EFFECTIVENESS OF ROADSIDE BEST MANAGEMENT PRACTICES …
Transcript of EFFECTIVENESS OF ROADSIDE BEST MANAGEMENT PRACTICES …
University of Rhode Island University of Rhode Island
DigitalCommons@URI DigitalCommons@URI
Open Access Dissertations
2021
EFFECTIVENESS OF ROADSIDE BEST MANAGEMENT PRACTICES EFFECTIVENESS OF ROADSIDE BEST MANAGEMENT PRACTICES
(BMPS) ON MAINTAINING STORMWATER QUALITY THROUGH (BMPS) ON MAINTAINING STORMWATER QUALITY THROUGH
MONITORING AND MODELING MONITORING AND MODELING
Khurshid Jahan
Follow this and additional works at: https://digitalcommons.uri.edu/oa_diss
EFFECTIVENESS OF ROADSIDE BEST MANAGEMENT PRACTICES (BMPS)
ON MAINTAINING STORMWATER QUALITY THROUGH MONITORING AND
MODELING
BY
KHURSHID JAHAN
A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
IN
BIOLOGICAL AND ENVIRONMENTAL SCIENCES
UNIVERSITY OF RHODE ISLAND
2021
DOCTOR OF PHILOSOPHY DISSERTATION
OF
KHURSHID JAHAN
APPROVED:
Dissertation Committee:
Major Professor Soni Mulmi Pradhanang
Thomas B. Boving
Arthur J. Gold
Brenton DeBoef
DEAN OF THE GRADUATE SCHOOL
UNIVERSITY OF RHODE ISLAND
2021
ABSTRACT
A comprehensive assessment of the stormwater BMPs’ in terms of flow volume
reduction, peak flow attenuation and runoff pollutant concentration reduction was made
to evaluate the effectiveness of existing stormwater best management practices (BMPs)
on the campus of the University of Rhode Island, RI, USA.
Urban stormwater runoff became a pressing issue in the later part of the 1980s
with the recognition that point source controls were not sufficient to protect and restore
water quality. In the last 15 years, the observed annual total rainfall in the study area
was usually greater than long-term average yearly rainfall, suggesting the climate is
towards wetter conditions. In addition to changing climate conditions, the areal extent
of urban and suburban land uses also increased in Rhode Island, i.e., there is a clear link
between increased stormwater runoff and the increasing extent of impervious surfaces
in build-over areas. Compared to predevelopment conditions, impervious surfaces can
lead to a rapid rise in peak flows in systems receiving stormwater runoff. About 10% to
15% of the total impervious areas in the USA are parking lots. This percentage is ex-
pected to increase in the future. Changing drainage patterns can cause floods, channel
erosion but also carries the potential for the decreased baseflow and streambed altera-
tions.
Roadside best management practices (BMPs) are techniques or methods that aim
to prevent or reduce the overall negative impacts of stormwater runoff flow and improve
the quality of stormwater runoff cost-effectively. These BMPs can be characterized into
three types based on their performance: (i) source control, (ii) flow control and (iii)
runoff treatment. Structural stormwater treatment practices are the most common type
of water quality control BMP in Rhode Island. They include 1) wet vegetated treatment
systems, 2) infiltration practices, 3) filtering systems, 4) green roofs, and 5) open chan-
nel practices. Therefore, it is essential to integrate monitoring and modeling of the treat-
ment of water stormwater pollutants by these BMPs and to evaluate water quality risks
of surface runoff along roadsides. Therefore, the proposed study was conducted to eval-
uate the effectiveness of the roadside BMPs under peak flow conditions as a function of
runoff depth and in terms of runoff pollutant reduction. In addition, a stormwater man-
agement model was applied, and the various types of Low Impact Development (LID)
control were analyzed.
This study conducted a four-pronged analysis based on 1) RS-GIS based SCS-
CN model developed; 2) EPA SWMM 5.1 for hydrologic and hydraulic simulation; 3)
parameter estimation using the Bayesian approach, and 4) predictive modeling through
Artificial Neural Network.
Specifically, this study focuses on:
(i) EPA SWMM5.1 model set up for runoff and stormwater pollutants,
(ii) Calibration and validation of the developed model
(iii) Estimation of parameter uncertainty of SWMM using Bayesian statistics
(iv) Application of Artificial Neural Network (ANN) is predicting on the of
important storm water pollutants
The developed model confirmed the significant role of LID in reducing runoff
depth and peak flow. The established LID structures (permeable pavement, bioretention,
and vegetative swale) are effective in runoff depth, peak flow, and pollutants reduction
(Total suspended sediments, Nitrate-N, and Orthophosphate-P) for the smaller rainfall
intensities (1-yr, 2-yr, 5-yr, and 10-yr) or rainfall design. The structures are not much
effective in terms of 50-yr and 100-yr rainfall design.
In conclusion, this study provides development and evaluation of the stormwater
management model in assessing the effectiveness of stormwater best management
practices and provides a decision-making tool for the stormwater managers and planners.
v
ACKNOWLEDGMENTS
I pay my humble gratitude to The Almighty, The Most Merciful and Beneficent,
who conferred his endless blessings in completing this Ph.D. journey.
This thesis arises in the part of “Philosophy of Doctoral in Environmental and
Life Sciences” by the Department of Geosciences at the University of Rhode Island,
USA. It is a delight to pass on my appreciation to all in my modest affirmation.
First and foremost, I am incredibly grateful to my supervisor Dr. Soni M
Pradhanang, Associate Professor, Department of Geosciences, URI, for her controlled
supervision, advice, and guidance from the beginning and for giving me extraordinary
experiences throughout the work. Above all, and most importantly, she offered me
unwavering encouragement and support in a variety of ways. Her truly scientist intuition
has made her a constant oasis of ideas and passions in science, which inspired me and
enriched my growth as a researcher and a scientist. I am indebted to her more than I can
express.
A deep feeling of gratitude to my committee member, Dr. Thomas B. Boving,
Professor, Department of Geosciences, for his pivotal role in the study design, fieldwork,
advice, and support that made this work accomplished. I gratefully acknowledge my
other committee member Dr. Arthur J. Gold, Professor, Department of Natural
Resources Sciences, for his timely advice and suggestions with kindness, enthusiasm,
and dynamism. His involvement with his originality has triggered and nourished my
intellectual maturity that I will benefit from for a long time to come.
Many thanks go particularly to Dr. Rebecca Brown, Professor, Department of
Botany, for contributing to the field preparation and her valuable advice in the biological
vi
discussion and provide necessary references. In addition, I am thankful to Dr. Vinca
Craver, Department of Civil Engineering, for her permission to use her lab for necessary
lab experiments. I am also grateful to Dr. Dawn Cardace, Professor, Department of
Geosciences, for her special encouragement and precious time reviewing my article.
I like to acknowledge the financial support for this study provided by Rhode
Island Department of Road and Transport (RIDOT) SPR-234-2362, and the University
of Rhode Island. I would like to thank URI College of the Environment and Life
Sciences (CELS) for its academic support throughout my graduate study.
I convey heartfelt gratitude to Dr. Anne Veeger, Vice Provost for Academic and
Faculty Initiatives, University of Rhode Island, and Dr. Jacqui Tisdale, Assistant Dean,
Students Office, for their constructive help and kind support for my family and me
throughout the challenging time in this Ph.D. journey.
I want to show my gratitude to the Department Chair, Dr. Brian Savage, Dr.
David Fastovsky, and all the professors of the Geoscience Department, University of
Rhode Island, for their suggestions, inspiration, and guidance. Special thanks to Dr.
Gavino Puggioni, Department of Computer Science and Statistics, for his outstanding
support and teaching during all the statistics courses I took.
I am also thankful to Julie Smallridge, the department's office staffs, and my
fantastic lab mates for their cordial help. I am grateful to my dearest friends Michaela,
May, Greta for being tremendously caring and providing me a breathing space. Thanks
to Tim Sharman for his help with the field work.
Where would I be without my family? A well-deserved special mention to my
parents Late Md. Khurshaduzzaman and Rasheda Zaman for their inseparable support
vii
and prayers. I could remember my deceased father, who was reading an article in a
newspaper about a Ph.D. graduate woman named after me back in the ’90s and wished
me to be a Ph.D. graduate one day.
Words come up short to specific my appreciation to my husband Tanjil, whose
dedication, love, encouragement, and unwavering confidence in me keep my dream
alive. I owe him for unselfishly letting his intelligence, passions, and ambitions collide
with mine. I am also thankful to my pretty girl Tanzeela Tasneem Tiyana and my
beloved son, Tahfeem Ahmed Tahan, for their extraordinary patience, endurance, and
unconditional love. I am a blessed mom having such wonderful and caring kids.
Thanks to my father- in-law, Mustafa Abu Bakar Siddique, and mother-in-law,
Shirin Begum, for supporting me.
Last but not the least, Dr. Nisa Khan, President of IEM LED Lighting
Technologies, I don’t know which word will fit to express my feelings to her. I had a
long-cherished dream of higher study. It is she with whom my dream, my effort, came
into reality. She makes me learn how to enhance concepts, ideas & thinking
academically, professionally, and personally. Her inspiration, motherly affection, and
support shape the pillar for my dream. Her unconditional love and protective care in
joys and sorrows make my journey smoother and prepare me to take new challenges. I
am grateful to Dr. Fred Heisman for his outstanding support and love towards my family
and me.
Finally, I would like to thank everyone important to the successful realization of
the thesis and express my apology that I could not mention personally one by one.
ix
PREFACE
This Ph.D. dissertation is written in a manuscript format adhering to the
guidelines of University of Rhode Island Graduate School. There are six main chapters
in this dissertation and one supplementary chapter documenting data.
The first chapter, entitled “ Surface Runoff Responses to Suburban Growth: An
Integration of Remote Sensing, GIS, and Curve Number” authored by Khurshid Jahan,
Soni M Pradhanang, and Mohamad Ehsan Bhuiyan, is a published article in the Land
Journal (Land 2021, 10(5), 452; https://doi.org/10.3390/land10050452). The second
chapter entitled “Modeling Runoff Quantity with different LID structures in the
Suburban Watershed using Stormwater Management Model (SWMM)” authored by
Khurshid Jahan, Soni M Pradhanang, and Thomas Boving in under review in the
Hydrological Sciences Journal-IAH. The third chapter entitled “Performance
Evaluation of LID Controls for Runoff Quality Using SWMM in Suburban Watershed”
is authored by Khurshid Jahan, Soni M. Pradhanang, and Rebecca Brown is in
preparation for the part II submission to the Hydrological Sciences Journal-IAH. The
fourth chapter is titled “Predictive Uncertainty Estimation of Stormwater Management
Model using Bayesian Markov Chain Monte Carlo approach” and is authored by
Khurshid Jahan, Soni M Pradhanang, and Russell D. Briggs and is prepared for
Hydrological Processes Journal. The fifth chapter entitled “Predicting Runoff Chloride
Concentrations in Suburban Watersheds Using an Artificial Neural Network (ANN)”
authored by Khurshid Jahan and Soni M Pradhanang is a published article in the
Hydrology Journal ((Hydrology 2020, 7(4),
80; https://doi.org/10.3390/hydrology7040080)
x
The sixth chapter contains major conclusions, the significance of the study and
provides future research directions associated with this dissertation. The supplementary
chapter entitled “How effective are the bioswales in improving water quality? A case
study of a sub-urban catchment” provides an overview of field and laboratory analysis
data along with statistical analysis.
xi
TABLE OF CONTENTS
ABSTRACT ……………………………………………………………….………….ii
ACKNOWLEDGEMENT………………………………………………….………….v
DEDICATION………………………………………..……………………………...viii
PREFACE……………………………………………………………………………. X
LIST OF TABLES ..................................................................................................... xvii
LIST OF FIGURES ..................................................................................................... xx
CHAPTER 1 ................................................................................................................. 1
Manuscript І ................................................................................................................. 1
1. Introduction ................................................................................................................ 3
2. Study Area .................................................................................................................. 8
3. Methodology ............................................................................................................ 10
3.1. Surface Runoff Model ....................................................................................... 10
3.2. Integrated RS-GIS Approach to Surface Runoff Modeling .............................. 12
3.2.1. Land-Use and Land Cover Type Using Remote Sensing ........................... 12
3.2.2. Hydrogeological Parameter Determination Using GIS .............................. 15
Derivation of Soil Data ............................................................................................. 16
Derivation of Slope, Elevation, and Stream Order Data ...................................... 16
Rainfall Data......................................................................................................... 17
3.2.3. Hydrological Modeling within GIS ............................................................ 19
4. Results and Discussion: ........................................................................................... 20
4.1. Suburban Growth in the Study Area ................................................................. 20
4.2. Impact of Suburban Growth on Surface Runoff ............................................... 22
4.3. Impact of Suburban Growth on Rainfall-Runoff Relationship ......................... 26
5. Conclusions .............................................................................................................. 27
References .................................................................................................................... 29
xii
CHAPTER 2 ............................................................................................................... 36
Manuscript ІІ .............................................................................................................. 36
1. Introduction .............................................................................................................. 38
2. Study Area and Data Collection ............................................................................... 42
2.1 Study Area .......................................................................................................... 42
2.2 Local Precipitation ............................................................................................. 43
2.2.1. Observed Rainfall Data ............................................................................... 43
2.2.2. Rainfall Data Preparation............................................................................ 44
2.3 Drainage Pipeline Data ....................................................................................... 45
2.4 Elevation, Sub-catchment, Soil Infiltration, Land use and % Impervious Surface
.................................................................................................................................. 47
3. Methodology ............................................................................................................ 49
3.1. Hydrological Model Selection ........................................................................... 49
3.3 Model Calibration and Validation ...................................................................... 53
3.4.2 Permeable Pavement (PP) Scenario ............................................................. 55
3.4.3 Bio-Retention (BR) Cell Scenarios.............................................................. 56
3.4.4 Vegetative Swale (VS) Scenarios ................................................................ 58
3.5 Assessment of the LID envrionmental benefits ............................................. 58
3.5.1 Calculation of Runoff and Peak Reduction ............................................... 59
3.5.2 Calculation of LID environmental benefits ................................................. 59
4. Result and Discussion .............................................................................................. 60
4.1 SWMM Calibration and Validation ................................................................... 60
4.2 Results in Baseline Scenario .............................................................................. 63
4.3 Runoff Volume Reduction ................................................................................. 64
4.4 Peak Discharge Reduction .................................................................................. 66
4.5 Environmental Benefits .................................................................................... 68
5. Conclusions .............................................................................................................. 69
References .................................................................................................................... 71
CHAPTER 3 ............................................................................................................... 79
xiii
Manuscript III ............................................................................................................ 79
Abstract ........................................................................................................................ 80
1. Introduction ............................................................................................................ 81
2. Methodology .......................................................................................................... 86
2.1 Study Area .......................................................................................................... 87
2.3 Data collection, source, and preparation ........................................................ 93
2.3.1 GIS data ....................................................................................................... 93
2.3.2 Sample collection and analyses ................................................................... 95
2.4 Scenarios of the developed model ...................................................................... 96
2.1.4 LID Selection and Developed Scenarios ..................................................... 96
2.5. Model performance evaluation .......................................................................... 98
2.5.1. Sensitivity analysis ...................................................................................... 98
2.5.2 Calibration and Verification of SWMM Models .......................................... 99
3. Result and Discussion ............................................................................................ 101
3.1 Model Performance Evaluation ........................................................................ 101
3.1.1 Parameters sensitivity analysis .................................................................. 102
3.1.2 Model Calibration and Validation ............................................................. 106
3.2 LID implementation on stormwater quality ..................................................... 111
3.2.1 Comparison of stormwater management scenarios .................................. 111
4. Conclusion ........................................................................................................... 117
CHAPTER 4 ............................................................................................................. 126
Manuscript IV .......................................................................................................... 126
Abstract ...................................................................................................................... 127
1. Introduction .......................................................................................................... 128
2. Study Area, Data & Materials .............................................................................. 133
3. Material & Methodology...................................................................................... 135
3.1 SWMM Model ................................................................................................. 135
xiv
3.2 Objective Function ........................................................................................... 136
3.3 Bayesian Approach and Error Modeling .......................................................... 138
3.4 Markov Chain Monte Carlo (MCMC) ............................................................. 140
3.5 Sensitivity Analysis (SA) ................................................................................. 141
4. Findings and Discussions ....................................................................................... 144
4.1 Sensitivity Analysis .......................................................................................... 144
4.2 Model Efficiency .............................................................................................. 148
4.3 Parameter Uncertainty ...................................................................................... 150
4.4 Predictive Uncertainty ...................................................................................... 151
5. Conclusion ............................................................................................................. 154
CHAPTER 5 ............................................................................................................. 161
Manuscript V ............................................................................................................ 161
1. Introduction ............................................................................................................ 163
2. Materials and Methods ........................................................................................... 166
2.1. Artificial Neural Network (ANN) ................................................................... 168
2.2. Back Propagation (BP) Algorithm .................................................................. 169
2.2. Curve Fitting Algorithm .................................................................................. 170
2.4. Density Distribution algorithm ........................................................................ 170
2.5. Model Structure ............................................................................................... 171
2.5.1. ANN Parameter Selection: Hidden Layers and Nodes ............................. 173
2.5.2. ANN Parameter Selection: Learning Rate and Momentum ..................... 173
2.5.3. ANN Parameter Selection: Initial Weights ............................................... 174
2.5.4. ANN Parameter Selection: Selection of Input Variables.......................... 174
2.5.5. ANN Parameter Selection: Data Partition ................................................ 175
2.5.6. ANN Parameter Selection: Model Performance Evaluation .................... 176
3. Results and Discussion ........................................................................................... 177
3.1. Model Output .................................................................................................. 177
3.2. Curve Fitting Analysis ..................................................................................... 179
3.3. Density Distribution of the Predicted Chloride Concentration ....................... 181
xv
3.4. Cross Validation Based on Snow and Precipitation Events ............................ 182
4. Conclusion ............................................................................................................. 186
References .................................................................................................................. 189
CHAPTER 6 ............................................................................................................. 193
CONCLUSION ...................................................................................................... 193
Scope of the work ..................................................................................................... 196
CHAPTER 7 ............................................................................................................. 197
Supplementary ......................................................................................................... 197
Abstract ...................................................................................................................... 198
1. Introduction .......................................................................................................... 199
2. Site Description .................................................................................................... 202
3. Materials and methods ........................................................................................... 204
3.1 Study Design .................................................................................................... 204
3.2 Rainfall pattern and size ................................................................................... 205
3.3 Monitoring Data Collection and Analysis ........................................................ 205
3.4 Turbidity ........................................................................................................... 207
3.5 Chloride ............................................................................................................ 209
3.5 Statistical Analysis ........................................................................................... 210
3.5.1 One-way ANOVA ..................................................................................... 211
3.5.2 Two-way ANOVA ..................................................................................... 211
3.6 Reduction Rate ................................................................................................. 214
5. Result and Discussion ............................................................................................ 214
5.1 Nitrate-N ........................................................................................................... 215
5.1.1 analysis on one-way and two-way ANOVA ................................................. 215
5.1.2 Reduction rate of the Nitrate-N ................................................................. 220
5.2 Orthophosphate-P (PO4-P) ............................................................................... 222
5.2.1 analysis on one-way and two-way ANOVA ................................................. 222
xvi
5.2.2 Reduction rate of the Orthophosphate-P (PO4-P) ...................................... 227
5.3 Total Suspended Sediments (TSS) ................................................................... 229
5.2.1 Analysis of One-way and Two-way ANOVA............................................... 229
5.2.2 Reduction rate of the Total Suspended Sediments (TSS) .......................... 233
6. Conclusion ............................................................................................................. 235
References .................................................................................................................. 236
xvii
LIST OF TABLES
TABLE PAGE
Chapter One
Table 1. General characteristics of four candidate image scenes…….…….…..….….10
Table 2. General characteristics of four candidate image scenes………………….….15
Table 3. Satellite-detected suburban expansion in South Kingstown, Rhode Island…21
Chapter Two
Table 1: Point Rainfall Frequency (pf) estimates with 90% confidence intervals and
supplementary information (NOAA Atlas 14, volume 10, version 3)……….....….…44
Table 2: Categories of the rainfall………………………….…………………………47
Table 3. Summary of Subbasins Characteristics …….……………………………….50
Table 4: Summary of LID Characteristics……………………………………………57
Table 5: SWMM calibration parameter………………………………………………61
Table 6: Comparison between three calibrations run………………...……………….63
Table 7. Percentage of runoff and peak reduction per unit area under six rainfall
reoccurrence scenarios…………………………………..……………………………65
Table 8: Environmental benefits for different LID structures………………………..68
xviii
Chapter Three
Table 1: Percentage of runoff depth and peak reduction per unit area under six rainfall
re-occurrences ………………………………………..……………………….….…..85
Table 2: Input parameters of the sub-catchment for the developed SWMM-LID……88
Table 3: Buildup and Wash off and parameters of pollutants on different land-use
types …………………………………………………………….…………………....92
Table 4: Data source of the required parameter of SWMM-LID……….….….…..…94
Table 5: Parameters of three LID controls used in the LID-SWMM Simulations.......97
Table 6: Model sensitive Parameter and their ranking based on the sensitivity
analysis…………………………………………………………………..…………..106
Table 7: Adjusted SWMM calibration surface layer and soil parameters………..…107
Table 8: Model Calibration and validation output………………...………………...108
Chapter Four
Table 1: Model Parameter and Ranges used for Calibration and Uncertainty
Analysis……………………………………………………………………………...138
Table 2: Model Parameter Settings………………………………………………….148
Table 3: Table 3: Model Calibration and Validation Output……………….……….149
Chapter Five
Table 1. ANN model output for all the sites…………….…………………………..178
Table 2. Summary of the curve fitting assessment…………………………….……181
xix
Chapter Seven [Supplementary]
Table 1: Monitoring Strategies……………………………...………………………203
Table 2: Types of rainfall based on rainfall intensity……………………………….205
Table 3: Summary of the parameters and analytical method used, including detection
limits (MDL)…………………………………………………………….…………..206
Table 4: Analysis of variance for linear regression………….………...……………218
Table 5: Categorical variables……………………………………………………….213
Table 6: Model selection based AIC for Nitrate-N, Orthophosphate-P, and
TSS…………………………………………………………………….………….....216
Table 7: Nitrate-N response on treatment type………………….…………………..218
Table 8: Orthophosphate-P response on treatment type……………………….……226
Table 9: TSS response on treatment type…………………………………...……….232
xx
LIST OF FIGURES
FIGURE PAGE
Chapter One
Figure 1. A glimpse of flood events from 2010 (a,b) and 2013 (c,d) in the conducted research area. Source: 2010 flood [56] and 2013 flood [60]……………………….….7
Figure 2. The geographical location for the study site: South Kingstown, RI, USA….9
Figure 3. The Process of Land-use and land cover Change Detection………...……..14
Figure 4. Hydrogeological parameters for the study area: (a) Stream order, (b) Hydro-logic soil group (HSG), (c) Slope, (d) Elevation…………………………………..…18
Figure 5. A schematic representation of the geographic information system (GIS) based soil conservation service curve number (SCS-CN) model in estimating the sur-face runoff…….……………………………………………….……………………...20
Figure 6. Land type change detection from 1994 to 2020…………………..………..22
Figure 7. Runoff depth change detection and area from 1994 to 2020…………….....23
Figure 8. Storm Runoff depth of 2020, 2014, and 1994……………………………...25
Figure 9. Rainfall-runoff relationship for the study period……………..…….....……27
Chapter Two
Figure 1: (a) SWMM, the stormwater Water Management Model, program
configuration (b) Overview of the SWMM structure indicating linkage among the
computational blocks. Receiving water simulation is by external programs (Adapted
from Huber and Dickinson, 1988)…………………..………….…………….……….40
Figure 2: Land-use changes based on 36 years of data in the study area adopted from
the recent research paper (Jahan et al. 2021)………………………….………….…..41
Figure 3: Location Map of the Study Area…………………………...………………43
Figure 4. Typical design rainfall processes of different return periods of 1-, 2-, 5-, 10-,
25-, 50-, and 100- yrs. (PDS: Precipitation Data Server) ……………………………45
Figure 5. Elevation and flow path, Soil infiltration rate, Land use, and the percentage of
impervious surface in the study area………………………………….…..…………..46
xxi
Figure 6: Sub-catchments, Nodes, and Streamflow Delineation of the watershed into
sub-basins of the study area with nodes, streamflow, and outlets…………..……..…48
Figure 7: Time series of the observed (green) and simulated (red) discharge [m3/s] for
Run 1, 2, and 3 during the calibration period. The precipitation rate [mm/sec] for the
corresponding time of discharge is included on the secondary axis (blue)……...……62
Figure 8: Simulated runoff processes in four LID structures with different return-
period events: (a) One year; (b) Two-year; (c) five-year (d) ten-year (e) 50-year, and
(f) 100 year…………………………………………………………………...….……64
Figure 9. Hydrological effectiveness of each LID per unit area……………...………65
Figure 10: Environmental benefits and its standard error for each LID structure in
different rainfall intensities. The standard error of the mean was applied to construct a
confidence interval (95%) for the environmental benefits for each of the LID
structures. ………………………………………………………………………….....69
Chapter Three
Figure 1: Subdivision of the conducted study area for developed SWMM-LID……..88
Figure 2: Flowchart of applied SWMM-LID framework………………………...…..93
Figure 3: a) Model runoff results of sub-catchments for hydrological parameters; b)
Model runoff results of the outfall of the catchment concerning hydraulic parameters.
