target:
•TSS
Data Collection & Analysis
• Study area review
• CIS data: land use, stream, DEM, BMP sites.
• Watershed and BMP information/data
• Compile monitoring data (calibration/validation)
Project Setup
• BMP representation: placement, configuration, and cost
• LAND/WATERSHED Representation
• Routing network
• Assessment point(s)
• Test system application (externally calibrated model)
Put Optimization Processor to Work!
• Select decision variables (BMP dimensions)
• Select assessment points (BMP/Outlet locations)
• Select evaluation factors, control targets (end points)
Results Analysis and Representation (Post-Processor)
• Cost-effectiveness curve
• Alternative solutions
• Optimum BMP selection (spatial, by cost interval)
Figure 4-2. Upper North Branch Oak Creek Watershed case study road map.
4.1.3. Project Setup
Available sources of information including land use maps, local pollutant source characterization reports,
and water quality assessment reports for the larger Oak Creek watershed, were used to set up the
4-108
-------
SUSTAIN project and identify potential locations for management practices. The setup for management
analysis also required the selection of a typical precipitation year for use in comparing alternatives and
assessing downstream impacts. From the analysis of precipitation data, the hydrologic year 2001
(10/1/2000-9/30/2001) was determined to represent the average conditions in the watershed. Total
rainfall depth for the year was close to the long-term average for the area. At the same time, both
precipitation depth and intensity distribution during that year were relatively close to the long-term
statistical average distribution. Figure 4-3 presents a graph of average annual precipitation at Milwaukee
Airport for water years 1988-2002.
—•—Milwaukee Mitchell Airport (WI5479)
Figure 4-3. Average annual precipitation volume at Milwaukee Airport for water years 1988-2002.
Although the water year 2001 was not the closest year to the average annual value over the 15 -yr
evaluation period, it was the most recent average year among the set and had a typical rainfall magnitude
and intensity distribution. Figure 4-4 shows rainfall volume and intensity distribution for wet intervals
occurring in water year 2001. In the figure, the volume and intensity percentile ranges are based on the
record of storms occurring over the 15-yr period. A year with a perfect typical distribution would have
the same number of precipitation intervals in each bin.
14
Rainfall Intensity
•Rainfall Volume
•Rainfall Volume
Figure 4-4. Rainfall volume and intensity wet-interval distribution for water year 2001.
4-109
-------
The previously developed model was successfully calibrated and validated for the Oak Creek watershed,
inclusive of the land use types considered in this case study (SEWRPC 2007). Using SUSTAINS external
modeling function, these previously developed land use time series were imported directly from the
calibrated model. Figure 4-5 and Figure 4-6 show the modeled versus observed flow at the Oak Creek
U.S. Geological Survey (USGS) gage for the selected water year 2001. Figure 4-7 and Figure 4-8 are
examples of modeled versus observed TSS for water years 1995-2001 and water year 2001, respectively.
^M Avg Monthly Rainfall (in.)
Avg Observed Flow (10/1/2000 to 9/30/2001 )
Avg Modeled Flow (Same Period)
900
800
700
500
400
200 - - -
100 -
. J _ _ _ _ L
_ I I _
-iivtl
- 2
4 J
- 6 £
10 ro
Q
- 12
14
0
Oct-00 Nov-00 Dec-00 Jan-01 Feb-01 Mar-01 Apr-01 May-01 Jun-01 Jul-01 Aug-01 Sep-01
Date
Figure 4-5. Comparison of daily flow at Model Outlet 58 with USGS 04087204 at South Milwaukee.
100
i
I
80
1 60
0
T3
| 40
0
D)
i 20
^
0
Avg Flow (10/1/2000 to 9/30/2001)
Line of Equal Value
Best-Fit Line
y = 0.9228)^ + 3.4356
20 40 60 80
Average Observed Flow (cfs)
100
Avg Monthly Rainfall (in.)
-Avg Observed Flow (10/1/2000 to 9/30/2001)
•Avg Modeled Flow (Same Period)
100
10 11 12 1 2 3 4 5 6 7 8 9
Figure 4-6. Comparison of monthly flow at Model outlet 58 with USGS 04087204 at South
Milwaukee.
The land use time series from the calibrated watershed model were exported as unit-area hydrographs and
pollutographs for each modeled land use type. Table 4-3 is a summary of annual average outflow and
TSS load by modeled land use in Oak Creek on a unit-area basis.
4-110
-------
600 -,
500
400
300
200
100
— Modeled (Reach 58) •
Observed (OAK Gage OC-05)
li , Ml
tiiii *
i '
•
Jl Am.
ll
;
• \
>! I
•
I]
A i
.1
lkli
O-94 M-95 D-95 J-96 F-97 S-97 A-98 N-98 J-99 J-00 A-00 M-01
Figure 4-7. Modeled vs. observed TSS (mg/L) at Oak Creek gage OC-05, water years 1995-2001.
Modeled (Reach 58)
Observed (OAK Gage OC-05)
O-OO N-00
M-01 A-01
Figure 4-8. Modeled vs. observed TSS (mg/L) at Oak Creek gage OC-05, water year 2001.
Table 4-3. Summary of Modeled Annual Average Outflow and TSS Load in Oak Creek Watershed
Land Use ID
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Land Use Name
GRASS_B
GRASS_C
GRASS_D
FOREST
CROP_B
CROP C
CROP D
PASTURE_B
PASTURE_C
PASTURE_D
WETLAND
ULTRA LOW
RESIDENTIAL
COMMERCIAL
INDUSTRIAL
GOVTJNSTIT
TRANS_FREE
Area
(acres)
1,183
7,782
231
1,087
380
1,395
127
156
693
110
1,270
58
608
2,095
403
106
229
Flow
(acre-in./yr)
11.8
11.2
11.1
11.4
12.5
11.7
11.5
13.7
12.6
12.3
9.2
22.3
22.3
22.3
22.3
22.3
22.3
TSS
(Ib/acre/yr)
117
120
156
105
460
1,253
2,278
30
118
263
303
527
511
784
913
529
949
4-111
-------
4.1.4. Optimization and Results Analysis
For the purposes of optimization, the study area was divided into 11 subwatersheds. Figure 4-9 shows the
land use distribution in the study area, overlain with the modeled subwatersheds.
Legend
— Streams
1 Subwatershed
Water
Developed Open Space
Low Intensity Residential
High Intensity Residential
High Intensity Commercial
Bare Rock/Sand/Clay
Deciduous Forest
Evergreen Forest
Mixed Forest
Evergreen Shrubland
Gra sslan d/H erbaceous
Pasture/Hay
Row Crops
Wetlands
Emergent Herbaceous Wetlands
Upper North Branch Oak Creek Landuse
and Subwatershed Delineation
Figure 4-9. Land use distribution in the modeled subwatersheds.
4-112
-------
Figure 4-10 shows how the subwatersheds were classified into six groups on the basis of prevailing land
uses, suitable management practices, and location in the watershed. Groups 1 through 4 are labeled in the
map as A, B, C, and D. For those areas, BMP placement was applied to urban land uses. Subwatershed 6
was singled out for special consideration as the fifth group because it is not part of the drainage area of
the other larger subwatersheds. Because of its small size, BMPs in subwatershed 6 were optimized
directly during the tier-2 optimization. The remaining subwatersheds (2, 3, 5, 7, and 9) were mostly non-
urban and were assumed to need no additional management and were not evaluated for BMP placement
because the objective of this analysis was to target TSS loads from urban areas.
L/ ^r1" ^^L i-ix
Figure 4-10. Subwatershed grouping for two-tiered optimization.
BMP Representation
For this case study, both distributed and centralized BMP options were evaluated. Distributed BMP
options include rain barrels, bioretention, and porous pavement. The centralized BMP option is an
infiltration basin.
To improve computational efficiency, the aggregate BMP option (as described in Section 3.3.3), was
used. An aggregate BMP consists of a series of process-based optional components, including on-site
interception, on-site treatment, routing attenuation, and regional storage/treatment. The aggregate BMP
component evaluates storage and infiltration characteristics from multiple BMPs simultaneously without
explicit recognition of their spatial distribution and routing characteristics within the selected watershed.
For this case study example, the aggregate BMP had four component BMPs—rain barrels (on-site
interception), bioretention (on-site treatment), porous pavement (on-site treatment), and dry infiltration
4-113
-------
basin (regional treatment). Figure 4-11 is a schematic diagram of aggregate BMP components, drainage
areas to BMPs, and BMP-to-BMP routing networks. Figure 4-12 lists the area percentage of each land
use that contributes to each of the aggregate BMP components.
Low-Density-Residential
High-Density-Residential
TT = II I :
-••••••••• •• •••• !•••••!
^^1 Rain Barrel [^^
r. ™—f
^^^^^^^^^^^^^^^ Bioretention
I | Impervious
^] Pervious
Dry Infiltration Basin
Outlet
Figure 4-11. Aggregate BMP schematic.
* Aggregate BMP Landuse Distribution
HPIx
Select Sutawatershed |j T
1 1
Land Use Distribution (%)
k
<
Landuse Group/Info Type lArea (ac.) RamBarreH (%) | BioRelentionBasiiJ PorousPavementl (%)
BMPID
Category
BMPType
1 2 |3
water/wetland Pervious 1 3. 93
Open Impervious
Open Pervious
COM Impervious
COM Pervious
Hiqh-Density-Residential Impe
Hiqh-Density-Residential Pervi
Low-Densitv'-Residential lmper\
Low-Dens itv'-Re s i d e nti al Pe rvi c
Downstream ID
19.5E
4.89
0.22
0.15
1.B
4.81
9.32
83.92
On-Site Interception
RamBarrel
.00
.00
.00
.00
.00
30.00
.00
40.00
.00
2
On-Site Treatment
BioRetentionBasin
.00
.00
.00
65.00
.00
70.00
.00
60.00
.00
4
On-Site Treatment
PorousPavement
.00
.00
.00
35.00
.00
.00
.00
.00
.00
2
DryPond! (%) I Outlet (%)
4
0
Regional Storage Outlet
DryPond
.00
.00
.00
.00
50.00
.00
30.00
.00
.00
0
Outlet
100.00
100.00
100.00
.00
50.00
.00
70.00
.00
100.00
0
>
Save
Close
Figure 4-12. Aggregate BMP land use distribution.
As shown in Figure 4-11, the rain barrel component collects runoff from rooftops (as part of the
impervious surfaces) in low- and high-density residential areas. Bypass from the rain barrel is routed to
bioretention, together with runoff from the non-rooftop impervious surfaces in low- and high-density
residential areas and impervious areas that could be subjected to the porous pavement option in
commercial areas. It was assumed that the parking lot portions of the commercial impervious area could
be converted to porous pavement as a treatment option. In addition to the distributed BMPs (i.e., rain
barrel, bioretention, and porous pavement), a centralized facility—dry infiltration basin—was also a
candidate for consideration. In addition to outflow from the bioretention component, the centralized
facility receives outflow from part of the pervious areas in high-density residential and commercial land
4-114
-------
uses. Outflow from the centralized facility is routed to the watershed outlet. The other areas that are not
routed to any aggregate BMP components are assigned to drain directly to the watershed outlet.
To run the optimization analysis, the user must define decision variables that will be used to explore the
various possible BMP configurations. For this analysis, the decision variables are the number of fixed-
size units of the distributed BMP types and surface area for the centralized BMP type (dry infiltration
basin). Because the decision variable values range from zero to a maximum number depending on the
drainage area, it is possible for one component in the treatment train to never be selected. During the
optimization scenario, if the BMP number or size value of zero is selected, that point will act as a transfer
node in the network (i.e., inflow = outflow), and the associated cost that is a function of the number of
BMPs or BMP surface area will not set to zero.
The physical configuration parameters, infiltration, and water quality simulation parameters of each BMP
components are listed in Table 4-4.
Table 4-4 BMP Parameters
Parameter
Rain
Barrel
Bioretention
Porous
Pavement
Dry Infiltration
Basin
Physical Configuration
Unit size
Design drainage area (acre)
Substrate depth (ft)
Underdrain depth (ft)
Ponding depth (ft)
28ft3
0.02
N/A
N/A
4
60ft2
0.1
2.5
1
0.5
0. 1 acre
0.1
2
1
0.2
Max: 3,000 ft2
N/A
1
1
Orifice height: 0.5
Weir height: 4
Infiltration*
Substrate layer porosity
Substrate layer field capacity
Substrate layer wilting point
Underdrain gravel layer porosity
Vegetative parameter, A
Underdrain background infiltration rate**
(in./hr),/c
Media final constant infiltration rate (in./hr),/c
N/A
N/A
N/A
N/A
N/A
N/A
N/A
0.5
0.3
0.15
0.5
0.6
0.5
3
0.5
0.2
0.05
0.5
1
0.5
3
0.4
0.3
0.15
0.5
0.6
0.5
1
Water Quality***
TSS 1st order decay rate (I/day), k
TSS filtration removal rate, Prem (%)
0.2
N/A
0.2
85
0.2
60
0.2
85
' Source: Tetra Tech 2001.
** Soil map shows the majority background soil has hydrologic soil group of C; therefore, 0.5 in./hr background infiltration rate is
assumed.
*** Based on calibration using University of Maryland monitoring data (Tetra Tech 2003).
BMP Cost
Cost estimation is a critical component because the optimization process needs the data to evaluate and
compare cost-effectiveness of one scenario relative to the others. Table 4-5 presents the cost functions for
the BMP types used in this case study (rain barrels, bioretention, porous pavement, and a dry infiltration
basin).
4-115
-------
Table 4-5. BMP Cost Functions for the Case Study
BMP Type
Rain Barrel
Bioretention
Porous Pavement
Dry Infiltration Basin
Cost Function
$15/ft3
$67 ft3
$12/ft2
Cost = 12.4 x J/0760
V is the volume of the basin in ft3
Reference
BMP Cost database
BMP Cost database
BMP Cost database
CASQA Stormwater BMP
Handbook (CASQA 2003)
Optimization
A two-tiered optimization approach was applied to this case study. As previously described, the same set
of management actions (aggregate BMP) was applied to subwatersheds A, B, C, and D. Tier-1
optimization analyses for these areas result in a unique, cost-effectiveness curve for each. A second
round of optimization (tier-2) is then performed and assessed at the most downstream outlet of the
watershed. Tier-2 decision variables include discrete points (representing a combined management
options) along the tier-1 cost-effectiveness curves and BMP options for subwatershed 6. For
subwatershed 6, the optimization is allowed to select distributed or centralized BMPs, or a combination of
both. The optimization objective is to maximize TSS load reduction at the watershed outlet and minimize
the cost of implementation.
Tier -1 Optimization for Subwatershed A
This section describes the process for developing tier-l cost-effectiveness curves using subwatershed A as
an example. Figure 4-13 shows the BMP placement and routing network for tier-1 subwatershed A. The
total area of subwatershed A is 515 acres. The subwatershed is further subdivided into six subareas, with
sizes ranging from 36 to 138 acres. This subwatershed was intentionally subdivided into subwatersheds
of about 100 acres in size because preliminary testing has shown that given an hourly time step, the
aggregate BMP approach closely resembles a fully articulated network in areas of around 100 acres (see
Section 3.3.6). The aggregate BMP treatment train previously described in the BMP Representation
section is applied to each of the six subareas. The watershed outlet (J4 in Figure 4-13) is designated as
the assessment point, and TSS annual load is used as the evaluation factor.
Aggregate BMP
Junction
Assessment
Figure 4-13. Aggregate BMP arrangement in tier-1 subwatershed A.
4-116
-------
Tier-1 Optimization Results
The full set of results of all tier-1 optimization runs are summarized in Figure 4-14 as small gray circles
with near-optimal solutions in larger orange circles along the upper-left frontier. Each near-optimal
solution represents one combination of decision variables, including the number of rain barrels,
bioretention units, area of porous pavement, and size of infiltration basin for each subarea of
subwatershed A. Figure 4-14 also highlights five solutions as green circles (numbered 1 to 5) selected to
be the tier-2 search domain and their associated costs and TSS load reductions are summarized in Table
4-6. These specific solutions were selected to account for the full range of achievable TSS reduction.
30%
• All Solutions
O Cost-Effectiveness Curv
Selected Solutions
$0.5
$1.0
$1.5
$2.0
$25
$30
$35
$4.0
$4.5
$5.0
Cost($ Million)
Figure 4-14. Tier-1 cost-effectiveness curve for subwatershed A with the selected solutions.
Table 4-6. Selected Tier-1 Solutions on the Cost-Effectiveness Curve
Solution ID
1
2
3
4
5
Cost
($ million)
0.90
1.37
1.84
2.83
4.09
TSS Load Reduction
(%)
27.3
29.9
32.1
32.7
33.4
Figure 4-15 shows the BMP cost distribution for all the near-optimal solutions on the cost-effectiveness
curve with respect to the cost of four BMP options (rain barrel, bioretention, porous pavement, and dry
infiltration basin). The graph also reveals the BMP selection preference with increase in total cost. The
figure shows that the dry pond is the most cost-effective choice and, thus, is the first option to be fully
used throughout the range. The next choices are bioretention and rain barrel. Porous pavement was the
last to be considered suggesting that it is the least cost-effective BMP type among the four types
considered in this case study for TSS load reduction.
The same procedure was applied to develop the tier-1 cost-effectiveness curves for Subwatersheds B, C,
and D. The four curves, together with that for subwatershed 6 and the remaining untreated areas, become
the input decision variables for the tier-2 optimization analysis.
4-117
-------
POROUSPAVEMENT
RAINBARREL
BIORETENTION
DRYPOND
Effectiveness (% Reduction)
Figure 4-15. Composition of best solutions on tier-1 cost-effectiveness curve for subwatershed A.
Tier-2 Optimization
The tier-2 optimation analysis is assessed at the outlet of the Oak Creek watershed. The decision
variables include the selected points along the four tier-1 cost-effectiveness curves and BMP selection for
subwatershed 6. The untreated areas are included as a fixed boundary condition of TSS load and not
involved in the optimization runs. Since runoff from the untreated areas is expected to be relatively clean,
the water would boost the assimilative capacity of streams in and downstream of the watershed. The
inclusion of this relatively clean water will influence the optimization results.
Figure 4-16 shows a schematic of the tier-2 analysis network with the objective to develop the cost-
effectiveness curve that meents TSS reduction load goals at the outlet. The analysis involves stream
routing. The four most downstream segments (J3-J4, J4-J10, J10-J13, and J6-J12) are simulated as
trapezoidal channels, while shorter segments are assumed dummy conduits considering the travel time of
those segments are likely less than an hour. This provides the benefit of reducing computation time.
Tier-2 Optimization Results
The optimization analysis resulted in the tier-2 cost-effectiveness curve shown in Figure 4-17. The cost-
effectiveness curve suggests that the maximum achievable TSS load reduction, given the objectives and
constraints associated with the study, is approximately 30 percent. To further examine the cost-effective
solutions, three selected solutions are highlighted in Table 4-7.
Table 4-7 shows that for the lowest tier-2 reduction target of 16 percent, only the most downstream
subwatersheds C and D were treated. In addition, rain gardens were preferred in subwatershed 6. The
most cost-effective solution that achieves a 16 percent load reduction at the outlet costs $1.85 million. At
the next selected tier-2 target of 23 percent, the cost would increase to $2.75 million because
subwatershed B is added to the list. Finally, for a 30 percent load reduction, all four tier-1 subwatersheds
were selected for treatment.
4-118
-------
Figure 4-16. Schematic of tier-2 analysis network.
g
c
1 load reduc
i
c
(8
£
10
0
30%, $3.64 million
^^V"^
l^J9eeS89^^. ^0^ 8oOQ ^>0«W o o9oo a
23%, $2.75 million
^ y
^m 9»«8^© 90 oo o0 oo „ 0
16%, $1.85 million
O Best Solutions
• Selected Solutions
4 2.4 4.4 6.4 8.4 10.4 12
Cost ($ million)
.4
Figure 4-17. Tier-2 cost-effectiveness curve.
4-119
-------
Table 4-7. Selected Tier-2 Best Solutions
Tier-2 TSS Load % Reduction Target
Total Cost ($million)
Tier-1 Subwatersheds
TSS Load % Reduction Allocation
Subwatershed 6
A
B
C
D
Distributed Rain Barrels (#)
Distributed Rain Gardens (ft2)
Distributed Porous Pavement (acre)
Dry Infiltration basin (acre)
16
1.85
0
0
27.3
27.3
0
2,100
0
0
23
2.75
0
27.3
27.3
27.3
0
2,100
0
0
30
3.64
27.3
27.3
27.3
27.3
0
2,100
0
0
A closer examination of the three tier-2 cost-effective solutions suggests that it is more cost-effective to
provide load reductions at lower subwatersheds than at upper subwatersheds. This is because stream
segments through natural processes will inherently provide some load reduction benefit from settling or
transport routing. Hence, placing BMPs at upper reaches to reduce loads is not as cost-effective as at
downstream reaches since the same investment in urban areas close to the assessment point at the
watershed outlet would yield a greater reduction. In other words, why spend money to reduce a load
when it can be reduced for free in stream transport? It is assumed that the streambanks are stable and
erosion does not take place. Had streambank erosion been factored in as a problem, the types of BMPs
selected upstream to control peak discharge or pollutant load might have resulted in a completely
different set of solutions. Note that SUSTAIN does not include a streambank erosion simulation
mechanism. This case study example further highlights the importance of careful formulation of the
problem and understanding of the associated implications and findings of the results.
It is also interesting to observe that although higher levels of treatment options were available from
among tier-1 subwatersheds (as shown in options 2 to 5 in Table 4-6), none of them were included in the
the tier-2 optimal solutions due to unfavorably higher marginal costs. Furthermore, the load reduction in
stream segments mitigated the need for additional treatment.
It is important that readers examine the specific conclusions drawn from this exercise in the context of
this specific problem formulation and the imposed assumptions and constraints. Had the same problem
been formulated slightly differently, it might have resulted in a completely different set of solutions.
When working with a relatively large watershed, hydrological and water quality responses to various
BMP treatment options can be complex and result in several competing effects that could cloud intuitive
interpretation. SUSTAIN provides a framework for quickly exploring the response to multiple problem
formulations and examining the impact of the inherent cost and BMP performance assumptions. Using
the framework to examine the responses at several intermediate nodes in a complex network would help
in selecting the best solutions.
4.1.5. Summary
This case study has demonstrated: (1) how SUSTAIN can be linked to an existing watershed model, (2)
using aggregate BMPs, and (3) using a tiered watershed optimization approach for identifying the cost
effectiveness of BMP solutions. In summary, the aggregate BMP approach is a simplified approach for
preserving the physically based response of distributed BMP types that, preliminary tests have shown can
4-120
-------
reduce computational effort without compromising accuracy when used appropriately. Also, the tiered
approach is an efficient way of disaggregating an optimization problem for large watersheds into
manageable units for decision making BMP placement. Finally, the results-interpretation process further
highlights the fact that SUSTAIN is a tool, not an advisor. Users must interpret model results and weigh
them in light of user-specified assumptions, problem formulation, and optimization goals. Its application
must be preceded by an intimate understanding of the study area and the influential factors affecting
decision making for stormwater management.
4.2. Little Rocky Run Watershed
Fairfax County is in Northern Virginia and is part of the Greater Washington, DC, metro area. The
county has undergone significant growth over the years and is considered to be almost completely built-
out according to the county's comprehensive development plan (Fairfax County 2007). In many areas,
stormwater management involves retrofitting existing facilities and incorporating new stormwater
treatment for older areas in which treatment has not previously existed. To assist with planning efforts,
the county is developing watershed management plans to assess watershed conditions, identify
stormwater management needs, and prioritize future storm water/BMP implementation efforts. The
Fairfax County Department of Public Works and Environmental Services (DPWES) previously developed
watershed management plans for 11 of the 30 watersheds in the county and began developing watershed
management plans for the remaining 19 watersheds in 2006. The detailed plans incorporate an
assessment of current and future watershed conditions and problem areas, identify the county's structural
and nonstructural needs, analyze stormwater management and BMP options, and prioritize recommended
stormwater projects on the basis of measurable goals. County staff are using watershed plans and
technical tools to address stormwater planning needs, MS4 and TMDL requirements, Chesapeake Bay
nutrient- and sediment- reduction goals, and local stakeholder concerns. Hydrologic, hydraulic, and water
quality models (SWMM, HEC-RAS, and other systems) were developed to assess watershed conditions
and quantify the benefits of various stormwater management practices. The models provide the
foundation for the watershed planning process. Little Rocky Run is included in the group of watershed
management plans being developed by Fairfax County.