[In SWMM, time patterns allow external dry weather flow to vary periodically. They
comprise of a set of adjustment factors applied as a multiplier (SWMM 5.1
manual]…………………………………………………………………………..…..104
Figure 4: Buildup and Wash-off parameter sensitivity analysis a) TSS; b) Nitrate-N; and
(c) Orthophosphate-P…………………………………………………......…………105
xxii
Figure 5: Calibration and Validation of the developed SWMM (a) Run 1 in Sub-
Catchment 3 and (b) Run 2 in Sub-catchment 6…………………………………….108
Figure 6: The observed and predicted TSS concentration (a) Site 1 (Agricultural Land)
(b) Site 5 (Parking Lot)…………………………………………………...…………109
Figure 7: The observed and predicted Nitrate-N concentration (a) Site 1 (Agricultural
Land) (b) Site 5 (Parking Lot)…………………………………………………….…110
Figure 8: The observed and predicted Orthophosphate-P concentration (a) Site 1
(Agricultural Land) (b) Site 5 (Parking Lot)……………………………..…………110
Figure 9: Performance of Bioretention for six different sites (a) TSS, (b) NITRATE-N,
and (c) Orthophosphate-P………………………………………..….………………113
Figure 10: Performance of Permeable Pavement for six different sites (a) TSS, (b)
NITRATE-N, and (c) Orthophosphate-P…………………………………….……..114
Figure 11: Performance of vegetative swale for TSS, NITRATE-N, and
Orthophosphate-P removal…………………………………………………………116
Chapter Four
Figure 1: (a)- Traditional Model, (b)- Uncertainty Model (adapted and modified from
[15]………………………………………………………………………….……….130
Figure 2: Location Map of the Study Area…………………………...……………..134
Figure 3: a) Model runoff sensitivity analysis of sub-catchments for a) hydrological
parameters; b) the outfall of the catchment concerning hydraulic parameters [In SWMM,
time patterns allow external dry weather flow to vary periodically. They comprise of a
xxiii
set of adjustment factors applied as a multiplier (SWMM5.1
Manual)]…………………………………………………………………….……….145
Figure 4: Sensitive Parameters Response of SWMM Quantity Module (a) Main
Response of the selected parameter (b) first-order SI indices represents the response of
each of the variables (c) Total Effects SI indices measure the portion of variability that
is due to total variation in each input or variable (d) Main effects for each input variables
which obtained as a by-product of the variance analysis with mean 90% error interval
(e) Uncertainty distribution of Sensitive parameters…………………………….….146
Figure 5: Sensitive Parameters Response of SWMM Quality Module (a) Main Response
of the selected parameter (b) first-order SI indices represents the response of each of the
variables (c) Total Effects SI indices measure the portion of variability that is due to
total variation in each input or variable (d) Main effects for each input variables which
obtained as a by-product of the variance analysis with mean 90% error interval (e)
Uncertainty distribution of Sensitive parameters……………………………………148
Figure 6: (a) Posterior distribution of Intercept (b) 10000 Iteration of Monte Carlo
Simulation with the random progression of the intercept of the model (c) Posterior
distribution of Slope (d) 10000 Iteration of Monte Carlo Simulation with the random
progression of the slope of the model……………………….…………………….…150
Figure 7: (a) Predictive Uncertainty Band (b) Monte Carlo simulation of the
data…………………………………………………………………………….…….152
Figure 8: Predictive Uncertainty Band Monte Carlo simulation of a) TSS, b) Nitrate-
N, and c) Ortho phosphate P…………………………….…………………………..154
xxiv
Chapter Five
Figure 1. Chloride concentration trends from 1992 to 2012 in the USA show regional
differences (modified from United States Geological Survey Report, 2014)…….…165
Figure 2. The map illustrates the location of the study area in Rhode Island, USA. Red
bounding box shows the exact location of the study site shown above (Google Earth
screenshot- on the right)………………………..…………….. …………………….166
Figure 3. Variability in observed chloride concentrations in runoff are shown, from
2016 to 2019………………………………….……………………………………..167
Figure 4. This diagram illustrates the structure of a conventional feed-forward back-
propagation neural network model……………………………….…………………172
Figure 5. This diagram illustrates the structure of the artificial neural network (ANN)
model developed in this work……………………………………………..………..172
Figure 6. Curve fitting assessment between ANN output and target data of six differ-
ent sites. Site 1, 3, 4, and 5 showed a similar trend, not scattered data. On the other
hand, site 2 and 6 has scattered data point, but 80% of the dataset were in the confi-
dence interval…………………………………………………………..….……….180
Figure 7. Density distributions (ANN outputs) for chloride concentrations at all sites
reveal shifting means. Spatial distribution covered <50% for sites 1, 2, 3, and 4
whereas sites 5 and 6 covered >70% of predicted chloride……….…….…..…….183
xxv
Figure 8. Precipitation and snow events (a) 2017, (b) 2018, and (c) 2019 in South Kings-
town, RI. (a) total four significant snow events (>10 cm) and low to moderate precipi-
tation events appeared during winter in 2017; (b) three significant snow (>10 cm) events
(among them one was the worst >25 cm) and low to moderate with high-frequency pre-
cipitation occurred during winter in 2018; and (c) only one significant snow and low to
moderate precipitation happened during the winter in
2019……………………………………………………………………………….....184
Figure 9. Predicted concentrations for three sites reflect heavy Cl inventory in Winter
2018…………………………………………………………….……………………185
Chapter Seven [Supplementary]
Figure 1: Roadside Best Management Practices Sites within the URI Campus..…...203
Figure 2: An overview of Turbidity in the Six different monitoring sites…………..208
Figure 3: An overview of Chloride concentration in the Six different monitoring
sites………………………………………………………………………….…….…209
Figure 4: Statistical parameter of the different types of rainfall…………………….216
Figure 5: Statistical parameter of the different types of seasons…………...……….216
Figure 6: Residuals and Normal Q-Q plot of the developed model……….………...218
Figure 7: Nitrate-N response to treatment types in each of the sites………...…...…219
Figure 8: Comparison of Nitrate concentration reduction between control and treatment
site. Blue colored symbol represented the reduction (%), and the red color symbol
xxvi
represented the negative reduction for the treatment site. The dot lines separate Pre-
treatment and post-treatment periods………………………..……………………….222
Figure 9: Statistical parameter of the different types of rainfall…………………….223
Figure 10: Statistical parameter of the different types of season……………………224
Figure 11: Residuals and Normal Q-Q plot of the developed model………….…….225
Figure 12: Ortho-phosphate response to treatment types in each of the sites ……....226
Figure 13: Comparison of Ortho-phosphate concentration reduction between control
and treatment site. Blue colored symbol represented the reduction (%), and the red
color symbol represented the negative reduction for the treatment site……………..228
Figure 14: Statistical parameter of the different types of rainfall…………….……..230
Figure 15: Statistical parameter of the different types of season……………...…….231
Figure 16: Residuals and Normal Q-Q plot of the developed model………………..232
Figure 17: TSS response to treatment types in each of the sites………………...…..233
Figure 18: Comparison of TSS concentration reduction between control and treatment
site. Blue colored symbol represented the reduction (%), and the red color symbol
represented the negative reduction for the treatment site………….....……………..234
1
CHAPTER 1
Manuscript І
Surface Runoff Responses to the Suburban Growth: An Integration of Remote
Sensing, GIS, and Curve Number
Published in Land 2021, 10(5), 452
Khurshid Jahan1, Soni M. Pradhanang1,*, Md Abul Ehsan Bhuiyan2
1Department of Geosciences, University of Rhode Island, Kingston, RI 02881
2Department of Civil and Environmental Engineering, University of Massachusetts,
Amherst, MA 02115, USA
*Correspondence: [email protected] (S.M.P.)
2
Abstract: Suburban growth and its impacts on surface runoff were investigated using
the soil conservation service curve number (SCS-CN) model, compared with the inte-
grated advanced remote sensing and geographic information system (GIS)-based inte-
grated approach, over South Kingstown, Rhode Island, USA. This study analyzed and
employed the supervised classification method on four Landsat images from 1994,
2004, 2014, and 2020 to detect land-use pattern changes through remote sensing appli-
cations. Results showed that 68.6% urban land expansion was reported from 1994 to
2020 in this suburban area. After land-use change detection, a GIS-based SCS-CN
model was developed to examine suburban growth and surface runoff estimation. The
developed model demonstrated the spatial distribution of runoff for each of the studied
years. The results showed an increasing spatial pattern of 2% to 10% of runoff from
1994 to 2020. The correlation between runoff co-efficient and rainfall indicated the sig-
nificant impact of suburban growth in surface runoff over the last 36 years in South
Kingstown, RI, USA, showing a slight change of forest (8.2% area of the total area) and
agricultural land (4.8% area of the total area). Suburban growth began after 2000, and
within 16 years this land-use change started to show its substantial impact on surface
runoff. We concluded that the proposed integrated approach could classify land-use and
land cover information to understand suburban growth and its potential impact on the
area.
Keywords: urbanization; suburban growth; land-use and land cover
3
1. Introduction
Urbanization and suburban growth increase challenges to the surface water bod-
ies, including flooding, channel erosion, water quality degradation, biodiversity, and
climate [1,2]. The current global environmental change pattern based on land use and
land cover is of concern [3]. Urbanization, particularly uncontrolled expansion associ-
ated with urban sprawl, is fueled by population growth and increasing demand for resi-
dential areas. Most notably in the environmental sector, rapid urbanization degrades
watershed functions and reduces agricultural and forest lands [3–5]. Eventually, the
quick changes in land-use and land cover impact the annual water balance locally and,
potentially, regionally [6–8], with severe consequences on the frequency, volume, and
peak rates of surface runoff. The proportion of rainfall that becomes surface runoff in-
creases along with increases in imperviousness of a watershed. Infrastructural develop-
ments such as building construction, residential development, lack of green areas, and
impervious surfaces such as parking lots accelerate runoff [9]. An increase in impervi-
ous surfaces and built-up land is more vulnerable to flooding than the surrounding en-
vironment. Higher runoff volumes lead to increased occurrences of flood and expansion
of floodplains. Therefore, land-use changes associated with urban development, vege-
tation, and hydrologic conditions are the major factors that affect urban flooding in
many ways [10], leading to environmental degradation. [8,11]
Recent investigations have focused on characterizing land-use changes and their
adverse effects on landscape characteristics that generate flood hazards, including inun-
dation and erosion [11]. Therefore, information on land-use and land cover (LULC),
4
which greatly influence hydrologic applications and water quality [12–14], aids in fully
understanding urbanization and its impact on local hydrology and water quality [15,16].
Urban growth has various stages, and each stage affects local hydrology in different
ways [8–14]. For example, during the first stage of urbanization, hydrology is changed
due to the removal of trees and vegetation [14,17], leading to decreased interception,
evapotranspiration, and sedimentation. During the second stage of urbanization, houses,
commercial buildings, streets, culverts, and parking lots increase imperviousness,
thereby, affecting and changing the water balance. Specifically, this stage increases
storm flows and runoff depth and degrades surface and groundwater quantity and qual-
ity in urban and suburban areas. The decrease in infiltration often leads to increased
runoff flashiness and peak discharge for even small-sized storms. As a result, flood be-
come a significant concern for highly and moderately urbanized and suburban areas.
The combination of a remote sensing (RS) and a geographic information system (GIS)
approach can viably assess such hydrologic changes [18–22] and produce a LULC de-
tection map for assessing the detailed changes in an area [23]. RS has made an immense
contribution to detecting change in the LULC which has helped researchers to think
about the impact of LULC modifications [23,24].
Remote sensing (RS) data products are cost-effective and readily available as
inputs to hydrologic and watershed models [25–27]. The RS technique gathers multi-
spectral, multiresolution, and multitemporal data or images and then transforms the im-
ages into information for urban land-cover datasets [7]. Multispectral bands play a vital
role in numerous urban growth studies, emphasizing the necessity for advanced land-
5
use and land cover change information for remote sensing optical satellite imagery ap-
plication [28–30]. Multispectral satellite images have been used as source data for
LULC change and water body detection applications since the 1960s [31]. Multispectral
remote sensing includes the acquisition of visible, near-infrared, and short-wave infra-
red images in many broad wavelength bands [32]. Different materials reflect and absorb
differently at different wavelengths, and this absorbance and emissivity characteristic
are used to detect LULC changes [33].
Multispectral images are primarily applied in the detection of urban land-use
changes rather than hyperspectral images [34,35]. Digital data in the form of satellite
images enable accurate computing of various LULC categories and help maintain the
spatial data infrastructure, essential for monitoring urban expansion and land-use studies
[36]. Consequently, RS and GIS data products are effectively used in the watershed,
town, urban, and regional planning [37]. Since most surface runoff modeling and land-
cover detection parameters are geographic data, integration of RS and GIS techniques
is expected to be more effective in evaluating the impacts of urban LULC. GIS has been
extensively applied in hydrologic models [27,38,39] because of its spatial analysis func-
tion, especially for model data preparation, model input parameters extraction, or model
output visualization [39].
The Soil Conservation Service Curve Number (SCS-CN) method, one of the
most widely used in rainfall-runoff modeling, was developed by the Natural Resources
6
Conservation Service (SCS), U.S. Department of Agriculture (USDA) [40–42]. Numer-
ous studies [43–48] showed that the SCS-CN method was developed beyond its original
scope and turned into an integral part of simulation models. This method was applied
for different landscape structures, soils, and climate conditions [49–53]. These research
outputs indicated that the SCS-CN runoff method could be used effectively for large
and small watershed areas. This study combined RS and GIS to measure stormwater
runoff and establish a relationship between rainfall and the stormwater runoff coeffi-
cient (i.e., the ratio of surface runoff to total rainfall). Stormwater runoff coefficient, a
crucial parameter in hydrology, is frequently used to examine the impacts of urban
LULC, which leads to urban runoff generation [54]. Stormwater runoff, pollutants man-
agement, and subsequent accumulation in soils, surface water, and groundwater pose
significant challenges to many Federal and State agencies. These challenges warrant a
better assessment of the impacts of suburban growth on runoff changes and local hy-
drology.
Due to rapid urbanization and its impact on the environment, researchers often
concentrate on highly urban areas to detect LULC changes and degradation rather than
focusing on suburban areas. Because of population demand and the interest in and need
for infrastructural development, suburban land-use is also gradually changing over time,
turning into urban areas. Only a limited number of studies have been done on suburban
LULC change detection, and no research has been conducted in the selected suburban
area. A recent study in this suburban area has revealed high chloride concentration in
the impervious zone due to road salt during the winter season [12]. The study area also
7
experienced two significant flood events within a short duration; one happened in March
2010, and another occurred in March 2013 (Figure 1) [55–62]. The 2010 flood event
resulted from the combination of the March Nor’easter storm and reservoir manage-
ment, while the 2013 flood was a precipitation event. The flood duration for 2010 and
2013 was three days and one day, respectively, but the damage amount was massive.
The magnitude of these flood events and damages has cost the urban and suburban in-
frastructure, which led us to investigate the land-use changes and assess the impact of
suburban growth on surface runoff.
Figure 1. A glimpse of flood events from 2010 (a,b) and 2013 (c,d) in the conducted
research area. Source: 2010 flood [56] and 2013 flood [60].
This study included an integrated approach to estimate the surface runoff and
examine suburban growth effects on LULC pattern change. The main objectives are: (a)
8
to detect suburban LULC changes using satellite RS and GIS and to study spatial pat-
terns of suburban growth; (b) to examine the effect of such suburban growth on surface
runoff generation, and (c) to detect the most affected area within the urban watershed.
2. Study Area
This study is focused on a suburban municipality (South Kingstown) in southern
Rhode Island, USA (Figure 2). According to the definition of ‘suburban’ [63], the pop-
ulation density is usually between 1000–1200 individuals per square kilometer. In our
study, the population is slightly lower, i.e., 206.2 per square kilometer according to the
2010 census, than the population for defined suburban areas. The study area is predom-
inantly flat, low relief, and the annual average rainfall is approximately1.34 m [62].
Based on bore log data, the groundwater level in this area is 3.12 m from the surface,
and the upper part of the aquifer is dominated by sand with some gravel [64,65]. The
geology of the aquifer, combined with the shallow groundwater table and the prevalence
of impervious surfaces, classifies this system’s water as a high-risk area for pollution-
induced by surface runoff [65].
This study used four multispectral (MS) satellite images at 30 m spatial resolution
from USGS EarthExplorer [63]. Specifically, to ensure the original image’s spectral
quality, we used MS imagery, which includes standard red-green-blue (RGB) channels
and narrow spectral channels from near- or middle-infrared regions’ reflectance spectra
[29,30]. Four candidate image scenes were chosen from four different years, and their
respective features are presented in Table 1. It is noted that each image scene was
9
radiometrically corrected using a relative radiometric correction method [66] with Erdas
Imagine. The scanline error correction was also conducted for the Landsat 7 ETM im-
age. Scanline error correction is required for LANDSAT 7 ETM images except for
LANDSAT 5 TM. Striping is one of the limitations for LANDSAT 7 ETM images. The
striping arises due to the scanline corrector (SLC) failure in 2003 [33,35]. Finally,
Figure 2. The geographical location for the study site: South Kingstown, RI, USA.
10
Surface runoff estimates were obtained from these satellite image scenes applying GIS
application.
Table 1. General characteristics of four candidate image scenes.
S.No.
Date of
Images
Satellite Sensors
Resolution
(m)
Bands
Thermal
Band
1. 09-19-1994 LANDSAT-5 TM 30 7 6
2. 09–14-2004 LANDSAT-5 TM 30 7 6
3. 09-18-2014 LANDSAT-7 ETM 30 8 6
4. 08-09-2020 LANDSAT-8 OLI 30 11 10 and 11
3. Methodology
3.1. Surface Runoff Model
In this research, we applied the Soil Conservation Service (SCS) method in rainfall–
runoff modeling. The SCS method utilizes several significant factors such as soils, wa-
tershed characteristics, i.e., slope, elevation, shape, and land-use over the study area
[67,68]. The other two noteworthy factors that affect runoff are rainfall duration and
intensity. The SCS-CN model was extensively used to determine the CN, ranging from
0–100 [67]. The estimates of surface runoff depend on the potential retention in the
catchment. Surface runoff is largely impacted by three factors, i.e., interception, surface
retention, and infiltration, which vary for different soil types. The SCS-CN equation is
mathematically represented as Equation (1):
11
Q =(P − 0.2S)2
(P + 0.8S) (1)
Here, Q indicates storm runoff which is estimated from rainfall (P), and S specifies
maximum potential storage and is defined as,
S = (1000
CN) − 10 (2)
Here, CN indicates the runoff curve number of a hydrologic soil group–landcover com-
plex. Two parameters are required to solve the equation: rainfall (p) and CN. The rain-
fall data is collected from the National Oceanic and Atmospheric Administration
(NOAA). CN is used to estimate the runoff from rainfall, ranging between 30 to 100,
based on soil properties of land types (Table 2). In this study, rainfall-runoff depth is
estimated for nine different kinds of land type: (1) Agricultural land, (2) Commercial
land, (3) Forest, (4) Grass and pasture, (5) Residential, (6) Industrial, (7) Open Space,
(8) Parking lot and street, and (9) Water. The permeability characterizes the hydrologi-
cal soil group (HSG) (A, B, C, D). The infiltration rate is higher in group A (>0.30 in/h)
or >7.62 mm/h) even when the soil is thoroughly wetted, while group D has the lowest
permeability and infiltration rate (0–0.05 in/h or 01.27 mm/h) for the runoff. Group B
(0.15–0.30 in/h or 3.81–7.62 mm/h) and C (0.05–0.15 in/h or 1.27–7.62 mm/h) soils are
intermediate between groups A and D. Group A and D consist of sand and clay, respec-
tively. Rango et al. (1983) [68] claimed only a 5% error in land-cover estimates from
Landsat data at the basin level and a much higher error at the cell level. A composite
CN can be computed for the different land-uses for a watershed using the following
equation:
12
CNc = CN1A1+ CN2A2+ ………+CNiAi+ ….+ CNnAn
∑ Aini=1
(3)
where CNi is the curve number of area i, Ai is the area of each LULC for area i and n is
the number of land uses. In this study, CN is calculated for each land class using ArcGIS
10.6 from the vector soil dataset.
3.2. Integrated RS-GIS Approach to Surface Runoff Modeling
Integrated RS-GIS approaches are used for surface runoff modeling. The analy-
sis comprises three main parts: (a) derivation of the land class of the study area using
RS, (b) hydrological parameter determination applying GIS, and (c) runoff modeling
using GIS. For the land class derivation, four MS image scenes were selected from four
different years. The land class type and soil information provided the hydrological curve
number, one of the key parameters needed for the hydrologic models. Hydrological pa-
rameters (directly related to runoff calculation), such as maximum storage, were deter-
mined using the curve number. The land-use types were used as independent variables
for the proposed methodology. Lastly, the runoff was determined using precipitation
and the maximum storage dataset. In the succeeding sub-sections, all three processes
are described in detail.
3.2.1. Land-Use and Land Cover Type Using Remote Sensing
We used land-use and land cover patterns for four different years from 1994 to
2020 at the same and almost the same time (Table 2). For all the images, we considered
less than 10% cloud cover. Initially, all images were corrected using a common Univer-
sal Transverse Mercator (UTM) coordinate system [69]. Every image was then radio-
metrically corrected according to the Jensen method [67]. Scanline error correction was
13
done for the 2014 images. For the land-use and landcover derivation, supervised classi-
fication with a maximum likelihood algorithm was applied [70,71]. The MS image for
2020 was collected from Landsat 8 OLI (Operational Land Imager), and scanline error
correction was not required for this image.
A supervised classification method was used for image classification. This
method usually requires a priori knowledge of each of the land types. A group of training
data sets was collected to identify each of the land types. Four different classification
algorithms were available to proceed with the supervised classification. In this study,
we applied the maximum likelihood approach for land-use change analysis. The algo-
rithm uses the spectral signature of the pixels from the training dataset to classify the
whole image. The spectral signature file used the training information to define the sta-
tistics, such as the mean and variance of each land type. Every training dataset consists
of at least 5 to 10 data points [71]. The more data points represented, the greater the
accuracy of the classification. In this study, we collected 12 to 15 data points for each
of the land classes. The process was subsequently applied to 1994, 2004, 2014, and 2020
(Figure 3).
Every classified image was superimposed by land-use and land cover shapefiles
from Anderson land classes to determine the accuracy of the classification [70], which
were used to categorize the land use. Land use and land cover vector files are available
for 1995, 2004, and 2011 in the Rhode Island Geographic Information System (RIGIS).
Therefore, the classified images are evaluated with the publicly available land-use and
14
land cover vector datasets. 1995, 2004, and 2011 data were used to evaluate 1994, 2004,
and 2014 land classification. Since the current land-use and land cover shapefile for
2020 is not available, no image accuracy evaluation was done for 2020 land classes.
In this study, residential, commercial, industrial, and parking lot and street areas were
considered urban class types as they are also directly related to population growth. A
total of nine land-use classes (Table 2) were derived from the images. In terms of using
multispectral perspective, three-band combinations: Green (band 2), Red (band 3), and
near-infrared (band 4) were used to derive the land classification. A flow chart is pre-
sented in Figure 3 to describe the detailed process.
Figure 3. The Process of Land-use and land cover Change Detection.
15
The classified suburban growth image was overlaid with the vector files to calibrate the
suburban expansion area. All the raster files were created using 30 m cell size, and this
cell size area was calculated for every land class. The town boundary vector file was
utilized to detect the suburban expansion for the particular year.
To detect urban expansion in the suburban area, curve number (CN) values were
used (Table 2). Table 2 represents the CN value for different land types and hydrologic
soil groups.
Table 2. General characteristics of four candidate image scenes.
Land Type A B C D
Agricultural Land 64 75 82 85
Commercial 89 92 94 95
Forest 30 55 70 77
Grass/Pasture 39 61 74 80
Residential 60 74 83 87
Industrial 81 88 91 93
Open Space 49 69 79 84
Parking and paved spaces 98 98 98 98
Water/Wetlands 0 0 0 0
Adapted from 210-VI-TR-55, Second Ed., June 1986 [64.]
3.2.2. Hydrogeological Parameter Determination Using GIS
In this step, we prepared soil and precipitation images to generate hydrogeolog-
ical parameters, and the overall processes are discussed below:
16
Derivation of Soil Data
The soil data is prepared from a digital soil survey with a detailed soil geographic
data level, jointly developed by the Rhode Island Soil Survey and National Cooperative
Soil Survey, and was downloaded from the Rhode Island Geographic Information Sys-
tem (RIGIS) [72]. The map data contains detailed information such as hydrological
group, shape area, shape length, and soil name. For this study, we applied hydrologic
soil group (HSG) (A, B, C, and D) (Figure 4b) information (Table 2) for each land type.
All four types of HSG were found in the study area. As shown in Figure 4b, the HSG in
group B occupied about 29% (61 sq. km) of the total area. About 22% (48 sq. km) of
the total area belonged to group D. However, groups A and C occupied approximately
17.7% (37 sq. km) and 9.1% (19 sq. km), respectively. Considering the HSG proportion,
the study mainly consisted of a moderate infiltration rate (group B) of 3.81 to 7.62
mm/h. A raster file is then generated for the study area based on the hydrologic group
using the runoff curve number (CN) values. The maximum storage of the area was then
calculated using the raster calculator.
Derivation of Slope, Elevation, and Stream Order Data
A slope is an essential parameter of watershed characteristics representing an
angle between the inclined ground surface and the horizontal plane. A slope map was
produced (Figure 4c) using a DEM with a 30 m resolution for the study area applying
the ‘surface’ option from the spatial analyst tool of ArcGIS. The slope map showed the
slope variation in degree. About 75% of the area had a slope between (0–5) degrees,
and 18% had a slope of (5.1–15) degrees, mainly in the southern part of the study area.
17
Consequently, elevation (Figure 4d) and stream order (Figure 4a) maps were also de-
veloped to represent the overall scenario of the hydrogeological parameter of the study
area. This area’s elevation ranges from 0 to 323 m and has predominantly first and sec-
ond-order streams.
Rainfall Data
Rain data was derived from the Parameter-elevation Regressions on Independent
Slopes Model (PRISM) [73–75] climate mapping system. Since only one USGS
weather station was available in the study area, there were insufficient neighboring sta-
tions to create the spatial distribution of rainfall in a given area. Therefore, the PRISM-
derived high-resolution precipitation data were considered for this study. PRISM pre-
cipitation products are spatially gridded at 4 km resolution. Elevation is the primary
variable that controls the precipitation pattern [76–78]. PRISM data accounts for eleva-
tion impacts and, therefore, could be used reliably for this study.
18
Figure 4. Hydrogeological parameters for the study area: (a) Stream order, (b) Hydro-
logic soil group (HSG), (c) Slope, (d) Elevation.
19
For this study area, 10 PRISM stations are projected and used for the analysis.
The kriging application [79] was used to generate the spatial distribution image by as-
signing average yearly rainfall in the coverage with a 30 m cell size for the raster.
Kriging is expressed as:
ZK = ∑ ƛiZini=1 (4)
where Zk is an estimate by kriging, ƛ is a weight for Zi and Zi is a variable. The weight
is determined to ensure unbiasedness [79].
PRISM data helped to calculate the storage for the entire study area for each of the land
classes.
3.2.3. Hydrological Modeling within GIS
We applied GIS for the runoff modeling of this suburban watershed (Figure 5).
Multiple images (land cover and soil images) were used to construct the CN image.
Each area’s CN value was estimated using USDA 1972 [67] standard SCS values (Table
2). Potential maximum storage, S, was derived using map algebra application of GIS
for the entire time series. Then, storm runoff depth images were prepared from rainfall
and potential maximum storage images using Equation (1). Four images for 1994, 2004,
2014, and 2020 were created for four different years applying the same methodology.
The resulting runoff images were reclassified into runoff ranks. Suburban growth and
its development were most prominent in the potential maximum storage images.
We then developed the relationship between rainfall-runoff coefficients for this
study area. The runoff coefficient was calculated for each of the selected years’ ten
largest storm events based on the rainfall amount. A runoff coefficient curve was then
20
constructed as a function of the flood size. A significant change of the suburban effect
based on the runoff coefficient pattern was observed over time.
Figure 5. A schematic representation of the geographic information system (GIS)
based soil conservation service curve number (SCS-CN) model in estimating the sur-
face runoff.
4. Results and Discussion:
4.1. Suburban Growth in the Study Area
Suburban growth represents the expansion of residential, commercial, industrial,
and roads and parking lots, and results indicate an increasing trend, as shown in Table
3. Land type change detected both areal expansion and reduction. The 36 years analysis
results (Table 3) showed that the suburban area expanded by about 68.6% (6564 acres)
within this administrative region.
Simultaneously, the agricultural area increased considerably (by approximately
5%), and the forest area decreased by nearly 8.2%. The proportion of urban area to the
21
total land area was 18%, 20.4%, 29%, and 31% in 1994, 2004, 2014, and 2020, respec-
tively. According to land-use change detection, from 1994 to 2004, suburban areas did
not expand. Differences in LULC change from 1994 to 2020 from satellite image anal-
ysis are presented in Table 3.
Table 3. Satellite-detected suburban expansion in South Kingstown, Rhode Island.
Land Type
Calculated Area (Acre) Change Detection of the Area
1994 2020 Area (Acre) %
Urban 9569 16,133 6564 68.6
Agriculture 2684 2814 130 4.8
Forest 18,440 16,934 −1506 -8.2
Grass/Pasture 6662 710 −5952 -89.3
In this study, suburban land-use expansion (Figure 6) is the main predictor in
analyzing the surface runoff pattern. Suburban growth is directly related to percentage
imperviousness. Increasing imperviousness leads to an increase the suburban runoff.
22
Figure 6. Land type change detection from 1994 to 2020.
In the study area, suburban expansion started from 2000 to 2020 along with con-
siderable population growth [75]. Suburban institutional development also exhibited the
overall expansion of the suburban area. For suburban growth, the main driving factor is
population growth and residential zone demand [4–6]. The suburban area expanded by
68.6% during this time; if the trend continues, the suburban area will quickly convert to
an urban area.
4.2. Impact of Suburban Growth on Surface Runoff
The impact of suburban expansion on surface runoff was examined by compar-
ing the measured runoff volume from 1994 to 2020. With the growth in suburban ex-
23
pansion, runoff is also expanded over time. This study found that suburban growth in-
creased surface runoff. The GIS-based SCS-CN model was used to evaluate the surface
runoff for the years 1994, 2004, 2014, and 2020. For each of the years, runoff areas for
different runoff depths were calculated. The resulting images of runoff depth showed
the runoff depth changes in this area (Figure 7).
The spatial distribution of modeled runoff (Figure 8) showed the runoff change in
2014 and 2020 compared to 1994. A moderate runoff depth (Figure 7) and runoff area
increased over time in the study area. The runoff increase was relatively low from 1994
to 2004 and higher in 2014; even in 2020, the runoff depth increased significantly (by
about 15% on average) compared to 2014 (Figure 7).
Figure 7. Runoff depth change detection and area from 1994 to 2020.
24
We then ranked runoff from 1 to 9 based on the runoff depth, indicating the lowest
runoff depth as rank 1 (i.e., a decrease or no change in runoff depth) and the highest
rank as 9 (i.e., zones having the highest runoff depth of 9 mm in 2020). Each of these
ranks is a continuous and discontinuous extension expressed with a different color in
the raster file for 1994, 2014, and 2020. The yellow color represents the runoff depth of
3 to 4 mm in 1994 and 2014. Noticeably, yellow holds a major portion of the study area
in 1994 (17.2% of total area) and 2014 (16.6% of the total area). However, in 2020, the
noticeable portion of the site has a runoff depth of 4 to 5 mm, indicating that the area
had a higher runoff depth 2020. In 2020, the maximum runoff depth was reported as 9
mm, which was not observed in the previously investigated years (1994, 2004, and
2014). A visual and detailed interpretation of the study area’s areal extent and spatial
occurrence was created by aggregating categories 6 to 8 for 2014 (Figure 8). Due to a
small amount of suburban growth between 1994 and 2004, the spatial distribution of
runoff depth for 2004 did not show much change relative to 1994. The modeled runoff
spatial distribution of each year (1994, 2014, and 2020) reflected a suburban expansion
in this area. The land-use classification showed that urban land type for 2020 increased
by approximately 8.3% from 2014, and these land changes led to an increase of runoff
depth from 2014.
The total area of runoff depth under the ranks increased by around 21.7% from
1994 to 2020. Despite the decrease in rainfall, we still found an increasing trend in run-
off depth. This was higher in the northwestern part of the study site. This northwestern
area was also becoming more vulnerable due to expansion in suburban growth and land
25
use. A correlation between the distributed runoff area and the suburban expansion area
was examined using the ordinary least square (OLS) tool from the spatial statistical op-
tion in ArcGIS. The result showed a strong positive correlation between two mapped
patterns with multiple r-values of 0.63 (average) (p < 0.05), where the correlation value
indicated an increasing trend along with the increasing suburban growth in the study
area.
Figure 8. Storm Runoff depth of 2020, 2014 and 1994.
26
4.3. Impact of Suburban Growth on Rainfall-Runoff Relationship
The runoff coefficient was measured according to the ten highest rainfall events
for each of the years. The runoff coefficient range varied based on imperviousness, rain-
fall depth, duration, and intensity [15]. Figure 9 showed the relationship between rainfall
and runoff coefficient for the study area. A higher runoff coefficient was expected when
there was more rainfall volume over the suburban area. According to the SCS model,
the rainfall and runoff coefficient relationship is governed by maximum potential stor-
age. The impervious zone has low or no potential storage, and the runoff coefficient
value showed a strong relationship in these areas. This strong relationship illustrates the
effects of suburban growth in this study area. The two dynamic variables are the subur-
ban growth rate and the maximum potential storage. A correlation between the suburban
growth rate and the maximum potential storage variables generated r = 0.45 in 1994, r
= 0.68 in 2014 and r = 0.71 in 2020. During the entire period from 1994 to 2020, the
runoff coefficient increased by about 0.28. This increasing trend indicated that urbani-
zation played a vital role in the rainfall-runoff relationship. Thus, urbanized areas are
more prone to increased runoff and flooding events because lower potential storage of-
ten implies that the same amount of rainfall may generate more runoff depth depending
on imperviousness [13,14]. Furthermore, the land-use change detection showed about
68.6% increased suburban growth from 1994 to 2020 (Table 3). The standard deviation
of the runoff coefficient, ranging from 0.16 to 0.021 for these 36 years, indicates subur-
ban growth.
27
Figure 9. Rainfall-runoff relationship for the study period.
5. Conclusions
We developed an integrated approach using an RS and GIS-based SCS-CN
model to assess suburban growth influences in surface runoff. The combined effort of
RS-GIS confirms it to be an efficient tool for suburban growth analysis. By applying
this methodology, we developed a linkage between suburban growth and surface runoff
through spatial analysis, which showed a positive and significant correlation.
Land-use and land cover changes are examined through remote sensing using
four different Landsat images applying supervised classification. We emphasized the
change detection in the urban land type. The analyzed output showed that urban areas
expanded by about 68.6% from 1994 to 2020, whereas forest (decreased by only 8.2%)
28
and agricultural (decreased by only 4.8%) land types showed a little change in this area.
Despite this slight change in forest and agriculture, significant urban land change im-
pacted the surface runoff, and this change confirmed suburban growth in suburban ar-
eas. The output raster file of land use changes from remote sensing was used as an input
parameter along with soil and precipitation in the GIS-based SCS-CN model.
The GIS-based SCS-CN approach was applied to develop the model for estimat-
ing runoff for four different years. The significant advantage of employing the GIS ap-
plication in rainfall–runoff modeling is that more accurate sizing and calculation can be
achieved compared to traditional methods. We relied on GIS analysis to detect the sub-
urban growth effect on surface runoff in the study areas. Surface runoff and its areal
extension were also examined from the analysis. The modeled output indicated that run-
off depth had increased from 2%–10% from 1994 to 2020. The spatial distribution of
the surface runoff also signifies a moderate to significant effect of suburban growth.
Furthermore, the increasing trend of runoff coefficient with time and rainfall events in-
dicates the positive impact of suburban growth.
The integrated approach worked successfully in suburban areas and could arrive
at the stormwater quality impacts in this study. Stormwater quality degradation could
be the scope of work for further analysis. The two primary factors that were predictors
of the surface runoff are rainfall and the hydro group soils (HSG) group. From 1994 to
2020, both runoff depth and spatial extension of runoff areas enlarged. Due to the in-
crease of suburban expansion in the northwestern part of the study area, the runoff depth
29
primarily increased. This integrated approach is a valuable tool to analyze suburban
growth and its impact on surface runoff by developing rainfall–runoff modeling, espe-
cially for suburban areas. This research’s final output strongly supports the impact of
urbanization impact in this suburban region on surface runoff. Further research should
be focused on stormwater management in terms of quality and quantity to minimize the
future environmental impact in suburban areas.
Author Contributions: For this research, K.J. developed the research framework, designed and
carried out the analysis of results, and wrote the original paper. S.M.P. contributed to the inter-
pretation of results, resources, supervision, project administration. M.A.E.B. conceptualization,
figures preparation, interpretation of results, writing, review & editing and contributed to the
development of the paper. All authors have read and agreed to the published version of the
manuscript.
Funding: The research study is funded by the Rhode Island Department of Transportation, SPR-
234-2362, and the University of Rhode Island.
Acknowledgments: Authors are thankful to the Rhode Island Department of Transportation
(RIDOT) for the project fund. We also thank Arthur Gold and Yeqiao Wang, Professors of the
University of Rhode Island, for their valuable comments and recommendations.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Grimm, N.B.; Faeth, S.H.; Golubiewski, N.E.; Redman, C.L.; Wu, J.; Bai, X.; Briggs, J.M.
Global Change and the Ecology of Cities. Science 2008, 319, 756–760, doi:10.1126/sci-
ence.1150195.
2. Pradhanang, S.; Jahan. K. Urban Water Security for Sustainable Cities in the Context of
Climate Change; Chapter 14, Water, Climate and Sustainability; John Wiley & Sons, Inc.:
Hoboken, NJ, USA, 2021; pp. 213–224, doi:10.1002/9781119564522.ch14.
30
3. Hegazy, I.R.; Kaloop, M.R. Monitoring urban growth and land use change detection with
GIS and remote sensing techniques in Daqahlia governorate Egypt. Int. J. Sustain. Built
Environ. 2015, 4, 117–124, doi:10.1016/j.ijsbe.2015.02.005.
4. UN-HABITAT (United Nations Human Settlement Programme). The State of the African
Cities 2008—A Framework for Addressing Urban Challenges in Africa. Nairobi; United
Nations Human Settlements Programme: 2008; ISBN:978-92-1-132015-2.
5. Sankhala, S.; Singh, B. Evaluation of urban sprawl and land use land cover change using
remote sensing and GIS techniques: A case study of Jaipur City, India. J. Int. J. Emerg.
Technol. Adv. Eng. 2014, 4, 66–72.
6. Abustan, I.; Sulaiman, A.H.; Wahid, N.A.; Baharuddin, F. Determination of rainfall-runoff
characteristics in an urban area: Sungai kerayong catchment, Kuala Lumpur. In Proceed-
ings of the 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK,
August 31 – September 5, 2008; Available online: https://citeseerx.ist.psu.edu/view-
doc/download?doi=10.1.1.517.1745&rep=rep1&type=pdf (January 2009).
7. Weng, Q. Modeling urban growth effects on surface runoff with the integration of remote
sensing and GIS. J. Environ. Manag. 2001, 28, 737–748.
8. Kibler, D.F. (Ed.) Urban Stormwater Hydrology; American Geophysical Union: Washing-
ton, DC, USA, 1982.
9. Goudie, A. The Human Impact on the Natural Environment, 3rd ed.; The MIT Press: Cam-
bridge, MA, USA, 1990.
10. Komrad, C.P. Effect of Urban Development on Floods. U.S. Geological Survey 2003,
USGS Fact Sheet FS-076-03. Available online:
https://pubs.usgs.gov/fs/fs07603/pdf/fs07603.pdf (accessed on).
11. Rogger, M.; Agnoletti, M.; Alaoui, A.; Bathurst, J.C.; Bodner, G.; Borga, M.; Chaplot, V.;
Gallart, F.; Glatzel, G.; Hall, J.; et al. Land use change impacts on floods at the catchment
scale: Challenges and opportunities for future research. Water Resour. Res. 2017, 53,
5209–5219, doi:10.1002/2017wr020723.
12. Jahan, K.; Pradhanang, S. Predicting Runoff Chloride Concentrations in Suburban Water-
sheds Using an Artificial Neural Network (ANN). J. Hydrol. 2020, 7, 80, doi:10.3390/hy-
drology7040080.
13. Niemczynowicz, J. Urban hydrology and water management—Present and future chal-
lenges. Urban Water 1999, 1, 1–14, doi:10.1016/s1462-0758(99)00009-6.
14. Fletcher, T.; Andrieu, H.; Hamel, P. Understanding, management and modeling of urban
hydrology and its consequences for receiving waters: A state of the art. Adv. Water Resour.
2013, 51, 261–279, doi:10.1016/j.advwatres.2012.09.001.
15. Isah, T. Remote Sensing Studies on Urban Change Detections. J. Int. J. Comput. Sci. Inf.
Technol. Res. 2015, 3, 62–71, ISSN 2348-120X. Available online: www.research-
publish.com (accessed on).
31
16. Oyinloye, R.O.; Adesina, F.A. Some aspect of the growth of Ibadan and their Implications
for socio-economic development. J. Ife Soc. Sci. Rev. 2006, 20, 113–120.
17. McGrane, S.J. Impacts of urbanization on hydrological and water quality dynamics, and
urban water management: A review. Hydrol. Sci. J. 2016, 61, 2295–2311,
doi:10.1080/02626667.2015.1128084.
18. Ehlers, M.; Jadkowski, M.A.; Howard, R.R.; Brostuen, D.E. Application of SPOT data for
regional grow than analysis and local planning. J. Photogramm. Eng. Remote Sens. 1990,
56, 175–180.
19. Treitz, P.M.; Howard, P.J.; Gong, P. Application of satellite and GIS technologies for land-
cover and land-use mapping at the rural-urban fringe: A case study. J. Photogramm. Eng.
Remote Sens. 1992, 58, 439–448.
20. Harris, P.M.; Ventura, S.J. The integration of geographic data with remotely sensed im-
agery to improve classification in an urban area. J. Photogramm. Eng. Remote Sens. 1995,
61, 993–998.
21. Yeh, A.G.O.; Li, X. Urban growth management in the Pear River delta an integrated remote
sensing and GIS approach. J. ITC J. 1996, 1, 77–85.
22. Yeh, A.G.O.; Li, X. An integrated remote sensing and GIS approach in the monitoring and
evaluation of rapid urban growth for sustainable development in the Pearl River Delta,
China. Int. Plan. Stud. 1997, 2, 193–210, doi:10.1080/13563479708721678.
23. Pandey. A.; Sahu, A.K. Generation of curve number using Remote Sensing and Geographic
information system. J. Geospat. World 2009. Available online: https://www.geospatial-
world.net/article/generation-of-curve-number-using-remote-sensing-and-geographic-in-
formation-system/ (accessed on).
24. Gajbhiye, S. Estimation of Surface Runoff Using Remote Sensing and Geographical Infor-
mation System. J. Sci. Technol. 2015, 8, 113–122, doi:10.14257/ijunesst.2015.8.4.12.
25. Engman, E.T.; Gurney, R.J. Remote sensing in hydrology. In Remote Sensing Application
Series; xiv 225; Chapman & Hall: London, UK, 1991.
26. Drayton, R.S.; Wilde, B.M.; Harris, J.H.K. Geographic information system approach to
distributed modeling. J. Hydrol. Process. 1992, 6, 36–368.
27. Mattikalli, N.M.; Devereux, B.J.; Richards, K.S. Prediction of river discharge and surface
water quality using an integrated geographic information system approach. J. Int. J. Remote
Sens. 1996, 17, 683–701.
28. Radke, J.; Andra, S.; Al-Kofani, O.; Roysan, B. Image change detection algorithms: A
systematic survey. J. IEEE Trans. Image Process. 2005, 14, 291–307.
29. Bhuiyan, A.E.; Witharana, C.; Liljedahl, A.K.; Jones, B.M.; Daanen, R.; Epstein, H.E.;
Kent, K.; Griffin, C.G.; Agnew, A. Understanding the Effects of Optimal Combination of
Spectral Bands on Deep Learning Model Predictions: A Case Study Based on Permafrost
32
Tundra Landform Mapping Using High-Resolution Multispectral Satellite Imagery. J. Im-
aging. 2020, 6, 97, doi:10.3390/jimaging6090097.
30. Bhuiyan, A.E.; Witharana, C.; Liljedahl, A.K. Use of Very High Spatial Resolution Com-
mercial Satellite Imagery and Deep Learning to Automatically Map Ice-Wedge Polygons
across Tundra Vegetation Types. J. Imaging 2020, 6, 137, doi:10.3390/jimaging6120137.
31. Nagne, A.D.; Dhumal, R.K.; Vibhute, A.D.; Nalawade, B.D.; Kale, K.V.; Mehrotra, S.C.
Advantages in Land use classification of urban areas from Hyperspectral data. Int. J. Eng.
Tech. 2018, 4, ISSN:2395-1303. Available online: http://www.ijetjournal.org (accessed
on).
32. Witharana, C.; Bhuiyan, M.A.E.; Liljedahl, A.K. Big Imagery and high-performance com-
puting as resources to understand changing Arctic polygonal tundra. Int. Arch. Photo-
gramm. 2020, 44, 111–116.
33. Science Mission Directorate. “Wave Behaviors” NASA Science. J. Natl. Aeronaut. Space
Adm. 2010. Available online: http://science.nasa.gov/ems/03_behaviors (accessed on).