SUSTAIN was used to evaluate the influence of various BMP scenarios for both a flood control target (10-
yr design storm peak flow rate) and a water quality target [total phosphorus (TP) reduction of 40 percent].
In terms of SUSTAIN functionality, this case study demonstrates the following:
• The use of the internal SWMM land use time series generation option
• Multiobjective optimization formulation for flood control and water quality objectives
4.2.1. Project Setting
In the western portion of the county (Figure 4-18), the Little Rocky Run watershed is approximately 7
square miles in area, with an average imperviousness of 30 percent. Land use is primarily residential (61
percent), followed by open space (19 percent), and areas occupied by institutional and public facilities (18
percent). The watershed includes approximately 20 miles of stream channel and 13 regional ponds. The
watershed was selected for development of a SUSTAIN case study because of several factors, including
the availability of detailed watershed planning data and a recently completed and calibrated SWMM
design storm model. Flood control in peak flow reduction and improvements in sediment and nutrient
levels are two primary stormwater management goals for this recently developed watershed.
4-121
-------
^^^^•^^^r**
Figure 4-18. Little Rocky Run watershed in Fairfax County, Virginia.
4.2.2. Data Collection and Analysis
This SUSTAIN case study application began with a review of watershed characteristics followed by
assembling information related to the existing watershed modeling activities for the area, including the
simulated time series outputs and model parameters used in the existing watershed models. Available
sources of information included land use maps, pollutant source information, existing BMP information,
local BMP cost data, and the existing Little Rocky Run SWMM model. Figure 4-19 shows the selected
watershed for the case study application—a mixed land use area in the headwaters of the Little Rocky
Run watershed.
4.2.3. Project Setup
The next step was to transfer the compiled model parameters into the LAND module (based on SWMM5)
in SUSTAIN for model setup. Model setup also involved land use reclassification, assigning flow and
water quality time series to corresponding land use types, delineation of BMP tributary areas, and
selection of assessment points. The established model configuration was then used to evaluate different
treatment scenarios to measure relative impacts, perform optimization analysis, and finally, interpret and
present results. Figure 4-20 shows a roadmap of this case study and its objectives.
4-122
-------
— Stream
•LRR SWMM Calibration Location
CH LRR Watershed
- LRR SWMM Subbasins
OStudy Area Subbasins
Little Rocky Run Watershed
SWMM Calibration Locations and Study Area Location Map
0 0.125 0.25
Figure 4-19. Selected study area in Little Rocky Run watershed.
Case Study Objectives
Question to be
answered:
What is a cost-effective
management strategy
that simultaneously
provides flood control
and water quality benefit
for a small urban
watershed?
Control Targets:
• Peak flow reduction
• Total Phosphorus load
reduction
Data Collection & Processing
• CIS data: land use, stream, DEM, BMP sites.
• Existing SWMM model
• Potential applicable BMP types and configurations
• Compile monitoring data (cal bration/validation)
Scenario Evaluation
• Establish desired spatial scale for evaluation
• Integrate existing BMP implementation plans
• Assess unique land use conditions
• Build routing network
Put Optimization Processor to Work!
• Select decision variables (BMP dimensions)
• Select assessment points (BMP/Outlet locations)
• Select evaluation factors, control targets (end points)
Results Analysis and Representation (Post-Processor)
• Alternative solutions
• Optimum BMP selection (spatial, by cost interval)
Figure 4-20. Little Rocky Run case study road map.
4-123
-------
Existing SWMM Modeling
A comprehensive SWMM (version 5) 2-, 10-, and 100-yr design storm model for the Little Rocky Run
watershed was developed for watershed planning purposes (Tetra Tech 2009). The model calibration was
performed at five locations following the county's approved methodology (Tetra Tech 2008). These
locations were selected from a consideration of land use, soils, slope, and major confluences along the
mainstem. Calibration was performed beginning with the most upstream location and proceeding
downstream to the watershed outlet. At each calibration location, SWMM model predictions were
compared to the peak flows estimated by the Anderson and USGS methods (Anderson 1970; USGS
2001), which are documented in the county's SWMM calibration methods (Tetra Tech 2008).
The Anderson approach was developed for estimating peak runoff rates from watersheds with varying
degrees of development and design storms with recurrence intervals of 2.33 years to 100 years. The
regression relationship in the approach was derived from analysis of flood data at eighty-one watersheds
in and around Washington, DC with sizes ranging from 0.00034 mi2 to 570 mi2. The imperviousness of
these watersheds varies from less than 1 percent to 100 percent, and the watershed conditions from
natural to fully developed (Anderson 1970).
In the Anderson approach, the peak flow rate from a watershed is expressed as a function of five
independent variables, including the watershed area, length, slope, percentage of imperviousness, and the
type of drainage collector system.
Alternatively, the USGS (2001) has developed regression equations, using urban runoff data from 199
basins in 56 cities and 31 states, to relate urban peak flow rates with basin characteristics for various
recurrence intervals at different regions. The regression equations were built into a computer program
called National Flood Frequency (NFF, Version 3.2) for peak flow predictions from recurrence interval
storms ranging from 2 to 500 years, along with the corresponding confidence interval of the estimations.
The SWMM calibration was performed using the 2-, 10-, and 100-yr 24-hour design storms. The
watershed GIS data were used to generate the input data parameters, summarized in Table 4-8, for the
Anderson approach and the USGS method. Model calibration locations are shown in Figure 4-19.
Table 4-8. List of Input Parameters for Both the Anderson and the USGS Methods
SWMM
Junction
3
13
18
28
32
Area
(A)
(mi2)
0.56
2.19
3.54
6.19
7.18
Index of
Slope
(S)
(ft/mi)
53.93
49.72
39.48
42.71
32.67
Longest
Flow
Path (L)
(mile)
0.98
2.35
3.94
6.47
8.89
Lag
Time
(T)
(hr)
0.33
0.52
0.71
0.90
1.12
%
Imper-
vious
(I)
26.96
22.40
25.57
28.89
25.00
Imper-
vious
Co-
efficient
(K)
1.40
1.34
1.38
1.43
1.38
Flood-Frequency Ratio
(R)
2-yr
1.00
1.00
1.00
1.00
1.00
10-yr
1.84
1.89
1.85
1.82
1.86
100-yr
3.92
4.12
3.98
3.84
4.00
Development
Factor
(USGS
Method)
11.00
10.00
10.00
11.00
9.00
Previous SWMM calibration studies of the Little Rocky Run watershed (Tsihrintzis and Hamid 1998;
Zaghloul 1983) focused on the parameters of depression storage, infiltration parameters, and slope at each
calibration location. The calibration target was based on the relationship between the Anderson approach
peak flow estimation and the USGS method peak flow estimation range, as follows:
• When the computed peak flows from the Anderson approach fall within the range USGS
estimation range, the target flow for calibration is the USGS estimated range
4-124
-------
• When the estimated flows from the Anderson approach is outside the flow range from the USGS
method, the target flow for calibration is ±20 percent of the Anderson approach estimates
The calibration process ends when a good match is reached between the SWMM predicted peak flow rate
and the estimates from the Anderson/USGS methods. Table 4-9 compares the peak flows from the USGS
method, the Anderson approach, and the calibrated SWMM.
Table 4-9. Comparison of Predicted Peak Flows (ft3/s) in the Little Rocky Run Watershed
SWMM
Junction
o
J
13
18
28
32
The USGS Method
2-yr
124-510
278-1,138
387-1,585
678-2,782
564-2,316
10-yr
271-1,113
600-2,460
819-3,361
1,388-
5,692
1,188-
4,872
100-yr
400-2,300
897-5,163
1,222-7,038
2,051-
11,809
1,785-
10,275
The Anderson Method
2-yr
341.9
801.4
1,056.0
1,549.3
1,506.7
10-yr
629.0
1,511.1
1,957.2
2,823.0
2,801.2
100-yr
1,338.8
3,299.1
4,197.9
5,944.9
6,027.0
SWMM
2-yr
289.7
488.5
612.5
933.7
945.3
10-yr
580.1
1,133.4
1,457.3
2,260.5
2,308.0
100-yr
1,021.4
2,157.4
2,839.7
4,346.2
4,498.2
An example calibration plot is shown in Figure 4-21. It shows the predicted peak flows by three methods
on the same graph. The figure shows that the calibrated SWMM model peak flow rates are within the
USGS peak flow estimation range for 2-, 10-, and 100-yr 24-hour design storms.
2500 -i
5T
_0)
g
o
0)
Q_
D Est mated peak flow range by USGS method
A Estimated peak flow rate by Anderson method
• Calibrated SWMM prediction of peak flow rate
A
2-year
A
10-year
Design storm frequ
^
•
100-year
jncy
Figure 4-21. SWMM calibration results at junction 3 of the Little Rocky Run watershed.
Validation of Land Module in SUSTAIN
Figure 4-19 shows the selected study area for the case study application—a mixed land use area in the
headwaters of the Little Rocky Run watershed. The study area is approximately 240 acres, and the land
use distribution of the area size and imperviousness are presented in Table 4-10.
To ensure that SUSTAIN's internal land simulation module would reproduce the results of the calibrated
SWMM model, SUSTAIN was set up for the same area using the SWMM input parameters from the
calibrated Little Rocky Run model (Table 4-11). A 10-yr, 24-hour design storm was used to compare the
simulated surface runoffs using the internal land simulation option in SUSTAIN and standalone SWMM
model.
4-125
-------
Table 4-10. Land Uses of the Case Study Area in the Little Rocky Run Watershed
Land Use Groups
Transportation
Residential and Other Urban
Open Space
Total
Area
(acres)
65.4
95.9
79.5
240.8
Percent of Total
Area
(%)
27
40
33
100
Imperviousness
(%)
52
21
10
25
The study area was subdivided into three subareas on the basis of the type of treatment received (Figure
4-22). The areas that received quantity control onlt were grouped into subarea B1. The areas that use
BMPs to provide water quality treatment were grouped into subarea C. Lastly, the areas not served by
stormwater management facilities or received waivers because of the construction of downstream
stormwater facilities were grouped into subarea D. Fairfax County (2007a and 2007b) include additional
information on the subarea delineation and SWMM modeling approach.
Table 4-11. Major Input Parameters for Modeling of Three Subareas
Subarea
Area (acre)
Width (ft)
Slope (%)
Imperviousness (%)
Impervious Manning's n
Impervious Depression Storage (in.)
Pervious Manning's n
Pervious Depression Storage (in.)
C
4.82
21.37
1.31
24.95
0.015
0.1
0.266
0.2
Bl
36.63
162.43
1.31
26.05
0.015
0.1
0.269
0.2
D
167.64
743.38
1.31
22.41
0.011
0.04
0.282
0.1
* - 4
* * ii
•off en SI "J1*" i,r-" *w-E»J, 'ttratt
Figure 4-22. Representation of three subareas.
4-126
-------
Figure 4-23 shows the hydrograph generated at the outlet of the study area by the standalone SWMM
model. The same model configuration was replicated in SUSTAIN using the internal land simulation
module. Figure 4-24 shows a one-to-one comparison of the computed flows on the two separated
hydrographs generated. The perfect duplication of peak flow computations is a good confirmation that
the SWMM model codes were correctly incorporated into the Land Module of SUSTAIN.
^ SWMM 5 - CaseStudy_LRR.inp - [Graph - Node Outlet Total Inflow]
r"V- File Edit Viev
Report Wind:™ Help
Map
Categories
Title/Noies
Options
Climatology
- Hydrology
Ram Gages
Subcalchmenis
Aquifers
Snow Packs
RDII Holographs
r. Hydraulics
=• Nodes
Junctions
Outfalls
Dividers
Storage Units
-- ® Links
Transects
Controls
±i Quailtji
Time Patterns
Map Labels
|:>t i-
Node Outlet Total Inflow
Auto-Lengtti Off CfS Jjjj^ 729% X,Y: 5138.167. 7559.095
Figure 4-23. SWMM-generated hydrograph at the outlet of the study area.
SWMM versus SUSTAIN
20 30 40 50
Flow (cfs) Simulation in SUSTAIN
Figure 4-24. SWMM versus ^[/.ST/iCV-generated hydrograph comparison.
4-127
-------
4.2.4. Optimization and Results Analysis
The optimization problem can be mathematically expressed as:
Objective:
n
Minimize ^Cost(BMPt)
i=l
Subject to:
\l — \J, target
LTP > 40%
where
BMPj = BMP (i.e., bioswales, bioretention, and wet pond) configuration decision variables associated
with location /',
Q = the computed 10-yr design storm peak flow at the pond outlet,
Qtarget= the target value of the 10-yr peak flow rate at the pond outlet, and
LTP = the computed amount of TP load reduction percentage at the study area outlet.
To facilitate the BMP evaluation, the study area was grouped into three subbasins, one for each of the
three major land use types: transportation (27 percent), residential and other urban (40 percent), and open
space (33 percent). The land use distribution of the area size and imperviousness are presented in Table
4-10. This configuration was used to optimize the placement of BMPs to reduce peak flow rate and
nutrient load.
SWMM input parameters used to represent the physical characteristics of a subbasin for runoff
computations include area, width, slope, percent imperviousness, Manning's n for both pervious and
impervious surfaces, depression storage for both impervious and pervious surfaces, percentage of
impervious surfaces with zero depression storage, subarea internal routing method and percentage, and
the Horton infiltration parameters. Below is a summary of how each of the input parameters was
generated. Table 4-12 summarizes the numerical values of the SWMM input parameters for the three
subbasins.
Area—the surface area of a subbasin, calculated in a GIS environment using area summary functions.
Width—the width of a subbasin, which, according to the SWMM User's Manual, is calculated by
dividing the subbasin area by the longest flow path. The longest flow path is automatically generated
using ArcHydro.
Slope—slope for a subbasin is calculated as rise over run, in which the run represents the longest flow
path, and rise represents the elevation difference between the starting and ending points of the longest
flow path.
Percent imperviousness—the percent imperviousness of a subbasin is calculated by dividing the total
planimetric impervious area (i.e., building, roadway, parking lot, and sidewalk) by the total area of the
subbasin.
4-128
-------
Manning's n—the Manning's n values for both impervious and pervious surfaces are estimated on the
basis of land surface characteristics. The area of each land use type in a subarea is used to calculate the
area-weighted Manning's n for the whole subarea.
Depression storage—the depression storage is a volume that must be filled prior to the occurrence of
runoff on both pervious and impervious areas. In the absence of users' defined values, default values of
0.2 in. for pervious surface and 0.1 in. for impervious surface are suggested.
Percentage of impervious surface with zero depression storage—a default value of 25 percent is used
in the initial model setup as suggested in the SWMM manual.
Table 4-12. Subbasin SWMM Parameters
Subbasin
Land Use Groups
Area (acre)
Width (ft)
Slope (%)
Imperviousness (%)
Impervious Manning's n
Impervious Depression Storage (in.)
Pervious Manning's n
Pervious Depression Storage
1
Transportation
65.4
145.22
1.31
51.6
0.015
0.11
0.266
0.23
2
Residential and
Other Urban
95.9
404.77
1.31
20.5
0.015
0.11
0.266
0.23
3
Open Space
79.5
237.19
1.31
9.5
0.015
0.11
0.266
0.23
Table 4-13 is a summary of the hydrologic soil group distribution in the case study area from county-
provided SSURGO data. Note that a major portion (90.6 percent) of subbasin 3 (open space) has D soils,
while subbasins 1 (transportation) and 2 (residential and other urban) have a mixture of B and D soils.
Table 4-13. Hydrologic Soil Group Distribution in Study Area Subbasins
Subbasin 1
Subbasin 2
Subbasin 3
Soil Group B
(%)
56.35
40.09
8.64
Soil Group C
(%)
6.14
5.47
0.81
Soil Group D
(%)
37.51
54.44
90.55
The Green-Ampt infiltration equation was used to simulate the infiltration loss. The equation requires
that three parameters be specified. As a conservative assumption, the parameters for D soils are assumed
for all three subbasins:
• The average capillary suction (Su) = 3 in.
• Hydraulic conductivity (Ks) = 0.5 in./hr
• The initial moisture deficit (IMD) = 0.25 (fraction)
4-129
-------
As described in Section 3.2.3, SUSTAINuses the buildup and washoff processes for water quality
simulation. The calibrated SWMM model for the Little Rock Run watershed focused only on flows and
did not include water quality components. Pollutant loads from the watershed were estimated using a
simplified spreadsheet tool, STEPL (provide a reference). Nevertheless, there are several SWMM
modeling studies in Fairfax County in which the buildup and washoff parameters were determined
(Behera et al. 2006). These studies were reviewd and representative buildup and washoff coefficients
were derived as shown in Table 4-14 and used in SUSTAIN for estimating TP concentrations.
Table 4-14. TP Buildup and Washoff Parameter Values
Name
Residential Impervious
Residential Pervious
Transportation Impervious
Transportation Pervious
Open Space Impervious
Open Space Pervious
Buildup
B=Min(Ch C2f3)
Cj
8.0
6.3
7.4
6.3
7.3
6.0
C2
0.045
0.031
0.04
0.031
0.032
0.030
C3
0.523
0.42
0.429
0.42
0.388
0.394
Washoff
W= CjqC2B
Cj
0.83
0.70
0.76
0.70
0.70
0.692
C2
1.38
1.49
1.403
1.49
1.42
1.465
BMP Representation
Three BMP types considered in the analysis included bioswale, bioretention, and a regional wet pond.
Bioswales and bioretention facilities are distributed BMPs to treat runoff from local impervious areas
with bioswales for highways and roads and bioretentions for residential and other urban areas. Regional
wet ponds are centralized BMPs and typically are large facilities designed to treat large drainage areas.
An existing regional wet pond was located at the study area outlet and, therefore, receives runoff from the
entire study area. Figure 4-25 shows a flow chart diagram of the BMP simulation network. Table 4-15
lists the BMP parameters that were used in the model.
Bioswales
Bioswales are modified vegetated swales that use bioretention media beneath the swale to improve water
quality, reduce the runoff volume, and reduce the peak runoff rate. In addition to provide drainage
conveyance function as traditional drainage swales, bioswales use vegetation, compost, and/or riprap to
enhance infiltration, water retention, and removal of silt, nutrient and pollutants through a variety of
physical, chemical, and biological processes.
Bioretention
Bioretention cells, also known as rain gardens, are small-scale, shallow vegetated depressions that provide
water quality benefits by rapid filtering through soil media, biological and chemical reactions in the soil
matrix and root zone, and infiltration into the underlying subsoil. Bioretentions can be designed as on-
line or off-line facilities with respect to the stormwater conveyance.
Regional Wet Pond
As a traditional practice that is mainly used for flood control, regional ponds were constructed in rapidly
urbanizing areas in Fairfax County. There is a regional pond at the outlet of the case study area. The
pond was designed to attenuate the peak discharge for 2-, 10-, and 100-yr 24-hour design storms.
4-130
-------
Highway and Roads
I
Bioswale
Residential and Other Urban
1
Bioretention
\
Regional
Other Area
1
' Outlet
Wet Pond ""^ ^
Figure 4-25. BMP placement schematic in the Little Rocky Run case study area.
Table 4-15 BMP Parameters
Parameter
Bioswale
Bioretention
Regional Wet
Pond
Physical Configuration
Maximum size (acre)
Drainage area (acre)
Substrate depth (ft)
Underdrain depth (ft)
Ponding depth (ft)
0.36
25.5
2
N/A
0.5
0.33
33.6
2.5
1
0.5
2.24
241
N/A
N/A
Orifice height: 1
Weir height: 4.75
Infiltration*
Substrate layer porosity
Substrate layer field capacity
Substrate layer wilting point
Underdrain gravel layer porosity
Vegetative parameter, A
Underdrain background infiltration rate
(in./hr), fc
Media final constant infiltration rate (in./hr),
fc
0.5
0.3
0.15
0.5
0.6
0.5
3
0.5
0.3
0.15
0.5
0.6
0.5
o
6
N/A
N/A
N/A
N/A
N/A
N/A
N/A
Water Quality**
TP 1st order decay rate (I/day), k
TP filtration removal rate, Prem (%)
0.2
65
0.2
65
0.2
N/A
* Source: Tetra Tech 2001
** Based on calibration using University of Maryland monitoring data (Tetra Tech 2003)
BMP Cost
For the case study, annualized life cycle cost functions summarized in Table 4-16 were used to estimate
costs of BMPs considered. Annualized life cycle costs include the initial installation cost, as well as
maintenance and replacement costs.
4-131
-------
Table 4-16. Annualized BMP Cost Function
BMP Type
Bioretention
Bioswale
Wet pond
Annualized Cost Function
$ 1.03/ft2 per year
$0.67/ft2 per year
Cost = 0.12 x 24.5 x F0705 ($ per year)
V is the volume of the wet pond in ft3
Reference
Fairfax County BMP Factsheets (Fairfax
County 2005)
Fairfax County BMP Factsheets (Fairfax
County 2005)
CASQA Stormwater BMP Handbook
(CASQA 2003).
Bioretention Cost
Table 4-17 shows the breakdown of installation and annualized costs for a typical bioretention cell with a
surface area of 900 ft2. A bioretention cell is assumed to have a life span of 25 years, at which point it
will be removed and replaced.
Table 4-17. Annualized Life Cycle Cost for a Bioretention Cell with a Surface Area of 900 ft2
Item
Installation*
Mulching and
Debris Removal
Replace Vegetation
Remove & Replace
Total Cost
Annualized Cost
Required Cost per Year (2005 Dollars)
0
10,000
10,000
1
350
200
550
2
350
200
550
3
350
200
550
4
350
200
550
5
350
200
550
6
350
200
550
7
350
200
550
8
350
200
550
9
350
200
550
10
350
200
550
...
25
10,000
10,000
$925/year (includes replacement in year 25)
Source: Fairfax County BMP Factsheets
"The developer cost, which is not included in the annualized cost.
Bioswale Cost
Table 4-18 shows the installation cost and annualized costs for a bioswale with a surface area of 900 ft2.
A bioswale is also assumed to have a life span of 25 years, at which point it will be removed and
replaced.
Table 4-18. Annualized Life Cycle Cost for a Bioswale with a Surface Area of 900 ft2
Item
Installation*
Mowing
Reseeding/Replanting
Remove & Replace
Total Cost
Annualized Cost
Required Cost per Year (2005 Dollars)
0
10,000
10,000
1
100
100
200
2
100
100
200
3
100
100
200
4
100
100
200
5
100
100
200
6
100
100
200
7
100
100
200
8
100
100
200
9
100
100
200
10
100
100
200
...
25
10,000
10,000
$600/year (includes replacement in year 25)
Source: Fairfax County BMP Factsheets
*The developer cost, which is not included in the annualized cost.
4-132
-------
Wet Pond Cost
Wet Pond costs were estimated as described below.
Construction Cost
Cost = 24.5 x Fa70i(see Table 4-16) (4-1)
where
Cost ($) = the cost for construction, design, and permitting, and
V= volume of the wet pond in ft3.
Converting the capital cost to annualized capital cost by assuming a 20-yr life span and 0.05 annual
interest rate gives
Annualized Capital Cost ($/yr) = 0.08 x 24.5 x Va705 (4-2)
Maintenance Cost
The mid-range of maintenance cost is approximated to be four percent of capital cost per year.
Annualized Maintenance Cost ($/yr) = 0.04 x 24.5 x V0'705 (4-3)
Annualized Life Cycle Cost (including capital and maintenance cost)
Annualized Cost = 0.12><24.5x V0'705 (see Table 4-16) (4-4)
where
Annualized Cost ($) = the cost of construction, design, permitting, maintenance, and replacement,
and
V = the volume of the wet pond in ft3.
Simulation Period and Evaluation Factors
As previously described, the objectives of this optimization case study are to provide flood control up to
the 10-yr design storm and achieve a minimum of 40 percent reduction of TP load. Because of local
water quality improvement needs and the ultimate downstream effect on the Chesapeake Bay, reduction
in nutrient loading is an important objective for the stormwater management in Fairfax County. BMP
effectiveness was measured at the designated assessment point at the study area outlet.
The use of continuous simulation provides modelers with an opportunity to capture the dynamic
responses of BMPs under various storm conditions. However, among the large potential sets of data, it is
often possible to identify one set of data that can reasonably represent either the long-term average
treatment performance of BMPs or a critical condition that is most closely associated with the nature of
the problem being studied. From the analysis of precipitation data, the calendar year 1994 (January 1,
1994-December 31, 1994) was determined to be representative of average hydrologic conditions in the
watershed. The total rainfall depth for the year was close to the 17-yr average of 43.3 in. In addition,
both precipitation depth and intensity distribution were relatively close to the long-term statistical average
distribution. Figure 4-26 shows the average annual precipitation at the Washington Dulles International
Airport for calendar years 1990-2006.
As mentioned previously, calendar year 1994 had the most typical rainfall magnitude and intensity
distribution from among the four highlighted years that were close to the 17-yr average. Figure 4-27
shows the rainfall volume and intensity distribution for wet intervals occurring in calendar year 1994 at
4-133
-------
Washington Dulles International Airport. In the figure, the volume and intensity percentile ranges
correspond to storms occurring over the 17-yr period. A year with a perfect typical distribution would
have the same number of precipitation intervals in each bin.
-Washington Dulles Intl AP (448309)
888888
Figure 4-26. Annual precipitation at the Washington Dulles International Airport.
Rainfall Intensity
Rainfall Volume
Figure 4-27. Volume and intensity distribution of storm events in 1994.
Existing Condition with Regional Wet Pond
The study area includes an existing regional wet pond that was designed to attenuate the peak discharge
from 2-, 10-, and 100-yr storm events. Using SUSTAIN, the simulated 10-yr, 24-hour design storm peak
flows and TP annual loads are summarized in Table 4-19. Note that peak flow values are reasonably
compared to the original SWMM model results assessed at the outlet of the case study area.
Table 4-19. 10-Yr Design Storm Peak Flows and TP Annual Load under Existing Conditions
Model
SUSTAIN
Fairfax County SWMM
10-yr, 24-hr Design Storm Peak Flow
(cfs)
Pond inflow
108.3
107.7
Pond outflow
53.4
50.4
TP Annual Load
(Ib/yr)
Pond inflow
78.2
N/A
Pond outflow
59.8
N/A
4-134
-------
Optimization Setup
Bioswales and bioretention cells were recognized as feasible BMPs for the case study area to attenuate
peak flow and reduce pollutant load. The decision variables used in the study included the size of surface
area of bioswales and bioretention cells, and the storage capacity of the wet pond. Although the size of an
existing regional pond was treated as a decision variable, it is not to suggest that the existing pond size
can be reduced. Instead, the benefit of adding supplemental storage capacity in the form of bioswales and
bioretentions could demonstrate a net increase to the overall on-site storage capacity, thereby yielding
additional flood control capacity for the wet pond. The difference between the required versus the
optimized size equals the net gain in flood control capacity for the wet pond. Two optimization
objectives were defined in this exercise: (1) to maintain the 10-yr design storm peak flow at the pond
outlet, and (2) to reduce the TP load by 40 percent.
Optimization Results
Using Scatter Search techniques, a near-optimal solution was identified after approximately 300 model
runs. Figure 4-28 shows the near-optimal solution that meets both the 10-yr peak flow (top portion) and
TP load reduction (bottom portion) targets. This solution is presented alongside a scenario without any
BMPs (PostDev) and the existing condition with only the existing wet pond (Existing). This best solution
carries an annualized life cycle cost of $64,400 per year.
Output Dracloiy ID \SUSTAINVData
CSl'«!.]|TP.-Load|
Plot of Evaluation Factors
PrftDev PostDev Existing
Cmt Inloimatian
Besl 1 $64.339.7
D \SUSTAIN\Data
Plot of Evaluation Factors
i 11 ii i
II I
PostDev ' Existing
Cost Inloimatian
Be«1 SE4.3337
Figure 4-28. Results showing the benefits of BMPs.
4-135
-------
Figure 4-29 presents all the solutions examined during the search process and the near-optimal solutions
are highlighted in orange. The blue points indicate the cost and peak flow rate of all the solutions
evaluated. The small number of blue points located to the left of the optimal solutions and below the flow
rate target line are the solutions that meet the peak flow target, but not the water quality target. Table
4-20 compares the best solution with the existing scenario and lists the selected BMP decision variable
values for the best solution. It shows that, relative to existing condition life cycle cost (related to the
existing wet pond), to achieve the 10-yr peak flow and the TP load reduction targets, an additional
$35,000 per year would be needed. This additional investment for water quality improvement would
require 0.36 acre (15,700 ft2) of bioswales to treat runoff from roads and highways and 0.04 acre (1,740
ft2) of bioretention area to treat residential and other urban areas. The best solution also indicates that the
new BMPs for water quality improvement yield a 17 percent net savings in terms of flood mitigation
storage volume at the regional wet pond.
f 80
|
£
j
o
u. 60
|
0_
*
VIA* A MAA * * i A . A * * AA
AMAAAAA i»*4 A 1 AA 4k ^vA / AA
r *»A
1 A A * .A? ' IIP** ^rVf
* * ** **
4 •S'^l • •• '
• V*f *\lff
/ • f Ni l4JUi*l4i.4» i
'• • + ***** **£*!**
.• i. " • • • " .jTli*1**
•*• '^ABtlA*
• • •" JU^AAAAM^^^
• •
|i , A All solutions- lOyrpeakflow
• ' • A Best solutions -lOyrpeakflow
10-yr peak flow target
• All solutions-TP load reduction
f | Best solutions - TP load reduction
, — TP reduction target
1 • •
40-
i*
c
fi)
*°
Q.
I
C
a-.
o
P
10
0 10000 20000 30000 40000 50000 60000 70000 80000
Annual ized Life Cyde Cost ($fyr)
Figure 4-29. Domain of optimization searches and identified best solutions.
Table 4-20. Cost-Effective Solution Details
Annualized cost ($/yr )
10-yr peak flow (cfs)
TP annual load reduction (%)
Existing Scenario
29,017
53.4
23.5
Best Solution
64,400
53.0
40.7
BMP
Regional pond area (acre)
Bioswale area (acre)
Bioretention area (acre)
2.24
0
0
1.86
0.36
0.04
4-136
-------
4.2.5. Summary
This case study has demonstrated: (1) a verification of the internal SWMM land module against the
standalone SWMM 5 model and the use of the internal model to generate flow and pollutant loads from
land, and (2) use of the Scatter Search optimization technique to find a near-optimal solution, given
multiple control targets of peak flow and TP load reduction. Using optimization to guide decision making
has been demonstrated to provide meaningful insights into the hydrologic response and benefits of BMPs
for stormwater management. SUSTAIN is a powerful tool for decision-making. However, the outcomes
from a SUSTAIN application would very much depend on the user-defined goals and assumptions.
Hence, the results must be interpreted considering the user-specified assumptions, problem formulation,
and defined optimization constraints. As stated before, the SUSTAIN application must be accompanied by
an intimate understanding of the study area and the identification of influential factors that affect the
decision making to achieve stormwater management goals.
4-137
-------
Chapter 5 References
Anderson, D.G. (1970). Effects of Urban Development on Floods in Northern Virginia. Geological
Survey Water- Supply Paper, 2001-C. U.S. Geological Survey, Washington, DC.
Anderson, E.A. (1973). National Weather Service River Forecast System — Snow Accumulation and
Ablation Model. NOAA Technical Memorandum, NWS HYDRO-17. U.S. Department of
Commerce, National Oceanic and Atmospheric Administration, Washington, DC.
ASCE (American Society of Civil Engineers). (1982). Gravity Sanitary Sewer Design and Construction.
ASCE Manual of Practice No. 60. American Society of Civil Engineers, New York, NY.
ASCE (American Society of Civil Engineers). (1993). Criteria for Evaluation of Watershed Models.
American Society of Civil Engineers, Task Committee of the Watershed Management
Committee. Journal of Irrigation and Drainage Engineering 1 19(3):429-442
Behera, P.K, B.J., Adama, and Y.L. James. (2006). Runoff Quality Analysis of Urban Catchments with
Analytical Probabilistic Models. Journal of Water Resources Planning and Management 132
Bicknell, B.R., J.C. Imhoff, J.L. Kittle, A.S. Donigian, Jr., and RC. Johanson, Jr. (1997). Hydrological
Simulation Program— FORTRAN, User's manual for Version 11. EPA/600/R-97/080. U.S.
Environmental Protection Agency, National Exposure Research Laboratory, Athens, GA.
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.
CASQA (California Stormwater Quality Association). (2003). CASQA Stormwater BMP Handbook: New
Development and Redevelopment . California Stormwater Quality Association, Menlo Park, CA.
Chocat, B., P. Krebs, J. Marsalek, W. Rauch, and W. Schilling. (2001). Urban drainage redefined: from
Stormwater removal to integrated management. Water Science and Technology 43(5):61-68.
Deb, K., A. Pratap, S. Agarwal, and T. Meyarivan. (2002). A fast and elitist multi -objective genetic
algorithm: NSGAII. IEEE Transactions on Evolutionary Computation 6(2): 182-197.
Deletic, A. (2001). Modelling of water and sediment transport overgrassed areas. Journal of Hydrology
248(1-4): 168-182
Donigian, A.S., Jr., J.C. Imhoff, B.R Bicknell, and J.L. Kittle. (1984). Application Guide for
Hydrological Simulation Program— FORTRAN (HSPF). EPA-600/3-84-065. U.S. Environmental
Protection Agency, Environmental Research Laboratory, Athens, GA.
Donigian, A.S., and J.T. Love. (2003). Sediment calibration procedures and guidelines for watershed
modeling. In WEF Specialty Conference Proceedings. WEF TMDL 2003, November 16-19,
2003, Chicago, IL.
Dorn, J.L., and S. Ranjithan. (2003). Evolutionary Multiobjective Optimization in Watershed Water
Quality Management, in Evolutionary Multi-Criteria Optimization, Lecture Notes in Computer
Science (LNCS), eds. C.M. Fonseca, P.J. Fleming, E. Zitzler, K. Deb, and L. Thiele, Vol. 2632,
pp. 692-706, Springer- Verlag, Berlin.
5-138
-------
Deb, K. (2001). Multi-Objective Optimization Using Evolutionary Algorithms. Isted. Wiley-Interscience
series in systems and optimization. John Wiley and Sons, New York, NY.
Fair, M.F., J.C. Geyer, and D.A. Okun. (1968). Water and Wastewater Engineering. John Wiley and
Sons, New York, NY.
Fairfax County. (2005). Fact Sheets for Candidate LID BMPs for the Fairfax County, Virginia Public
Facilities Manual. Prepared for Fairfax County Department of Public Works and Environmental
Services by Fairfax County Low Impact Development Center, Fairfax, VA.
http://www.lowimpactdevelopment.org/fairfax.htm. Accessed July 24, 2009.
Fairfax County. (2007a). The Comprehensive Plan for Fairfax County, Virginia. 2007 edition.
Department of Planning and Zoning, Planning Division, Fairfax, VA.
Fairfax County. (2007b). Preliminary Comments and Thoughts: SWMM Models for Johnny Moore,
Pohick, Accotink, Little Rocky Run, Horsepen Creek, andSugarlandRun. Fairfax County, VA.
Glover, F., M. Laguna, and R. Marti. (2000). Fundamentals of Scatter Search and Path Relinking. Control
and Cybernetics 29(3):653-684.
Glover, F. (1977) Heuristics for Integer Programming Using Surrogate Constraints. Decision Sciences
8(7): 156-166.
Goldberg, D. (1989). Genetic Algorithms in Search, Optimization, and Machine Learning. Addison-
Wesley, Reading, MA.
Hamon, W.R. (1961). Estimating Potential Evapotranspiration. Journal of the Hydraulic Division.
Proceedings of the American Society of Civil Engineers. 87:107-120.
Harrell, L.J. (2001). Evolutionary algorithm-based design of a system of wet detention basins under
uncertainty for watershed management. In Proceedings of the EWRI World Water and
Environmental Resources Congress, May 20-24, 2001, Orlando, FL.
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.
Hydrocomp, Inc., and AQUA TERRA Consultants. (1996). Modifications to HSPFfor High Water Table
and Wetlands Conditions in South Florida: Conceptual Analysis, Preliminary Design and
Implementation, and Initial Testing. Report to South Florida Water Management District, West
Palm Beach, FL.
James, W., W.C. Huber, R.E. Pitt, RE. Dickinson, L.A. Rosener, J.A. Aldrich, and W.R. James. (2002).
SWMM4 User's Manual (User's Guide to the EPA Storm water Management Model and to
PCSWMM/ University of Guelph and CHAI, Guelph, Ontario, Canada.
Kadlec, R.H., and R.L. Knight. (1996). Treatment Wetlands. CRC Press, Lewis Publishers, Boca Raton,
FL.
Laguna, M., and R. Marti. (2002). The OptQuest Callable Library to Appear in Optimization Software
Class Libraries, eds. S. Voss and D.L. Woodruff, pp. 193-218. Kluwer Academic Publishers,
Boston, MA.
Lai, F-h., T. Dai, J. Zhen, J. Riverson, Jr., K. Alvi, and L. Shoemaker. (2007). SUSTAIN—An EPA BMP
Process and Placement Tool for Urban Watersheds. In Proceedings of the Water Environment
Federation 2007 Total Maximum Daily Loads (TMDL) Meeting, June 24-27, Bellevue, WA.
Linsley, R.K., J.B. Franzini, D.L. Freyberg, and G. Tchobanoglous. (1992). Water-Resources
Engineering. 4th ed. McGraw-Hill, New York, NY.
5-139
-------
Maidment, D.R., ed. (1993). Handbook of Hydrology. McGraw-Hill, New York, NY.
Minnesota Stormwater Steering Committee. (2005). Minnesota Stormwater Manual. Version 1.1.
Minnesota Pollution Control Agency, St. Paul, MN.
Munoz-Carpena, R. (1993). Modeling hydrology and sediment transport on vegetative filter strips. Ph.D.
dissertation. North Carolina State University, Raleigh, NC.
Munoz-Carpena, R., C.T. Miller, and J.E. Parsons. (1993). A quadratic Petrov-Galerkin solution for
kinematic wave overland flow. Journal of Water Resources Research 29(8):2615-2627.
Munoz-Carpena, R., and J.E. Parsons. (2003). VFSMOD-W—Vegetative Filter Strips Hydrology and
Sediment Transport Modeling System, Model Documentation and User's Manual. University of
Florida, Agricultural and Biological Engineering, Gainesville, FL.
Neitsch, S.L., J.G. Arnold, J.R Kiniry, J.R. Williams, and K.W. King. (2002). Soil and Water
Assessment Tool Theoretical Documentation, Version 2000. Grassland, Soil and Water Research
Laboratory, Agricultural Research Service and Blackland Research Center, Texas Agricultural
Experiment Station, Temple, TX.
NRDC (Natural Resources Defense Council). (2006). Rooftops to Rivers: Green strategies for controlling
Stormwater and combined sewer overflows.
http://www.nrdc.org/water/pollution/rooftops/rooftops.pdf Accessed July 2009.
Partheniades, E. (1962). A Study of Erosion and Deposition of Cohesive Soils in Suit Water. Ph.D.
dissertation. University of California, Berkeley, CA.
Persson, J., N.L.G. Sommes, and T.H.F. Wong. (1999). Hydraulics efficiency of constructed wetlands and
ponds. Water Science and Technology 40(3):291-300.
Perez-Pedini, C., J. Limbrunner, and R.M. Vogel. (2005). The Optimal Location of Infiltration-Based
BMPs for Stormwater Management. Journal of Water Resources Planning and Management
131(6): pp. 441-448
Rawls, W.J., D.L. Brakensiek, and N. Miller. (1983). Green-Ampt infiltration parameters from soils data.
Journal of Hydraulic Engineering 109(1):62-70.
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 Division, National
Risk Management Research Laboratory, Cincinnati, OH.
SEWRPC (Southeastern Wisconsin Regional Planning Commission). (2007). A Regional Water Quality
Management Plan Update for the Greater Milwaukee Watersheds. Planning Report NO. 50
Southeastern Wisconsin Regional Planning Commission in cooperation with the Milwaukee
Metropolitan Sewerage District, Wisconsin Department of Natural Resources, and the U.S.
Geological Survey.
Tetra Tech. (2001). Low-Impact Development Management Practices Evaluation Computer Module,
User's Guide. Prepared for Prince George's County, Maryland, by Tetra Tech, Inc., Fairfax, VA
Tetra Tech. (2003). Validating the Low-Impact Development Management Practices Evaluation
Computer Module. Prepared for Prince George's County, Maryland, Department of
Environmental Resources, by Tetra Tech, Inc., Fairfax, VA
Tetra Tech. (2008). Proposed Calibration Method for County SWMM Models. Prepared for Fairfax
County, Virginia, by Tetra Tech, Inc., Fairfax, VA
Tetra Tech. (2009). SWMM Model Calibration for Little Rocky Run Watershed. Prepared for Fairfax
County, Virginia, by Tetra Tech, Inc., Fairfax, VA
5-140
-------
Tetra Tech and USEPA (U.S. Environmental Protection Agency). (2002). The Loading Simulation
Program in C++ (LSPC) Watershed Modeling System—Users' Manual.
Tollner, E.W., B.J. Barfield, C.T. Haan, and T.Y. Kao. (1976). Suspended sediment filtration capacity of
simulated, rigid vegetation. Transactions oftheASAE 19(4):678-682.
Tsihrintzis V.A., and R. Hamid. (1998). Runoff quality prediction from small urban catchments using
SWMM. HydrologicalProcess 12:311-329.
University of British Columbia-Department of Mechanics. (No date). Fluid Dynamics Course Notes.
http://batman.mech.ubc.ca/~mech380/notes/openchannel/weirs/broadc.html. Accessed October
2005.
USER (U.S. Bureau of Reclamation). (2001). Water Measurement Manual. U.S. Department of the
Interior, Bureau of Reclamation, Washington, DC.
http://www.usbr.gov/pmts/hydraulics_lab/pubs/wmm/chap07_03.html. Accessed August 2009.
USEPA (U.S. Environmental Protection Agency). (1983). Results of the Nationwide Urban Runoff
Program (NURP). Vol. 1, NTIS PB 84-185552), U.S. Environmental Protection Agency, Water
Planning Division, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (1991). Guidance for Water Quality-Based Decisions:
The TMDL Process. EPA 440/4-91-001. U.S. Environmental Protection Agency, Office of Water,
Washington, DC.
USEPA (U.S. Environmental Protection Agency). (1997). Compendium of Tools for Watershed
Assessment and TMDL Development. EPA 841-B-97-006. U.S. Environmental Protection
Agency, Office of Water, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (1999a). 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 (U.S. Environmental Protection Agency). (1999b). HSPFParm: An Interactive Database of
HSPFModel Parameters, Version 1.0. EPA-283-R-99-004. U.S. Environmental Protection
Agency, Office of Water. Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2000). BASINS Technical Note 6. Estimating
Hydrology and Hydraulic Parameters for HSPF. EPA-823-ROO-012. U.S. Environmental
Protection Agency. Office of Water, Washington, DC.
http://www.epa.gov/waterscience/basins/docs/tecnote6.pdf
USEPA (U.S. Environmental Protection Agency). (2004a). Stormwater Best Management Practice
Design Guide. EPA/600/R-04/121. U.S. Environmental Protection Agency, Office of Research
and Development, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2004b). The Use of Best Management Practices
(BMPs) in Urban Watersheds. EPA/600/R-04/184. U.S. Environmental Protection Agency,
Office of Research and Development, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2004c). Report to Congress: Impacts and Control of
CSOs andSSOs. EPA-833-R-04-001. U.S. Environmental Protection Agency, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2006). BASINS Technical Note 8. Sediment Parameter
and Calibration Guidance for HSPF. U.S. Environmental Protection Agency, Office of Water,
Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2007). Memorandum: Using Green Infrastructure to
Protect Water Quality in Stormwater, CSO, Nonpoint Source and other Water Programs.
5-141
-------
Benjamin Grumbles, EPA's Assistant Administrator for Water. U.S. Environmental Protection
Agency, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2008a). Handbook for Developing Watershed Plans to
Restore and Protect Our Waters. EPA-841-B-08-002. U.S. Environmental Protection Agency,
Office of Water, Nonpoint Source Pollution Branch, Washington, DC.
USEPA (U.S. Environmental Protection Agency). (2008b). A Screening Assessment of the Potential
Impacts of Climate Change on Combined Sewer Overflow (CSO) Mitigation in the Great Lakes
and New England Regions. EPA/600/R-07/033F. U.S. Environmental Protection Agency,
Washington, DC.
USGS (U.S. Geological Survey). (2001). The National Flood-Frequency Program—Methods for
Estimating Flood Magnitude and Frequency in Rural Areas in Virginia, 2000. USGS Fact Sheet
023-01. U.S. Geological Survey, Office of Surface Water, Reston, VA.
Wischmeier, W.H., and D.D. Smith. (1965). Predicting Rainfall Erosion Losses from Cropland East of
the Rocky Mountains. Agricultural Handbook No. 282. U.S. Department of Agriculture,
Washington, DC.
Wong, T.H.F., and P.P. Breen. (2002). Recent advances in Australian practice on the use of constructed
wetlands for stormwater treatment. In Global Solutions for Urban Drainage, Proceedings of the
Ninth International Conference on Urban Drainage, eds. E.W. Strecker and W.C. Huber.
American Society of Civil Engineers, September 2002, Portland, OR. CD-ROM,.
Wong, T.H.F., H.P. Duncan, T.D. Fletcher, and G.A. Jenkins. (2001). A unified approach to modelling
urban stormwater treatment. In Proceedings of the Second South Pacific Stormwater Conference.
June 27-29, 2001, Auckland, New Zealand.
Zaghloul, N.A. (1983). Sensitivity analysis of the SWMM Runoff-Transport parameters and the effects of
catchment discretisation. Advances in Water Resources 6:214-223
Zhen, X., and S.L. Yu. (2004). Optimal location and sizing of stormwater basins at watershed scale.
Journal of Water Resources Planning and Management 130(4):339-347
5-142
-------
Appendix A. Needs Analysis and Technical Requirements
This appendix presents an evaluation of technical needs for developing a computer framework to assist
stormwater management professionals in planning for BMP implementation in urban watersheds that
achieves the desired source water and water quality protection cost effectively. The objective is to define
the need for a system that can address both placement and selection of management practices in urban
areas. The major programs targeted for SUSTAIN applications include urban watershed planning,
stormwater management, and TMDL implementation. SUSTAIN must also be applicable to additional
programs such as MS4s, the stormwater Phase IINPDES permit program, and source water protection.
Each program requires the evaluation of key management questions and consideration of related
indicators. Source water protection studies will need to address water supply protection; typically
including eutrophication related indicators (i.e., phosphorus) and sediment. For TMDLs, the key
indicators will be dictated by the waterbody's designated use (e.g., primary contact, warm water fishery)
and the type of pollutants causing the impairment (i.e., metals, nutrients, fecal coliforms). The needs
analysis addresses the various watershed protection programs by identifying three general categories of
questions typically asked in urban management projects:
• What are the parameters for measuring the benefit or impact of management to protect source
waters?
• What is the difference in performance between management options/scenarios including one or
more practices?
• Which management alternatives will achieve environmental targets at the lowest cost?
These three questions are discussed in the Needs Analysis section below. It is followed by a discussion of
specific technical requirements for building SUSTAIN.
A.I. Needs Analysis
SUSTAIN was designed to answer three needs analysis questions. For each question, specific capabilities
required are included to show how individual elements work to meet overall project objectives.