34. Witharana, C.; Bhuiyan, M.A.E.; Liljedahl, A.K.; Kanevskiy, M.; Epstein, H.E.; Jones,
B.M.; Daanen, R.; Griffin, C.G.; Kent, K.; Jones, M.K.W. Understanding the synergies of
deep learning and data fusion of multispectral and panchromatic high resolution commer-
cial satellite imagery for automated ice-wedge polygon detection. ISPRS J. Photogramm.
Remote Sens. 2020, 170, 174–191.
35. Mohanrajan, S.N.; Loganathan, A.; Manoharan, P. Survey on Land Use/Land Cover
(LU/LC) change analysis in remote sensing and GIS environment: Techniques and Chal-
lenges. Environ. Sci. Pollut. Res. 2020, 27, 29900–29926, doi:10.1007/s11356-020-09091-
7.
36. Mukherjee, S. Land use maps for conservation of ecosystems. J. Geog. Rev. India 1987, 3,
23–28.
37. Sharma, S.K.; Gajbhiye, S.; Nema, R.K.; Tignath, S. Assessing vulnerability to soil erosion
of a watershed of tons River basin in Madhya Pradesh using Remote sensing and GIS. J.
Int. J. Environ Res Dev 2014, 4, 153–164.
38. Berry, J.K.; Sailor, J.K. Use of a Geographic Information System for storm runoff predic-
tion from small urban watersheds. J. Environ. Manag. 1987, 11, 21–27.
39. Dongquan, Z.; Jining, C.; Haozheng, W.; Qingyuan, T.; Shangbing, C.; Zheng, S. GIS-
based urban rainfall-runoff modeling using an automatic catchment-discretization ap-
proach: A case study in Macau. Environ. Earth Sci. 2009, 59, 465–472,
doi:10.1007/s12665-009-0045-1.
40. Mishra, S.K.; Pandey, A.; Singh, V.P. Special Issue on Soil Conservation Service Curve
Number (SCS-CN) Methodology. J. Hydrol. Eng. 2012, 17, 1157,
doi:10.1061/(asce)he.1943-5584.0000694.
33
41. Singh, P.K.; Mishra, S.K.; Berndtsson, R.; Jain, M.K.; Pandey, R.P. Development of a
Modified SMA Based MSCS-CN Model for Runoff Estimation. Water Resour. Manag.
2015, 29, 4111–4127, doi:10.1007/s11269-015-1048-1.
42. Vojtek, M.; Vojteková, J. GIS-based Approach to Estimate Surface Runoff in Small Catch-
ments: A Case Study. Quaest. Geogr. 2016, 35, 97–116, doi:10.1515/quageo-2016-0030.
43. Soulis, K.X.; Valiantzas, J.D. SCS-CN parameter determination using rainfall-runoff data
in heterogeneous watersheds. The two-CN system approach. J. Hydrol. Earth Syst. Sci.
Discuss. 2011, 8, 8963–9004.
44. Mishra, S.K.; Singh, V.P. Long-term hydrological simulation based on the Soil Conserva-
tion Service curve number. Hydrol. Process. 2004, 18, 1291–1313, doi:10.1002/hyp.1344.
45. Soulis, K.X.; Dercas, N. Development of a GIS-based spatially distributed continuous hy-
drological model and its first application. J. Water Int. 2007, 32, 177–192.
46. Zhan, X.Y.; Huang, M.L. ArcCN-Runoff: An ArcGIS tool for generating curve number
and runoff maps. J. Environ. Model. Softw. 2004, 19, 875–879.
47. Moretti, G.; Montanari, A. Inferring the flood frequency distribution for an ungauged basin
using a spatially distributed rainfall-runoff model. Hydrol. Earth Syst. Sci. 2008, 12, 1141–
1152, doi:10.5194/hess-12-1141-2008.
48. Adornado, H.A.; Yoshida, M. GIS-based watershed analysis and surface runoff estimation
using curve number (C.N.) value. J. Environ. Hydrol. 2010, 18, 1–10.
49. Wilcox, B.P.; Rawls, W.J.; Brakensiek, D.L.; Wight, J.R. Predicting runoff from Range-
land Catchments: A comparison of two models. Water Resour. Res. 1990, 26, 2401–2410,
doi:10.1029/wr026i010p02401.
50. Holman, I.P.; Hollis, J.M.; Bramley, M.E.; Thompson, T.R.E. The contribution of soil
structural degradation to catchment flooding: A preliminary investigation of the 2000
floods in England and Wales. Hydrol. Earth Syst. Sci. 2003, 7, 755–766, doi:10.5194/hess-
7-755-2003.
51. Romero, P.; Castro, G.; Gomez, J.A.; Fereres, E. Curve number values fo olive orchards
under different soil management. J. Soil Sci. Soc. Am. J. 2007, 71, 1758–1769.
52. Tedela, N.H.; McCutcheon, S.C.; Rasmussenn, T.C.; Hawkins, R.H.; Swank, W.T.; Camp-
bell, J.L.; Adams, M.B.; Jackson, R.; Tollner, E.W. Runoff curve numbers for 10 small
forested watersheds in the mountains of the eastern united states. J. Hydrol. Eng. 2012, 17,
1188–1198.
53. Nagarajan, M.; Basil, G. Remote sensing- and GIS-based runoff modeling with the effect
of land-use changes (a case study of Cochin corporation). Nat. Hazards 2014, 73, 2023–
2039, doi:10.1007/s11069-014-1173-9.
54. Pandit, A.; Gopalakrishnan, G. Estimation of annual storm runoff coefficients by continu-
ous simulation. J. Irrig. Drain. Eng. 1996, 122, 211–220.
34
55. USGS. Elevation of the March-April 2010 Flood High Water in Selected River Reaches in
Rhode Island by Phillip, J. Zarriello and Gardener, C. Bent; U.S. Department of the Inte-
rior. U.S. Geological Survey: 2011. Available online:
https://pubs.usgs.gov/of/2011/1029/pdf/ofr2011-1029-508.pdf (accessed on).
56. Kenyon Grist Mill. Flood 2010: Photo Documentary (Before, During, and After). All Con-
tent Copyright—Official Website of Kenyon Grist Mill LLC 21 Glen Rock Road/P.O. Box
221 West Kingston, RI 02892. 2013. Available online: https://www.kenyonsgrist-
mill.com/flood2010.html (accessed on).
57. The Independent. Towns, Residents Close to Recovering From Historic Flood by Chris
Church. Independent.com. 2011. Available online: https://www.independentri.com/lo-
cal/article_71f7559f-a7ec-537d-b380-80ac5e5c3ddb.html (accessed on).
58. SRI. A Look Back at the Flood of the Century. Southern Rhode Island Newspaper. 2011.
Available online: https://www.ricentral.com/a-look-back-at-the-flood-of-the-century/arti-
cle_c8e44c9d-a7f9-59f8-ac2b-f345b5f0e0a9.html (accessed on).
59. Gorman, G. Rhode Island Flood Zones-South KingstownRI Real Estate has them too.
Rhode Island Coastal Real Estate. 2012. Available online: https://www.rihouse-
hunt.com/ri-blog/rhode-island-flood-zones-south-kingstown-ri-real-estate-has-them-
too/?doing_wp_cron=1614615852.4240589141845703125000 (accessed on).
60. The Independent, Heavy Rains Brings Flooding, Road Closures. Reporter Stephen Green-
well. Independentri.com. 2014. Available online: https://www.independentri.com/inde-
pendents/south_county/article_35880c91-006d-5e2b-b32b-af311135fe9e.html (accessed
on).
61. Patch the Community Corner. Flash Flood Watch in Narragansett and South Kingstown by
Erin Tiernan. Community Corner. 2013. Available online: https://patch.com/rhode-is-
land/narragansett/flash-flood-watch-in-narragansett-and-south-kingstown (accessed on).
62. ECO RI News. Rising Waters Threaten Region’s Cultural Treasures. ECO RI News. 2015.
Available online: https://www.ecori.org/climate-change/2015/1/5/rising-waters-threaten-
regions-culture (accessed on).
63. USDA. Economic Research Service Using Data from U.S. Census Bureau, 2013. Available
online: https://www.ers.usda.gov/data-products/ (accessed on).
64. U.S. Climate Data. 2019. U.S. Climate Data 2021 | Version 3.0 | by Your Weather Service.
Available online: https://www.usclimatedata.com/ (accessed on).
65. South Kingstown Source of Water Assessment (SKSWA), University of Rhode Island Co-
operative Extension. 2003. Available online: http://cels.uri.edu/rinemo/asses-
ments/sk_source_water_assesment.pdf (accessed on).
66. USGS Earth Explorer Landsat Archive (1980–2019). Available online: https://earthex-
plorer.usgs.gov (accessed on 10 February 2019).
67. Jensen, J.R. Introductory Digital Image Processing: A Remote Sensing Perspective, 2nd
ed.; Prentice-Hall: Hoboken, NJ, USA, 1996.
35
68. USDA—Soil Conservation Service. National Engineering Handbook; Sec. 4. Hydrology;
USDA: Washington, DC, USA, 1972.
69. Rango, A.; Feldman, A.; George, T.S.; Ragan, R.M. Effective use of landsat data in hydro-
logic models. JAWRA J. Am. Water Resour. Assoc. 1983, 19, 165–174, doi:10.1111/j.1752-
1688.1983.tb05310.x.
70. Camps-Valls, G.; Tuia, D.; Gómez-Chova, L.; Jiménez, S.; Malo, J. Remote Sensing Image
Processing. Synth. Lect. Image Video, Multimedia Process. 2011, 5, 1–192,
doi:10.2200/s00392ed1v01y201107ivm012.
71. Hütt, C.; Koppe, W.; Miao, Y.; Bareth, G. Best Accuracy Land Use/Land Cover (LULC)
Classification to Derive Crop Types Using Multitemporal, Multisensor, and Multi-Polari-
zation SAR Satellite Images. Remote Sens. 2016, 8, 684, doi:10.3390/rs8080684.
72. Samaniego, L.; Schulz, K. Supervised Classification of Agricultural Land Cover Using a
Modified k-NN Technique (MNN) and Landsat Remote Sensing Imagery. Remote Sens.
2009, 1, 875–895, doi:10.3390/rs1040875.
73. Anderson, J.R.; Hardy, E.E.; Roach, J.T.; Witmer, R.E. A Land Use and Land Cover Clas-
sification for Use with Remote Sensor Data; (USGS Professional Paper 964); Government
Printing Oce: Washington, DC, USA, 1976, doi:10.3133/pp964.
74. USDA. Urban Hydrology for Small Watershed TR-55. In 210-VI-TR-55, 2nd; USDA:
Washington, DC, USA, 1986.
75. Daly, C. Guidelines for assessing the suitability of spatial climate data sets. Int. J. Clim.
2006, 26, 707–721, doi:10.1002/joc.1322.
76. Bhuiyan, M.A.E.; Yang, F.; Biswas, N.K.; Rahat, S.H.; Neelam, T.J. Machine Learning-
Based Error Modeling to Improve GPM IMERG Precipitation Product over the Brahma-
putra River Basin. Forecasting 2020, 2, 248–266, doi:10.3390/forecast2030014.
77. Bhuiyan, M.A.; Nikolopoulos, E.I.; Anagnostou, E.N. Machine learning–based blending
of satellite and reanalysis precipitation datasets: A multiregional tropical complex terrain
evaluation. J. Hydrometeor. 2019, 20, 2147–2161.
78. Bhuiyan, M.A.E.; Nikolopoulos, E.I.; Anagnostou, E.N.; Quintana-Seguí, P.; Barella-
Ortiz, A. A nonparametric statistical technique for combining global precipitation datasets:
Development and hydrological evaluation over the Iberian Peninsula. Hydrol. Earth Syst.
Sci. 2018, 22, 1371–1389.
79. Journel, A.G.; Huijbregts, C.J. Mining Geostatistics; Academic Press: Cambridge, MA,
USA, 1978; 600p.
36
CHAPTER 2
Manuscript ІІ
Modeling Runoff Quantity with different LID structures in the Suburban
Watershed using Stormwater Management Model (SWMM)
Under review in Hydrological Sciences Journal-IAH, June 2021
Khurshid Jahan1, Soni M Pradhanang1,*, and Thomas Boving1, 2
1Department of Geosciences, University of Rhode Island, RI, Kingston, USA 2Department of Civil and Environmental Engineering, University of Rhode Island, RI,
Kingston, USA
*Correspondence: [email protected]
37
Abstract
This study represents the application of the stormwater management model
(SWMM) in simulating stormwater runoff quantity (both peak flow and runoff
depth) in a suburban watershed area. Further, it investigates the comparative per-
formance of different low impact development (LID) measures. The Chipuxet wa-
tershed in Rhode Island (USA) was selected to develop an optimized LID struc-
ture for the suburban watershed using SWMM (Ver.5.1). The focus is on enhanc-
ing the treatment for both peak flow and stormwater runoff depth by LID struc-
tures during 1, 2, 5-, 10-, 50-, and 100-year rainfall return periods. The compre-
hensive effectiveness of different LID structures was measured for 4 different sce-
narios without any LID, permeable pavement (PP), bioretention (BR), or vegeta-
tive swale (VS). A detailed analysis was then performed for runoff peak reduction
and the runoff volume or depth reduction for PP, BR, and VS. The results show
that the LID structures effectively control the peak flow and runoff depth of storm
events for all six different return periods. The study revealed that both BR and PP
could control the peak and runoff volume in the suburban watershed. Finally, en-
vironmental benefits for each of the LID structures were calculated for the differ-
ent return periods of rainfall.
Keywords: stormwater management model; suburban watershed; low impact de-
velopment, analytical hierarchy process; bioretention; permeable pavement, bios-
wales.
38
1. Introduction
Rapid urbanization occurred in the United States in the early twentieth century [1]
that changed the land utilization from forest or agricultural uses to suburban or urban
lifestyles. Since then, large-scale urbanization has led to rapid infrastructure
development resulting in increased impervious surface areas. Most of the urban and
suburban land surfaces with high impervious surfaces do not sufficiently allow
stormwater and snowmelt to infiltrate into the ground. Further, the expansion of
impervious areas accelerates rainwater convergence that causes a higher volume of
runoff and peak flows [2-4]. The consequences of the increased runoff and the peak
flows have changed the natural hydrologic cycle [5,6] and increased urban flooding,
runoff pollution, water environment deterioration while damaging local ecology [7-12].
The US Environmental Protection Agency (EPA) specified in a 1988 report to Congress
that urban stormwater runoff is the third and fourth most extensive cause of water
quality impairment of the nation's lakes and rivers, respectively [13, 14]. In another
report in 1992, the EPA stated that the urban runoff is the second and third-largest source
of the impairment of the lakes, estuaries, and rivers, respectively [15,16]. Therefore,
assessing the impact of urbanization is essential for developing plans to reduce
stormwater runoff impact on hydrologic processes.
To reduce stormwater runoff impacts on the hydrologic process, many researchers
have broadly applied stormwater management models, especially in urban areas [17-
31]. The most widely used urban water quantity and quality model is the Storm Water
Management Model (SWMM) developed by the EPA in 1971 [32]. Since then, it has
39
undergone several significant updates for improved outputs [33]. The most recent
version, SWMM5.0, was released in early 2014 [34,35]. It can simulate hydrology,
hydraulics, and surface runoff quality of watersheds [36]. The model consists of
modules that usually apply to simulating different components of the hydrologic cycle
[37]. The governing principle for simulating runoff is the conservation of mass and
momentum equations for gradually varied and unsteady flow, known as the Saint
Venant flow equation. SWMM is a dynamic, comprehensive hydrological rainfall-
runoff simulation model that is usually used for a single event or transient simulations
of all aspects of the urban and suburban hydrologic and quality cycles [32,38,39]. This
model comprises four computational or building blocks, i.e., 1) runoff, 2) storage/
treatment, 3) transport, and 4) extran (Extended transport), as shown in Figure 1. The
RUNOFF block generates hydrographs and pollutographs. The primary input
parameters required to simulate hydrographs are rainfall hyetographs and the sub-
catchment physical characteristics. SWMM5.0 has modernized both the model’s
structure and its graphical user interface (GUI), making SWMM easier to use and more
accessible to a new generation of engineers and water resources specialists. However,
it is complex for the general public or planners with no modeling experience [19]. The
modified and the most recent version of SWMM (5.1.010) has many subroutines added
to analyze impacts of green infrastructure (GI) through low-impact development (LID)
components [37,38,40-48].
40
Figure 1: (a) SWMM, the stormwater Water Management Model, program
configuration (b) Overview of the SWMM structure indicating linkage among the
computational blocks. Receiving water simulation is by external programs (Adapted
from Huber and Dickinson, 1988 [49])
SWMM with different LID structures has been used widely for different regions
and countries, such as the USA and China. It must be noted that this study focuses on
the predominantly suburban watershed with mixed land use where the effectiveness of
SWMM coupled with LID has not been considered.
LID systems are characterized by a sequence of micro-scale stormwater devices or
structures with varying design parameters [50]. The LID techniques infiltrate and trap
pollutants in stormwater at the source and minimize imperviousness of both urban and
suburban areas [51,52]. Examples of LID structures include green roofs, permeable
pavements, bioretention cells, concave greenbelts; vegetative swales, and rain gardens
[32]. Several studies have reported the effectiveness of LID systems for maintaining
water quantity and quality in urban areas. However, there are no studies that primarily
41
focus on suburban areas. An investigation has been done on the growth in suburban
watersheds showing significantly increased surface runoff in terms of quantity and
quality [53,54]. This study analyzed suburban growth based on 36 years of land use and
land cover changes using the soil conservation service curve number (SCS-CN) model
[54]. In addition, the investigated area experienced two significant flood events in
March 2010 and March 2013 [55-62] (Figure 2), which resulted in significant
infrastructural damages. Both suburban growth (Figure 2) and flood events warrant us
to investigate the stormwater management techniques in this suburban area.
Figure 2: Landuse changes based on 36 years of data in the study area adopted from
the recent research paper [54]
42
Suburban areas are becoming equally important as urban areas to investigate
stormwater management. The research herein was conducted in a suburban
environment, and its purpose is to develop a stormwater management model (SWMM)
for controlling stormwater runoff depth and peak flow with LID structures. This
research tests the effectiveness of 4 different scenarios, which are reducing the peak
runoff and runoff depth by developing the stormwater model for the suburban watershed
in the absence of LID, permeable pavement (PP), bioretention (BR), and vegetative
swale (VS). A comprehensive evaluation of the different LID scenarios is assessed in
this study based on an analytic hierarchy process (AHP) [63, 64]. The AHP is a multi-
criteria programming technique for decision-making in complex environments. The
AHP works by developing sets of priorities for alternatives managements and the
criteria used to judge those alternative managements [65]. Environmental benefits such
as reduction in runoff volume and peak are considered in the prioritization and
determination of construction and infrastructural development projects [19,50,64].
2. Study Area and Data Collection
2.1 Study Area
The Chipuxet watershed is in southern Rhode Island, USA, and covers 36.93 sq.
mi (95.64 sq. km) (Figure 3). The watershed provides most of the public water supply
(used in agriculture, recreational use, and drinking water) of two towns. The Chipuxet
River that drains it is mainly recharged by groundwater [66]. The watershed
characteristics are typical for many similar suburban watersheds in the eastern United
States.
43
Figure 3: Location Map of the Study Area
Agricultural, drinking water, and recreational and current water demands affect water
quantity and quality and impact the recharge and runoff characteristics [67].
2.2 Local Precipitation
2.2.1. Observed Rainfall Data
Rainfall-runoff modeling with EPA-SWMM requires rainfall data as input to
the model. There are two National Oceanic and Atmospheric Administration (NOAA)
rain gauge stations (WBANNO: 54796 and 54797) in and around the Chipuxet
watershed (Figure 4). Both stations record data at daily 5-mins intervals. The collected
time-series data for 2006 to 2020 are continuous, having no missing data. Among this
time series data set, the 2017 and 2018 records were chosen to be the SWMM-5 model
calibration and validation, respectively.
44
2.2.2. Rainfall Data Preparation
Precipitation data server (PDS) based typical precipitation processes for six
rainfalls return occurrence intervals (1, 2, 5, 10, 25, 50, and 100 years) with five-minute
precipitation resolution and a minimum of 24h duration was investigated. All data,
including point precipitation frequency estimates with 90% confidence intervals and
supplementary information, were obtained from the NOAA Atlas (Table 1).
The rainfall depth ranges are from 3 inches (80 mm) to 8.59 inches (195 mm) of
24-hr precipitation for 1-yr to 100-yr (Figure 4). This study considered only one, two,
five, ten, fifty, and hundred years return periods to analyze the LID structures scenario.
Table 1: Point Rainfall Frequency (pf) estimates with 90% confidence intervals and
supplementary information (NOAA Atlas 14, volume 10, version 3).
Average
Recurrence
Interval,
Year
5-
min
10-
min
15-
min
30-
min
60-
min
2-
hr
3-hr 6-hr 12-
hr
24-
hr
1 8 12 14 19 24 32 37 48 61 74
2 10 14 17 23 30 39 45 58 73 88
5 13 18 21 30 38 50 59 74 92 111
10 15 22 25 36 45 60 70 88 108 129
25 19 26 31 43 56 73 85 107 130 155
50 21 30 35 49 63 83 96 121 147 175
100 24 34 40 55 71 93 108 135 164 195
45
PDS-based depth-duration-frequency (DDF) curves
Latitude: 41.74800, Longitude: -71.56180
Figure 4. Typical design rainfall processes of different return periods of 1-, 2-, 5-, 10-,
25-, 50-, and 100- yrs. (PDS: Precipitation Data Server)
2.3 Drainage Pipeline Data
Sewer mains and interceptors for public sewer systems from the State’s Utility data
[67] generally are only for pipes with 10 inches (0.25meter) or greater diameter.
Polyvinyl Chloride (PVC) sewer pipe and fittings are primarily used in the drainage
having 20 feet in length and 0.83 feet diameter. Due to the low to moderate topographic
slope (Figure 5), no overland flow occurs in this area unless heavy rainfall (> 32 mm/day;
Table 2), runoff water moves along the roads or pavements toward the low-lying areas
and flows into rivers. Consequently, the roads as the drainage channels are the central
part of the drainage system in this study (Figure 5).
46
Figure 5. Elevation and flow path, Soil infiltration rate, Land use, and the percentage
of impervious surface in the study area
47
Table 2: Categories of the rainfall
Rainfall Categories Rate of the precipitation
Light rain 0 - 4 mm/day
Light-Moderate 4 – 16 mm/day
Moderate - Heavy rain 16 – 32 mm/day
Heavy rain 32 - 64 mm/day
Heavy- Torrential rain 64 – 128 mm/day
Extreme or Torrential rain >128 mm/day
Source: Alpart et al., 2002
2.4 Elevation, Sub-catchment, Soil Infiltration, Land use and % Impervious Surface
Elevation data, including Digital Topographic Map (DTM) 1:2000 and Remote
Sensing Images (RSI) 1:2000, are required inputs for the SWMM model. Watershed
delineation (Figure 6) for the Chipuxet Watershed extracted from Digital Elevation
Model (DEM) file data and worked on in ArcGIS 10.6.1. Discharge outlet points can be
either nodes of the drainage system or a sub-catchment. Sub-basins are divided into
pervious and impervious subareas (Figure 6). Surface runoff can infiltrate into the upper
soil zone of the pervious subarea but not through the impervious subarea. Impervious
areas are divided into two subareas - one that contains depression storage and another
that does not. Runoff flow from one subarea in a sub-catchment can be routed to the
other subarea or both subareas.
Percent imperviousness is calculated from the land-use 2011 data, and perviousness
is also validated with the impervious data from RIGIS. The percent of imperviousness
48
is relatively high in the western part of the watershed, ranging from 9% to 46.5%. The
2001 Land use and land cover was updated here to assess the land types.
Figure 6: Sub-catchments, Nodes, and Streamflow Delineation of the watershed into
sub-basins of the study area with nodes, streamflow, and outlets.
49
A total of 49 sub-catchments were delineated. Each of the sub-catchment was
divided into two subareas: an impervious area, a pervious area. The areas and slopes of
each sub-catchment were calculated using ArcGIS. Then ArcGIS files were converted
into .inp files through inp.PINS application.