1. What are the parameters for measuring the benefit or impact of management? What is the target
value to achieve?
To select an optimal condition and compare the benefits of various management practices or
combinations of practices, a performance measure or indicator must be selected to use for evaluation. In
examining environmental conditions in urban areas, multiple performance measures or indicators of
condition are recommended. The specific performance measures vary depending on the designated use of
the water body (warm water fishery, cold water fishery, recreation) and the condition of the water body.
For example, multiple factors or stressors might influence a warm-water fishery. Some potential
stressors are changes in hydrology measured as peak flow and frequency of 1-yr stream flow events,
elevated nutrient concentrations, elevated sediment concentrations, and higher summer temperatures.
Each of these stressors can be measured using performance measures such as peak flow, flow volume,
temperature, and nutrient concentration. Predictive models can use these performance measures as output
A-143
-------
values for optimization and selection of alternatives. A specific value or target can be set as a goal. For
example, the temperature target might be set as a maximum of 85 F. Targets can be set on the basis of
water quality standards or using expert examination of water quality conditions. Multiple stressors
typically affect urban streams. Table A-l provides a summary of the most commonly used performance
measures and the specific parameters used. The selection of one or more performance measures suitable
for the local conditions is appropriate for evaluating the benefits of management.
Table A-l. Summary of Recommended Indicators and Measurement Units
General Performance
Measure
Hydrology
Sediment
Water Quality
Ecological measures
Specific Performance Measure
Flow
Total sediment
Total suspended sediment
Total solids
Pollutant
Nitrogen (NO3, NH3, TKN)
Phosphorus (TP, PO4)
Metals (typically zinc, lead,
arsenic, manganese, aluminum)
Pathogens
Dissolved oxygen
Temperature
Others — typically not modeled,
habitat condition, species
diversity, stream condition, fish
quantity and diversity
Measurement (units)
Volume (ft3)
Frequency (x/yr of selected peak, volume)
Duration (hr)
Concentration (mg/L)
Load (tons/year, tons/month)
Concentration (mg/L)
4-day average concentration (mg/L)
Load (loads per day, month, or year)
Pathogens — geometric mean (cfu/mL)
Dissolved oxygen — daily minimum,
average
daily
Summer mean, 7-day average
2. What is the difference in performance between management options/scenarios including one or
more practices ?
To determine optimal solutions for a complex watershed, SUSTAIN needs to address multiple locations
and practices in various combinations throughout the watershed. It must be sensitive to conditions like
the following:
a. For each practice, SUSTAIN needs to be able to simulate the selected suite of performance
measures. The framework must be capable of evaluating changes in performance measures on
the basis of an unbiased evaluation of individual practice performance for a range of structural
and nonstructural practices. Typical structural and nonstructural practices that the framework
would evaluate can be classified into three general categories of BMPs using the mode of
application. These categories are Point BMPs, Linear BMPs, and Area-Based BMPs. Examples
of each type are shown in Table A-2.
A-144
-------
Table A-2. Typical Structural and Nonstructural Practices by Mode of Application
Point BMPs
Dry extended detention
pond
Wet retention pond
Shallow marsh
Extended detention wetland
Submerged gravel wetland
Organic filter
Sand filter
Bioretention
Infiltration trench
Porous pavement
Dry swale
Wet swale
Inlet devices
Baffle box
Oil-grit separator
Linear BMPs
Vegetated Buffer Strips
Riparian Zone Restoration
Area-Based BMPs
Fertilizer management
Impervious area minimization
Disconnected impervious areas
Site level water management
Soil management
Street Sweeping
Multiple practices in various combinations need to be considered. Figure A-l provides a
schematic of some of the potential combinations that SUSTAIN needs to evaluate. The options
include various combinations of land areas, BMPs, conduits (pipes), or stream reaches (RCH).
Some swales or buffers are illustrated by land-to-land series (number 4). Series of two or more
BMPs might need to be considered (number 9).
= Evaluation Point)
(7) LAND SMPconduit
BMP H BMP H RCH I
(9)
Figure A-l. Typical management configurations.
c. The location of individual or multiple practices relative to a water body or receiving water might
vary as well. Some BMPs are located on-site, with very small drainage areas; other BMPs, such
as stormwater ponds, are located closer to stream systems and have larger drainage areas.
3. Which management alternatives will achieve environmental targets at the lowest cost?
A-145
-------
The effectiveness of BMP options (sizes, locations) in achieving the desired water quality goal must be
compared based on costs. The scenarios can also be evaluated based on the cost required to achieve the
desired environmental condition. For each individual practice or combination of practices, the system
will include a method for estimating the costs of construction and O&M (Heaney et al. 2002).
A.2. Technical Requirements
Specific technical requirements were defined for the identified needs. For example, consideration of the
full set of indicators (hydrology, sediment, pollutants, and ecological impact) requires simulation of
dynamic hydrology and time-varying loads of sediment and pollutants, and potentially other ecological
indicators such as temperature or relationships to biological indicators. Evaluating the implications of
various configurations of management practices requires the ability to consider the performance of
individual and multiple practices and the sensitivity of each of those practices to their relative location in
the network. It is usually necessary to simulate longer time periods and storm sequences to demonstrate
response to a wide set of forcing conditions. The technical requirements for SUSTAIN are listed below.
• Simulate hydrologic response and a level of detail sufficient for analysis of a hydrograph (peak
flow and volume)
• Simulate multiple pollutant types, including nutrients (nitrogen and phosphorus), pathogens (fecal
coliform bacteria, Escherichia coli [E. coli]) and metals (e.g., zinc, aluminum)
• Simulate fate and transport of pollutants at a time step suitable for evaluating short-duration and
long-duration impacts consistent with evaluation of acute and chronic surface water criteria
• Simulate multiple size classes of sediment for input to management structures
• Simulate other habitat stressors, such as temperature
• Simulate in-stream dissolved oxygen based on inputs of biological oxygen demand, sediment
oxygen demand, nutrient loads, and other environmental factors
• Evaluate urban and mixed land uses, including pervious and impervious areas
• Consider a full range of management practices at a similar level of spatial resolution and
technical detail
• Consider distributed or small-scale upstream management practices, practices in series, and larger
downstream facilities
• Link watershed management to downstream measures of environmental conditions (e.g.,
dissolved oxygen in a river, nutrient concentration in a lake or estuary) outside the immediate
vicinity of a study area
Consequently, specific modeling procedures and algorithms were determined to fulfill the objectives of
SUSTAIN. For example, simulation of hydrologic response requires that the model support the
examination of rainfall/runoff processes at a level of detail sufficient to plot a time variable hydrograph
and/or a pollutograph. Supporting model applications at multiple scales is essential for the SUSTAIN
application. Scale may vary widely depending on the location and size of a watershed. The need to
provide a modular modeling system and multiple scale applications govern the software and system
designs.
A. 2.1. Spatial Scale
One dominant technical requirement of SUSTAIN is the ability to site management practices at multiple
scales. The way that BMPs are placed at different spatial levels, i.e. on-site, sub-regional, and regional
(Figure A-2), influences the overall cost-effectiveness of the storm water controls system (Zhen 2002). In
A-146
-------
an urban setting, examples of the on-site scale are building lots and neighborhoods with a drainage area of
less than 10 to 100 acres. The recently promoted LID technologies are normally applied on a micro or
on-site scale because the major design consideration is to retain and treat runoff near its source. Typical
BMPs used for LID include bioretention/rain gardens, rain barrels, filter strips, grass swales, infiltration
trenches, and detention or retention ponds. They operate at one point within a landscape, and treat runoff
from a certain drainage area. Other types of BMPs that are not necessarily associated with LID, such as
riparian buffers are linear by nature, and function by intersecting the landscape immediately adjacent to
streams. Area-based BMPs, such as reduced/disconnected imperviousness and street sweeping, represent
changes in human behavior and activity which may occur at many different scales.
Conventional BMPs collect runoff at hydrologic junctions farther downstream, at a level typically
associated with the sub-regional scale. The sub-regional scale or township-level drainage areas are on the
order of 100 to 5,000 acres. At this scale the benefits of management are often measured by the impact
on receiving streams, lakes, or other larger waterbodies. The regional scale, which is the largest
evaluation level, represents a county-level drainage area that is typically greater than 5,000 acres.
(a) On-site BMPs (b) Sub-regional BMPs (c) Regional BMPs
Figure A-2. BMP placement at various spatial levels: (a) on-site; (b) sub-regional; and (c)
regional.
The system may ultimately be applied in a tiered or nested application (Figure A-3). More detailed small
scale applications could be combined and evaluated on a larger scale to develop optimal solutions.
Various combinations of watersheds might be used to provide a manageable level of detail and maintain
computational efficiency. To address the technical requirement for multi-scale simulation, the landscape
modeling, which provides the hydrologic and water quality time series data for simulation of BMPs,
should be able to represent various spatial resolutions. The spatial and temporal resolution of SUSTAIN
also needs to vary according to the type, location, and spatial density of the BMPs evaluated. The model
needs to provide an unbiased evaluation of on-site, sub-regional, and regional BMPs to provide input
appropriate for optimization and comparative analysis of management plans.
A. 2.2. System and Modeling Requirements
From the defined technical requirements, modeling procedures or algorithms and system requirements
were identified. System requirements are organized into four areas: (1) operational system features, (2)
watershed/landscape simulation, (3) BMP simulation, and (4) stream conveyance simulation. While
evaluating the candidate modeling algorithms, some of the practical constraints, limitations, and
capabilities of each alternative were considered. Also considered were the simulation options and
flexibility of the application. Each system requirement category is described in more detail below.
Operational Requirements
SUSTAINmust provide a framework for long-term simulation of the landscape, management practices,
and hydrological system. The overall system provides the linkages between the land activities, the
management practices, and the stream or hydrologic network. The system must also provide the utilities
to support the placement and sizing of BMPs, developing watershed simulation networks that may
include sequences of land parcels, management practices, and stream reaches. Several operational
A-147
-------
requirements are placed on this system. For example, the system should operate at a short or variable
time step sufficient to represent hydrologic and pollutant loading pollutographs, typically 1 hour or less.
The system should support placement of BMPs of various types (i.e., linear stream buffers,
impoundments), calculation of the associated drainage area, and construction of networks of land uses,
BMPs, and streams or pipe conveyances. SUSTAIN should be configured to simulate small
subwatersheds or cells to a minimum size of approximately 1 acre. The system should be able to
represent larger complex watersheds by subdivided smaller subwatershed units. To provide
computational flexibility, the ability to define a mixture of larger and smaller units should be considered.
SUSTAIN should also have the ability link to other external models, either watershed models for inputs of
flow and pollutant time series or receiving water models. External linkage to receiving water models will
facilitate examination of downstream environmental condition. For example, an evaluation of
management scenarios to control nutrients in a watershed could be linked to a lake model for the purposes
of evaluating in-lake chlorophyll-a.
Two-Tier Analysis
Target Load
Reduction
Figure A-3. Tiered watershed application.
Watershed/Landscape Simulation
The watershed/land simulation includes the algorithms to process water, sediment, and pollutant routing
on the landscape. The technical requirements include a continuous simulation and small simulations time
steps. The algorithms to represent these processes must also be of sufficient detail to evaluate changes in
surface management and physical site characteristics that can be used as management variables. The
algorithms for the following simulations are needed to meet the technical requirements:
• Physically based infiltration simulation (e.g., Green-Ampt)
• Overland flow routing/hydrograph generation
• Pollutant accumulation and washoff
• Sediment detachment and transport
• Land-to-land flow routing
• Groundwater interaction
A-148
-------
BMP Simulation
A wide range of BMPs, both structural and nonstructural, needs to be evaluated by SUSTAIN. The
simulation methods must provide an unbiased evaluation of the effectiveness of BMPs. Nonstructural
management practices can include minimizing impervious areas, augmenting soil infiltration capacity
through lawn management, recycling of roof runoff (e.g., using rain barrels), and disconnecting
impervious surfaces (e.g., rain gutter outlets). Nonstructural management practices may also include
source controls such as minimizing or reducing fertilizer and pesticide applications. Nonstructural
practices can be evaluated by adjusting impervious areas, changing pollutant accumulation rates (e.g.,
changes in fertilizer application rates), or changing surface roughness characteristics (e.g., vegetative
management). Nonstructural practices are area-based since the dominant management is spatially
distributed.
Structural practices involve the placement and construction of a facility that captures and manages runoff
from a site. Structural practices are typically point-based, since they are in a specific location and
manage runoff captured from a defined drainage area. The typical practices use various combinations of
storage, infiltration, filtration, biological processes, and hydrologic separation to provide control of
hydrology and remove or reduce sediment and pollutants. Table A-3 provides a summary of the
dominant and secondary functional processes employed in various structural management practices.
Some management practices employ additional processes, identified as optional on the table, depending
on the specific design features and the site conditions. For example, a stormwater detention facility might
use infiltration as well as deposition/settling if the site has permeable soils with sufficient infiltration
capacity. Table A-3 shows that many practices use similar processes to achieve flow, sediment, and
water quality control. The table also identified the need for a management practice modeling system that
can simulate these key processes, including storage/detention, infiltration, filtration, biological
uptake/conversion, and hydrodynamic separation. The technical requirement to simulate these processes
supports the selection of the algorithms for simulation of BMPs.
The following specific capabilities are recommended:
• Process-based simulation of retention and detention types of management with, at a minimum,
first order decay and settling
• Time series simulation of point-based structural management practices that considers runoff
routing and hydrodynamic separation
• Area-based practices, including surface cover management, through the use of watershed/
landscape analysis
• Linear practices such as riparian buffers by routing surface and sub-surface runoff/pollutants from
one land unit to the next
Stream Conveyance Simulation
The stream routing and conveyance network component provides a linkage between
subwatershed/landscape units, management practices, and other direct discharges within an urban
watershed. The stream conveyance module is used to route runoff, sediment, and pollutants through a
stream network, which is often present in an urban watershed. The rigor of simulation for the stream
portion is related to the dominant processes present in urban streams. Key features include settling,
resuspension, and decay (i.e., fecal coliform) and changes in the stream channel (i.e., stream bank erosion
or degradation). Therefore, during conveyance in a stream, the module should consider settling,
resuspension, and decay processes. Accounting for stream bank erosion should be considered as an
option as well. Larger waterbodies, including rivers, lakes, and tidal waters might require more detailed
simulation of chemical and biological processes. These systems can best be simulated through external
A-149
-------
linkage to several comprehensive receiving water models such as Environmental Fluid Dynamics Code
(EFDC; Hamrick 1992) and WASP (Wool et al. 2003).
Table A-3. Types of Structural BMPs and Major Processes
Structural BMP Types
Dry Extended Detention
Pond
Wet Retention Pond
Shallow Marsh
Extended Detention
Wetland
Submerged Gravel
Wetland
Organic Filter
Sand Filter
Bioretention
Infiltration Trench
Porous Pavement
Dry Swale
Wet Swale
Buffer Strip
Baffle Box
Inlet Devices
Oil-Grit Separator
Storage
Detention
+
+
+
+
+
0
0
0
0
-
0
0
-
+
-
+
Infiltration
(o)
(o)
(o)
(o)
(o)
(+)
(+)
(+)
+
+
(o)
(o)
+
-
-
-
Filtration
-
-
-
-
+
+
+
+
(o)
(o)
-
-
(o)
-
+
-
Biological
Uptake and
Conversion
-
0
+
0
+
0
0
+
0
-
-
0
0
-
(o)
-
Structure-
Facilitated
Hydrodynamic
Separation
-
(o)
(o)
(o)
-
-
-
-
-
-
-
-
-
+
(+)
+
Note: () optional function; + major function; o secondary function; - insignificant function
Definitions of the process groupings:
! Storage detention: detaining water
! Infiltration: infiltrating water to the ground
! Filtration: passing water through a porous medium
! Biological uptake and conversion: reducing nutrients and other pollutant as aquatic plants and microorganisms use them
for growth
! Structure-facilitated hydrodynamic separation that considers physical design features: separating insoluble pollutants
(solids, oil, and floatables) by introducing physical or hydrodynamic forces, e.g. baffles, whirlpool effect
A-150
-------
Appendix B. Model Evaluation and Selection
B.I. Introduction
This appendix provides a summary of the targeted evaluation and selection of public-domain software in
accordance with the design requirements of SUSTAIN. Currently available models and modeling
frameworks were identified and evaluated according to their technical capabilities and software systems.
The review effort focused on identifying key models that addressed one or more of the needed algorithms
or analysis methods. The purpose of the review was to identify candidate models or portions of models
for integration or adaptation into SUSTAIN.
B.2. Overview of Available Models
The review of available models followed a structured process based on the results of the technical needs
analysis. Generally the review focused on publicly available models and modeling systems, although
proprietary models that have been published and have relevant capabilities were included for comparative
purposes. The following are some example considerations in the selection of available models for review:
• Is the model in the public domain, and how easily adaptable, current, and available is the source
code for the model?
• Is the model well established with an extensive application history and record?
• Is the model appropriate for small to mid-size urban watersheds?
• How rigorous are its algorithms in simulating watershed processes?
• Is the model relevant for pollutants present in urban areas?
• Does the system include interface capabilities or linkages that could be relevant to the SUSTAIN
design?
The selection of models for review focused on identifying models that could have relevance to one or
more areas of SUSTAIN. For this reason some models with specialized features that are not typically used
in urban environments (e.g., WAMView [SWET 2002]) were included. The emphasis was on selection
of models that are in the public domain or are available for distribution without charge. Other proprietary
models with limited information on model algorithms and documentation were excluded from the
analysis (i.e., Mike-SHE, MOUSE [DHI Inc. Web site]). The set of models selected for review is listed in
Table B-l and profiles for each model are provided in Section B.3. A distinct set of evaluation factors
was developed for watershed models, BMP systems, and interface and software platforms.
B-151
-------
Table B-l. Available Models Reviewed
Watershed Models
SWMM, HSPF, LSPC, WAMview, WARMF, SLAMM, P8 UCM,
ANSWERS, CASC2D, KINEROS, WEPP, DR3M-QUAL, SWAT,
AnnAGNPS, AGNPS, GWLF
Systems: BASINS, EPA TMDL Toolbox
BMP Models
Prince George's County BMP Module, P8
UCM, VFSMOD, MUSIC, DMSTA,
SWMM, BMP AM
Systems: LIFE
B. 2.1. Watershed Model Evaluation Factors
The following factors were identified for evaluation of available watershed models. The evaluation
results are summarized in Table B-2. These factors are closely aligned with general modeling
considerations and the four major categories of simulation needs (i.e., land, reach, conduit, and BMP).
• At what spatial scale (cell, field, catchment, subwatershed, or watershed) is the modeling
application most suitable?
• At what time scale (continuous or event-based) is the simulation performed, and what is the
minimum applicable computation time step?
• What land uses (urban and nonurban) can be simulated? Are point sources addressed?
• How rigorous are its algorithms for hydrology simulation, how is the rainfall-runoff simulation
performed, and is groundwater interaction included?
• How rigorous are its algorithms at water quality (pollutant loading) simulation? How does it
address sediment, nutrients, and other pollutant loading generation, transport, and transformation,
if included?
In landscape or watershed models, an essential feature is how the area is segmented. For evaluation
purposes segmentation was defined as four distinct options:
• Catchment (CM): Capable of simulating multiple watersheds and subwatersheds
• Cell: Watershed area represented as a network of cells. Flow is routed from cell to cell
• Field: Limited to a small single simulation unit, typically a field or monitoring plot
• Watershed (Wsh): Limited to single watershed for each model simulation
B. 2.2. BMP Technical Evaluation Factors
The following factors were considered in BMP model evaluation as summarized in Table B-3.
• What types of BMPs can be addressed?
• What pollutant removal processes and mechanisms are simulated?
• What algorithms are applied for flow routing and pollutant removal process simulation?
• What water quality constituents can be simulated?
B. 2.3. Model Interface Evaluation Factors
The model interface features of the models were evaluated, using the following factors:
• What GIS features, if any, are incorporated?
B-152
-------
• How is the subwatershed/channel network represented?
• Data management utilities
• Model code
• Interface code
Table B-4 and Table B-5 contain model interface evaluations for watershed models.
B.3. Evaluation and Review of Available Models
This section provides an evaluation and review of available models. The models reviewed are organized
into three groups—landscape models, BMP models, and comprehensive modeling systems. Within each
group, the models are sequenced based on their expected relevance to urban management analysis.
Regardless of their position in the sequence, all models reviewed might have specific features that could
prove useful for the development of SUSTAIN. A narrative discussion is provided below for each model,
including key features, capabilities, special techniques, and software capabilities. The narrative
description supports earlier summary tables (Table B-2 to Table B-5). Section B.4 provides further
discussion of the strengths and weaknesses of the reviewed models and identifies the models for
integration into the SUSTAIN design.
B. 3.1. Landscape Model Reviews
Landscape models are models that simulate land-based hydrology and water quality, and provide
sediment and pollutant loading estimates. Many of these models also incorporate some of the features of
BMP models (i.e., simulation of various management practices) and stream conveyance systems. The
following landscape models were reviewed for potential integration of components and interface with
SUSTAIN.
SWMM
The Stormwater Management Model (SWMM) is a dynamic rainfall-runoff simulation model developed
by EPA and primarily applied to urban areas, for single-event or long-term (continuous) simulation using
various time steps (Huber and Dickinson 1988). It was developed for the analysis of surface runoff and
flow routing through complex urban sewer systems. The last official version was 4.4h. SWMM5 is a
completely revised and updated release of SWMM. However, SWMM5 will continue to be expanded
with new functions, particularly a quality routine.
In SWMM, flow routing is performed for surface and subsurface conveyance and groundwater systems,
including the options of nonlinear reservoir channel routing and fully dynamic hydraulic flow routing.
By choosing the fully dynamic hydraulic flow routing option, SWMM can simulate backwater,
surcharging, pressure flow, and looped connections. SWMM has a variety of options for quality
simulation, including the traditional buildup and washoff formulation, as well as rating curves and
regression techniques. The Universal Soil Loss Equation (USLE) is included to simulate soil erosion.
SWMM incorporates first-order decay and particle-settling mechanisms in pollutant transport simulations,
including the option of a simple scour-deposition routine in conduits. Storage, treatment, and other BMPs
can also be simulated. A more detailed description of its BMP simulation capabilities is provided in the
next section.
B-153
-------
Table B-2. Watershed Model Evaluation Summary
Criteria
Land Uses
Time Scale
Hydrology
Pollutant Loading
Pollutant Routing
Urban
Rural
Point Sources
Continuous
Single Event
Time Step
Runoff
Flow Routing
Base/low
Sediment
Nutrients
Others
Transport
Transformation
Operation Unit
Public Domain
SWMM
•
4
•
•
•
V
•
•
•
•
•
•
4
O
CM/Cel
1
Y
to
a.
!/5
M
•
•
•
•
•
V
•
4
•
•
•
•
•
•
C
M
Y
U
a.
!/5
-J
•
•
•
•
•
V
•
4
•
•
•
•
•
•
C
M
Y
WAMview
4
•
4
•
•
V
•
4
•
•
•
•
<
<
CM/Cel
1
N
WARMF
•
•
•
•
V
4
•
•
•
•
•
CM
N
SLAMM
•
-
•
-
V
O
O
•
•
4
-
CM
Y
s
u
P
90
a.
•
-
•
Hou
r
O
O
•
•
0
-
CM
Y
ANSWERS
-
-
V
4
-
•
-
4
-
Cel
1
Y
CASC2D
-
-
V
•
•
-
-
•
-
Cel
1
Y
KINEROS
-
-
-
V
4
-
-
-
•
-
Fiel
d
Y
£
W
£
-
-
•
V
4
-
-
-
4
-
Fiel
d
Y
DR3M-QUAL
•
•
•
-
V
4
O
-
•
-
CM
Y
H
<
£
in
0
•
•
Day
4
•
•
•
4
HR
U
Y
AnnAGNPS
-
•
•
-
Da
y
O
-
•
•
-
CM
Y
AGNPS
-
•
-
•
Even
t
O
-
-
•
-
Cell
Y
a
§
•
4
•
-
Day
-
O
-
0
-
Ws
h
Y
Sources: LSPC (Tetra Tech 2002), WAMview (SWET 2002), WARMF (Chen et al. 1999; Chen et al. 2001 ;Weintraub et al. 2001), P8 UCM (Walker 1990), ANSWERS (Bouraoui et al.
1993), CASC2D (Ogden 2001), KINEROS (USDA2003; Woolhiser et al. 1990), WEPP (Flanagan and Nearing 1995), DR3M-QUAL (Alley etal. 1982a; Alley etal. 1982b), SWAT
(Neitsch et al. 2001), AnnAGNPS (AnnAGNPS 2000), AGNPS (Young et al. 1986), GWLF (Haith et al. 1992).
•High *M edium °Low - Not Incorporated
1 Ongoing work links WASP with the model
V Variable simulation time step
CM = Catchment: Capable of simulating multiple watersheds and subwatersheds.
Cell = Watershed area is represented as a network of cells. Flow is routed from cell to cell.
Field = Limited to small, single simulation unit, typically a field or monitoring plot.
Wsh = Watershed: Limited to single watershed simulation.
B-154
-------
Table B-3. Summary of BMP Models and Capabilities
Model
Prince
George's
County BMP
Module
P8UCM
VFSMOD
DMSTA
MUSIC
SWMM
WETLAND
Types of BMP
Detention basin
Infiltration practices
(e.g., infiltration trench,
dry well, porous
pavement)
Vegetative practices
(e.g., wetland, swale,
filter strip, bioretention)
Detention basin
Infiltration practices
Swale/buffer strip
Manhole/splitter
Vegetative filter strip
Wetland
Detention basin
Detention basin
Infiltration practices
Vegetative practices
Detention basin
Infiltration practices
Detention basin
Wetland
Processes/
Mechanisms
Storage
Infiltration
Overflow/outlet flow
Decay process
Soil media pollutant
removal
Storage
Infiltration
Overflow/outlet flow
Settling/decay
Infiltration
Overland flow routing
Sediment transport
Storage
Seepage (in & out)
Evapotranspiration
Phosphorus cycle
Storage
Infiltration
Decay
Infiltration
Sedimentation
First-order decay
Storage
Infiltration
Nutrients cycling (C,
N,P)
Sediment deposition,
resuspension,
decomposition.
Dissolved oxygen
influx
Microbial and
vegetative activities
(growth and death)
Algorithms
Storage routing
Holtan's equation
Weir/orifice flow
First-order decay
Linear reservoir
Green- Ampt method
Second-order decay
Particle removal scale
factor
Green- Ampt method
Kinematic wave
University of
Kentucky algorithm
Storage-stage
CSTR in series
Dynamic phosphorus
cycling
CSTR in series
First order decay (k1-
C* model)
Horton and Green-
Ampt methods
Camp's theory for
quiescent condition
and Chen for
turbulence
Water budget
ET: Pan data or
Thornthwaite's
method
Monod kinetics
Constant vegetative
growth rate
Freundlich isotherms
forP
sorption/desorption
First-order
mineralization
Water Quality
Constituents
User-defined
pollutants
Sediment
User-defined
pollutants
Sediment
Phosphorus
User-defined
pollutants
User-defined
pollutants
Nitrogen
Phosphorous
Carbon
DO
Sediment
Bacteria
Reference
Prince
George's
County
(2001)
Walker
(1990)
Munoz-
Carpena and
Parsons
(2003)
Kadlec and
Walker
(2003)
Wong (2002)
Huber and
Dickenson
(1988)
Lee (1999)
Lee et al.
(2002)
B-155
-------
Table B-3. (Continued)
Model
VAFSWM
REMM
Types of
BMP
Detention
basin
Wetland
Vegetative
buffer strip
Processes/ Mechanisms
Storage
Infiltration
Particle settling
Adsorption to plant and
substrate
Vegetative uptake
Infiltration
Evaportransporation
Surface and subsurface flow
routing
Nutrients cycling (C, N, P)
Erosion
Sediment transport
Algorithms
Water budget
ET: user specified rate
CSTR in series
First-order kinetics
(adsorption, plant uptake)
Green- Ampt equation
ET: modified Penman
Monteith equation, and Darcy
Buckingham equation
Storage routing
Darcy 's equation
Nutrient cycling: Century
Model
Nitrification: First-order
Weir/orifice flow
Erosion: USLE
Sediment transport: Einstein
and Bagnold equations
Water Quality
Constituents
User-defined
pollutants
Sediment
Sediment
Nutrients (C,
N,P)
Reference
Yu, Fitch
and Earles
(1998)
SEWRL,
USDA-
ARS
(1999)
SWMM has been applied to address various urban water quantity and quality problems in many locations in
the United States and other countries (Donigian and Huber 1991; Huber 1992). In addition to its use in
developing comprehensive watershed-scale planning, typical uses of SWMM include predicting CSOs,
assessing the effectiveness of BMPs, and providing time series input to dynamic receiving water quality
models (Donigian and Huber 1991.)
HSPF
Hydrological Simulation Program-FORTRAN (HSPF) is a comprehensive package developed by EPA for
simulation of watershed hydrology and water quality for both conventional and toxic organic pollutants
(Bicknell et al. 1997). This model can simulate the hydrologic and associated water quality processes on
pervious and impervious land surfaces and in streams and well-mixed impoundments. HSPF incorporates the
Agricultural Runoff Management (ARM) model and Nonpoint Source Runoff (NPS) model into a watershed
analysis framework that includes fate and transport in one-dimensional stream channels. It allows the
integrated simulation of land and soil contaminant runoff processes with in-stream hydraulic and sediment-
chemical interactions. The result of this simulation is a time history of the runoff flow rate, sediment load, and
nutrient and pesticide concentrations, along with a time history of water quantity and quality at any point in a
watershed.
HSPF simulates three sediment types (sand, silt, and clay) in addition to a single organic chemical and
transformation products of that chemical. Further, the in-stream model assumes that the receiving waterbody
is well mixed with width and depth and is thus limited to well-mixed rivers and reservoirs. The transformation
and reaction processes include hydrolysis, oxidation, photolysis, biodegradation, volatilization, and sorption.
Sorption is modeled as a first-order kinetic process in which the user must specify an adsorption and
desorption rate and an equilibrium partition coefficient for each of the three solids types. Resuspension and
settling of silts and clays (cohesive solids) are defined in terms of shear stress at the sediment water interface.
The model computes the capacity of the system to transport sand at a particular flow. Settling and/or scouring
are defined by the difference between the sediment load in suspension and the transport capacity.
B-156
-------
Table B-4. Watershed Model Interface Evaluation Summary
Features
CIS for Setup
Data
Management
Network
Interface Code
Model Code
SWMM
N/A
v4: Text files
v5: Database
Text files
v4: Table
v5: Graphical
C
v4: FORTRAN
v5:C
HSPF
Arc View
(BASINS)
WDM
Graphical
(BASINS)
Avenue
FORTRAN
LSPC
Arc View
Access
Textfiies
Graphical
VB/Avenue
C++
WAMview
Arc View
Database
Text files
Graphical
Avenue
FORTRAN,
VB
WARMF
Arc View
Database
Text files
Graphical
VB
FORTRAN
V
SLAMM
N/A
Text files
N/A
VB
B
P8UCM
N/A
Text files
Table
FORTRAN
FORTRAN
ANSWERS
GRASS
Database
Text files
Graphical
AML
FORTRAN
Table B-5. Watershed Model Interface Evaluation Summary
Features
CIS for Setup
Data
Management
Network
Interface Code
Model Code
CASC2D
N/A
Text files
CIS
Text files
N/A
C
KINEROS
N/A
Text files
Text file
N/A
FORTRAN
WEPP
N/A
Access
Textfiies
N/A
VB
FORTRAN
DR3M-QUAL
N/A
WDM
Text files
Text file
N/A
FORTRAN
SWAT
Arc View
dBASE
Text files
CIS
Avenue
FORTRAN
AnnAGNP
S
Arc View
Access
Text files
CIS
VB
Avenue
FORTRAN
AGNPS
N/A
Text files
Graphical
FORTRAN
V
FORTRAN
GWLF
AVGWLF
Tt
Extension
Text files
Graphical
3 Avenue
VB
Sources: LSPC (Tetra Tech 2002), WAMview (SWET.2002), WARMF (Chen et al. 1999; Chen et al. 2001 ;Weintraub et al.2oo 1)> P8 UCM (Walker 1990), ANSWERS (Bouraoui et al.
1993), CASC2D (Ogden 2001), KINEROS ( USDA 2003; Woolhiser et al. 1990), WEPP ( Flanagan and Nearing 1995), DR3M-QUAL (Alley etal. 1982a; Alley etal. 1982b), SWAT
(Neitsch et al. 2001), AnnAGNPS (AnnAGNPS 2000), AGNPS (Young et al. 1986), GWLF (Haith et al. 1992).
BASINS-Better Assessment Science Integrating Point and Nonpoint Sources, WDM-Watershed Data Management, VB-Visual Basic, AML-ARC Macro Language, SWMM v4-
Version 4.0, SWMM vS-SWMM Version 5.0.
B-157
-------
The model has been extensively used for both screening-level and detailed analysis. The Chesapeake Bay
Program used HSPF to model total watershed contributions of flow, sediment, nutrients, and associated
constituents to the tidal region of the bay (Donigian et al. 1990, Donigian and Patwardhan 1992). Moore
et al. (1992) describe an application to model BMP effects on a Tennessee watershed. Scheckenberger
and Kennedy (1994) discuss how HSPF can be used in subwatershed planning. Donigian et al. (1996)
describe the use of HSPF to identify and quantify the relative pollutant contributions from both point and
nonpoint sources and to evaluate agricultural BMPs for the LeSueur Basin of southern Minnesota.
LSPC
The Loading Simulation Program in C++ (LSPC) is a watershed modeling system that includes
streamlined HSPF algorithms for simulating hydrology, sediment, and general water quality on land, as
well as a simplified stream transport model (Tetra Tech. and USEPA 2002). The model, based on the
Mining and Data Analysis System (MDAS) methodology, was specifically developed to handle large,
complex watersheds (with 1,000 or more subwatersheds) and to support TMDL development for such
cases. The key advantage of LSPC is that it has no inherent limitations in terms of modeling size or
model operations. In addition, the Microsoft Visual C++ programming architecture allows for seamless
integration with modern-day, widely available software such as Microsoft Access and Excel.
This dynamic watershed model provides the linkage between source contributions and in-stream
response. It is used to simulate watershed hydrology and pollutant generation and transport, as well as
stream hydraulics and in-stream water quality. LSPC is capable of simulating flow, sediment, metals,
nutrients, pesticides, and other conventional pollutants, as well as temperature for both pervious and
impervious lands. The reach routing module also simulates fate and transport of these pollutants through
a stream network. Table B-6 lists the HSPF modules that are currently supported in the LSPC watershed
model.
In addition to LSPC's data management and programming platform features, the model was also designed
with specific tools to support and assist in the development of TMDLs for areas affected by nonpoint
and/or point sources. The TMDL tools allows for evaluation of land use-level and point source-level
loads, evaluation of load reduction options, and comparison of baseline versus alternative scenario results.
Table B-6. HSPF Modules Supported in the LSPC Watershed Model
Simulation
Type
Land-based
processes
In-stream
processes
HSPF Module
PWATER
IWATER
SNOW
SEDMNT
PWTGAS
IQUAL
PQUAL
HYDR ADCALC
CONS
HTRCH
SEDTRN
GQUAL
HSPF Module Description
Water budget for pervious land
Water budget for impervious land
Incorporates snowfall and snowmelt into water budget
Production and removal of sediment
Est. water temperature, dissolved gas concentrations
Simple relationships with solids and water yield
Simple relationships with sediment and water yield
Hydraulic behavior, pollutant transport
Conservative constituents
Heat exchange, water temperature
Behavior of inorganic sediment
Generalized quality constituent
B-158
-------
WAMView
Watershed Assessment Model (WAM) is a GIS-based model that allows engineers and land use planners
to interactively simulate and assess the environmental effects of various land use changes and associated
land use practices (SWET 2002). WAM was originally developed with an Arc/Info interface for the
entire Suwannee River Water Management District (SRWMD; 19,400 km2 of northern Florida) and has
since been customized for St. Johns River Water Management District (SJRWMD) in northeast Florida to
accommodate its special regional and geological characteristics. The SJRWMD version includes an
Arc View interface, and thus it is called WAMView. WAMView provides hourly time series of flow,
TSS, and nutrients for all the contributing watersheds. The simulated hydrologic parameters include
source cell surface and groundwater flow, and stream reach daily flow; simulated water quality
parameters are suspended solids, sediment N, sediment P, soluble N, soluble P, BOD, bacteria, and toxics.
The model provides water quality daily outputs at source cells, subbasins, and stream reaches. An effort
is under way to link WAMView to the WASP model.
The water quality assessments are accomplished using two methods. The first method provides spatial
assessment using impact indices, and the second uses detailed hydrologic and water quality transport
modeling. The method used depends on the watershed assessment parameter of interest. The indexing
approach is used for parameters that are hard to quantify and that are also directly associated with
pollutant transport, while the modeling approach addresses the major pollutants of sediment and nutrients.
Both approaches provide outputs at the source cell, sub-basin, and basin outlet levels. Both approaches
use the watershed characteristic data from existing GIS coverage to determine the appropriate input data
(indices for index approach and model parameter sets for the modeling approach). These data are used to
calculate the combined impact of all watershed characteristics for a given grid cell. Once the combined
impact at each unique cell within a watershed is determined, the cumulative impact for the entire
watershed is determined by first attenuating the constituent to the subbasin outlets and then calculating an
area-weighted ranking/index at the attenuated load generated at each cell. Constituents are attenuated
based on the flow distances (overland flow route to nearest waterbody, through wetlands or depressions,
and within streams to the subbasin outlet), flow rates in each related flow path, and types of wetlands or
depression encountered. The contaminant transport modeling is accomplished by first simulating all the
unique grid cell combinations of land use, soils, and rain zone by using a unique cell model that contains
several source cell models, including GLEAMS (Knisel 1993), EAAMOD (SWET 1999), a wetland
module, and an urban module. The unique cell model, also called the BUCSHELL (Basin Unique Cell
Shell Program) model, operates on square grid cells with a typical size of 1 hectare (100 m x 100 m). The
cell model simulates the daily flow and constituents from each unique cell within the watershed using one
of the four submodels unique to WAMView, e.g., GLEAMS, EAAMOD, URBAN, and WETLAND,
depending on land uses and soil. The time series outputs for each grid cell are routed and attenuated to
the nearest stream and then routed through the stream using WAMView's BLASROUTE (Basin Land
and Stream Routing) module. The BLASROUTE module predicts flow, stage, and water quality. It
routes through a stream network with attenuation, also routes through depression and wetlands. The
model uses linear reservoir flow routing, and applies attenuation based on flow rate, characteristics of
flow path, and flow distance. It also allows outlet stage and concentration definition with backflow.
WAMView is limited for SUSTAIN because of its development and application emphasis on rural areas.
However, the cell-based representation and model configuration process provide potential benefits for
assessing the localized loading and spatial implications in the placement of BMPs.
WARMF
WARMF (Watershed Analysis Risk Management Framework) was developed by Systech Engineering,
Inc., as a decision support system for calculating TMDLs (Chen 1999). The GIS map-based tool contains
five interconnected modules: Engineering, Data, Knowledge, TMDL, and Consensus. In WARMF, a
B-159
-------
watershed is divided into a network of land catchments, stream segments, and stratified lakes. The
engineering module is a dynamic watershed simulation model that calculates daily runoff, nonpoint
source loads, groundwater flow, and hydrology and water quality of river segments and stratified
reservoirs. The data module contains meteorological, air quality, point source, reservoir release, and flow
diversion data. The nonpoint source loads are routed together with point source loads to predict water
quality in rivers and lakes. The simulation models embedded in the WARMF engineering module were
adapted from well-established simulation codes. The main computing engine was taken from the
Integrated Lake-Watershed Acidification Study (ILWAS) model. The ILWAS model divides a watershed
into land catchments, stream segments, and lake layers. Land catchments are further divided into canopy
and soil layers. These watershed compartments are connected to form a network for hydrologic and water
quality simulations.
The hydrologic model simulates the processes of canopy interception, snowpack accumulation and
snowmelt, infiltration through soil layers, evapotranspiration from soil, exfiltration of groundwater to
stream segments, kinematic wave routing of stream flows, and flow routing of reservoirs. Such detailed
simulations track the flow paths of precipitation from canopy through soil layers and streams to lakes.
Along each flow path, the chemistry module performs mass balance and chemical equilibrium
calculations to account for the processes of dry deposition to the canopy, nitrification of ammonia on the
canopy, ion leaching from sap to the canopy surface, washoff by through-fall, ion leaching by snowmelt,
and the soil processes, e.g., litter fall, litter breakdown, litter decay, nitrification, denitrification, cation
exchange, anion adsorption, weathering, and nutrient uptake.
The algorithms of WARMF were derived from many available codes. Algorithms for snow hydrology,
groundwater hydrology, river hydrology, lake dynamics, and mass balance for acid base chemistry were
based on the ILWAS model. Algorithms for erosion, deposition, resuspension, and transport of sediment
were adapted and modified from ANSWERS. The pollutant accumulation on land surface was modified
from SWMM. Instead of using export coefficients, an algorithm for mixing and washoff was used to
simulate the processes that generate nonpoint source loading. The first-order decay of coliforms and
BOD and its impact on dissolved oxygen follow the techniques used in traditional water quality models.
The sediment adsorption-desorption of pesticides and phosphorus and the kinetics of nutrients and algal
dynamics were adapted from WASPS.
WARMF provides step-by-step roadmaps for calculating TMDLs and for building consensus. WARMF
also offers GIS-generated maps, tables, and graphing capabilities. In addition, the costs/benefits of
pollutant trading, stakeholders, alternative ranking, and the nominal scores of rankings are calculated at
the watershed scale. These tools can be used for management analysis at the watershed scale. Support
for site-scale, land-use-specific, and subwatershed-level analyses is limited.
The major limitation of WARMF is that it is not a public domain model. WARMF is also oriented to
rural land areas. The management and alternatives analysis is limited to watersheds, and simulation of
multiple levels of controls by subwatershed/land use requires repeated simulation. The strength of
WARMF is detailed representation of chemical processes, especially with respect to metals and pH.
SLAMM
The Source Loading and Management Model (SLAMM) was originally developed to better understand
the relationships between sources of urban pollutants and runoff quality (Pitt 1993). SLAMM is strongly
based on actual field observations, with minimal reliance on pure theoretical processes that have not been
adequately documented or confirmed in the field. It has been continually expanded since the late 1970s
and now includes a wide variety of source area and outfall control practices (infiltration practices, wet
detention ponds, porous pavement, street cleaning, catch basin cleaning, and grass swales). Beginning
with version 5, SLAMM is Windows-based and thus is called WinSLAMM.
B-160
-------
The model performs continuous mass balances for particulates and dissolved pollutants and for runoff
volumes. Runoff is calculated by a method developed by Pitt (1987) for small-storm hydrology. Runoff
is based on rainfall minus initial abstraction, and infiltration is calculated for both impervious and
pervious areas. Triangular hydrographs, parameterized by a statistical approach, are used to simulate
flow. Exponential buildup and rain washoff, as well as wind removal functions, are used in computing
runoff pollutant loadings. Water and sediment from various source areas are tracked as they are routed
through treatment devices. SLAMM is mostly used as a planning tool to better understand sources of
urban runoff pollutants and the effectiveness of their control.
SLAMM is capable of considering many stormwater controls that affect source areas, drainage systems,
and outfalls, for a long series of rainfall events. The program considers how particulates filter or settle
out in control devices. Particulate removal is calculated based on the structural design characteristics.
Storage and overflow of devices are also considered. At the outfall locations, the characteristics of the
source areas are used to determine pollutant loads in solid and dissolved phases. Another ability of
SLAMM is to accurately describe a drainage area in sufficient detail for water quality investigations, but
without requiring a great deal of superfluous information that field studies have shown to be of little value
in accurately predicting discharge results. SLAMM also applies stochastic analysis procedures to more
accurately represent actual uncertainty in model input parameters to better predict the actual range of
outfall conditions (especially pollutant concentrations). Like all stormwater models, SLAMM needs to be
accurately calibrated and then tested (verified) as part of any local stormwater management effort.
The major limitation of SLAMM is that it is strongly based on a statistical approach that uses the current
available field observations; therefore, it is not a process-based model. Some of the key features of the
model have potential for incorporation into SUSTAIN. For instance, the algorithms and data used for
addressing source control could be applied to SUSTAIN.
ANSWERS
The Areal Nonpoint Source Watershed Environment Response Simulation (ANSWERS) model is a
comprehensive model developed to evaluate the effects of land use, management schemes, and
conservation practices or structures on the quantity and quality of water from agricultural or rural
watersheds (Beasley 1986). It was among the first generation of distributed watershed models, which
allow for a better analysis of spatial as well as temporal variabilities of pollutant sources and loads. It was
initially developed on a storm event basis to enhance the physical description of erosion and sediment
transport processes in agricultural watersheds. Data preparation for ANSWERS is rather complex,
especially when watersheds are large. The output routines, however, are quite flexible and results are
available in several tabular and graphical forms. The program has been used to evaluate management
practices for agricultural watersheds and construction sites primarily in Indiana. It has been combined
with extensive monitoring programs to evaluate the relative importance of point and nonpoint source
contributions to Saginaw Bay in Michigan. This application involved the computation of unit area
loadings under different land use scenarios for evaluation of the tradeoffs between load allocations (LAs)
and wasteload allocations (WLAs). Recent model revisions include improvements to the nutrient
transport and transformation subroutines (Dillaha et al. 1988). Bouraoui et al. (1993) describe the
development of a continuous simulation version of the model.
The main limitation of ANSWERS is its emphasis on erosion and sediment transport in rural areas, which
are not tested for primarily urban areas.
CASC2D
The Cascade 2 Dimensional (CASC2D) sediment model is a fully unsteady, physically based, distributed-
parameter, square-grid, two-dimensional, infiltration-excess (Hortonian) hydrologic model for simulating
the response of a watershed subject to rainfall (Ogden 2001). Major processes simulated include
B-161
-------
continuous soil-moisture accounting, rainfall interception, infiltration, surface and channel runoff routing,
soil erosion, and sediment transport. Raster (square grid) is the computational unit. CAS2D allows the
user to select a grid size (typically 30-200 m) that appropriately describes the spatial variability in all
watershed characteristics. CASC2D is physically based and solves the equations of conservation of mass
and energy to determine the timing and path of runoff in the watershed. CAS2D applies Green and Ampt
with or without a redistribution method for infiltration simulation; an explicit finite-difference, two-
dimensional, diffusive-wave method for overland flow routing; and options of explicit one-dimensional,
diffusive-wave or implicit dynamic-wave channel routing. The empirical Kilinc and Richardson (1973)
soil erosion model as modified by Julien (1995) is applied in CASC2D to determine the sediment
transport from one overland flow grid cell to the next. CASC2D employs Yangs' (1973) method to
routing sand-size sediment in stream channels. Silt and clay size sediment are assumed to be transported
with flow; deposition or erosion of silt and clay within the channels is neglected (Ogden 1998). The
physically based distributed model is superior in simulation of runoff process at small scales within the
watershed. As a spatially distributed model, CASC2D offers the capability of determining the value of
any hydrologic variable at any grid point in the watershed at the expense of requiring significantly more
input than traditional approaches. CASC2D can accept spatially varied hydrologic parameter input or
rainfall input; however, because of the extensive data amounts required, data uncertainty may result in a
non-unique calibration.