3. Methodology
3.1. Hydrological Model Selection
Researchers developed many methods for evaluating the effects of LID to study
highly urbanized watersheds; among them, hydrological models are considered the most
accurate method [23,26, 69-76]. Hydrological models, such as the Long-Term
Hydrologic Impact Assessment-Low Impact Development (L-THIA-LID) model [Liu
et al., 2015]; the improved SCS-CN model [77,78]; the System for Urban Storm-water
Treatment and Analysis Integration (SUSTAIN) model [79]; and the SWMM [78, 80-
82] all are well-established models to evaluate LID structures. In this study, the most
updated SWMM version 5.1 was applied for surface runoff simulation because it
provides realistic peak discharge and surface runoff simulation [82], and it is
supplemented with a LID module [22, 29,32]. SWMM with different LID structures
has been used widely for different regions and countries, such as the USA and China. It
must be noted that this study focuses on suburban and semi-urban-mixed watersheds
where the effectiveness of SWMM coupled with LID has not been well understood.
SWMM is open access and a flexible numerical model that simulates the stormwater
runoff quantity and quality based on the physical processes of surface runoff, infiltration,
surface ponding, and flow routing.
50
3.2 Input Parameterization
For any model simulation, proper selection of the initial parameter values is most
crucial. Previous studies related to SWMM [16,18] indicated that the width (W) of the
region, impervious coefficient, Manning coefficient (n), and depression storage (D) for
both the impervious and pervious areas are sensitive parameters for runoff simulation.
Width (W) depends on the shape and size of the watershed area and its topography and
underlying surface and can be calculated according to Equation (1) [18],
W = K * (0.2 <K<5) …………………………………………………………..(1)
where K is the shape correction coefficient, and A is the area of the watershed. K is
crucial for calculating W and was calibrated based on the measured geometric
characteristics. According to the shape of the study area, the initial value of K was
assigned as 1.1 in this study.
The volume of the surface runoff is directly related to the impervious coefficient.
In this study, all the initial k values were used based on the Rhode Island Stormwater
Design and Installation Standards Manual, 2010, and then the corresponding impervious
coefficients were obtained (Table 3).
Table 3. Summary of Sub-catchments Characteristics
Sub-basin
ID
Area
Sq.km
%
Imp
Slope N-Imp N-Per S-Imp S-Per Manning's
Roughness
SC_1 0.35 7.0 15 0.01 0.1 0.02 0.25 0.055
SC_2 0.40 15.9 8.5 0.01 0.1 0.02 0.25 0.055
SC_3 0.40 4.0 10 0.01 0.1 0.02 0.25 0.12
SC_4 1.20 0.0 8.5 0.01 0.1 0.02 0.25 0.12
51
SC_5 4.27 128 6 0.01 0.1 0.02 0.25 0.065
SC_6 1.23 61.5 12 0.01 0.1 0.02 0.25 0.03
SC_7 2.04 40.8 8 0.01 0.1 0.02 0.25 0.12
SC_8 1.16 23.1 6 0.01 0.1 0.02 0.25 0.12
SC_9 0.07 0.0 7.5 0.01 0.1 0.02 0.25 0.12
SC_10 1.31 26.2 15 0.01 0.1 0.02 0.25 0.12
SC_11 1.01 10.1 8 0.01 0.1 0.02 0.25 0.12
SC_12 0.30 9.1 7 0.01 0.1 0.02 0.25 0.12
SC_13 1.41 49.3 7 0.01 0.1 0.02 0.25 0.12
SC_14 0.72 28.9 12 0.01 0.1 0.02 0.25 0.055
SC_15 1.08 0.0 12 0.01 0.1 0.02 0.25 0.12
SC_16 0.09 3.8 12 0.01 0.1 0.02 0.25 0.023
SC_17 0.94 56.3 9 0.01 0.1 0.02 0.25 0.023
SC_18 0.36 1.8 8 0.01 0.1 0.02 0.25 0.12
SC_19 0.82 77.8 7.5 0.01 0.1 0.02 0.25 0.011
SC_20 1.37 0.0 7 0.01 0.1 0.02 0.25 0.12
SC_21 0.47 2.4 10 0.01 0.1 0.02 0.25 0.12
SC_22 0.67 0.0 7 0.01 0.1 0.02 0.25 0.12
SC_23 1.14 28.6 14 0.01 0.1 0.02 0.25 0.055
SC_24 1.72 8.6 13 0.01 0.1 0.02 0.25 0.12
SC_25 1.73 0.0 4 0.01 0.1 0.02 0.25 0.12
SC_26 0.34 0.0 11 0.01 0.1 0.02 0.25 0.055
SC_27 1.23 12.3 10 0.01 0.1 0.02 0.25 0.055
SC_28 0.98 4.9 3 0.01 0.1 0.02 0.25 0.12
SC_29 7.55 113 12 0.01 0.1 0.02 0.25 0.065
SC_30 0.69 27.4 12 0.01 0.1 0.02 0.25 0.12
SC_31 3.65 0.0 3 0.01 0.1 0.02 0.25 0.12
SC_32 2.31 46.3 7 0.01 0.1 0.02 0.25 0.055
52
SC_33 1.44 0.5 8 0.01 0.1 0.02 0.25 0.035
SC_34 0.81 8.1 12 0.01 0.1 0.02 0.25 0.012
SC_35 0.63 15.9 15 0.01 0.1 0.02 0.25 0.055
SC_36 0.20 0.0 4 0.01 0.1 0.02 0.25 0.12
SC_37 0.99 0.0 6.5 0.01 0.1 0.02 0.25 0.12
SC_38 0.02 0.0 9 0.01 0.1 0.02 0.25 0.055
SC_39 0.72 28.8 8 0.01 0.1 0.02 0.25 0.023
SC_40 1.86 27.9 14 0.01 0.1 0.02 0.25 0.065
SC_41 0.46 0.2 8 0.01 0.1 0.02 0.25 0.035
SC_42 3.45 41.4 12 0.01 0.1 0.02 0.25 0.055
SC_43 1.79 17.9 11 0.01 0.1 0.02 0.25 0.012
SC_44 0.61 18.4 9 0.01 0.1 0.02 0.25 0.055
SC_45 2.14 21.4 7 0.01 0.1 0.02 0.25 0.1
SC_46 0.69 13.7 8 0.01 0.1 0.02 0.25 0.055
SC_47 2.15 43.1 9 0.01 0.1 0.02 0.25 0.055
SC_48 1.21 0.0 6 0.01 0.1 0.02 0.25 0.12
SC_49 0.41 18.3 9 0.01 0.1 0.02 0.25 0.023
The impervious and pervious areas influence the sub-surface properties, and initial
values are all empirical. The study area is moderately low in elevation does not vary
much; the initial n of impervious areas (channels) was set as 0.25, and the initial n of
pervious areas were set as 0.9, according to the manual of SWMM 5.0 [32]. For the
model initialization, D of the impervious and pervious areas was 1.7 mm and 6 mm,
respectively, based on the point frequency estimate presented in Table 2.
53
For the model calibration and validation, the initial sensitive parameters were
considered the optimization parameters, and the Horton model was applied here for the
infiltration processes. The infiltration (f) was calculated as Equation (2) [18,22],
f = fc + (f0 − fc) e−kt ………………………………………………………………(2)
where f0 and fc are the minimum and maximum infiltration rates (mm/sec), respectively,
and k is the attenuation coefficient [-]. Although fc and f0 are less sensitive than the k
and W parameters, they are critical in infiltration calculation. The other SWMM
parameters, such as cross-section size, the characteristics of the drainage pipeline
(channel), and slope, were calculated based on the DEM, and investigation data (Table
3). The drainage plan was collected from the stormwater management plan report on
the South Kingstown area [84].
3.3 Model Calibration and Validation
SWMM was calibrated with discharge data from 2017. Three individual storm
events were considered for model calibration and validation with the discharge data
from USGS 01117350 Chipuxet river at West Kingston, RI. Calibration was conducted
for three different model sets up (Run 1, Run 2 and Run 3) in three different sub-
catchments. The adjusted validation parameter is presented in Table 5.
Figures 7 shows the observed, and simulated discharge hydrograph for the three
different models. The optimal parameter values for the study area were selected (Table
5) according to the Nash-Sutcliffe Efficiency (NSE) [85].
NSE = 1 − ∑ (Qsim,t−Qobs,t)2n
t
∑ (Qobs,t−Q͞obs)2nt
…………………………..…………….…………(3)
54
where Qsim,t [m3/s] is the predicted discharge, Qobs,t [m
3/s] is the observed discharge, Q¯
obs [m3/s] is the mean of the observed discharge and n is the number of time steps and t
is the time. NSE ranges between −∞ and 1, where 1 is considered a perfect fit for any
model where NSE is sensitive to extremes [86]. Dongquan et al. 2009 [87] suggests that
an NSE value of 0.5 or greater is acceptable for SWMM simulation. The coefficient of
determination, R2, an established measure in statistics quantifying the model’s fit, is
described as:
R2 = 1 −∑ (Pj−P^j)2m
j=1
∑ (Pj−P͞j)2mj=1
……………………………………………………..………..(4)
where Pj and Pjˆ are the observed and estimated values, Pj¯ is the observed mean, and
m is the number of discharge peaks. The denominator is the total variance in the
observed data, and the numerator is the variance between the simulated and observed
data. The closer R2 is to 1, the larger proportions of the observed variability can be
explained with the simulations of estimated values [88]. As predicting runoff peaks is
one of the main purposes of this suburban hydrological models, R2 of observed and
predicted runoff peaks provide the measure of performance measures.
3.4 LID Selection and Design Scenarios
SWMM 5.1 includes modules for LID components including permeable pavement
(PP), bio-retention cells (BR), and vegetative swales (VS). These are the most common
LID in the study according to the Stormwater Management Plan [84] and the Rhode
Island Stormwater Design and Installation Standards Manual [89]. The SWMM-LID
implementation considers the landscaping schemes and how the local constructions are
planned to mean that the Land use criteria, local drainage, and construction plans are
55
taken into consideration. SWMM-LID preserves the original terrain, limits the ratio of
impervious surface areas, and avoids the direct connection of impervious areas. Also,
suitable LID structure selection is made according to local conditions in terms of both
technical and social/economic factors, and the specific stormwater problem of the
selected area. In this study, two large storm events in 2010 and 2013 and their impacts
on this area and the suburban growth [54] problems are considered before implementing
the LID structures for the developed SWMM model. These two aforementioned storms
resulted in large flood and infrastructural damages.
Scenarios for PP, BR, and VS were developed, considering storm design events (one,
two, five, ten, fifty, and hundred year). The storm, soil, and LID design parameters of
each LID (Table 4) measurement was assigned according to the soil infiltration rate,
land use, and elevation, and local technical regulation [88]. The peak infiltration rate in
determining runoff capture is computed based on the input parameters of the selected
LID structures.
3.4.1 Baseline Scenario
The calibrated SWMM having no LID at the outfall served as the baseline
scenario and was compared to the model that included PP, BR, and VS.
3.4.2 Permeable Pavement (PP) Scenario
Properly maintained permeable pavements are highly effective for parking lots or
where the percentage of impervious surfaces is elevated. In this model, permeable
pavements were set in the University of Rhode Island parking lot and the municipal
56
sidewalks. The PP ratio of the sidewalks was set as 90% and that of the residential
community roads was 70%. The surface layer thickness was approximately 2 cm with
0.65 porosity; the entire layer thickness was 50.8 cm, and 1.3 cm/hr was considered as
the infiltration rate according to the Rhode Island stormwater management technical
regulation [89]. Slopes (<9%), percentage of underlying soil (Clay content <20%,
Silt<40%) were assigned based on [84]. Table 4 represents all input parameters of the
selected LID structures.
3.4.3 Bio-Retention (BR) Cell Scenarios
Bioretention (BR) reduces runoff volume, and peak discharge rate and it is also
capable of reducing the pollutant levels [33, 34]. BR is primarily effective in low slope
areas (5% to 15%), such as this study area.
BR cells were in the northwest part of the watershed. The required parameter values
(Table 4) used in the model were based on the Rhode Island stormwater management
structure manual. The depth of the surface storage was 21 cm with 90% vegetation
coverage; the soil thickness was 12 inches with 0.15 porosity. According to the duration
depth and frequency curve, the precipitation depth is 8 cm. The field capacity was 0.2,
and the wilting point was 0.2. The conductivity coefficient of the soil layer was 6, with
1.1 cm/hr hydraulic conductivity. The gravel/stone layer height was 25 cm with a 20.32
suction head of media layer. The other required parameter i.e., surface roughness 0.5
(manning’s, n), ponding layer (6 – 9) cm, was taken based on [84].
57
Table 4: Summary of LID Characteristics
Parameter Permeable
Pavement
Bioretention
Cell
Vegetative
Swale
Surface Layer
Storage Depth (cm) 2 21 18
Vegetation (%) 0 90 100
Media Layer
Depth (cm) 50.8 31 12.7
Porosity (%) 0.65 0.15 0.25
Field Capacity (%) n/a 35 12
Hyd.cond (cm/hr) 2 3.6 0.85
Cond.slope n/a 6 15
Suction head n/a 20.32 2.5
Stone Layer
Depth (cm) 18-25 25 86
Void ratio (v/s) 0.65 67 67
Soil hyd.cond.
(cm/hr)
2.3 1.3 1.1
Underdrain-free
Coefficient 3.8 4 5
Exponent 0.5 0.5 0.5
Offset (cm) 3.1 0 10.1
Underdrain-controlled
58
Coefficient 2.7 0.55 0.25
Offset (cm) 21-39 2.9 4.7
3.4.4 Vegetative Swale (VS) Scenarios
VS are considered an important type of LID due to their design and effectiveness.
Because of its dense grass cover, the roughness of the swales is significantly higher than
that of the streambed conduits; simultaneously, VS also provides good infiltration. VS
is also applicable to water quality treatment through sedimentation and biological
uptake. VS was set in six Sub-catchments (SC 7,13,17,21,39, and 49) where the slope
is in between 7% to 14% and the area of %impervious was greater than 10 sq.km of the
study area (Table 3). The total swale length was 500 m, and the total storage depth of
the surface layer was approximately 18 cm. The Manning’s n for the surface bed was
taken as 0.5 while a value of 0.24 was used for the swales. The hydraulic conductivity
of saturated soil was 0.85 cm/hr, and the soil porosity was 0.25. All the input parameters
are represented in Table 4.
3.5 Assessment of the LID envrionmental benefits
LID structures are usually cost-effective and robust approaches for stormwater
management bring social, economic, public health, and environmental benefits to
communities. It typically influences the physical environment by reducing impervious
surfaces and constructing natural habitat and permeable surfaces. In this study,
comprehensive environmental benefits are assessed based on two sub-indicator runoff
reduction and peak reduction. It was simulated to compare the reduction rates before
and after the addition of LID scenarios.
59
3.5.1 Calculation of Runoff and Peak Reduction
Runoff volume and peak reduction is calculated for each of the LID structures
to assess its effectiveness using equation 4 [89],
R (%) = Q0−QLID
Q0 x 100(%) ……………………………………………………. (5)
Where R is the reduction rate in %, Q0 (cm/s) is the simulated runoff without LID
structure, and QLID (m3/s) is the simulated runoff with LID.
3.5.2 Calculation of LID environmental benefits
The benefits of each LID were measured in this study, applying the analytical
hierarchy process (AHP). The individual benefits were calculated using a weighted
summation of indicator values applying equation 6:
Bim= ∑ Wkik Ikm …………………………………………………………………… (6)
where Bim is the benefit i in the scenario m, Wki refers to the calculated weight of
indicator k in benefit i, and lkm is the normalized value of the indicator k in scenario m.
Linear normalization makes the indicator values dimensionless, equation 7:
lkm = Xkm
∑ XkmMm=1
…………………………………………….……………………… (7)
where Xkm is the original value of the indicator k in scenario m, and M is the total
number of the LID scenario; in this study M is 4.
The performance in the AHP analysis is a product of relative weight of the
category and estimated the sub-indicator of runoff reduction and peak reduction for each
60
of the LID scenarios. The comprehenssive environmental benefit was determined for
four different LID scenarios applying four different weight indicator (20%, 40%, 60%,
and 80%). Indicators represents the area increased for each of the LID controls. For
example, if the permeable pavement area is increased 20% then how much runoff
volume and peak could be reduced compared with the original value. Each of the LID
structures area is increased under the same hydrological (rainfall, duration of rainfall,
and intensity), landuse, soil infiltration rate, and the topographical (slope, elevation).
Runoff volume and peak discharge reduction were assessed then for each of the LID
structures (PP, BR, and VS) under different weight indicator.
4. Result and Discussion
4.1 SWMM Calibration and Validation
The obtained NSE in run 1 is 0.73, in run 2 is 0.58, and for run 3 it is 0.64 (Table
5). The R2 is 0.88, 0.68, and 0.73 for run1, 2 and 4 respectively (Table 5).
61
Table 5: SWMM Calibration Parameter
Calibration Run 1 (Sub-catchment 24) :NSE = 0.73
Parameter Adjustment Measured Run 1 Run 2 Run 3
% Impervious ….... x 1.0 X 0.23 X 0.25
Depression Storage (mm) ….... 2.032 1.524 2.54
fo & fc ….... 1 1 1
Sub-catchment Width ….... 1 1 1
Impervious "n" ….... 0.02 0.02 0.02
Pipe "n" ….... 0.013 0.013 0.013
Depth of Runoff (mm) 1.15 5 1.8 1.15
Maximum Flow (cms) 5.3 4.43 6.9 7.1
Time of Max.Flow 9:45 9:40 9:45 9:40
62
Calibration Run 2 (Sub-catchment 18) :NSE = 0.59
Calibration Run 3 (Sub-catchment 33) :NSE = 0.64
Figure 7: Time series of the observed (green) and simulated (red) discharge [m3/s] for
Run 1, 2, and 3 during the calibration period. The precipitation rate [mm/sec] for the
corresponding time of discharge is included on the secondary axis (blue).
63
Table 6: Comparison between three calibrations run
Calibration of SWMM Sub-catchment NSE R2
Run 1 Sub 24 0.73 0.88
Run 2 Sub 18 0.59 0.68
Run 3 Sub 33 0.64 0.73
4.2 Results in Baseline Scenario
The runoff processes under different LID scenarios (Figure 8) were simulated in
the SWMM model and their corresponding runoff depths (Tab.7) were found to be
proportional to the return period, i.e., with the increasing return period, the runoff depth
is increased. Among the four different LID scenarios simulated, BR and PP showed the
highest impact in terms of runoff volume and peak reduction in the study area.
(a) (b)
(c)
(d)
64
Figure 8: Simulated runoff processes in four LID structures with different return-
period events: (a) One year; (b) Two-year; (c) five-year (d) ten-year (e) 50-year, and
(f) 100-year
4.3 Runoff Volume Reduction
The runoff reduction effectiveness (Figure 9) was evaluated through the
percentages of runoff reduction calculation and analysis (Table 8), considering the four
LID structures. Among all, PP is the most effective in runoff reduction, especially when
the return period of precipitation is short i.e., 1-year and 2-years. For PP, the runoff
reduction percentages ranged from 31% to 78% in all types of rainfall events, and the
peak reduction percentages range was 28 %to 70%. The reduction rate is much higher
in the one-year and two-year return period rainfall event than rest of the rainfall return
periods. It means, over the time the performances are gradually decreasing, and this
output is the same for all applied LID structures in this study.
(f) (e)
65
Table 7. Percentage of runoff and peak reduction per unit area under six rainfall
reoccurrence scenarios.
LID Measures Runoff Reduction/ha (%) Peak Reduction/ha (%)
1
yr
2
yr
5
yr
10
yr
50
yr
100
yr
1
yr
2
yr
5
yr
10
yr
50
yr
100
yr
Permeable Pavement (PP) 78 65 58 58 45 31 70 61 45 45 32 28
Bio-retention cell (BR) 70 70 55 52 48 35 72 71 50 41 33 24
Vegetative Swale (VS) 60 65 45 45 36 28 55 52 40 40 23 19
Figure 9. Hydrological effectiveness of each LID per unit area.
BRs were effective for the study area, especially where the slope is relatively low.
Hydraulic gradient and the gravel pack design are the most impactful BR design
parameters. Besides, 90% vegetation cover has a significant impact in reducing the
runoff volume. BR showed the same characteristics that PP exhibited on reducing runoff
66
percentages, with the longer return period (runoff reduction rate ranged from 35% to
70%). This result suggests that BR cells are effective in various rainfall intensities.
Applying various measures or structures changes the state and process of the runoff in
part of the study area, which likely balances some of the overall effectiveness.
In terms of runoff reduction, VS was the least effective among all three LIDs
evaluated herein, with the runoff reduction ranging from 28% to 60%. Location and the
percentages of the impervious area are likely the most important factors determining the
relative performance of VS. Usually, VS are effective in flat land where the slope is <
10% [91-93], but the selection of the location and the proper design is essential for
getting the most significant outcomes.
Overall, all three LID structures can reduce the runoff volume between 28% to 78%
during rainfall events with different return periods. It is noticeable that the volume
reduction range is higher for shorter return periods and that LIDs are less effective if the
rainfall events in the return periods is higher than five years. The effectiveness of runoff
reduction in all scenarios decreased as the rainfall intensity increased. Proper
maintenance of the LID structures could increase their performance for long period of
time.
4.4 Peak Discharge Reduction
The peak discharge reduction was calculated to evaluate the effectiveness of the
LID structures (permeable pavement, bioretention cells, and vegetative swales) (Table
8). All LID structures showed significant outputs in terms of runoff volume and peak
67
reduction for the study area (Figure 9). The calculated percentage peak flow reduction
of the LID scenarios (Table 8) was based on the rainfall return periods assessing and
comparing their suitability. The analysis showed that storage volumes within BR cells
could be depleted during significant storm events. In 2013, Qin et al. [94] suggested that
BR performance significantly varies from storm to storm in urban watersheds, and in
the suburban study, it showed the same performance for BR. In this study, LID
structures were designed based on 1, 2, 5, 10, 50, and 100 years return period of rainfall
intensity. The performance of each LID structure decreased with increased storm
intensities. VS were less effective for peak reduction (19% - 55%) compared to BR and
PP, although VS had an effective storage capacity (18 cm) in the study area. The reason
for the lower VS effectiveness could be related to their size and frequency which
influences their total water storage capacity. For peak reduction, permeable pavement
(28% - 70%) and bioretention cells (24% - 72%) are most effective (Table 8 and Figure
9) for the selected location.
In summary, the proper implementation of LID structures was far from ideal for
solving the interval problems of flood in the suburban study areas. Previous research
supports these analyses [95]. A regional flood risk map should be prepared based on the
simulation results of the ponding depth. Public awareness needs to be created for the
inhabitants in the low-lying areas which are also the high-risk areas for the regions
studied.
68
4.5 Environmental Benefits
Environmental benefits were represented by two corresponding sub-indicator of
runoff volume and peak flow reduction, where it measured their benefits after
implementing the LID structures. The comprehensive environmental benefits of PP, BR,
and VS for six different rainfall return periods are represented in Table 8.
Table 8: Environmental benefits for different LID stuctures
Types of LID Facilities
Rainfall Return Period (%)
1 yr 2 yr 5 yr 10 yr 50 yr 100 yr
Permeable Pavement (PP) 75 64 54 54 41 30
Bio-retention Cell (BR) 71 70 56 48 43 31
Vegetative Swale (VS) 58 54 43 43 32 25
The reduction rates of runoff volume and peak increased with the increase of the
size and numbers of the LID structures (Figure 10). Again, the effectiveness of the LID
structures decreased with the increase of rainfall intensities; it is also applicable for the
environmental benefits (Figure 10).
0 20 40 60 80 100
1 yr
2 yr
5 yr
10 yr
50 yr
100 yr
Environmental Benefit (%)
Vegetative Swale (VS) Bio-retention Cell (BR) Porous Pavement (PP)
69
Figure 10: Environmental benefits and its standard error for each of the LID structures
in different rainfall intensities. The standard error of the mean was applied to construct
a confidence interval (95%) for the environmental benefits for each of the LID structures.
Finally, the following equation are applied to determine the environmental benefits for
three selected LID structures in terms of six different rainfall return period:
Environmental Benefits (%) = 0.667 * runoff volume reduction + 0.333* peak reduction.
The reduction rates of LID facilities for different rainfall intensities per unit area
(sq. km) were ranked as follows: BR > PP > VS. Bioretention cell showed a consistent
effectiveness in every rainfall intensity even for the 100 years return period (31% ± 5%).
5. Conclusions
Unlike previous studies which applied SWMM to mostly urbanized watersheds,
this study extended the SWMM model to a data-sparse suburban area. Because of the
importance of suburban sprawl in most countries, combined with an increasing
likelihood of flooding events in the future, more studies of stormwater management in
suburban watershed are needed. Input data covered rainfall data for different return
periods and the digital input of the drainage pipe networks was the basis for delineating
49 sub-catchments. Three different LID structures were considered, i.e., permeable
pavement, bioretention cell, and vegetative swales. The model calibration and validation
were conducted for three set ups i.e., run 1, run 2, and run 3, in three different sub-
catchments. The respective NSE values were 0.73, 0.59, and 0.64, with an average
value of 0.65. The average R2 value was 0.76. Both NSE and R2 indicated a good fit of
the SWMM model in this study area. Further, four additional scenarios were modeled
70
with the ‘no LID’ serving as the baseline to which PP, BR, and VS LIDs were compared
to by evaluating their response to runoff volume reduction and the peak discharge
reduction. The simulations indicate that all three LID structures were maximum
efficient during one-year to five-year rainfall events only. During more severe storms,
PP and BR exhibited a significant reduction in their effectiveness controlling runoff
volume and peak discharge. This study suggests that the LID performance is location
specific with design and input parameters (Table 4) being are the controlling factors.
Environmental benefit analysis also suggest that LID structures are effective for the
suburban areas. Most importantly, permeable pavement and bioretention showed the
significant impact in terms of runoff and peak discharge reduction for this suburban area.
Moreover, through the proper and regular management and monitoring of various
LID structures, the effectiveness in various aspects could make additional improve
under all scenarios. It is noted that the effectiveness of the LID structures is inversely
correlated to the rainfall intensity in terms of all the selected aspects. This study has also
revealed that for the suburban area bioretention cell, and the permeable pavement are
the applicable LID structure for managing the peak and runoff volume.
Acknowledgments:
The authors acknowledge the funding obtained from the Rhode Island Department of
Transportation (RIDOT)- SPR-234-2362, USDA Multistate Hatch Grants S1063 and S1089,
and the College of Environment and Life Sciences, University of Rhode Island. This research
and associated analysis are done through Hydro-Systems and Water Quality Lab (Pradhanang
Lab) assistance. The authors like to special thank William Brown, Meteorologist from NOAA's
National Centers for Environmental Information, for providing data support, especially for the
suburban area. The authors are also thankful to the CHI for providing a PCSWMM fellowship.
71
Conflicts of Interest: There is no conflict of interest.
References
1. Lumen. Urbanization and its challenges, Module 4: Urbanization and Immigration,
https://courses.lumenlearning.com/atd-hostos-ushistory/chapter/urbanization-and-its-
challenges/, 2014.
2. Hatt, B. E., Fletcher, T. D., Walsh, C. J., and Taylor, S. L.: The influence of urban density
and drainage infrastructure on the concentrations and loads of pollutants in small streams,
Environ. Manage., 2004, 34, 112–124, https://doi.org/10.1007/s00267-004-0221-8.
3. Leopold, L. B., Wolman, M. G., and Miller, J. P.: Fluvial Processes in Geomorphology,
Geographical J., 1995, 131.
4. Liu, Y., Ahiablame, L. M., Bralts, V. F., and Engel, B. A.: Enhancing a rainfall-runoff
model to assess the impacts of BMPs and LID practices on storm runoff, J. Environ. Man-
age., 2015, 147, 12–23, https://doi.org/10.1016/j.jenvman.2014.09.005.
5. Jacobson, C. R.: Identification and quantification of the hydrological impacts of impervi-
ousness in urban catchments: a review, J. Environ. Manage., 2011, 92, 1438–1448,
https://doi.org/10.1016/j.jenvman.2011.01.018.
6. Zhang, X., Guo, X., and Hu, M.: Hydrological effect of typical low impact development
approaches in a residential district, Nat. Haz., 2016, 80, 389–400,
https://doi.org/10.1007/s11069-015-1974-5.
7. Paule-Mercado, M.A., Lee, B.Y., Memon, S.A., Umer, S.R., Salim, I. Influence of land
development on stormwater runoff from a mixed land use and land cover catchment. Sci.
Total Environ. , 2017, 599, 2142–2155. https://doi.org/10.1016/j.scitotenv. 2017.05.081.
8. Zhou, Q. A review of sustainable urban drainage systems considering the climate change
and urbanization impacts. Water, 2014, 6 (4), 976–992. https://doi.org/10.3390/ w6040976.
9. Arnold, C. L. and Gibbons, C.J.: Impervious surface coverage – The emergence of a key
environmental indicator, J. Am. Plann. Assoc., 1996, 62, 243–258,
https://doi.org/10.1080/01944369608975688.
10. Beckers, A., Dewals, B., Erpicum, S., Dujardin, S., Detrembleur, S., Teller, J., Pirotton,
M., and Archambeau, P.: Contribution of land use changes to future flood damage along
the river Meuse in the Walloon region, Nat. Hazards Earth Syst. Sci., 2013, 13, 2301–2318,
https://doi.org/10.5194/nhess-13-2301-2013.
11. Claessens, L., Hopkinson, C., Rastetter, E., and Vallino, J.: Effect of historical changes in
land use and climate on the water budget of an urbanizing watershed, Water Resour. Res.,
2006, 42, 446–455, https://doi.org/10.1029/2005wr004131.
12. Pradhanang, S.; Jahan. K. Urban Water Security for Sustainable Cities in the Context of
Climate Change; Chapter 14, Water, Climate and Sustainability; John Wiley & Sons, Inc.:
Hoboken, NJ, USA, 2021; pp. 213–224, doi:10.1002/9781119564522.ch14.
72
13. U.S. EPA EPA Journal: Nonpoint Source Pollution: Runoff of Rain and Snowmelt, Our
Biggest Water Quality Problem. 1991, Volume 17, Number 5.
14. Novotny, V. and Olem, H. Water Quality Ð Prevention, Identi®cation, and Management
of Di€use Pollution, 1994, Van Nostrand Reinhold, New York.
15. US EPA. National Water Quality Inventory: 1990 Report to Congress. Office of Water,
1992. EPA 503/99-92-006.
16. Lee, G.Fred and Jones-Lee, Anne. Impact of Municipal and Industrial Non-Hazardous
Waste Landfills on Public Health and the Environment: An Overview. Prepared for Cali-
fornia EPA Comparative Risk Project, Sacramento, CA. 1994. Available at:
http://www.gfredlee.com/cal_risk.htm
17. Tsihrintzis,V.A. and Hamid, R. Modeling and management of urban stormwater runoff
quality: a review, Eat. Resour. Manage. 1997, EWRA, 11(2), 137-164.
18. Elliot, A.H., Trowsdale, S.A. A reviewof models for low impact urban stormwater drain-
age, Environ. Modell. Softw., 2007, 22.394-405.
19. Chow . MF; Yusop. Z.; Toriman. M. E. Modelling runoff quantity and quality in tropical
urban catchments using stormwater management model. Int. J. Environ. Sci. Technol.,
2012, 9:737-748. DOI 10.1007/s13762-012-0092-0
20. Barbosa. AE; Fernandes. JN; David. LM Key issues for sustainable urban stormwater man-
agement. Water research, 2012, 46, 6787-6798
21. Jia. H.; Lu Y.; Yu. SL; Chen. Y. Planning of LID-BMPs for urban runoff control: The case
of Beijing Olympic village. Separation and Purification Technology, 2012, 84, 112-119.
Doi: 10.1016/j.sepper.2011.04.026
22. Lee J.G.; Seivakumar A.; Alvi. K.; Riverson. J.; Zhen. J.X. A watershed-scale design op-
timization model for stormwater best management practices. Environmental modeling &
software., 2012, 37, 6 – 18. Doi: 10.1016/j.envsoft. 2012.04. 011
23. Baek. SS; Choi. DH; Jung J.W.;Lee. HJ; Lee.H.. Optimizing low impact development
(LID) for stormwater runoff treatment in urban area, Korea: Experimental and modeling
approach. Water Research, 2015, 86, 122-131. Doi. 10.1016/j.watres. 2015.08.038.
24. Park. M.; Chung. G.; Yoo. C.; Kim. JH Optimal design of stormwater detention basin using
the genetic algorithm. KSCE Journal of Civil Engineering, 2012, 16(4): 660-666. DOI.
10.1007/s12205-012-0991-0
25. Luan, Q.; Fu, X.; Song, C.; Wang, H.; Liu, J.; Wang, Y. Runoff effect evaluation of the
through SWMM in typical mountains, Low- Lying urban areas: A case study in China.
Water, 2017, 9, 439, doi 10.3390/w9060439
26. Dietrich. A.; Yarlagadda. R.; Gruden. C. Estimating the potential benefits of green storm-
water infrastructure on developed sites using hydrologic model simulation. Environmental
Progress & Sustainable Energy, 2016, (Vol 00, Doi 10.1002/ep
73
27. Welker. A. L.; Asce. M.; Mandarano. L.; Greising. K.; Mastrocola.K. Application of a
monitoring plan for stormwater control measures in the Philadelphia region. J. Envi-
ron.Eng., 2013, 139.1108-1118. ISSN 0733-9372/2013/8-1108-1118/
28. Jia, H.; rossWang, X.; Ti, C.; Zhai, Y.; Field, R.; Tafuri, A.N.; Cai, H.; Shaw, L.Y. Field
monitoring of a LID-BMP treatment train system in China. Environ. Monit. Assess., 2015,
187, 373.
29. Dietz, M.E., Clausen, J.C. Stormwater runoff and export changes with development in a
traditional and low impact subdivision. J. Environ. Manag., 2008, 87 (4), 560-566
30. Rosa, D.J. Clausen. J.C., Dietz, M.E. Calibration and verification of SWMM for low im-
pact development, JAWRA J. Am. Water Resour. Assoc., 2015, 51 (3), 746-757
31. Rossman, L.A., EPA’s stormwater management model user’s manual. E.P.A./600/R-
05/040, Office of research development, 2010, Cincinnati, OH.
32. Metcalf & Eddy, Inc. Stormwater management model, Volume 1 – Final Report;
11024DOC07/71; Water quality office, Environmental protection agency, July 1971,
Washington, DC, USA.
33. United States Environmental Protection Agency (US EPA), Preliminary data summary of
urban stormwater best management practices, EPA-821-R-99-012, July 1971, Office of
Water, Washington, DC, Washington State Department of Ecology (WA DOE).
34. EPA. Stormwater Management Model User’s Manual Version 5.0. National risk manage-
ment research laboratory office of research and development US Environmental Protection
Agency, 2015, OH 45268
35. Niazi M, Chris Nietch, Mahdi Maghrebi, Nicole Jackson, Brittany R. Bennett, Michael
Tryby, Arash Massoudieh.J Sustain Water Built Environ. Author manuscript; available in
PMC 2020 Jun 30.Published in final edited form as: J Sustain Water Built Environ., 2017
Jan 24; 3(2): 10.1061/jswbay.0000817. doi: 10.1061/jswbay.0000817
PMCID: PMC7326159
36. USGS. Elevation of the March-April 2010 Flood High Water in Selected River Reaches in
Rhode Island by Phillip J. Zarriello and Gardener C. Bent. US Department of the Interior.
US Geological Survey. 2011. https://pubs.usgs.gov/of/2011/1029/pdf/ofr2011-1029-
508.pdf
37. Tsihrintzis, A.V., Hamid, R., Runoff quality prediction from small urban catchments using
SWMM. J. Hydrogeological Processes, 1998, Vol 12, 311-329
38. Gironas. J., Roesner,LA, Davis, J., Rossman, L.A., Supply. W. Stormwater management
model application manual. National risk management research laboratory, 2009, Office of
Research and development, US Environmental Protection Agency, Cincinnati, OH.
39. Rossman, L.A. Storm Water Management Model User’s Manual Version 5.1, 2014,
USEPA: Washington, DC, USA
74
40. Warwick, J.J., and Tadepalli, P. Efficacy of SWMM Application,J.Wat. Resour.Plan.Man-
age., ASCE, 1991,117, 352-366
41. Abustan. I. Modelling of phosphorus transport in urban stormwater runoff. Ph.D. Disser-
tation, School of civil and Environmental Engineering ,1998 The university of New South
Wales, Australia
42. Barco. J., Wong. KM Automatic calibration of the US EPA SWMM model for a large
urban catchment. J Hydraul. Eng. 2008, 134 (4): 466-474.
43. Temprano, J., Arango, O., Cagiao, J., Suarez. J., Tejero.I., Stormwater quality calibration
by SWMM: a case studyin northern Spain. Water SA, 2006, 32(1): 55-63.
44. Hood M. Reihan A., Loigu. E., Modelling urban stormwater runoff pollution in Tallinn,
Estonia.In: UWM International symposium on new directions in urban water manage-
ment, 2007, Paris, September 12-14.
45. Jang. S., Cho, M., Yoon.Y., Kim. S.Kim. G., Kim.L., Aksoy.H. Using SWMM as a tool
for hydrologic impact assessment. Desalinization , 2007, 212:344-356.
46. Nazahiyah. R., Zulkifli. Y., Abustan I. Stormwater quality andpollution load estimation
froman urban residential catchment in Skudai, Johor, Malaysia, Water Sci Technol., 2007,
56 (7): 1-9.
47. Park, S.Y., Lee, K.W., Park, I.H., Ha. SR Effect of the aggregation level of surface runoff
fields and sewer network for a SWMM simulation. Desalination, 2008, 226, 328-337.
48. Tan SBK., Chua. LHC., Shuy. EB., Lo. EYM., Lim LW. Performances of rainfall runoff
models calibrated over single and continuous storm flow events. J Hydrol. Eng., 2008, 13
(7): 597 – 607.
49. Huber, W. C., Dickinson, R. E. Storm Water Management Model, Version 4, User Manual,
1988, EPA-600/3-88-001a.
50. Fletcher, T.D.; Shuster, W.; Hunt, W.F. SUDS, L.I.D., BMPs, WSUD and more—The evo-
lution and application of terminology surrounding urban drainage. Urban Water J., 2015,
12, 525–542.
51. Graham, P.; Maclean, L.; Medina, D.; Patwardhan, A.; Vasarhelyi, G. The role of water
balance modelling in the transition to low impact development. Water Qual. Res. J. Can.,
2004, 39, 331–342.
52. Bai, Y.R.; Zhao, N.; Zhang, R.Y.; Zeng, X.F. Storm Water Management of Low Impact
Development in Urban Areas Based on SWMM. Water, 2019, 11, 33.
53. Jahan, K.; Pradhanang, S. Predicting Runoff Chloride Concentrations in Suburban Water-
sheds Using an Artificial Neural Network (ANN). J. Hydrol. 2020, 7, 80, doi:10.3390/hy-
drology7040080.
75
54. Jahan, K.; Pradhanang, S.M.; Bhuiyan, M.A.E. Surface Runoff Responses to Suburban
Growth: An Integration of Remote Sensing, GIS, and Curve Number. Land 2021, 10, 452.
https://doi.org/10.3390/land10050452
54. USGS. Elevation of the March-April 2010 Flood High Water in Selected River Reaches in
Rhode Island by Phillip J. Zarriello and Gardener C. Bent. US Department of the Interior.
US Geological Survey, 2011, https://pubs.usgs.gov/of/2011/1029/pdf/ofr2011-1029-
508.pdf
55. Kenyon Grist Mill. Flood 2010: Photo Documentary (Before, During, and After). All Con-
tent Copyright - Official Website of Kenyon Grist Mill L.L.C. 21 Glen Rock Road/P.O.
Box 221 West Kingston, RI 02892, 2013, https://www.kenyonsgrist-
mill.com/flood2010.html
56. The Independent. Towns, Residents Close to Recovering from Historic Flood by Chris
Church. Independent.com 2011, https://www.independentri.com/local/article_71f7559f-
a7ec-537d-b380-80ac5e5c3ddb.html
57. SRI. A Look Back at the Flood of the Century. Southern Rhode Island Newspaper, 2011,
https://www.ricentral.com/a-look-back-at-the-flood-of-the-century/article_c8e44c9d-
a7f9-59f8-ac2b-f345b5f0e0a9.html
58. Gorman, G. Rhode Island Flood Zones-South Kingstown RI Real Estate has them too.
Rhode Island Coastal Real Estate, 2012, https://www.rihousehunt.com/ri-blog/rhode-is-
land-flood-zones-south-kingstown-ri-real-estate-has-them-too/?do-
ing_wp_cron=1614615852.4240589141845703125000
59. The Independent, Heavy Rains Brings Flooding, Road Closures. Reporter Stephen Green-
well, Independentri.com, 2014, https://www.independentri.com/independ-
ents/south_county/article_35880c91-006d-5e2b-b32b-af311135fe9e.html
60. Patch the Community Corner. Flash Flood Watch in Narragansett and South Kingstown
by Erin Tiernan. Community Corner, 2013, https://patch.com/rhode-island/narragan-
sett/flash-flood-watch-in-narragansett-and-south-kingstown
61. ECO RI News. Rising Waters Threaten Region’s Cultural Treasures. ECO RI News, 2015,
https://www.ecori.org/climate-change/2015/1/5/rising-waters-threaten-regions-culture
62. Saaty, T. L. How to Make a Decision: The Analytic Hierarchy Process, European Journal
of Operational Research, 1970, 48,9-26. http://dx.doi.org/10.1016/0377-2217(90)90057-I
63. Li, Q.; Wang, F.; Yu, Y.; Huang, Z.C.; Li, MT; Guan, Y.T. Comprehensive performance
evaluation of LID practices for the sponge city construction: A case study in Guangxi,
China. J. Environ. Manag. 2019, 231, 10–20.
64. Schmoldt, D. L., J. Kangas, G. Mendoza, and M.Pesonen. The analytic Heirar-chy Process
in Natural Resource and Environmental Decision Making. DorDrecht, The Netherlands:
Kluwer Academic Publishers, 2001.
76
65. Wood-Pawcatuck,. Wood-Pawcatuck wild and scenic river, Chipuxet River, 2000
http://wpwildrivers.org/the-rivers/chipuxet-river/.
66. RI Water Resource Board, 2000. Chipuxet_Specific watershed.
http://www.wrb.ri.gov/documents/Chipuxet_specific%20watershed.pdf
67. RIGIS, 2019, http://www.edc.uri.edu/rigisspf/Metadata/Utilities/s44usl96.html.
68. Yang. Y; Chui, T.F.M. Optimizing surface and contributing areas of bioretention cells for
stormwater runoff quality and quantity management, Journal of Environmental Manage-
ment, 2018, 206,1090-1103. https://doi.org/10.1016/j.jenvman.2017.11.064
69. Rosa. D.J., Clausen. J.C., Dietz. EM, Calibration and verification of SWMM for low im-
pact development. Journal of the American Water Resources Association (JAWRA), 2015,
51(3): 746-757. Doi:10.1111/jawr.12272
70. Chui. T.F.M., Liu X., Zhan. W. Assessing cost-effectiveness of specific LID practice de-
signs in response to large storm events. Journal of Hydrology, 2016, 533, 353-364.
https://dx.doi.org/10.1016/j.jhydrol.2015.12.011
71. Lucas. CW and Sample. DJ Reducing combined sewer overflows by using outlet controls
for green stormwater infrastructure: case study in Richmond, Virginia. Journal of Hydrol-
ogy, 2015, 520, 473-488.https://dx.doi.org/10.1016/j.jhydrol.2014.10.029
72. Liu. Y., Cibin.R., Bralts. VF; Chaubey. I., Bowling L.C. Optimal selection and placement
of BMPS and LID Practices with a rainfall-runoff model. Environmental Modelling and
Software, 2016, 80, 281-296. http://dx.doi.org/10.1016/j.envsoft.2016.03.005
73. Kong. F. Ban. Y., Yin. H., James. P., Dronova. I. Modeling stormwater management at the
city district level in response to changes in land use and low impact development. Environ-
mental and Modelling & Software, 2017, 95, 132-142. http://dx.doi.org/10.1016/j.en-
vsoft.2017.06.021.
74. Liao. Z.L.He. Y., Huang. F. Wang. S., and HZU. Analysis on LID for highly urbanized
areas waterlogging control: demonstrated on the example of caohejing in Shanghai. Water
Science & Technology, 2013, 68.12, Doi.10.2166/wsi.2013.523.
75. Xie.J. Wu. C., Li.H., Chen. G. Study on stormwater management of grassed swales and
permeable pavement based on SWMM. Water, 2017. 9,840, doi:10.3390/w9110840.
76. Chronshey, R.G.; Roberts, R.T.; Miller, N. Urban hydrology for small watersheds (TR-55
REV.). Am. Soc. Civ.Eng. 1986, 55, 1268–1273.
77. Park, Y.S.; Enge, B.A.; Harbor, J. A web-based model to estimate the impact of best man-
agement practices. Water, 2014, 6, 455–471.
78. Shoemaker, L.; Riverson, J.; Alvi, K.; Rafi, T. SUSTAIN—A Framework for Placement
of Best Management Practices in Urban Watersheds to Protect Water Quality; Document
No. EPA-600-R-09-095, 2009, US Environmental Protection Agency (USEPA), Office of
77
Research and Development National Risk Management Research Laboratory: Cincinnati,
OH, USA.
79. Zhang, H. Low Impact Development of the Sponge Cities-the Research of the Typical
Mountain Runoff Effect. Master’s Thesis, 2016, Hebei University of Engineering, Handan,
China.
80. Marsalek, J.; Dick, T.M.;Wisner, P.E.; Clarke,W.G. Comparative evaluation of three urban
runoff models, Water Resour. Bull. AWRA 1975, 11, 306–328.
81. Villarreal, E.L.; Annette, S.D. Inner city stormwater control using a combination of best
management practices. Ecol. Eng., 2004, 22, 279–298.
82. Koudelak, P.; West, S. Sewerage Network Modelling in Latvia, use of InfoWorks CS and
Storm Water, Management Model 5 in Liepaja city. Water Environ. J. 2007, 22, 81–87.
83. Stormwater Management Plan (SMP), South Kingstown, Rhode Island. Town of South
Kingstown, July 17, 2020. BL Companies. BL Project Number: 18C6704.
https://www.southkingstownri.com/DocumentCenter/View/5219/Stormwater-Manage-
ment-Plan?bidId=
84. Nash, J.E. and Sutcliffe, J.V. River Flow Forecasting through Conceptual Model. Part 1—
A Discussion of Principles. Journal of Hydrology, 1970, 10, 282-290.
http://dx.doi.org/10.1016/0022-1694(70)90255-6
85. Krause, P., D. P. Boyle, and F. Bäse. “Comparison of different efficiency criteria for hydro-
logical model assessment”. In: Advances in geosciences, 2005, 5, pp. 89–97.
86. Dongquan, Z., C. Jining, W. Haozheng, T. Qingyuan, C. Shangbing, and S. Zheng. “GIS
based urban rainfall-runoff modeling using an automatic catchment-discretization ap-
proach: a case study in Macau”. In: Environmental Earth Sciences, 2009, 59 (2), p. 465.
87. Devore, J. L. and K. N. Berk. Modern Mathematical Statistics with Applications, 2007,
Thomson Brooks/Cole. 845 pp. ISBN: 9780534404734.