CASC2D development was initiated in 1989 at the Center for Excellence in Geosciences at Colorado
State University funded by the United States Army Research Office (ARO). The original version of
CASC2D has been significantly enhanced under funding from ARO and the U.S. Army Corps of
Engineers Waterways Experiment Station (USACEWES). USACEWES has selected CASC2D as its
premier two-dimensional surface water hydrologic model. CASC2D is also one of the surface-water
hydrologic models supported by the Watershed Modeling System (WMS), developed at Brigham Young
University. The GRASS GIS developed by the U.S. Army Construction Engineering Research
Laboratories can be used in the preparation of CASC2D data sets.
The limitations of CASC2D are as follows:
• CASC2D is a fully distributed, physically based, state-of-the-art hydrologic model, but with the
exception of sediment, it does not have an integrated water quality component
• Because the program uses a distributed scheme and physically based algorithms, application
requires extensive input data preparation and calibration
KINEROS
The Kinematic Runoff and Erosion (KINEROS) model is an event-oriented, physically based model that
describes the processes of interception, infiltration, surface runoff, and erosion from small agricultural
and urban watersheds (USDA 2003). The model represents a watershed by a sequence of planes and
channels and solves the partial differential equations describing overland flow, channel flow, erosion, and
sediment transport by using finite-difference techniques. The spatial variations of rainfall, infiltration,
runoff, and erosion parameters can be accommodated. KINEROS can be used to determine the effects of
various artificial features, such as urban developments, small detention reservoirs, or lined channels on
flood hydrographs and sediment yield. This model is suitable for small agricultural and disturbed urban
watersheds.
The following are the limitations of KINEROS:
• It is an event-based model
• It is primarily designed for small agricultural and disturbed urban areas
B-162
-------
• It simulates only sediment
WEPP
Developed by the U.S. Department of Agriculture's (USDA) Agricultural Research Service (ARS), the
Water Erosion Prediction Project (WEPP) model is a distributed-parameter, continuous-simulation model
developed to provide a new generation of soil erosion prediction technology (USDA NSERL 1995). The
model requires inputs for rainfall amounts and intensity; soil textural qualities; plant growth parameters;
residue decomposition parameters; effects of tillage implements on soil properties and residue amounts;
slope shape, steepness, and orientation; and soil erodibility parameters. Parameters used for predicting
erosion, including soil roughness, surface residue cover, canopy height, canopy cover, and soil moisture,
are updated daily. The basic output from WEPP consists of runoff and erosion summary information,
which can be produced on a storm-by-storm, monthly, annual, or average annual basis. The model output
files contain time-integrated estimates of runoff, erosion, sediment delivery, and sediment enrichment, as
well as the spatial distribution of erosion.
The limitations of WEPP are as follows:
• The emphasis of this model is on erosion and sediment simulation from pervious land areas;
therefore, it has limited applicability for evaluation of urban areas with significant impervious
areas
• The model simulates only sediment
SWAT
SWAT is a continuous-time, physically based river basin or watershed-scale model developed by the
USDA's ARS (USDA ARS, SWAT Web site) for agricultural watersheds. SWAT was developed to
predict the impact of agricultural land management practices on water, sediment, and agricultural
chemical yields in large, complex watersheds with varying soil, land use, and management conditions
over long periods of time using readily available inputs. The major components of SWAT are hydrology,
weather, erosion, soil temperature, crop growth, nutrients, pesticides, and agriculture management. A
flow routing component transports flow and loading from each subwatershed across subsequent
watersheds and allows for accumulation of subwatershed contributions. Model inputs are based on
geographic units comprising unique land use and soil characteristics. The SWAT inputs include land use,
land use practice, soil, climate, elevation and slope, stream network and morphology, water uses, and
point sources. The SWAT outputs include total nitrogen, phosphorus, and sediment loads from each
subwatershed and stream segment. SWAT accounts for sediment contributions from overland runoff
through the Modified Universal Soil Loss Equation (MUSLE), which provides increased accuracy,
compared with the original USLE method, when predicting sediment transport and yield. The model is
capable of simulating long time periods (over 100 years) while retaining its computational efficiency, and
it can link sediment contributions to specific source areas (i.e. subwatershed and/or land use areas).
Importantly, SWAT allows for the application of specific agricultural management measures to
geographic units. Management measures that can be applied to model units include varying planting and
harvest patterns, fertilization practices, and quality of manure nutrient content (via livestock feed).
The following are limitations of SWAT:
• The model is not suitable for urban land uses
• The model runs at a daily time step, and is not suitable for fast-responding urban drainage
B-163
-------
AnnAGNPS
The Annualized Agricultural Nonpoint Source Pollution (AnnAGNPS) model is a batch-process,
continuous-simulation, pollutant loading computer model written in standard FORTRAN 95 (AnnAGNPS
2000). The model is capable of simulating (1) water; (2) sediment by particle size class and source of
erosion; and (3) chemicals (nitrogen, phosphorus, organic carbon, and pesticides). Pollutant loadings are
generated from land areas (cells) and routed through stream systems on a daily basis. The rainfall-runoff
process is simulated using the Curve Number method, and sediment erosion is simulated using the USLE
method. The model simulates and tracks nutrients in both particulate form (combined with sediment) and
dissolved form. Special land use components such as feedlots, gullies, field ponds, and point sources are
included.
The following are limitations of AnnAGNPS:
• It is not suitable for urban watersheds
• It uses a daily time step
• The model applies empirical methods for rainfall-runoff and water quality simulations that are not
robust enough to handle shorter response processes
Single-Event AGNPS
Developed by the USDA's ARS, the Agricultural Nonpoint Source Pollution (AGNPS) model addresses
concerns related to the potential impacts of point and nonpoint source pollution on water quality (Young
et al. 1986). It was designed to quantitatively estimate pollution loads from agricultural watersheds and to
assess the relative effects of alternative management programs. The model simulates surface water runoff
along with nutrient and sediment constituents associated with agricultural nonpoint sources, as well as
point sources such as feedlots, wastewater treatment plants, and stream bank or gully erosion. The
rainfall-runoff process is simulated using the Curve Number method, and sediment erosion is simulated
using the USLE method. Single-event AGNPS simulates and tracks nutrients in both particulate form
(combined with sediment) and dissolved form. The available version of the model is event-based. The
structure of the model consists of a square-grid-cell system to represent the spatial distribution of
watershed properties. This grid system allows the model to be connected to other software such as GIS
and DEMs. This connectivity can facilitate the development of a number of the model's input
parameters.
The Single-Event AGNPS has the following limitations:
• It is not suitable for urban land uses
• The version currently available is event-based
• The model applies empirical methods for rainfall-runoff and water quality simulations that are not
robust enough to handle shorter response processes
GWLF
The Generalized Watershed Loading Function (GWLF) model was developed at Cornell University to
assess the point and nonpoint source loading of nitrogen and phosphorus from urban and agricultural
watersheds, including septic systems, and to evaluate the effectiveness of certain land use management
practices (Haith et al. 1992). One advantage of this model is that it was written with the express purpose
of requiring no calibration, making extensive use of default parameters. The GWLF model includes
rainfall/runoff and erosion and sediment generation components, as well as total and dissolved nitrogen
and phosphorus loadings. The rainfall-runoff process is simulated using the Curve Number method, and
sediment erosion is simulated using the USLE method. It simulates and tracks nutrients in both
B-164
-------
participate form (combined with sediment) and dissolved form. The model uses daily time steps and
allows analysis of annual and seasonal time series. The model also uses simple transport routing, based
on the delivery ratio concept. In addition, the simulation results can be used to identify and rank pollution
sources and evaluate basin-wide management programs and land use changes.
The limitations for application of GWLF to urban areas are as follows:
• It uses a daily time step
• The algorithms applied for hydrologic and water quality simulations are empirical, not process-
based, approaches
• It is a lumped single-watershed model that cannot represent a stream network
B.3.2. BMP Model Reviews
The following BMP models were evaluated as the candidate models to be incorporated into SUSTAIN.
Prince George's County BMP Module
The Prince George's County Department of Environmental Resources, Programs and Planning Division,
working with Tetra Tech, Inc., has developed a BMP evaluation module to assist in assessing the
effectiveness of BMP/Low Impact Development (LID) technology (Cheng 2002). This module uses
simplified process-based algorithms to simulate BMP control of either observed time series or modeled
flow and water-quality time series generated from runoff models such as HSPF. The design and
evaluation methodology for the BMP Module has five basic aspects: (1) the incorporation of input runoff
data, (2) design and representation of a site plan, (3) configuration of BMPs of various sizes and
functions, (4) schematic representation of flow routing through a network of BMPs, and (5) evaluation of
the impact of the site design and BMP configurations on hydrology and water quality. The module
platform provides interactive linkages between the first four design aspects. The BMP module's
assessment post-processor offers a series of evaluation methods for measuring the impact of the design
and BMP configurations on hydrology and water quality.
Under this methodology, two generalized conceptual models were developed to characterize the function
of a wide range of BMPs. These models have been categorized in the module as Class A and Class B
BMPs. Class A BMPs are those that retain water for some duration of time and have some means for
controlling outflow. Examples of Class A BMPs are stormwater detention and retention ponds or
reservoirs, catch basins, and bioretention cells. Class B BMPs are open channels whose stormwater
control is a function of the shape and channel characteristics. Examples of Class B BMPs are grass
swales and stream buffers zones. The physical processes represented in the BMP Module include
evapotranspiration and infiltration (using the Holtan-Lopez empirical infiltration equation), orifice
outflow (standard orifice equation), underdrain outflow, weir-controlled overflow or spillway (using weir
equations for sharp-crested rectangular and v-notch triangular options), BMP bottom slope and bottom
roughness (Manning's equation for open channel flow), underdrain filtration of pollutant, and general loss
or decay of pollutant (first-order loss equation). In addition to the physical design and placement of BMP
structures, the module offers the user the flexibility to define flow routing through a BMP or BMP
network; simulate Improved Management Practices (IMPs), such as reduced or discontinued
imperviousness through flow networking; and compare BMP controls against some defined benchmark,
such as a simulated predevelopment condition. Because the underlying algorithms are based on physical
processes, BMP effectiveness can be evaluated and estimated over a wide range of storm conditions,
BMP designs, and flow routing configurations.
B-165
-------
SWMM BMP Simulation Capabilities
The SWMM (version 4.4h and previous versions) is divided into four primary computational blocks or
modules. They include:
• Runoff (converting rainfall to runoff and generate nonpoint source runoff water quality time
series)
• Transport (kinematic wave flow routing and water quality routing through conveyance and
storage, applying first-order decay)
• EXTRAN (performing dynamic wave flow routing)
• Storage/treatment (simulating treatment and storage devices, applying storage routing, first-order
decay, and Camp's (1946) sedimentation theory to up to five settling velocity ranges)
The SWMM simulation of major BMP processes (storage, infiltration, first-order decay, and sediment
settling) is achieved by using one or a combination of the four blocks. The Storage/Treatment Block
offers the most flexibility in terms of simulating conventional stormwater treatment devices (e.g., ponds
and swales). The overland flow rerouting (land-to-land routing) options in the Runoff Block can be used
to mimic the parcel (individual lot)-level LID sites.
P8UCM
The Program for Predicting Polluting Particles Passage through Pits, Puddles, and Ponds, Urban
Catchment Model (P8 UCM), is used to model generation and transport of stormwater runoff pollutants in
an urban setting (Walker 1990). Calculations are performed on continuous water balances and mass
balances using hourly rainfall and daily air temperature time series. Primary applications of this model
are the evaluation of BMP site plans for compliance with treatment objectives expressed in terms of
removal efficiency for TSS. Secondary (and less accurate) predictions from this model are runoff quality,
loads, violation frequencies, water quality impacts due to proposed development, and loads generated for
driving receiving water quality models (Walker 1990). The model can simulate a variety of treatment
devices (BMPs), including swales, buffer strips, detention ponds (dry, wet, extended), flow splitters, and
infiltration basins. Methods applied in P8 include quasi-linear reservoir storage routing, Green-Ampt
infiltration equation, second-order reactions, and particle removal by use of a scale factor. Compared
with other models, second-order reaction simulation is a unique feature of P8; however, the lack of
parameter estimates for the second-order decay coefficient in the model and literature limits the
usefulness of such a method.
VFSMOD (Vegetative Filter Strip Model)
Vegetative Filter Strip Model (VFSMOD) is a field-scale, mechanistic, storm-based model designed to
route the incoming hydrograph and sedimentograph from an adjacent field through a vegetative filter strip
(VFS) and to calculate the outflow, infiltration, and sediment trapping efficiency (Mu3oz-Carpena and
Parsons 2003). The model handles time-dependent hyetographs, space-distributed filter parameters
(vegetation roughness or density, slope, infiltration characteristics), and different particle sizes in the
incoming sediment. VFSMOD consists of a series of modules simulating the behavior of water and
sediment in the surface of the VFS. The current modules available are shown in Table B-l and
summarized below:
• Green-Ampt infiltration module: A module for calculating the water balance in the soil surface
• Kinematic wave overland flow module: A one-dimensional module for calculating flow depth
and rates on the infiltrating soil surface
B-166
-------
• Sediment filtration module: A module for simulating transport and deposition of the incoming
sediment along the VFS
VFSMOD is essentially a one-dimensional model for the description of water transport and sediment
deposition along the VFS. The model can also be used to describe transport at the field scale (or field
edge) if flow and transport are mainly in the form of sheet flow (Hortonian) and the one-dimensional path
represents average conditions (field effective values) across the VFS.
RAINFALL
VFS OUTFLOW
Figure B-l. Schematic representation of VFSMOD.
VFSMOD uses a variable time step, chosen to limit mass-balance errors induced by solving the overland
water flow equation. The kinematic wave model selects the time step for the simulation, to satisfy
convergence and computational criteria for the finite element method, (Munoz-Carpena et al. 1993a,
1993b). The model inputs are specified on a storm basis. The model integrates the state variables after
each event to generate storm outputs.
MUSIC
The Model for Urban Stormwater Improvement Conceptualization (MUSIC) was developed by the
Cooperative Research Center (CRC) for Catchment Hydrology in Australia (Wong 2002). MUSIC is
designed to simulate urban stormwater systems operating at a range of temporal and spatial scales:
catchments from 0.01 km2 to 100 km2 and modeling time steps ranging from 6 minutes to 24 hours to
match the catchment's scale. MUSIC provides a user-friendly interface to allow complex stormwater
management scenarios to be quickly and efficiently created and the results to be viewed using a range of
graphical and tabular formats. The stormwater control devices that can be simulated in MUSIC include
ponds, bioretention, infiltration buffer strips, sedimentation basins, pollutant traps, wetlands, and swales.
Major algorithms applied in BMP simulation are a continually stirred reactors (CSTRs) in series model
and a first-order decay (k-C*) model (see Section 3.3 of main report).
LIFE
The Low Impact Feasibility Evaluation (LIFE) model is a continuous-simulation, physically based model
that simulates the hydrologic and hydraulic processes that take place in bioretention facilities, vegetated
swales, green roofs, and infiltration devices, as well as the effects of site fingerprinting and soil
compaction (Medina et al. 2003). The model also simulates runoff generation from all categories of land
cover, including roadways, landscaping, and buildings, over a variety of land uses and soil types. The
B-167
-------
LIFE model is a visually oriented, interactive tool developed on an Extend™ dynamic simulation
platform. The LIFE model is a proprietary model and its modeling details are not available for review.
IDEAL
The Integrated Design and Evaluation Assessment of Loadings (IDEAL) model is a spreadsheet model
for assessing the impact of BMPs in urban areas on discharge of water, sediment, nutrients, and bacteria
into streams (Barfield 2002). The model predicts effluent loads and concentrations of the above elements
coming from the watershed as impacted by vegetative filter strips, dry detention ponds, and wet detention
ponds. The IDEAL model is capable of estimating the runoff and pollutant loadings from urban areas,
categorized into pervious, impervious connected, and impervious unconnected areas. Flows and loadings
are summed and then directed to a pond that can be dry (no permanent pool) or wet (permanent pool).
The model routes these loadings through BMPs to determine pollutants removal efficiencies using
empirical technologies that have been experimentally validated. The model predicts single storm values
and converts them to average annual storm values using stochastic procedures. The IDEAL model is
designed to estimate BMP long-term pollutant removal efficiencies and is not intended to be used to give
accurate estimation on a storm event basis.
DMSTA (Dynamic Model for Stormwater Treatment Area) Model
DMSTA simulates daily water and mass balances in a user-defined series of wetland treatment cells, each
with specified morphometry, hydraulics, and phosphorus cycling parameters (Kadlec and Walker 2003).
Up to six treatment cells can be linked in series and/or parallel to reflect compartmentalization and
management to promote specific vegetation types. Each cell is further divided into a series of CSTRs to
reflect residence time distribution. Water-balance terms for each cell include inflow, bypass, rainfall,
evapotransportation, outflow, seepage in, and seepage out. Parameter estimates for the phosphorus
cycling model have been developed for various vegetation types. Water column storage, solid (biomass,
sorption) storage, uptake, recycle, and permanent burial processes are considered in dynamic phosphorus
cycling simulation. The model is coded in Visual Basic for Applications and the user interface is a
Microsoft Excel workbook.
WETLAND
The WETLAND model is a dynamic compartmental model to simulate hydrologic, water quality and
biological processes, and to assist the design and evaluation of wetland. The WETLAND model adopted
the continuous stirred-tank reactor (CSTR) prototype, and it is assumed that all incoming nutrients are
completely mixed throughout the entire volume. The model can simulate both free-water surface (FWS)
and subsurface flow (SSF) wetlands. The WETLAND model is constructed in a modular manner, and it
includes hydrologic, nitrogen, carbon, dissolved oxygen, bacteria, sediment, vegetation, and phosphorous
submodels. The hydrologic submodel uses a vertical dynamic water budget approach to calculate surface
storage, and carries out the computation at hourly time step. The factors considered in the hydrologic
model include inflow, precipitation, infiltration, and evapotranspiration. The Nitrogen submodel
simulates ammonification, immobilization, nitrification, denitrification, and peat accumulation, and
inclusion of NH3 volatilization, atmospheric deposition and N fixation in the modeling of overall N cycle
is optional. Sorption of NH4+to the soil and organic matter is not modeled because it is assumed that
sorbed NH4+ is still available to the attached microbes. The carbon model includes five variables:
biomass C, standing dead C, particulate organic C, dissolved organic C, and refractory C; The standing
dead C and biomass C is connected to the vegetation submodel. The dissolved oxygen submodel track
the oxygen influx from incoming stream flow, precipitation, reaeration from atmosphere, point sources,
and biomass flux. In addition, oxygen is assumed to be passed from vegetation stand to wetland bottom
at a constant rate during the growing season. The bacteria submodel accounts for all the microbial growth
and activity in the model. Both autotrophic and heterotrophic bacteria are modeled. Sedimentation is
modeled in the sediment sub-model. The processes simulated include inflow, outflow, deposition,
resuspension, and decomposition. Up to five different sediment classifications can be modeled. A simple
B-168
-------
vegetation submodel is included to simulate the biomass growth and death. The phosphorous submodel
considers four pools for the P cycle: particulate and dissolved for both the surface and bottom layer of the
wetland. Processes modeled in the phosphorous model include mineralization and additions from
biomass decomposition. Besides the hydrologic submodel, all the other submodels compute using daily
time step.
The strength of the WETLAND model lies on the linked Monod kinetics for the water quality variables,
also the model accounts for the seasonal variation by allowing users to change parameter values for
different season/time period. The weaknesses of this model include the completely mixed assumption,
which overlook the effect of the system shape, and the needs for extensive kinetic parameters.
VAFSWM
The Virginia Field Scale Wetland Model (VAFSWM) is a field scale model for quantifying the pollutant
removal in a wetland system. It includes a hydrologic subroutine to route flow through the treatment
system; Precipitation, evapotranspiration, and exchange with subsurface groundwater are considered in
the hydrologic balance. The model adopted a continuous stir tank system in series schema. VAFSWM
models mechanisms of settling, diffusion, adsorption to plants and substrate, and vegetative uptake for a
pollutant in dissolved and particulate forms in a two segment (water column and substrate), two state
(completely mixed and quiescent) reactor system by employing first-order kinetics. The governing
equations for quiescent condition are identical to that of turbulent condition, however far lower settling
velocities are assumed to account for the greater percentage of finer particles during the quiescent state.
The VAFSWM is a relatively simple model that includes the most dominant processes within the wetland
system. However, the users need to provide and calibrate the requisite kinetics parameters.
REMM
Riparian Ecosystem Management Model (REMM) has been developed as a tool that can help quantify the
water quality benefits of riparian buffers. REMM simulates the movement of surface and subsurface
water movement, sediment transport and deposition, nutrients transport, sequestration, and cycling, as
well as vegetative growth in riparian forest systems on a daily time step. In REMM, the riparian system
is considered to consist of three zones between the field and the water body. Each zone includes litter and
three soil layers, and a plant community that can have six plant types in two canopy levels. REMM can
be used to quantify nitrogen and phosphorous trapping in riparian buffer zone, determine buffer
effectiveness, investigate long-term fate of nutrients in buffer zones and, evaluate influence of vegetation
type on buffer effectiveness, and determine impacts of harvesting on buffer effectiveness.
The strength of REMM is its capability of simulating subsurface compartment, and the comprehensive
nutrients cycling. Comes with the complexity, one disadvantage of the model is the extensive data
requirement. REMM is still under development and has been continuously updated. Currently, a user
interface is being built to assist input and output data management.
B. 3.3. Modeling System Reviews
Several systems have been developed that include multiple models and software systems to facilitate data
storage, data preparation, model input file development, model application and linkages, and output post-
processing. These comprehensive systems have the potential for integration or communication with
SUSTAIN. As these systems continue to evolve, SUSTAIN will consider options to preserve compatibility
with these systems.
B-169
-------
BASINS
Better Assessment Science Integrating point and Nonpoint Sources (BASINS), developed by EPA, is a
multipurpose environmental analysis system for use by regional, state, and local agencies in performing
watershed and water quality based studies (USEPA 2001). BASINS has three major objectives: (1) to
facilitate examination of environmental information, (2) to support analysis of environmental systems,
and (3) to provide a framework for examining management alternatives.
BASINS integrates a GIS, national watershed and meteorological data, and state-of-the-art environmental
assessment and modeling tools into one convenient package. Originally released in 1996, with a second
release in 1998, a third in 2001, and a fourth in 2004, BASINS comprises a suite of interrelated
components. The current version is BASINS 4.0.
In a departure from previous versions, BASINS 4.0 databases and assessment tools run on a non-
proprietary, open source GIS system architecture (MapWindow). Its components work together to
support the user in performing various aspects of environmental analysis. The components include (1)
nationally derived databases with Data Extraction and Project Builder tools; (2) assessment tools
(TARGET, ASSESS, and Data Mining) that address large- and small-scale characterization needs; (3)
utilities to facilitate importing local data and to organize and evaluate data; (4) Watershed Delineation
tools; (5) utilities for classifying elevation (DEM), land use, soils, and water quality data; (6)Watershed
Characterization Reports that facilitate compilation and output of information on selected watersheds; (7)
an in-stream water quality model; (8) two watershed loading and transport models; and (9) a simplified
GIS-based, nonpoint annual loading model. Installed on a personal computer, BASINS allows the user to
assess water quality at selected stream sites or throughout an entire watershed. The software makes it
possible to quickly assess large amounts of point source and nonpoint source data in a format that is easy
to use and understand, as well as to prepare and set up watershed and in-stream transport models to
facilitate the TMDL analysis for waterbodies of concern.
A limitation of the current BASINS configuration is that data currently housed in the BASINS system is
typically too general to support detailed urban analysis. The system data would need be updated with
local data to facilitate application and provide higher resolution analysis necessary for SUSTAIN. For
more information, see the BASINS Web site (see http://www.epa.gov/waterscience/basins/).
EPA TMDL Modeling Toolbox
The TMDL Modeling Toolbox is a collection of models, modeling tools, and databases that have been
widely applied over the past decade in the development of TMDLs. The Toolbox takes those proven
technologies and provides the capability to more readily apply the models, analyze the results, and
integrate watershed loading models with receiving water applications (USEPA 2003). The design of the
Toolbox is such that each of the models is a standalone application. The Toolbox provides an exchange
of information between the models through common linkages. The modular design of the Toolbox allows
for additional models to be easily incorporated and integrated with the other tools. In addition, the
Toolbox provides the capability to visualize model results, a linkage to GIS and nongeographic databases
(including monitoring data for calibration), and the functionality to perform data assessments.