88. RISDISM. Rhode Island Stormwater Design and installation Standards Manual, 2010,
Rhode Island Department of Environmental Management and Coastal Resources Manage-
ment Council, Horsley Witten Group, Inc., Sandwich, MA with technical services from the
University of New Hampshire Stormwater Center, Durham, NH, and Loon Environmental,
LLC, Riverside, RI.
89. Kim. J., Lee. J., Song. Y., Han.H., Joo. J. Modeling the Runoff Reduction Effect of Low
Impact Development Installations in an Industrial Area, South Korea. Water, 2018, 10, 967;
doi: 10.3390/w10080967.
90. Bai. Y., Li. Y., Zhang. R., Zhao. N., Zeng. X. Comprehensive Performance Evaluation Sys-
tem Based on Environmental and Economic Benefits for Optimal Allocation of LID Facil-
ities. Water, 2019, 11, 341; doi:10.3390/w11020341
78
91. Wang, W. Hydrological effect assessment of Low Impact Development for urbanized area
based on SWMM., Master’s Thesis, 2011, Peking University, Beijing, China.
92. Li, Z.; Qin, H.; Xie, K. Hydrological effect analysis of Low Impact Development under
different rainfall conditions. China Water Wastewater, 2012, 28, 37–41.
93. Sun, J. Stormwater utilization and ecological design: The first phase of the Hannover
Kronsberg Project in Germany. Urban Environ. Des., 2017, 3, 93–96.
94. Qin. H-p.,Li.Z-x., Fu.G., The effects low impact development on urban flooding under dif-
ferent rainfall characteristics. J. Environ. Manag., 2013, 129, 577-585.
95. Chang, T.J. Connection of the LID and urban waterlogging prevention-Taiwan in subtrop-
ical rainy area as an example. In Proceedings of the 14th China Water Forum, Changchun,
Jilin, China, 26–27 August, 2016; Water Resources and Electric Power Press: Beijing,
China, 2016
79
CHAPTER 3
Manuscript III
Performance Evaluation of LID Controls for Runoff Quality Using SWMM
in Suburban Watershed
prepared for the part II submission to the Hydrological Sciences Journal-IAH,
July 2021
Khurshid Jahan1,*, Soni M Pradhanang1,*, and Rebecca Brown2
1Department of Geosciences, University of Rhode Island, RI, Kingston, USA 2Department of Plant Science and Entomology, University of Rhode Island, Kingston,
RI, USA
*Correspondence: [email protected] (K.J.), [email protected]
(S.M.P)
80
Abstract
Suburban growth impacts the surface runoff; therefore, stormwater runoff management
becomes a prime concern for suburban areas due to its impact on quantity and quality.
The study conducted in the suburban watershed to assess the effectiveness of the Low
Impact Development (LID) controls in terms of pollutant reduction. LID practices have
been proposed as a promising suburban management technology to control the runoff
pollutant or mitigate the environmental issues arising from increasing the impervious
cover. The water quality module of LID in the stormwater management model (SWMM)
was used through the RUNOFF block to develop the SWMM-LID model in this study.
A total of 42 individual rainfall events were used in the model simulation. Among them,
30 rainfall events were used for model calibration, and 12 events were used for model
validation. We evaluated the model performance for simulating total suspended
sediments (TSS), Nitrate-N, and Orthophosphate-P of the LID facilities for three
different scenarios i.e., Bioretention, Permeable Pavement, and the vegetative swale.
The selected LID controls occupied 11% of each sub-catchment, and the LID pollutant
removal efficiency was up to 27% to 83% for the TSS, 11% to 28% for Nitrate-N, and
14% to 27% for Orthophosphate-P. Rhode Island stormwater management group
already started their work also in the suburban areas. So, these studies will the associated
authority to implement the LID controls with a proper design and maintenance plan.
Keywords: Suburban growth, Low Impact Development, RUNOFF block, pollutant
removal efficiency, environmental issue, water quality module.
81
1. Introduction
Urban growth disturbs water resources and water quality [1]. Change of natural
hydrological systems due to urbanization can be seen in the form of increased runoff
rate and volume, decreased infiltration, groundwater recharge, baseflow, and water
quality deterioration in streams and rivers [2-7]. Hence, the urban land management to
address and mitigate the water-associated challenges, including the pollutants
generation and urban runoff in the environmental cycle, is a big challenge for urban and
suburban areas [4,7]. The U.S Environmental Protection Agency [8] considers nonpoint
sources as the leading threat to water quality in the Nation through the stormwater runoff.
Therefore, urban management practices such as Low Impact Development (LID)
emerged to control runoff and its pollutants near their source areas [9]. LID refers to
decentralized systems of small-scale treatment units located at or near the source area
[10]. The source areas consist of various urban land-use and land cover types, such as
streets, parking lots, construction, commercial, residential, recreational, and agricultural
landscapes. The impervious surfaces are largely responsible for the contribution of
runoff and pollutants [11]. On the other hand, pervious surfaces such as small sized
lawns, garden may also become essential contributors to pollutant sources during
significant storm events, although many stormwater management models ignore their
contribution [12], for example, SUDS, LTHIA-LID etc.
Stormwater runoff usually contains many different pollutants. Stormwater
monitoring and management programs in the urban catchment are crucial for
characterizing stormwater runoff quality and formulating an effective pollution control
82
strategy. However, stormwater quality monitoring is often too costly, time-consuming,
and laborious. As a result, computer models have become useful tools for simulating
pollutant transport and predicting stormwater pollutants based in available data. The
ultimate goals of urban water quality modeling are to characterize urban runoff, provide
input to the analysis of receiving water, determine the appropriate size of control
structures, and perform frequency analysis of water quality parameters [13]. A range of
climate mitigation measures is available for urban stormwater management. One is to
implement low-impact development (LID) practices [14]. Low impact development
(LID) is a stormwater management approach that mimics a site's natural hydrology as
the landscape is developed. The calculation of predevelopment hydrology is based on
native soil and vegetation. In the past decade, low impact development (LID) practices
have been proposed to deliver better hydrologic and environmental functions [10, 15].
LID technique provides various tools to reduce or control the hydrologic and water
quality effects of urbanization [10,16-18]. Many communities across the United States
are now implementing LID practices to meet stormwater management goals [1, 19, 20
Research groups formulated guidelines for the efficient implementation in different
locations of LID through modeling approaches, which help evaluate the LID
performance with various urban management scenarios [10, 21, 22].
Several research groups have developed numerical models to simulate the LID
process, e.g., Stormwater Management Model (SWMM) [23], System for Urban
Stormwater Treatment and Analysis (SUSTAIN) [24], Western Washington Hydrology
Model (WWHM) [25], and L-THIA Low Impact Development (L-THIA_LID) [26].
83
These models can simulate the effects of LID on the hydrological processes and have
also been widely applied to evaluate pollution reduction in urban areas [10, 27 – 30].
According to USGS 2019 [31] study, LID infrastructure can reduce the effects of
suburban development on streams by storing and infiltrating runoff on the landscape
before it reaches streams. The study found that LID controls in Maryland were
particularly effective at mitigating runoff for less than 20.32 mm rainfall events (1-year,
24-hour event equivalent to 66.04 mm of rainfall) but similar to the 1-inch (25.4 mm)
design criteria for all other stormwater practices.
LID performances differ from one rainfall return period to another, and therefore,
the runoff pollutant that each precipitation event contribute could vary. The main goal
of this paper is to develop a stormwater management model for the suburban catchment
to predict urban stormwater quality based on pollutant buildup and washoff concept. It
demonstrates that physically based models can be developed with various degrees of
complexity in different rainfall-runoff transformations depending on the desired levels
of model performance.
This research study aims to evaluate the effects of LID controls on hydrology
and water quality at a suburban watershed scale using EPA SWMM 5.1. The field data
were collected from 2016 to 2019. In this study, bioretention, permeable pavement, and
vegetated swale, among all types of LID controls, were selected to be employed to
impervious areas to reduce the effect of imperviousness on hydrology, and water quality
at the selected study area. The selected LIDs have the advantage of controlling the
84
stormwater quantity and quality at the source, reducing the flow volume, and thus
delaying the hydrologic response and reducing the pollutant load washed-off from urban
surfaces [31]. In this study, the focus is given in evaluating runoff quality in suburban
landscapes. However, few studies already proved that suburban growth also
significantly impacted quantity and quality in the suburban watershed [33,34]. There
are many benefits of LID controls in the urban and well as suburban catchments.
Previous studies have investigated the impact of different types of LID controls on urban
runoff flooding and water quality. For example, Hunt et al. 2008 [35] found the
bioretention cells can reduce the average peak flows by at least 45% in North Carolina
and Maryland during a series of rainfall events with high intensity. It has also been
proved in Dietz 2007 [36] study that bioretention can reduce sediment and nutrient up
to 99%. Another study carried out on vegetative swales by Ahiablame et al. 2012a [27]
that vegetative swale has an average retention range from 14% to 98% for nutrients and
TSS, and up to 93% for metals. Jackisch and Weiler [37] conducted a study on the
hydrological performance of permeable pavement vegetative swale, and bioretention
that revealed that the selected LIDs could capture 73% of surface runoff and the runoff
volume reduction range from 77 – 87%. Wilson et al. [38] compared the LID
performance in terms of water quality with the conventional development and the
outcome depicted that LID control water quality performance was way better than the
conventional systems. All the results confirmed that LID could replace the conventional
stormwater management systems. Many more similar research investigate the
hydrological and water quality execution of LIDs [39–44].
85
In this study, six different rainfall return periods and three individual scenarios
are analyzed. Six different rainfall return periods includes 1-yr, 2-yr, 5-yr, 10-yr, 50-yr
and 100-yr. The three developed scenarios are a) Bioretention (BR), b) Permeable
Pavement (PP), and c) Vegetative Swale (VS). The runoff quantity study conducted by
Jahan et al (unpublished) reported a better performance of PP and BR than VS in terms
of runoff depth and peak reduction [45], Jahan et al., 2021, unpublished]. Environmental
benefits were also measured for all three LIDs.
Table 1: Percentage of runoff depth and peak reduction per unit area under six rainfall
re-occurrences
LID Measures Runoff Reduction/ha (%) Peak Reduction/ha (%)
1
yr
2
yr
5
yr
10
yr
50
yr
100
yr
1
yr
2
yr
5
yr
10
yr
50
yr
100
yr
Permeable Pavement (PP) 78 65 58 58 45 31 70 61 45 45 32 28
Bio-retention cell (BR) 70 70 55 52 48 35 72 75 50 41 33 24
Vegetative Swale (VS) 60 65 45 45 36 28 55 52 40 40 23 19
Since this study aims at understanding how various rainfall intensities impact
runoff qualities, the various LIDs scenarios will allow us to evaluate how they perform
for higher rainfall intensities. Water quality parameters considered are Nitrate-N,
Orthophosphate-P, and Total suspended Sediments (TSS). The surface runoff quality
data were collected from the installed v-notch weir point at the study area; V-notch weir
mainly installed in the low flow zone to collect the stormwater runoff sample. For our
study site, we installed 300 V-notch weirs having 0.2 feet minimum head in all the six
86
monitoring sites. After collecting samples laboratory analysis was done in the
University of Rhode Island, Hydro-Systems and Water Quality Lab (Pradhanang Lab).
The stormwater data was collected for three years (from August 2016 to July 2019). To
develop the SWMM-LID model, the historical rainfall-flow data were also retrieved
from the on-site automated rain gauge station for model validation. Other necessary data
(drainage plan, conduit length, conduit roughness, installed LID design parameter
especially the BR, PP, and the VS) were collected via site inspection, literature review,
site-specific stormwater management report, and vector and raster maps. Sensitivity
analysis and model calibration were also carried out to identify the model's most
sensitive parameters. The significance of the study was to endorse the idea that LID
application in suburban areas can reduce stormwater pollutants and provide decision
support tool for retrofitting conventional stormwater management systems.
2. Methodology
This methodology explains the detailed process of SWMM-LID model
development for suburban stormwater runoff quality based on the buildup and washoff
concept [46, 47]. It also describes the model accuracy and performance with various
degrees of complexity due to heterogeneities in landscape shown by different water
quantity and quality parameter sensitivities Finally, the performance of the LID controls
are also demonstrated in terms of stormwater pollutant removal.
87
2.1 Study Area
This research was carried out in the Chipuxet watershed located in South
Kingstown, Rhode Island. The selected study area is split into 12 different sub-
catchments (Figure 1). The catchment has one rainfall gauge station (NOAA Station ID
54796), one automated stream gauge flow (USGS 01117350) run by USGS. Stormwater
quality monitoring sites are established near the parking lot, agricultural practice area
within the University of Rhode Island complex (Figure 1). Among Six monitoring sites,
four sites are adjacent to the agricultural land and two other sites are close to URI
parking lot. The conducted area is characterized with 10 to 20% of clay and 50-90% of
loam, silt loam, silt, or sandy clay loam. Soils consisting of clay, silt and sandy loam
have moderately low runoff potential when thoroughly wet and the soil infiltration range
is 0.15 – 0.30 in/hr [33]. The monthly average potential evapotranspiration data is
obtained from Northeast Regional Climate Center [48]. The average monthly potential
evapotranspiration estimated based on the NRCC’s adaptation of the MORECS model.
Stormwater runoff quantity and quality was measured with the samples collected
from the forty-two individual storm events from six different sites with a 600 ml bottle.
The samples were collected from 2016 to 2019. The area is a low elevated region having
a low to moderate slope. The slope of the study site ranged from 7 % to 10 %. The land-
use is considered suburban. The recent study has revealed suburban growth of about
68.7% from 1994 to 2020. Detailed characteristics of the selected research sites are
represented in Table 2.
88
Figure 1: Subdivision of the conducted study area for developed SWMM-LID
Table 2: Input parameters of the sub-catchment for the developed SWMM-LID
Sub_
ID
Area
(sq.k
m)
Width
(km)
P_Im
p
%Slo
pe
N_Imp N_Per D_Store
Imp
D_Store
_Per
Manning
’s N
Sub_1 0.08 0.03 0 7 0.01 0.1 0.02 0.25 0.023
Sub_2 0.03 0.01 0 7 0.01 0.1 0.02 0.25 0.023
Sub_3 0.10 0.05 0.05 8 0.01 0.1 0.02 0.25 0.023
Sub_4 0.03 0.01 0 9 0.01 0.1 0.02 0.25 0.02
Sub_5 0.09 0.04 0 7 0.01 0.1 0.02 0.25 0.12
Sub_6 0.04 0.02 0.038 10 0.01 0.1 0.02 0.25 0.12
Sub_7 0.17 0.07 0.07 10 0.01 0.1 0.02 0.25 0.12
89
Sub_8 0.13 0.06 0.035 8 0.01 0.1 0.02 0.25 0.12
Sub_9 0.08 0.03 0.014 7 0.01 0.1 0.02 0.25 0.12
Sub_10 0.09 0.04 0.01 7 0.01 0.1 0.02 0.25 0.023
Sub_11 0.15 0.07 0.04 9 0.01 0.1 0.02 0.25 0.023
Sub_12 0.16 0.07 0.05 10 0.01 0.1 0.02 0.25 0.12
P_Imp: % imperviousness, N-Imp: Mannings N of impervious area, N_Per: Mannings
N of pervious area, D_Store Imp: Depression storage on impervious area, D_Store_Per:
Depression storage on pervious area, Manning’s N: Manning’s coefficient
2.2 Model DescriptionThe EPA’s SWMM is a single event or continuous
hydrological and water quality simulation model developed principally for urban areas
[49]. The flowchart of the SWMM development for the study site is presented Figure 2.
SWMM consists of four different computational modules or blocks (runoff,
storage/treatment, transport, and extran) that can simulate various hydrological cycle
components such as rainfall, snow, interception, depression storage, infiltration, runoff
etc. The RUNOFF block was applied in this study to predict stormwater runoff at the
single rainfall events time scale.
Hydrological computations in the RUNOFF block are based on the theory of
nonlinear reservoirs. Time of concentration is computed based on the kinematic wave
theory. Two pervious area infiltration loss equations are available, the Horton and
Green-Ampt. The Green-Ampt equation (Equation 1) was used here. It has the
advantage over the Horton equation of using physically based parameters that can be
determined a priori [49]. These are the average capillary suction head, Su (mm), at the
90
wetting front, the initial moisture deficit (IMD) in mm/mm, and the saturated hydraulic
conductivity, Ks of the soil (mm/h). Therefore, this semi-theoretical equation was used
instead of the Horton equation because of the availability of relevant soil data from
pervious USGS studies and the local Natural Resources Conservation Service (U.S.
Department of Agriculture). The modified Green-Ampt equation proposed by Mein and
Larson (1973) [50] was used to estimate f1 (Rossman and Huber, 2016) [51]:
f1 = Ks(1+(𝜑− Ɵ1)(𝑑1+𝜓)
𝐹) ………………………………………………………….(1)
where f1 is the infiltration flux [cm/min], Ks is the saturated hydraulic conductivity of
the soil in LID (mm/min), φ is the soil porosity [-], Ɵ1 is the moisture content of the soil
in LID [-], 𝜓 is the suction head at the infiltration wetting front [mm], d1 is the ponding
depth on the surface [mm], and F is the cumulative infiltration volume per unit area
[mm]. Mein and Larson’s modified Green-Ampt equation suggests that the infiltration
rate equals rainfall intensity when the soil has not reached its saturation point [52].
The model simulates the rate of soil percolation (f2) in the soil layer through the
soil into the storage layer. The soil percolation can be a calculation based on Darcy’s
law which followed the same method employed in the groundwater module of the
SWMM [53]:
f2 = Ksexp (-HCO(φ-Ɵ2)), Ɵ2 > ƟFC ………………………………………………………………………….(2)
f2 = 0, Ɵ2 ≤ ƟFC
where, f2 is the soil percolation rate [cm/min], HCO is the decay constant derived from
the moisture retention curve [-], and θFC is the field capacity moisture content of the soil
[-].
91
The exfiltration rate (f3) from the storage zone into the native soil was estimated
for the storage layer. The following equation was used to calculate f3 [53]:
f3 = f2 – φ2𝜕𝑑2
𝜕𝑡, f3<K3 ……………………………………………………………………………………………………(3)
f3 = K3, f3 ≥ K3 …………………………………….………………………………………………… …………………..(4)
where, f3 is the exfiltration rate [mm/min], φ2 is the void fraction of the storage layer [-],
d2 is the water depth in the storage layer [mm], and K3 is the saturated hydraulic
conductivity of the native soil [mm/min].
Surface runoff is computed in SWMM for individual rainfall events, considering
land-use types and topography, accounting for antecedent moisture conditions,
infiltration losses in pervious areas, surface detention, overland flow, channel/pipe flow,
and constituents carried by runoff into inlets. Important input parameters are then the
catchment slope, pervious and impervious depression storage, channel and conduit
layout, geometry and properties, the Manning’s roughness coefficients for both
overland and channel flow, and rainfall intensity, among others.
The RUNOFF block can also simulate the quality of the runoff process within a
drainage basin and the routing of flows and contaminants along storm drain lines,
leading to calculating several inlet hydrographs and pollutographs. According to Huber
and Dickinson 1988 [49], there are several options for computing accumulation and
wash-off of pollutants in SWMM. The following a power-build-up equation was used
in this study:
Lt = 𝑄𝐹𝐴𝐶𝑇(3). 𝐷𝐷𝐹𝐴𝐶𝑇(2) ………..…………………………………………………………………………….(5)
92
where Lt is the pollutant concentration at each rainfall event t(mg/l); DD is the preceding
dry weather period (days), QFACT(2) is the power or exponent for the buildup
parameter (dimensionless); and QFACT(3) is the coefficient for the buildup parameter
(mg/l.day-QFACT(2)). Wash-off is a method of erosion and transportation of pollutants
from a catchment surface during runoff [53]. The following relationship describes the
relationship between wash-off (Mt) at each time step and the runoff rate Q.
Mt = RCOEF. Qt WASHPO …………………………………………………………………………………………….(6)
Where WASHPO is the exponent of the wash-off parameter (dimensionless); and
RCOEF (conc/event [mg/l/day]-WASHPO ) is the coefficient for the wash-off parameter.
Equations (5) and (6) are commonly used equations in SWMM-LID quality analysis.
Other equations also exist those are most complex. Based on Hubor and Dickinson
(1988) [45]. Equations (5) and (6) are the easiest to use when total runoff volumes and
pollutant concentration are accessible. For the selected study area three land types are
dominant and three pollutants (Nitrate-N; Total Suspended Sediments, TSS; and
Orthophosphate-P) are considered for developed model. The following Wash-off
parameters of pollutants (Table 3) are considered for this study.
Table 3: Buildup and Wash off and parameters of pollutants on different land-use types
Area Parameter Nitrate-N Orthophosphate-
P
TSS
Commercial area
(includes parking
lots, impervious
areas)
Accumulation rate coefficient
(1/event)
wash off coefficient
wash off exponent
0.23
0.004
1.8
0.15
0.004
1.7
0.3
0.004
1.3
93
Agricultural/green
area
Accumulation rate coefficient
(1/event)
wash off coefficient
wash off exponent
0.3
0.002
1.4
0.18
0.001
1.2
0.34
0.003
1.4
Roads Accumulation rate coefficient
(1/event)
wash off coefficient
wash off exponent
0.23
0.004
1.8
0.15
0.004
1.7
0.3
0.004
1.3
Figure 2: Flowchart of applied SWMM-LID framework
2.3 Data collection, source, and preparation
2.3.1 GIS data
GIS application is used in this study for the primary data preparation of the
SWMM-LID model. Watershed delineation, stream order generation, sub-catchment
characteristics (width, slope, %impervious, and % pervious) are the primary input
parameters for SWMM 5.1. The calculated physical parameter of the sub-catchments
was validated with USGS Streamstats database for the study area. The USDA National
94
Resources Conservation Service Soil Survey Geographic Database (SSURGO) was
used to determine the soils [54]. It contains the most detailed county-level data (Soil
Survey Staff, 2015), which is likely to yield better model results than the other available
soil database [55,56] and has been used extensively in hydrologic modeling studies [57
–60]. The developed GIS shapefile and its attributes were exported into inout file (.inp)
through inpPINs application. Finally, the .inp file was imported into the SWMM for
further analysis. Data sources are representation in Table 4.
Table 4: Data source of the required parameter of SWMM-LID
NO. Parameter Meaning Data Range Data Source
1 Manning-N Manning Coefficient 0.005 – 0.04 SWMM manual
2 N-Imperv Mannings N of impervious area 0.005 – 0.04 RIDSM report,
SWMM manual
3 N-Perve Mannings N of pervious area 0.1 – 0.8 RIDSM report,
SWMM manual
4 S-Imperv Depression storage on impervi-
ous area
0.2 ~ 2 (mm) RIDSM report,
SWMM manual
5 S-Perv Depression storage on pervious
area
2 ~ 10 (mm) RIDSM report,
SWMM manual
6 Pct-Zero Percent of impervious area with
no depression storage
50 ~ 80% RIDSM report, Cross
check with GIS
7 MaxRate Maximum rate on infiltration
curve
3 ~ 50 (mm/h) SWMM manual
8 MinRate Minimum rate on infiltration
curve
1~3 (mm/h) SWMM manual
95
9 Imperv (%) Percent of impervious area 10~90 (%) Google Earth, cross
check with GIS
10 Width Width of overland flow path Depends on the
area of sub-
catchment
GIS
11 Slope Average surface slope 3-5 (%) GIS
2.3.2 Sample collection and analyses
Total forty-two individual rainfall events were monitored for the study duration
(2016-2019). Due to the low variation of elevation of the study sites, flow could not be
effectively monitored. For the calibration of the streamflow, we used the USGS gauge
station (USGS 01117350). Water quality is a sample based on the rainfall amount,
which is greater than 12.7 mm or 0.5 inches. Grab samples (1 Lt each) were collected
from six individual sites for 42 rainfall events (Total 6*42) for the entire duration of the
research. Samples were analyzed for the following parameters and used to characterize
concentration:
- Total Suspended Solid (TSS)
- Nitrate-N (NO3-N+NO2-N)
- Orthophosphate-P (PO4-P)
- Turbidity
Stormwater runoff samples were also analyzed for Polycyclic Aromatic
Hydrocarbon (PAH) using Shimadzu GCMS following EPA 8270d method. However,
the analyses showed that all the samples had very low to negligible PAHs (the below
detection limit (BDL) concentration (<0.5 mg/l). The study area is a suburban area, and
96
gradually the suburban growth is expanding [33]. Therefore, for this developed
SWMM-LID model, we only considered TSS, -Nitrate-N, and Orthophosphate-P as
main water quality pollutants for this study.