The Toolbox allows for the steady-state/dynamic simulation of mass transport and water quality processes
in all types of surface water environments, including overland flow, small creeks, rivers, lakes, estuaries,
coastal embayments, and offshore areas. The Toolbox contains assessment tools, watershed models, and
receiving water models, including the following:
Assessment Tools
• Water Resources Database (WRDB)
B-170
-------
• Watershed Characterization System (WCS)
• WCS Sediment Tool
• WCS Mercury Tool
• WCS LSPC Tool
Watershed Models
• Loading Simulation Program in C++ (LSPC)
• Watershed Assessment Model (WAMView)
• Stormwater Management Model (SWMM)
Receiving Water Models
• A Dynamic, One-Dimensional Model of Hydrodynamics and Water Quality (EPDRivl)
• Stream Water Quality Model (QUAL2K)
• CONservational Channel Evolution and Pollutant Transport System (CONCEPTS)
• Environmental Fluid Dynamics Code (EFDC)
• Water Quality Analysis Simulation Program (WASP)
The Toolbox has a wide variety of included models and open architecture that facilitates linkages and
flexibility in application. A limitation of the system is the lack of specific models and tools for simulating
BMPs. Although LSPC and WAMView can be used to simulate BMPs, the systems do not include
detailed, process-based simulation capabilities or convenient tools to quickly set up and evaluate
alternative BMP management alternatives.
B.4. Discussion and Results of Model Review
A review was conducted of available models such as SWMM (Huber and Dickinson 1988; Huber 2001),
HSPF (Bicknell et al. 1993), and SLAMM (Pitt and Voorhees 2000), as well as publicly available
modeling systems, such as BASINS (USEPA 2001) and the TMDL Modeling Toolbox (Tetra Tech and
USEPA 2002). Based on this review, there is no single system or model with the flexibility and
capability to incorporate all the technical needs listed below for the SUSTAIN development.
• Ability to simulate hydrologic response and a level of detail sufficient for analysis of a
hydrograph (peak flow and volume)
• Ability to simulate multiple pollutant types, including nutrients (nitrogen and phosphorus),
pathogens [fecal coliform bacteria, Escherichia coli (E.Coli)] and metals (zinc, aluminum, etc.)
• Ability to simulate fate and transport of pollutants and evaluate both acute and chronic impacts
• Ability to generate sediment loading to streams
• Ability to simulate sediment transport in streams
• Ability to simulate multiple size classes of sediment for input to management structures
• Ability to simulate other habitat stressors, such as temperature
• Ability to simulate in-stream dissolved oxygen based on inputs of biological oxygen demand,
sediment oxygen demand, nutrient loads, and other environmental factors
• Ability to evaluate urban and mixed land uses, including pervious and impervious areas
B-171
-------
• Consideration of short and long time periods (single- and multiple-event simulation)
• Consideration of a full range of management practices at a similar level of spatial resolution
• Consideration of distributed or small-scale management practices and larger downstream
facilities
• Consideration of a series of management practices at various locations in the watershed
• Modeling of management practices on a time-variable basis consistent with the need to evaluate
hydrology and pollutant measures
• Ability to consider placement of management practices at any location in the watershed (e.g., at
various distances from waterbodies, at various stream orders)
• Ability to link watershed management to downstream measures of environmental condition (e.g.,
dissolved oxygen in a river, nutrient concentration in a lake or estuary) outside the immediate
vicinity of a selected study area
However, many models can provide portions of the needed features and algorithms. Comparison of the
available models and the technical needs supports selection of a subset of models for further
consideration, and their potential incorporation in SUSTAIN is organized according to the key components
identified in the preliminary design discussion. Table B-7 summarizes the strengths and weaknesses of
the selected watershed models in light of the SUSTAIN design requirements. Presented below is a
process-focused summary discussion of the models that supports the landscape and BMP model selection.
B.4.1. Watershed Models
The selection of watershed models for integration into SUSTAIN are discussed separately for hydrology,
sediment, pollutant loadings, and reach routing.
Hydrology
Several watershed models, including SWMM, SLAMM, HSPF, and LSPC, can provide time series
hydrology and pollutant loading at an hourly time step or less. This short temporal resolution is needed to
address small catchments and to provide concentration and load predictions and time series inputs to
management practices. This temporal resolution is necessary for the flexibility to predict the range of
hydrologic and water quality measures identified in the needs analysis. Some models, such as SWAT
(Neitsch et al. 2001), AnnAGNPS (AnnAGNPS 2000), AGNPS (Young et al. 1986), and GWLF (Haith et
al. 1992), are inappropriate because they use large time steps (1 day or greater) or insufficient description
of time-variable rainfall-runoff processes. Other models, such as CASC2D and KINEROS, use a grid-
based framework for distributed modeling of the watershed landscape. The grid-based formulation has
benefits for detailed simulation and sensitivity to the placement of management within the landscape.
However, its greatest limitations are high computational needs for larger watersheds and the availability
of spatially detailed data. The spatial detail can significantly increase the data preparation and setup time
for the model. Currently, CASC2D and KINEROS do not include water quality simulation capabilities.
Further evaluation is needed to determine whether cell- or grid-based modeling components can be
incorporated into SUSTAIN. The initial recommendation is to use pervious and impervious land
simulation routines from SWMM, HSPF, and/or LSPC.
Sediment
The HSPF and LSPC watershed models use a sophisticated process-based system to describe sediment
simulation for pervious areas and buildup/washoff for impervious areas. For pervious segments, sediment
is represented as a direct function of the rainfall intensity. The rainfall intensity determines the rate and
volume of material detached from an infinite soil matrix, while the scouring process determines the
B-172
-------
washoff and delivery of sediment to a stream segment. Scour can be used to represent gully erosion.
Because this process is energy-driven, the calibration changes with the time step and resolution of the
rainfall data driving the system. For impervious land surfaces, both HSPF/LSPC and SWMM use similar
approaches to simulate buildup and washoff of solids on the land surface. HSPF and SWMM allow the
user to apply special actions, such as street sweeping during the simulation, to assess the impact of such a
management activity on the overall delivery of solids from urban streets. SWMM allows three ways for
estimating sediment in runoff: (1) a rating curve, (2) a buildup and washoff approach, or (3) the USLE for
pervious surfaces.
If the methods described above are compared with another popular sediment estimation method such as
USLE (which is used by many of the models described in Section A.3), some limitations, in light of the
project requirements, are evident. The parameters feeding the USLE equation are based on long-term
assessments, and the results, though meaningful as a monthly or annualized loading estimate, fail to
adequately represent the detailed variability of individual storms or storms in series. In conclusion,
short/variable time step methods, such as those available in HSPF and LSPC for pervious areas and in
SWMM, HSPF, and LSPC for impervious areas, are better suited to satisfy the assessment objectives
outlined for SUSTAIN.
Pollutant Loading
Among the shorter/variable-time-step simulation models like SWMM, HSPF, and LSPC, buildup and
washoff of pollutants on a land surface is often used as the primary process for generating pollutant
loadings. In HSPF and LSPC, pollutants can also be represented as sediment-associated; therefore, some
of the pollutant mass will be considered as a fraction of the simulated sediment delivery. Base flow and
interflow concentrations in HSPF and LSPC are specified as constants, or they can be expressed as
monthly variable concentrations. SWMM does not allow for a variable buildup rate; however, it allows
the user to specify the equation and method used (power-linear, exponential, or Michaelis-Menton). As
with sediment, SWMM allows for pollutants to be specified as a function of the flow rating curve or by
using buildup and washoff. Pollutants can also be associated with sediment by expressing the mass as a
fraction of sediment. Simpler models, such as GWLF and P8, use a fixed concentration of a pollutant in
runoff and sediment, making them insensitive to changes in concentration or availability of pollutants
over time. These models also use daily or monthly time steps, and they cannot support the evaluation of
short-duration loading and impacts on stream systems. For pollutant loading, HSPF, LSPC, and SWMM
include the preferred techniques for integration into the SUSTAIN design.
Reach Routing
Landscape output must also be collected and routed via flow networks (channels and streams). Many
watershed models, including SWMM and HSPF, include stream routing modules. These routing
techniques, which involve some simulation of in-stream transport and pollutant transformation processes,
are sufficient for smaller streams with relatively short conveyance times (less than 1 day). Urban streams
typically have short retention times and limited opportunity for biological and chemical processes to
result in significant transformation of pollutants. Of the reviewed models, HSPF and LSPC reach routing
have the most detailed simulation capabilities for sediment and pollutant transport including sediment
deposition, scour, decay, and dynamic temperature simulation. SWMM's transport functions include
first-order decay and settling but do not include an option for temperature, biological transformation, or
algal growth. SWMM can simulate complex hydraulics using a fully dynamic wave method. For areas
with large, longer-retention-time river systems ortidally influenced systems, an external linkage (outside
SUSTAIN) can provide the ability to evaluate downstream impacts. Linkage with specialized receiving
water models, such as EFDC (Hamrick 1992) and WASP 6.0 (Wool et al. 2003), ultimately can be used
to consider the impacts of urban stormwater runoff on larger, more complex waterbodies. Specialized
receiving water models like WASP (Wool et al. 2003) are also best suited for evaluating eutrophication
processes and dissolved oxygen.
B-173
-------
Table B-7. Strengths and Weaknesses of Major Watershed Models
Model
Strengths
Weaknesses
SWMM
The best available public domain model for simulation of
sewer systems hydraulics:
Fully dynamic hydraulic routing
Hydraulic structure (manhole, weir, orifice, etc.) simulation
Overland flow routing between pervious and impervious areas
within a subcatchment
Various options for quality simulation: buildup and washoff,
rating curves, and regression techniques
Offers base flow simulation
Performs continuous simulation using variable time step
Considers only settling and first-
order decay in in-stream pollutant
routing and transformation
HSPF
Comprehensive simulation of watershed hydrology and
associated water quality processes on pervious and
impervious land surfaces
Capable of simulating the in-stream transfer and reaction
processes, including hydrolysis, oxidation, photolysis,
biodegradation, volatilization, sorption, and resuspension and
settling of cohesive and noncohesive solids
Performs land-to-land routing
Offers base flow and interflow simulation
Performs continuous simulation using variable time step
Does not perform fully dynamic
hydraulic flow routing
LSPC
Includes a streamlined set of HSPF subroutines and
algorithms
Simulation of watershed hydrology, and associated water
quality, processes on pervious and impervious land surfaces
No inherent limit to the size and scale of watershed modeling
Generalized in-stream water quality simulation, as well as
sediment associated land and in-stream processes
Performs continuous simulation using variable time step
Does not perform fully dynamic
hydraulic flow routing
WAMView
Grid based model with cell size down to 0.1 ha
Offers dynamic channel routing and allows outlet stage and
concentration definition with backflow.
Simulates wetland and depressions in the channel
Output overland, wetland, and stream load attenuation
mapped back to source cells
Source code and detailed
documentation is not available
Does not perform land to land
routing
CASC2D
Fully unsteady physically based distributed watershed model
at a user-specified resolution
Offers fully dynamic hydraulic channel routing
Uses diffusive wave method to route overland flow
Performs continuous simulation using variable time step
Only simulate sediment, not other
water quality constituents
Does not simulate subsurface flow
Fully physically based distributed
model; therefore, its application
requires extensive input data
preparation and calibration
Not suitable for urban watersheds
B-174
-------
B.4.2. BMP/LID Models
Simulation of BMPs varies between simplified representation of percent removal and partial or complete
representation of the processes of hydraulic controls, settling, and transformation of pollutants. A number
of available watershed models have the potential for use in BMP simulation (e.g., SWMM, HSPF, LSPC,
and SLAMM), but representation is achieved by custom adjustment of hydrologic and pollutant transport
parameters. Guidance for the application of watershed models such as SWMM and HSPF for simulation
of BMPs is limited. Consistent application is difficult, and in the absence of default data and documented
applications, intensive data collection and calibration are necessary. Some models, such as WAMView,
can be adjusted to represent land practice BMPs based on the USDA Curve Number guidance. Many of
the currently available, published BMP models are propriety (e.g., MUSIC) or have had limited release in
the public domain (e.g., BMP AM). Specialized BMP simulation tools such as VFSMOD (Munoz-
Carpena and Parsons 2003) focus on specific BMPs, in this case vegetative filter strips.
Most of the currently available systems have limited process simulation or lack guidance for the selection
and evaluation of management practices. Of the available systems, the Prince George's County BMP
Module provides capabilities to simulate a wide range of BMPs with particular emphasis on scale-scale,
distributed systems, using a process-based approach to address hydrology and pollutant removal. One
specialized need for BMP simulation is the ability to handle highly distributed management techniques
such as those employed in LID procedures. The Prince George's County BMP Module was designed
specifically to address LID simulation and networks with multiple management practices. The structure
of the BMP Module can facilitate the incorporation of additional BMP types and is suitable for linkage
with a variety of watershed and receiving water models. Prince George's County has provided the system
to users upon request and is willing to provide EPA with the code for adaptation and incorporation into
SUSTAIN.
For the process simulation of BMPs, the Prince George's County BMP Module, augmented by portions of
selected BMP processes provided by models such as SWMM, SLAMM, and P8, is recommended for
incorporation into SUSTAIN. In particular, BMP simulation techniques for stormwater ponds and
detention structures can be provided by SWMM. For BMPs such as riparian buffers, specialized
simulation techniques are also needed. Riparian buffers can be addressed by using the procedures in
VFSMOD (Munoz-Carpena and Parsons 2003) or by adapting the land-to-land transport routines used in
SWMM or HSPF.
B.5. Conclusions
The review of available models and BMP analysis systems confirms the initial selection in Task 1 of a
short list of models best suited to be included in the SUSTAIN system. The final recommended list of
models was based on an evaluation of the needs, the level of analysis included, the software capabilities,
and the availability of the code supporting the models. Each of these models provides essential software
tools; algorithms describing watersheds, receiving waters, or BMP processes; and a history of application
and testing. In addition, existing models can be linked with SUSTAIN for combined simulation of large,
complex watersheds and receiving waters. The selected models are the following:
• Watershed/landscape models: SWWM, HSPF, LSPC
• Stream conveyance and pollutant routing models: HSPF/LSPC stream routing and pollutant
transport functions, or SWMM routing and transport (SWMM5)
• Stream conduit (combined sewer overflow, or CSO) models: SWMM
B-175
-------
• BMP simulation models: Prince George's County BMP Module, including new algorithms for
detention ponds and structural options, and selected buffer zone simulation techniques from
VFSMOD
Development of the system will also require a framework manager, and supporting GIS tools,
optimization, cost estimation, and post-processing techniques. The relevant components of the selected
models, supporting algorithms, and tools will be integrated into a seamless framework that can provide
the required functionality.
B-176
-------
Appendix C. Summary of the Optimization Technical Panel Meeting
C.I. Background
Watershed and stormwater managers need modeling tools to evaluate how best to address environmental
quality restoration and protection needs in urban and developing areas. A place-based analysis system,
based on cost optimization, is essential to support government and local watershed planning agencies as
they coordinate efforts across the watershed to achieve desired improvements in water quality at a
minimum cost.
A two-day workshop was convened September 15-16, 2006, at the Fairfax, Virginia, office of Tetra Tech,
Inc., to bring together experts to discuss the current state-of-the-art in optimization concepts and methods
to support development of the optimization component in SUSTAIN. The invited experts included the
following:
• Dr. James P. Heaney (University of Florida)
• Dr. Manuel Laguna (University of Colorado)
• Dr. Arthur E. McGarity (Swarthmore College)
• Dr. S. Ranji Ranjithan (North Carolina State University)
• Dr. Christine A. Shoemaker (Cornell University)
• Dr. Richard M. Vogel (Tufts University)
• Dr. Laura J. Harrell (Old Dominion University)
Optimization decision variables include BMP locations, types and design configurations. Because there
can be an extremely large number of possible combinations of BMP choices that can meet desired water
quality and quantity constraints, strategies are needed to identify specific BMP options for
implementation from a vast output database. The primary objective of the workshop was to identify the
best strategies available for implementation in SUSTAIN. A secondary objective of the workshop was to
discuss and report issues related to cost estimating and in defining and quantifying the effectiveness of
individual BMPs or several BMPs in parallel or in series. This appendix is a summary of the workshop
discussion and recommendations.
C.2. Key Discussion Issues
The workshop focused on discussing and acquiring experts' knowledge on issues listed below in four
categories:
C. 2.1. General Issues
• Trend and focus - What are the current trends and focus in optimization research for watershed
planning?
C-177
-------
• Algorithm selection and evaluation - It was proposed to program two search algorithms in
SUSTAIN: 1) Scatter Search and 2) genetic algorithm. Which one is more robust in providing
placement decisions? Should other solution techniques be considered? How can it be confirmed
that global or near global solutions have been found?
C. 2.2. Optimization Approach
• Two-tier approach - Presumably a tiered optimization approach will facilitate placement of
BMPs in different spatial scales. Can a two-tier or cascading optimization approach work to
develop large scale solutions? BMPs may be placed at the site scale or subwatershed scale, but
overall control performances are evaluated at the watershed scale
• Top down vs. bottom up - Should a watershed optimization process be top-down (from the
watershed to subwatershed to site scales) or bottom-up?
C. 2.3. Computational Efficiency
• Aggregation of distributed BMPs - BMPs include distributed types such as green roofs,
bioretention basins, porous pavements and rain barrels. What are the most efficient solution
strategies and computational approaches to simulate and optimize hundreds of distributed BMPs?
How should the distributed BMPs be lumped (usually at parcel scales)? How should the BMP
clusters be represented by lumped hydrologic parameters (e.g., depression storages and
infiltration rates)?
• Simplified approach to derive effectiveness from multiple BMPs - BMPs can be in series or in
parallel in a given subwatershed. It will be computationally demanding if process simulations are
performed for each combination of treatment trains. Can experiments be performed to establish a
database for deriving a regression formula that can be used to estimate the pollutant load
reduction from all possible combinations of BMPs?
• Development of cost-effectiveness curves - What is the most efficient way to generate cost-
effectiveness curves (cost vs. effectiveness) when using meta-heuristic algorithms? The curve
can be derived from multiple costs vs. load reduction points by simulating multiple runs under a
range of load reduction targets. This option will be computationally time-consuming because a
large number of simulation runs may be required to derive multiple optimal solutions
C.2.4. Problem Formulation - Objectives, Constraints and Variables
• Pollution vs. flood control objectives - How to reconcile the potential conflicts between meeting
pollution control and flood control objectives? The pollution control effectiveness is usually
assessed by a continuous simulation, while flood control effectiveness is assessed by an event
simulation
• Multiobjective optimization - How should the objective equation be formulated?
• Future land use management - Is the future land use management a decision variable in the
BMP placement decision? In other words, should the land use planning and water quality
management be integrated? SUSTAIN is designed for placing BMPs in watersheds with known
existing or future land uses
• Cost estimating - For estimating the cost of BMPs, what will be the level of detail required to
maintain the consistency of decision parameters used in optimization analyses?
• Financial resources and implementation schedule - How to include constraints on financial
resources and schedules of BMP implementation in the optimization framework?
C-178
-------
C.3. Discussion Summary
This section summarizes the discussion and input from the invited experts, organized by the discussion
issues.
C. 3.1. General Issues
Trend and focus - What are the current trends and focus in optimization research for watershed
planning?
An emerging trend is to apply optimization techniques, especially meta-heuristic algorithms, to solve
stormwater management issues. Although a number of research projects have been completed in recent
years, most of them are conducted in academia and most of them were developed on a case-by-case basis.
There has not been any generic decision support system developed that can be used by a general public
practitioner to optimize size, type and locations of BMPs.
During the discussion, the application of neural networks and parallel computing for the purpose of
reducing search time was brought up. Although there are uncertainties that neural networks can
accurately represent the real simulation module with limited training process, it was suggested that they
can be used as a filter during the search process to avoid spending CPU time to evaluate bad solutions.
Parallel computing can be employed where a network of computers is available to use all possible
resources to obtain the search results in a shorter time.
A hybrid approach of combining traditional and meta-heuristic algorithms can be promising as traditional
algorithms are more efficient for reaching local optima and meta-heuristic algorithms have the advantage
of not being trapped at the local optima.
Algorithm selection and evaluation - It was proposed to program two search algorithms in SUSTAIN:
1) Scatter Search and 2) genetic algorithm. Which one is more robust in providing placement decisions?
Should other solution techniques be considered? How can it be confirmed that global or near global
solutions have been found?
There is no quick answer for the question of which algorithm is better than the other. In terms of solution
techniques, it was mentioned that Evolution Strategies are claimed to be faster at numerical optimization
than traditional Genetic Algorithms. A participant presented a stochastic RFB-Cornell radial basis
function approach and showed it converged significantly faster than a few other techniques for a
particular case study she conducted. The participant also suggested that the alternatives for optimization
algorithms need to be evaluated carefully since the simulation time can be substantial. It was also noted
that using commercial software Solver associated with spreadsheet analysis could be an efficient
alternative. Other participants also found commercial software useful for testing new search algorithms.
To address the question of how to confirm that global optima has been found, the experts agreed that,
theoretically, global optima cannot be proved when using meta-heuristic techniques. That is why the term
near optimal should always be used instead of optimal. However there are a few ways to help gain
confidence:
• Use a benchmark test case with known optima
• Compare and try different solution techniques
• Use commercial software to compare results
C-179
-------
Another way of looking at the near optima is that although it is not guaranteed to be the optima, they are
better than the other solutions that have been checked during the search process. This leads to the
suggestion that starting the search with a good solution might result in the near optimal solution faster. It
was pointed out that local optima can be proved by checking the derivative if the problem is continuous.
Other than one member, the invited experts appeared unfamiliar with the Scatter Search method. Two
experts both talked at several times about the potential utility of traditional dynamic programming
techniques.
The workshop experts demonstrated the following optimization applications that can be further explored:
One expert talked about the experience of using Genetic Algorithm (GA) to optimize the locations of
infiltration practices for reducing peak flow. A curve number (CN)-based distribution model was used to
simulate the hydrological responses and infiltration BMPs are represented as change of CN. A series
statistical analysis was performed to check if there is another way to identify the optimal BMP locations
without using optimization. The results were negative; this confirmed the need for applying optimization
techniques to get the cost-effective solutions for storm water management issues. It was also commented
that a decision support system does not necessarily provide BMP design details as part of the solution;
instead it is only necessary to suggest the general categories of BMPs and the expected treatment (i.e.,
infiltration and/or storage) capacity. In addition, sometimes simplified optimization such as Linear
Programming (LP) may give results that are comparable to GA solutions. The following web site was
suggested to download papers and manuscripts for more detail:
(http ://ase .tufts .edu/cee/faculty/vogel/bio .asp).
Another expert presented a spreadsheet optimization tool that used the Excel add-on optimization engine
Solver to find cost-effective BMPs. The BMPs were represented as a combination of on-site depression
storage (DS) and/or centralized storage/release systems.
A third expert showed Storm Water Investment Strategy Evaluator (StormWISE), which is a screening
level stormwater management optimization tool. This tool employs a top-down approach to prioritize
investment in subwatersheds for pollution control. The essential component of this tool is the generalized
pollutant-removal/cost functions for each land use in each subwatershed (first-stage). The functions are
then used for the second-stage optimization. As the pollutant-removal/cost functions are well-behaved, a
classical optimization technique, mixed integer/linear programming, is used to solve the second-stage
optimization problem. The following Web site (http://watershed.swarthmore.edu) has more details.
Two panelists pointed out the importance of providing diverse alternative solutions. One presented a case
study where an evolutionary algorithm was applied to obtain diversified alternative solutions that have
comparable objective values. It was emphasized that the approach was efficient because it was performed
along the search process for the main optimization problem so that it did not require to rerun the model.
Another expert also commented that there might often be multiple feasible solutions within a very small
percentage of benefit or cost range. In that case the system needs to identify the most diversified
alternatives that the user can choose from (using their own judgment).
C. 3.2. Optimization Approach
Two-tier approach - It is believed that a tiered optimization approach will facilitate placement of BMPs
in different spatial scales. Can a two-tier or cascading optimization approach work to develop large
scale solutions? BMPs may be placed at the site scale or subwatershed scale, but overall control
performances are evaluated at the watershed scale.
C-180
-------
Overall, the experts agreed that the tiered approach is promising; however, they foresee the obstacle of
daunting computation time if the meta-heuristic optimization algorithm is employed. A few ideas came
up during the discussion. The first group of ideas focused on reducing the complexity of the simulation
system by either employing a simpler and faster simulation approach or by using a generic cost-pollutant-
removal function to eliminate the needs of detailed BMP simulations. The second group of suggestions
focused on improving optimization efficiency. One expert mentioned the use of dynamic programming
(DP) for the second tier analysis. If applicable, DP can be more efficient than meta-heuristic algorithms.
However, it is recognized that implementing DP in a decision support system such as SUSTAIN, which is
intended to be applicable to many different cases, would be difficult because DP requires a case-by-case
problem formulation. Another suggested using neural networks as a filter during the optimization process
to avoid spending time in evaluating bad solutions.
Top down vs. bottom up - Should a watershed optimization process be top-down (from the watershed to
subwatershed to site scales) or bottom-up?
The top-down approach involves applying generalized cost-benefit functions (such as the pollutant-
removal/cost functions in Storm WISE) to prioritize the distribution of load reduction requirements at the
subwatersheds, given a target at the watershed level. The advantage of this approach is that an efficient
classical optimization algorithm can be used because the generalized cost-benefit functions are smooth
and convex. The challenge of this approach is to obtain reasonably accurate cost-benefit functions. If the
cost-benefit function is not accurate, the solutions can be skewed. Also, this approach does not explicitly
address BMP implementation details.
For the bottom-up approach, the search starts with the potential locations identified; therefore it explicitly
addresses the BMP implementation details. The downside of this approach, as commented on by one
expert, is that the amount of site-specific information and data required for specifying sites and potential
BMPs could be prohibitive. Also, the approach is simulation intensive and when it is applied for a large
watershed the computation time required can be extensive.
From discussions, a strategy that combines bottom-up and top-down procedures appears promising. The
overall optimization process can start with the top-down approach as applied in StormWISE using generic
cost-benefit functions to identify the high priority subwatersheds, then perform a detailed bottom-up
optimization search for each priority subwatershed to derive a more accurate and site-specific cost-benefit
curve. By doing so, the computation time is expected to be reduced because detailed
simulation/optimization is conducted only for the priority subwatersheds. The search process is then
completed with another round of top-down optimization using the cost-benefit functions derived from the
previous step.
C. 3.3. Computational Efficiency
Aggregation of distributed BMPs - BMPs include distributed types such as green roofs, bioretention
basins, porous pavements and rain barrels. What are the most efficient solution strategies and
computational approaches to simulate and optimize hundreds of distributed BMPs? How should the
distributed BMPs be lumped (usually at parcel scales)? How should the BMP clusters be represented by
lumped hydrologicparameters (e.g., depression storages and infiltration rates)?
One participant presented the approach of using aggregated depression storage to represent the site-scale
or distributed BMPs (such as green roofs, porous pavement, rain-gardens, etc.) at the catchment level.
Another suggested using response functions to represent distributed BMPs at the scale of a neighborhood
or region of an urban area. The response functions need to be in the form of simplified formulations
C-181
-------
derived from regressions or theoretical means. It was suggested that a highly detailed simulation model
driven by an optimizer can be used to generate data for curve fitting.
Simplified approach to derive effectiveness from multiple BMPs - BMPs can be in series or in
parallel in a given subwatershed. It will be computationally demanding if process simulations are
performed for each combination of treatment trains. Can experiments be performed to establish a
database for deriving a regression formula that can be used to estimate the pollutant load reduction from
all possible combinations of BMPs?
This topic was discussed under aggregation of distributed BMPs.
Development of cost vs. effectiveness curves - What's the most efficient way to generate cost-
effectiveness curves (cost vs. effectiveness) when using meta-heuristic algorithms? The curve can be
derived from multiple cost vs. load reduction points by simulating multiple runs under a range of load
reduction targets. This option will be computationally time-consuming because a large number of
simulation runs may be required to derive multiple optimal solutions.
One participant suggested that the cost-effectiveness curve can be developed in a continuous search at
various target values without stopping the search. The process can start with solving the optimization
problem with the highest target value. After getting the near-optimal solutions, relax the target and
resume the search. The previous solutions are kept and can be selectively used to construct the reference
set for the subsequent searches.
Simplification of the Channel/Pipe Routing Simulation
Channel/pipe routing is computationally extensive because it employs the kinematic wave flow routing
method. To reduce the computation burden, it is desirable to simplify the routing simulation during
optimization runs and only use the kinematic wave approach for evaluation runs. The possible simplified
routing options include, but are not limited to:
• Adopt the simple approach of steady flow routing (from the SWMM) for the optimization runs
• Pre-run the routing module with kinematic wave approach to build a stage-discharge relationship
and then use that relationship during the optimization runs
C. 3.4. Problem Formulation - Objectives, Constraints and Variables
Pollution vs. flood control objectives -How should the potential conflicts between meeting pollution
control and flood control objectives be reconciled? The pollution control effectiveness is usually assessed
by a continuous simulation, while flood control effectiveness is assessed by an event simulation.
One participant suggested to address flood control objectives by penalizing corresponding solutions if
flooding occurs during the long-term simulation.
Another expressed the idea of using goal programming. The approach should be to solve the event-based
flood control problem first and then, in most cases, the solution for the pollutant control will be
automatically included in it. Otherwise it is necessary to add extra dimensions in the optimization
problem formulation. Someone also mentioned that in urban land uses first-flush may be the main cause
of pollution, but in rural areas the larger storm events may be the major factor because of erosion. It was
commented that in suburban situations there will be a combination of both, therefore, both situations
should be addressed. It was suggested that one approach could be to include flood control considerations
as part of the screening stage of the analysis (i.e., narrow the search for water quality BMP's to the
subwatershed drainage areas where flood frequency is high). Another participant commented that
C-182
-------
although extreme events may be a major source of pollution or erosion, no BMPs are designed to handle
catastrophic events.
Multiobjective Optimization -How should the objective function be formulated?
The need for multiobjective optimization was recognized. Formulation was discussed in the context of
sequential analysis or various supplementation analyses of the near optimal solutions. No specific
recommendations were made on the solution of multiple objectives, although time and complexity
constraints were recognized.
Future land use management - Is the future land use management a decision variable in the BMP
placement decision? In other words, should the land use planning and water quality management be
integrated? SUSTAIN is currently designed for placing BMPs in watersheds with known existing or future
land uses.
It was noted that land use planning can have an implicit impact on the storm water management solutions.
For example, aggregating the development areas, which have a larger percentage of imperviousness, can
increase the cost-effectiveness of stormwater control practices.
Cost estimating - For estimating the cost of BMPs, what will be the level of detail required to maintain
the consistency of decision parameters used in optimization analyses?
One participant commented that the cost function is very important in decision-making and mostly
overlooked. LIDs make the cost estimation difficult because many LIDs have multiple purposes.
CAPITA, a wastewater treatment database, was mentioned for cost estimation. This database contains
realistic cost data for mostly conventional treatment units. It was also pointed out that it is difficult to
estimate the land cost. Another suggested that if the actual cost data were not available, then as long as
oo ? o
the relative costs were correct, the solutions would still be valid. It was suggested that the SUSTAIN
system allows the flexibility for users to use default data or enter locally derived cost information.
Another participant mentioned that it might be useful to use resources consumed as the surrogate for cost.
Financial resources and implementation schedule - Should constraints on financial resources and
schedules of BMP implementation be included in the optimization framework?
It was recognized that the system does not need to include schedules of BMP implementation because the
BMP options are discrete and solutions for the next target may not be inclusive of the solutions derived
under the current goal. An example was goven where there is a choice between large structures versus
small distributed systems. The funding limitation can drive the solution to either implementing the
distributed or the centralized systems, then the solutions are mutually exclusive. When there is a need for
next phase planning a separate optimization should be performed based on the future conditions.
It was commented that it is desirable to formulate the optimization problem as minimizing the cost
because if the constraint is the actual budget then the cost function needs to be accurate. Otherwise the
solution could be skewed.
C.4. Conclusions
The workshop included a thorough discussion of the tiered optimization approach, comparing top-down
and bottom-up search strategies. Expert opinions were gathered on how to prove if the optimization
solutions are good, if not the best, and how to evaluate and improve the search efficiency.
C-183
-------
In summary, the following items were identified as the major items worth considering in SUSTAIN
development and future improvement:
• Combine top-down and bottom-up search strategies for the tiered optimization
• Explore the use of classic optimization techniques, such as LP, Nonlinear Programming (NLP)
and DP, for the second tier top-down optimization
• Evaluate the employed optimization techniques by:
o using a benchmark test case with known optima
o comparing different solution techniques
o using commercial software to compare results (below are a few Web sites the experts
have mentioned):
• www.palisade.com Evolver
• www.solver.com Frontline Systems, Inc.
• www.mgc.ac.cn/genomecomp GenomeComp
• www. inria.fr/recherche/equipe s/dolphin. en .html Dolphin
• csmr.ca.sandia.gov/projects/opt.html Sandia
• Provide diverse near-optimal solutions
• Represent the distributed or site-scale BMPs using the hydrologic simulation parameters (i.e.,
depression storage and infiltration parameters)
• Explore the feasibility and options of applying the simplified channel/pipe routing approach for
optimization runs
• Explore the concept of relative cost
C-184
-------
Appendix D. Appendix References
Alley, W.M., and P.E. Smith. (1982a). Distributed Routing Rainfall-Runoff Model—Version II. Open File
Report 82-344. U.S. Geological Survey, Reston, VA.
Alley, W.M., and P.E. Smith. (1982b). Multi-event Urban Runoff Quality Model. Open File Report 82-
764. U.S. Geological Survey, Reston, VA
AnnAGNPS. Technical descriptions. (2000). http://www.sedlab.olemiss.edu/AGNPS.html. Accessed
2001.
Barfield, B.J., J.C. Hayes, K.F. Holbrook, B. Bates, J. Gillespie, and J. Fersner. (2002). IDEAL Model
User Manual. J.C. Hayes and Associates, Clemson, SC.
Beasley, D.B. Distributed parameter hydrologic and water quality modeling. (1986). In Agricultural
Nonpoint Source Pollution: Model Selection and Application, eds. Giorgini and F. Zingales, pp.
345-362.
Bicknell, B.R., J.C. Imhoff, J.L. Kittle, A.S. Donigian, Jr., and RC. Johanson, Jr. (1997). Hydrological
Simulation Program—FORTRAN, User's manual for Version 11. EPA/600/R-97/080. U.S.
Environmental Protection Agency, National Exposure Research Laboratory, Athens, GA.
Bouraoui, F., T.A. Dillaha, and S. Mostaghimi. (1993). ANSWERS 2000: Watershed Model for Sediment
and Nutrient Transport. ASAE Paper No. 93-2079. American Society of Agricultural Engineers,
St. Joseph, MI.
Camp, T.L. (1946). Sediment and the Design of Settling Tanks. Transactions of the American Society of
Civil Engineers, 111:895-936.
Chen, C.W., J. Herr, J., and L. Weintraub. (2001). Lessons learned from stakeholder process. Presented at
WEF TMDL Science Issues Conference, March 4-7, St. Louis, MO.
Chen, C.W., J. Herr, L. Ziemelis, R.A. Goldstein, and L. Olmsted. (1999). Decision Support System for
Total Maximum Daily Load. ASCE Journal of Environmental Engineering 125(7):653-659.
Cheng, M.-S., L. Coffman, J. Riverson, J. Shen, J. Ouyang, M. Lahlou, and L. Shoemaker. (2002). Low-
impact development management practices evaluation computer module. Presented at Water
Environment Federation Watershed Conference, February 24-27, Fort Lauderdale, FL.
DHI Inc. Web site, http://www.dhisoftware.com. Accessed 2004.
Dillaha, T.A., C.D. Heatwole, M.R Bennett, S. Mostaghimi, V.O. Shanholtz, and B.B. Ross. (1988).
Water Quality Modeling for Nonpoint Source Pollution Control Planning. Nutrient transport.
Report No. SW-88-02. Virginia Polytechnic Institute and State University, Department of
Agricultural Engineering, Blacksburg, VA.
Donigian, A.S., A.S. Patwardhan, and R.M. Jacobson. (1996). Wastershed Modeling of Pollutant
Contributions and Water Qaulity in the Le Sueur Basin of Southern Minnesota. In Proceedings,
Watershed '96. June 8-12, Baltimore, MD,.
Donigian, A.S., and W.C. Huber. (1991). Modeling of Nonpoint Source Water Quality in Urban and Non-
Urban Areas. EPA/600/3-91/039. U.S. Environmental Protection Agency, Washington, DC.
D-185
-------
Donigian, A.S., Jr., and A.S. Patwardhan. (1992). Modeling nutrient loadings from croplands in the
Chesapeake Bay Watershed. In Proceedings of Water Resources Sessions at Water Forum '92,
August 2-5, Baltimore, MD, pp. 817-822.
Donigian, A.S., Jr., B.R. Bicknell, L.C. Linker, J. Hannawald, C. Chang, and R. Reynolds. (1990).
Chesapeake Bay Program Watershed Model application to calculate bay nutrient loadings:
Preliminary Phase I findings and recommendations. Prepared by AQUA TERRA Consultants for
the U.S. Chesapeake Bay Program, Annapolis, MD.
Flanagan, D.C., and M.A. Nearing. (1995). USDA—Water Erosion Prediction Project: Hillslope profile
and water shed model documentation. NSERL Report No. 10. U.S. Department of Agriculture-
Agriculture Research Service, National Soil Erosion Research Laboratory, West Lafayette, IN.
Haith, D.A., R . Mandel, and RS. Wu. (1992). GWLF—Generalized Watershed Loading Functions,
version 2.0—User's manual. Cornell University, Department of Agricultural and Biological
Engineering, Ithaca, NY:.
Hamrick, J.M. (1992). A Three-Dimensional Environmental Fluid Dynamics Computer Code: Theoretical
and Computational Aspects. Special Report 317. The College of William and Mary, Virginia
Institute of Marine Science, Gloucester Point, VA.
Heaney, J. P., D. Sample, and L.Wright. (2002). Costs of Urban Stormwater Control EPA-600/R-02/021;
U.S. Environmental Protection Agency, Cincinnati, OH.
Huber, W.C. (1992). Experience with the USEPA SWMM Model for analysis and solution of urban
drainage problems. In Proceedings, Inundaciones YRedes De Drenaje Urbano, eds. J. Dolz, M.
Gomez, and J.P. Martin, pp. 199-220. Colegio de Ingenieros de Caminos, Canales Y Puertos,
Universitat Politecnica de Catalunya, Barcelona, Spain.
Huber, W.C. (2001). New options for overland flow routing in SWMM. In Proceedings of the EWRI
World Water and Environmental Resources Congress, May 20-24, Orlando, FL.
Huber, W.C. and R.E. Dickinson. (1988). Stormwater Management Model User's Manual, Version 4.
EPA/600/3-88/00la. U.S. Environmental Protection Agency, Athens, GA.
Julien, P.Y., B. Saghafian, and F.L. Ogden. (1995). Raster-based Hydrological Modeling of Spatially
Varied Surface Runoff. Water Resources Bulletin 31(3):523-536.
Kadlec, R.H., and W.W. Walker. (2003). Dynamic STA design model—Design model structure for
treatment wetland systems at low phosphorus concentrations for Everglades protection. Presented
at the Joint Conference on the Science and Restoration of the Greater Everglades and Florida Bay
Ecosystem, April 13-18, Palm Harbor, FL.
Kilinc, M., and E.V. Richardson. (1973). Mechanics of Soil Erosion from Overland Flow Generated by
Simulated Rainfall. Hydrology Papers No. 63., Colorado State University, Fort Collins, CO.
Knisel, W.G., ed. (1993). GLEAMS: Groundwater Loading Effects of Agricultural Management Systems,
Version 2.10. Publication No. 5. University of Georgia, Biological and Agricultural Engineering
Department, Coastal Plain Experiment Station, Tifton, GA.
Lee, E.R., (1999). SET-WET: A Wetland Simulation Model to Optimize NPS Pollution Control, Master
Thesis, Virginia Polytechnic Institute and State University, Department of Biological Systems
Engineering, Blacksburg, VA.
Lee, E.R., S. Mostaphimi, and T.M. Wynn. (2002). A Model to Enhance Wetland Design and Optimize
Nonpoint Source Pollution Control. Journal of American Water Resources Association 38(1): 17-
32.
D-186
-------
Medina, D.E., P. Graham, J. Thorpe, A. Patwardhan, and T. Hare. (2003). The Low Impact Feasibility
Evaluation (LIFE™) model: An interactive model for Low-Impact Development hydrology. In
Proceedings of the 2003 WEF TMDL Conference, November 16-19, Chicago, IL.
Moore, L.W., C.Y. Chew, R.H. Smith, and S. Sahoo. (1992). Modeling of Best Management Practices on
North Reelfoot Creek, Tennessee. Water Environment Research 64(3):241-247.
Munoz-Carpena, R. (1993a). Modeling hydrology and sediment transport on vegetative filter strips. Ph.D.
dissertation, North Carolina State University, Raleigh, NC.
Muiioz-Carpena, R., C.T. Miller, and J.E. Parsons. (1993b). A quadratic Petrov-Galerkin solution for
kinematic wave overland flow. Water Resources Research 29(8): 2615-2627.
Munoz-Carpena, R., and J.E. Parsons. (2003). VFSMOD-W Vegetative Filter Strips Hydrology and
Sediment Transport Modeling System: Model Documentation and User's Manual. Biological and
Agricultural Engineering, University of Florida, Gainsville.
Neitsch, S.L., J.G. Arnold, J.R. Kiniry, and J.R. Williams. (2001). Soil and Water Assessment Tool,
Theoretical Documentation, version 2000. U.S. Department of Agriculture, Agriculture Research
Service, Temple, TX.
Ogden, F.L. (2001) A Brief Description of the Hydrologic Model ofCASC2D. University of Connecticut,
Storrs, CT. http://www.engr.uconn.edu/~ogden/casc2d/casc2_desc.htm. Accessed 2004
Ogden, F.L., S.U.S. Senarath, and B. Saghafian, (1998). Use of Continuous Simulations to Improve
Distributed Hydrologic Model Calibration Uniqueness. University of Connecticut, Storrs, CT.
Pitt, R. (1987). Small storm urban flow and particulate washoff contributions to outfall discharges. Ph.D.
dissertation, University of Wisconsin, Civil and Environmental Engineering Department,
Madison, WI.
Pitt, R. (1993). Source Loading and Management Model (SLAMM). Presented at the National
Conference on Urban Runoff Management, March 30-April 2, Chicago, IL.
Pitt, R., and J. Voorhees. (2000). The Source Loading and Management Model (SLAMM), a Water
Quality Management Planning Model for Urban Stormwater Runoff. University of Alabama,
Department of Civil and Environmental Engineering, Tuscaloosa, AL.
Prince George's County. (2001). Low-Impact Development Management Practices Evaluation Computer
Module: User's Guide. Prepared for Prince George's County, MD, by Tetra Tech, Inc., Fairfax,
VA.
Scheckenberger, R.B., and A.S. Kennedy. The Use of HSPF in Subwatershed Planning. In Current
Practices inModelling the Management of Stormwater Impacts, ed. W. James, pp. 175-187.
Lewis Publishers, Boca Raton, FL.
SWET (Soil and Water Engineering Technology, Inc.) (1999). EAAMOD Technical and User Manuals.
Final reports to the Everglades Research and Education Center, University of Florida, Belle
Glade, FL.
SWET (Soil and Water Engineering Technology, Inc.). (2002). WAMView training manual. Developed
for EPA Region 4 training. Soil and Water Engineering Technology, Inc., Gainesville, FL.
Tetra Tech and USEPA (Tetra Tech, Inc., and U.S. Environmental Protection Agency). (2002). The LSPC
Watershed Modeling System—Users' Manual. Tetra Tech, Inc., Fairfax, VA, and U.S.
Environmental Protection Agency, Washington, DC.
USDA (U.S. Department of Agriculture). (2003). KINEROS—A kinematic runoff and erosion model.
http://www.tucson.ars.ag.gov/kineros/. Accessed 2004.
D-187
-------
USDA ARS (U.S. Department of Agriculture, Agricultural Research Service). SWAT Web site. U.S.
Department of Agriculture, Grassland, Soil and Water Research Laboratory, Temple, TX.
http://www.brc.tamus.edu/swat. Accessed 2004
USDA-ARS-Southeast Watershed Research Laboratory, (1999). Riparian Ecosystem Management
Model. U.S. Department of Agriculture, Agricultural Research Service, Tifton, GA.
http://www.tifton.uga.edu/remmwww/. Accessed 2004.
USDA NSERL (U.S. Department of Agriculture, National Soil Erosion Research Laboratory). (1995).
USDA Water Erosion Prediction Project (WEPP): Technical documentation. NSERL Report No.
10. U.S. Department of Agriculture, National Soil Erosion Research Laboratory, West Lafayette,
IN.
USEPA (U.S. Environmental Protection Agency). (2001). Better Assessment Science Integrating Point
and Nonpoint Sources—BASINS, Version 3.0, User's Manual. EPA-823-B-01-001. U.S.
Environmental Protection Agency, Office of Water, Washington, DC.
USEPA Region 4 (U.S. Environmental Protection Agency Region 4). (2003). TMDL toolbox.
http://www.epa.gov/region4/water/tmdl/tools/index.htm. Accessed 2004.
Walker, W.W., Jr. (1990). P8 urban catchment model program documentation, vl.l.
http://wwwalker.net/p8. Accessed 2004.
Weintraub, L.H.Z., C.W. Chen, and J. Herr. (2001). Demonstration of WARMF: a decision support tool
for TMDL development. Presented at WEF TMDL Science Issues Conference, March 4-7, St.
Louis, MO.
Wong, T.H.F., and P.P. Breen. (2002). Recent advances in Australian practice on the use of constructed
wetlands for stormwater treatment. In Global Solutions for Urban Drainage, Proceedings of the
Ninth International Conference on Urban Drainage, (CD-ROM) ed. E.W. Strecker and W.C.
Huber. American Society of Civil Engineers, September 2002, Portland, OR.
Wool, T.A., R.B. Ambrose, J.L. Martin, and E.A. Comer. (2003). Water Quality Analysis Simulation
Program (WASP), version 6.0 user's manual (draft). U.S. Environmental Protection Agency
Region 4, Environmental Research Laboratory, Athens, GA; USAGE Waterways Experiment
Station, Vicksburg, MS; Tetra Tech, Inc., Atlanta, GA.
Woolhiser, D.A., R.E. Smith, and D.C. Goodrich. (1990). KINEROS, A Kinematic Runoff and Erosion
Model: Documentation and user manual. ARS-77. U.S. Department of Agriculture, Agriculture
Research Service, Washington D.C.
Yang, C.T. (1973). Incipient motion and sediment transport. Journal of Hydraulics 99 (HY10): 1679-
1704.
Young, R.A., C.A. Onstad, D.D. Bosch, and W.P. Anderson. (1986) Agricultural Nonpoint Source
Pollution Model: A watershed analysis tool. U.S. Department of Agriculture, Agricultural
Research Service, Morris, MN.
Yu, S.L., G.M. Fitch, and T.A. Earles. (1998). Constructed Wetlands for Stormwater Management,
VTRC 98-R35RB. Virginia Department of Transportation Research Council, Charlottesville, VA.
Zhen, X.-Y. (2002). Development of Best Management Practice—BMP Placement Strategies at the
Watershed Scale. Ph.D. Dissertation, University of Virginia, Charlottesville, VA.
D-188
------- |