2.4 Scenarios of the developed model
2.1.4 LID Selection and Developed Scenarios
SWMM 5.1 allows to incorporate modules for LID components including
bioretention (BR), permeable pavement (PP), and vegetative swales (VS). In this study,
these three LID components are evaluated, and. the characteristics of the three selected
LID controls are represented in Table 4. The comprehensive effectiveness of different
LID structures on stormwater pollutant was measured for BR, PP, and VS. The
developed SWMM-LID model itself represented the vegetative swale (base scenario),
two more scenarios (bioretention, and permeable pavement) were developed in this
study. Scenarios were developed, considering storm design events (1-yr, 2-yr, 5-yr, 10-
yr, 50-yr, and 100-yr) and the efficiencies were measured The LID module of SWMM
(Table 5), BR, PP, and VS are represented for three layers (surface, pavement, and
storage layers), two (surface and storage), and one (surface) vertical layers respectively.
The PP and BR LIDS have an optional layer, i.e., underdrain system.
97
Table 5: Parameters of three LID controls used in the LID-SWMM Simulations
LID Controls System Components Parameter Value
Permeable Pavement Surface Layer Storage depth
Vegetative cover fraction
Surface roughness
Surface slope (%)
2
0
0.12
<9%
Pavement Layer Thickness (cm)
Void ratio
Impervious surface fraction
Permeability
18-25
0.65
0.038
2.3
Storage Layer Height (cm)
Void ratio
Filtration rate (2cm/hr)
50.8
0.65
2
Underdrain system Drain coefficient
Drain exponent
Drain offset height
3.8
0.5
3.1
Native soil Suction head
Conductivity
Initial deficit
n/a
2
n/a
Other Area (sq.km)
Width (km)
E.T. rate (simulation start from No-
vember)
0.04
0.02
0.71
Bioretention Cell Surface Berm height (cm)
Vegetative Volume Fraction
Surface Roughness (Manning’s n)
Surface Slope (%)
90
0.12
8
Soil Thickness (cm)
Porosity
Field Capacity
31
0.15
35
98
Wilting point
Conductivity (cm/hr)
Conductivity slope
Suction head (cm)
0.2
3.6
6
20.32
Storage Thickness (cm)
Void ratio
25
67
Drain Flow coefficient
Flow exponent
Offset height (cm)
4
0.5
10.1
Vegetative Surface Layer Berm Height (cm)
Vegetative Volume Fraction
Surface Roughness
Surface Slope
Swale side slope
18
100
0.24
7% - 10%
10%
The model simulation results were calibrated (first 30 rainfall events) and
validated (rest of the 12 rainfall events) against the measured data. Measured TSS,
Orthophosphate-P, and Nitrate-N concentrations for the current study were derived from
42 event-scale from six different sites. Other onsite water quality data for different land
cover types were not available
2.5. Model performance evaluation
2.5.1. Sensitivity analysis
Sensitivity analysis was performed to identify which parameters would be most
effective in minimizing differences between observed and predicted results [61].
Parameters were adjusted over a range of 50% of their original value while keeping all
99
other parameters unchanged, and the corresponding difference in runoff volume and
peak flow was calculated. Relative sensitivity was computed using equation (7):
Sensitivity = (𝜕𝑅
𝜕𝑃)(
𝑃
𝑅) ……………………………………………………………………………………………...(7)
where 𝛛R is the difference between the model input parameter and simulated parameter
value, 𝛛P is the difference between input and adjusted parameter value, R is the
simulated model output, and P is the input value of the parameter of interest [62].
2.5.2 Calibration and Verification of SWMM Models
SWMM 5.1 was developed with forty-two individual rainfall and calibrated with
entire sets of data from 2016 and 2019. Two individual calibration runs were obtained
with the daily discharge data from USGS 01117350 Chipuxet river at West Kingston,
RI. Calibration was conducted for two different model set up (Run 1 and Run 2) in two
different sub-catchments (sub-catchment 3 and 6) having two different types of land
types (agricultural and parking lot) and where the monitoring sites are located. For water
quality parameters calibration, we considered TSS, Nitrate-N, Orthophosphate-P for site
1 (site_1 is in the sub-catchment 3) and site 5 (Site 5 is in the sub-catchment 6). Among
42 individual events, 30 rainfall events were considered for the model calibration and
the rest of 12 rainfall events were applied for model validation. The calibration of
stormwater quality models usually is more complicated than quantity models due to the
build-up and wash-off methods process, sewer sediment transportation, pollutant
interactions, especially in the big urban watershed area [63, 64]. Also, the stormwater
quality is challenging to calibrate without sufficient on-site data [63, 64]. Due to
geographical characteristics (low relief and almost no variation in slope) of the study
100
area, the flow meter couldn’t collect enough samples. Therefore, manual depth
measurements were done for each storm. The Kingston Campus Master Plan, 2018 [65]
and Stormwater management plan 2020 [66] for the South Kingstown was used to
retrieve the required supplementary data i.e., detail drainage design, land-use and land
cover, location basis detail soil profile and texture, detail bioretention, permeable
pavement, and vegetative swale design, and location-based soil information. These data
were used for model input parameterization. The adjusted calibration parameter is
presented in Table 7.
The model performance on runoff peak and runoff quality simulation were tested
using various performance statistics. The optimal parameter values for the study area
were selected according to the Nash-Sutcliffe Efficiency (NSE) [67].
NSE = 1 − ∑ (Oi−𝑃𝑖)2n
i=1
∑ (𝑂i−O′)2ni=1
……………………………..………………………………(8)
The coefficient of determination, R2, an established measure in statistics quantifying the
model’s fit, is described as:
R2 = [ ∑ (𝑂𝑖
ni=1 −𝑂′)(𝑃𝑖−𝑃′)
√∑ (𝑂𝑖−𝑂′)2√∑ (𝑃𝑖−𝑃′)2ni=1
𝑛𝑖=1
] 0 ≤ R2 ≤ 1 ……………………………………..(9)
where n is the number of observations in the period under consideration, Oi is the i-th
observed value, 𝑂′ is the mean observed value, 𝑃𝑖 is the i-th model-predicted value, and
𝑃′ is the mean model-predicted value. NSE ranges between −∞ and 1, where 1 is
considered a perfect fit for any model where NSE is sensitive to extremes [68].
Dongquan et al. 2009 [69] suggest that an NSE value of 0.5 or greater is acceptable for
SWMM simulation. The denominator is the total variance in the observed data, and the
101
numerator is the variance between the simulated and observed data. The closer R2 is to
1, the larger proportions of the observed variability can be explained with the
simulations of estimated values [70].
To evaluate the developed model for the goodness of fit test root mean square
error (RMSE), the RMSE-observations standard deviation ratio (RSR) is calculated as
the RMSE and standard deviation ratio of measured data. The RMSE and RSR are
explained as follows
RMSE = √1
𝑛 ∑ (𝑃𝑖 − 𝑂𝑖)2𝑛
𝑖=1 ……………………………………….…………….. (10)
RSR = 𝑅𝑀𝑆𝐸
𝑆𝑇𝐷𝐸𝑉𝑜𝑏𝑠 …………………………………………………………………… (11)
where n is the number of observations in the period under consideration, 𝑃𝑖is the i-th
model-predicted value and 𝑂𝑖 is the i-th observed value, 𝑆𝑇𝐷𝐸𝑉𝑜𝑏𝑠 is the standard
deviation of the observed value. The ideal value for RSR is 0; however, it cannot be
expected to occur; otherwise, it would be a perfect model. So, values between 0 and 1
are acceptable for RSR when field-specific data are available for calibration. RSR varies
from the optimal value of 0 to a large positive value. The lower RSR, the lower the
RMSE, and the better the model simulation performance [71].
3. Result and Discussion
3.1 Model Performance Evaluation
Model performance evaluation is divided into two sections: (i) sensitivity
analysis and (ii) model calibration and validation.
102
3.1.1 Parameters sensitivity analysis
The sensitivity analysis showed that seven parameters namely rainfall, sub-
catchment area, percent imperviousness, percent slope, conduit roughness, conduit
length, and storage depth were highly sensitive to peak runoff and water quality
predictions. These seven parameters were used for the calibration of the SWMM-LID
model.
a) Sensitivity Analysis for Hydrological Parameters
Figure 3(a) presents the sensitivity analysis for the simulation of the hydrology
module in LID. Rainfall response showed a strong positive relation with the amount of
runoff generated. Rainfall always plays a dominant role in soil loss and sediment
delivery [72], affecting runoff. The analysis showed the relatively weak or low but
positive response of percent imperviousness on the model outcome. The infiltration
capacity decreased with the increase of the impervious surfaces. Figure 3a also indicates
that, as the rainfall receiving area increases, the output runoff amount from sub-
catchment increases almost proportionally. Drainage and rainfall amounts are the most
important variables to predict event loads associated with runoff volume and is reported
by another research [73, 74]. SWMM model output is also sensitive to the percent slope,
but the geographical characteristics of the study area have relatively low or no variation
of slope and elevation. According to El-Hassanin et al., (1993) [74] runoff-rainfall ratios
are high under the steep slopes and soil loss per unit of rainfall and per unit of runoff
increase as the slope gradient increase. This study shows similar model results. The
model showed high sensitivity to the change in slope showing increase in the runoff
with increase in slope; however, percent imperviousness sensitivity increased was
103
relatively lower. Since these parameters are related to the surface layer (Eq. 1), it can be
inferred that surface layer parameter are more sensitive than those related to the soil
layer (Eq. 2).
The peak runoff is directly proportional to the changes in imperviousness,
whereas it is inversely proportional to the D-Store impervious-pervious indicating that
the peak runoff will decrease with the reduction of percent imperviousness, and
depression storage. It should also be noted that in most cases the peak runoff is more
sensitive to the reduction of the D-Store impervious-pervious than the increase. In other
words, the peak runoff will only decrease 5% with the 30% increase in the D-Store
impervious-pervious, whereas it will increase 12% with the 30% reduction of D-Store
impervious-pervious. But in this study, the sensitivity analysis showed very minor
response (< 5% change) as the depression storage is less sensitive to the low slope areas.
Therefore, D-store impervious-pervious parameter was excluded from the model input
parameterization (Figure 3a).
(b) Sensitivity of Hydraulic Parameters
Figure 3(b) shows sensitivity of conduit roughness, conduit length, and the
storage depth to the runoff responses. The conduit roughness parameter of SWMM
stands for the Manning’s coefficient for the natural channel. Both conduit roughness
and conduit length respond inversely to total inflow volume. However, the peak runoff
showed a positive linear relationship with storage depth of the impervious. The positive
104
or negative responses of storage depth are largely dependent on the sub-catchment
characteristics.
Figure 3: a) Model runoff results of sub-catchments for hydrological parameters; b)
Model runoff results of the outfall of the catchment concerning hydraulic parameters.
[In SWMM, time patterns allow external dry weather flow to vary periodically. They
comprise of a set of adjustment factors applied as multiplier (SWMM 5.1 manual]
(c) Sensitivity of Buildup and Wash-off Parameters
Buildup and wash-off parameters are considered to complete the sensitivity
analysis for runoff pollutants (TSS, Nitrate-N, and Orthophosphate-P). The considered
buildup and wash-off parameters are accumulation rate coefficient (k), pollutant wash-
off coefficient (E1), and Pollutant wash-off exponent, E2. The initial value of the
parameter is mentioned in the Table 3. This analysis showed the %change in simulation
concentration due to the % change of buildup and wash-off parameters. Here, we range
of the parameter change is ±20%. The sensitivity analysis was performed by changing
each parameter while keeping other parameters constant. As shown in the Figure 4, it is
105
clearly observed that the pollutant wash-off exponent (E2) is the most sensitive
parameter for all three pollutants than other two considered sensitive parameters (k and
E1).
Figure 4: Buildup and Wash-off parameter sensitivity analysis a) TSS; b) Nitrate-N; and
(c) Orthophosphate-P
The results from a sensitivity analysis by Huang [75] and Wang [76] and the
sensitivity ranking of the hydraulic and hydrological parameters in SWMM are listed in
Table 6. The summary and the ranking of the considered seven sensitive parameters
presented in the Table 6 also showed the response the sensitive parameters with runoff.
106
Table 6: Model sensitive Parameter and their ranking based on the sensitivity analysis
Parameter Sensitivity Class Correlation Sensitivity Rank
Rainfall Very high Proportional 1
Sub-catchment Area Very high Proportional 2
Conduit Roughness High Inverse 4
Conduit Length Moderate Inverse 5
Storage Depth Moderate Proportional 5
% Impervious Low Proportional 6
% Slope Moderate Proportional 6
Pollutant wash-off coefficient Moderate Proportional 5
Pollutant wash-off exponent High Proportional 3
Accumulation rate coefficient Moderate Inverse 4
3.1.2 Model Calibration and Validation
SWMM-LID is developed with forty-two individual rainfall events. Thirty
individual events were considered for the model calibration, and the remaining twelve
rainfall events were used for the model validation. The Figure 5 (a, b) showed the
observed and simulated flow rates, and the adjusted values are represented in Table 7.
To calibrate flow, soil conductivity was lowered (Table 7) and higher percent
imperviousness was applied. The simulation run began in November and initial soil
moisture was considered nonzero, i.e., the model default of 0.001. According to the
SWMM Manual, soil moisture deficit value is a required parameter of flow simulation
using the Green Ampt Method. For the wet month, the soil moisture deficit value is
0.001. But if the simulation start period is a dry month, the model requires soil moisture
deficit to be calculated based on the rainfall, evapotranspiration, and soil types using an
107
in-built equation. Considering the simulation start period, the suction head was lowered
for the LID sub-catchment for better model performance. Higher soil suction would be
associated with lower soil moisture content [77]. To improve the relationship between
simulated runoff with observed runoff, Manning’s n for the LID control was lowered
from its initial value.
Table 7: Adjusted SWMM calibration surface layer and soil parameters
Parameter Adjustment Initial Value Calibrated value
(Run 1)
Calibrated value
(Run 2)
% Impervious x 1.0 X 0.23 X 0.25
Depression Storage (mm) 2.032 1.524 2.2
Sub-catchment Width 1 1 1
Impervious "n" 0.02 0.025 0.03
Pipe "n" 0.013 0.011 0.012
Soil Conductivity (mm/h) 13.97 5.7 4.9
Manning’s n for swale 0.24 0.15 0.14
Suction head (cm) 20.32 18.2 17.5
DStore-Perv 0.25 0.22 0.2
DStore-Imperv 0.02 0.017 0.018
The calibrated output showed that the model overestimates the flow for a few
events in both runs. Nevertheless, it is obvious from Figure 4 that, overall, the predicted
and observed hydrographs showed a good match according to the NSE, R2, and RSR
value (Table 8) for both runs [69-71].
108
Table 8: Model Calibration and validation output
Model
Evaluation
Calibration Validation TSS Calibration Nitrate-N Cali-
bration
Orthophosphate-
P Calibration
Run
1
Run
2
Run
1
Run 2 Site_1 Site_5 Site_1 Site_5 Site_1 Site_5
NSE 0.73 0.69 0.77 0.70 0.73 0.77 0.78 0.68 0.71 0.68
R2 0.75 0.64 0.84 0.7991 0.87 0.82 0.93 0.92 0.90 0.88
RSR 0.53 0.61 0.42 0.51 0.58 0.52 0.41 0.55 0.62 0.57
Figure 5: Calibration and Validation of the developed SWMM (a) Run 1 in Sub-
Catchment 3 and (b) Run 2 in Sub-catchment 6
Figures 6, 7 and 8 represent the calibration for TSS, Nitrate-N and
Orthophosphate-P. The output of the calibration is summarized in Table 8.
109
Figure 6: The observed and predicted TSS concentration (a) Site 1 (Agricultural Land)
(b) Site 5 (Parking Lot)
110
Figure 7: The observed and predicted Nitrate-N concentration (a) Site 1 (Agricultural
Land) (b) Site 5 (Parking Lot)
Figure 8: The observed and predicted Orthophosphate-P concentration (a) Site 1
(Agricultural Land) (b) Site 5 (Parking Lot).
111
3.2 LID implementation on stormwater quality
Once the model calibration and validations were completed for the study sites, the four
LID scenarios were tested using the model. Then, the model was run for the Nitrate-
NOrthophosphate-P1-yr, 2-yr, 5-yr, 10-yr, 50-yr, and 100-yr rainfall return periods. The
results of the model with the applied LIDs are shown in figures 9, 10, and 11.
According to the Kingston Campus Drainage Master Plan, 2018” [60] which is
prepared for the university of Rhode Island, the structural controls are generally required
to achieve 85% removal of total suspended solids (TSS), 30% removal of
Orthophosphate-P for discharges to freshwater systems, and 30% removal of Nitrate-N
for discharges. Based on this statement, the target was fixed for the pollutant
concentration reduction for each of the selected structural controls.
3.2.1 Comparison of stormwater management scenarios
The simulated stormwater management scenarios were compared with each other and
with the original model without and with LIDs. The LID structures retain stormwater at
the source and reduce the considered pollutant concentration. The total runoff and
pollutant concentration were calculated for each scenario and found that all applied LID
controls can reduce the pollutant concentration. The scenarios reduced the different
pollutants to varying extent, depending on the specific LID location and pollutant. The
target level was fixed for the three different pollutants.
(a) Bioretention
These results (Figure 9) indicated that in a suburban environment condition, bi-
oretention systems can reduce concentrations of all target pollutants (TSS, Nitrate-N,
and Orthophosphate-P). Additionally, bioretention can effectively reduce peak flow and
112
runoff volume from six different rainfall return period (1 yr, 2-yr, 5-yr, 10-yr, 50-yr,
and 100-yr) that was examined and described in the third chapter. The Figure 8 shows
the reduction of TSS, Nitrate-N, and Orthophosphate-P after implementing bioretention
controls. The reduction rates were measured for all six different rainfall return periods
and sites. The analysis showed that the TSS reduction rate met the satisfactory level for
1-yr (64% – 81%), 2-yr (60% - 75%), and 5-yr (52% – 68%) rainfall return period.
However, after that, the reduction level dropped (63% – 72%) compared to 1yr return
period.
113
Figure 9: Performance of Bioretention for six different sites (a) TSS, (b) Nitrate-N, and
(c) Orthophosphate-P
On the other hand, Nitrate-N bioretention showed better performance until 10 yr
(10% - 15%) rainfall return period following the target level 30%. The analysis showed
that the NITRATE-N reduction rate met the satisfactory level for 1 yr (19% – 28%), 2
yr (18% - 23%), and 5 yr (12% – 20%) rainfall return period. For the 50 yrs. rainfall
return period, the reduction level dropped from 53% to 90% compared to 1yr return
period.
The model output showed the low performance in terms of pollutant reduction
over time. Based on the Stormwater plan report, all the structural controls need mainte-
nance within a specific time frame. For bioretention, it requires operational maintenance
annually for this study area [61].
(b) Permeable Pavement
As shown in the figure 10, the reductions of TSS, NITRATE-N, and Orthophosphate-P
after implementation of permeable pavement were consistent up to 5-yr rainfall return
design. Total suspended sediments were significantly reduced by the permeable
pavement system for 1-yr (65% – 80%), 2-yr (61% - 70%), and 5-yr (53% – 67%)
114
rainfall return period. For the return period of 50-yrs and 100-yrs rainfall design the
reduction rate was 40% and 25% respectively. The accumulation of suspended
sediments in this system can lead to clogging [32], which can make the reduction rate
lower over the time.
Figure 10: Performance of Permeable Pavement for six different sites (a) TSS, (b)
Nitrate-N, and (c) Orthophosphate-P
115
Nitrate-N is highly mobile in soils and groundwater, and it is difficult to remove
from the system. The nitrate-N concentrations usually increase through PP when they
are normally drained for example, underdrain at the bottom of the cross-section, while
aerobic environments are dominant. In this study, the PP scenario showed higher
reduction rate of 22%n for the 5-yrs rainfall return period. However, the performance
rate declined for 10-yr, 50-yr, and 100-yr rainfall designs (average reduction rate <8%).
For the Orthophosphate-P, PP showed a better performance compared to the
nitrate-N having an average removal efficiency 22% for the 1-yr, 16% for 2-yr, 22% for
5-yr rainfall design from all the sites. However, the removal efficiency did not improve
for the higher intensity rainfall (for 50-yr and 100-yr rainfall design it showed the about
3 – 5% Orthophosphate-P removal efficiency).
(c) Vegetative Swale (VS)
In general, according to literature (Barret et al., 1993, and GKY and Associates
Inc., 1991), a well-designed, well-maintained swale system can remove up to 70% of
total suspended solids, 30% of total phosphorus, and 25% of Nitrate-N. In this study
similar output for thesuburban sites was found (Figure 11). Vegetative swales were
found to be very effective in reducing TSS concentration with the average upto 70% for
1-yr, 2-yr, and 5-yr rainfall design indicating that VS is capable to reduce TSS for 120
mm of rainfall depth. For 10-yr design rainfall it’s average removal rate was 47%.
116
Figure 11: Performance of vegetative swale for TSS, Nitrate-N, and Orthophosphate-P
removal
The analysis showed a consistent removal efficiency of VS for Nitrate-N for 2-
yr and 5-yr rainfall design whereas it performed maximum for 1-yr rainfall design for
all the sites. Effectivity of removing Nitrate-N for high intensity and longer period of
rainfall is lower (<4%) than any other LID controls.
117
For the Orthophosphate-P, VS showed not satisfactory efficiency of NITRATE-
N. Study showed an average removal efficiency 20% for the 1-yr, 13.5% for 2-yr, 8.3%
for 5-yr rainfall design from all the sites. However, the removal efficiency is not
significant when it gets higher intensity of rainfall, for 50-yr (<3%) and 100-yr (0.5%)
rainfall design.
4. Conclusion
Implementing BMPs and LID practices can reduce the adverse impacts of urban
growth on hydrology and water quality. However, this study emphasized modeling
stormwater quality impacts of BMPs and LIDs practices in suburban areas where
surface runoff has started to get disturbed through suburban growth. This Study
conducted to assess LID performance on runoff pollutant control reported that LID
performs significantly differently in different rainfall return with different intensities
US EPA SWMM 5.1.010 is applied to develop the model, and the 18 km2 suburban
catchment is divided into 12 sub-catchments. The developed model was calibrated and
validated both for quantity and quality using real-time data. A detailed sensitivity
analysis was also performed beforehand to find out the most sensitive parameter of the
developed model. The model was to simulate LID techniques at the sub-catchment scale
by applying bioretention, permeable pavement, and vegetative swale as an efficient
structural control for the study area. Each of the LID controls' performance was
measured for six different rainfall design that was derived from depth duration
frequency (DDF) curves estimated by NOAA for the study area.
118
The following assumptions could be made based on the model performance and the
results attained
• The sensitive parameters of the model are precipitation, sub-catchment area, %
imperviousness, slope, conduit roughness, conduit length, and D-Store impervious-per-
vious. A small change in the area, the depression storage will significantly change the
simulated runoff and the peak flow. %imperviousness, slope showed slight impact on
the runoff. However, the model is more sensitive to precipitation, area, and D-Store
impervious-pervious rather than the other four parameters. Also, the peak runoff would
be more affected when D-Store impervious-pervious decreased rather than increased. It
means that storage is inversely related to runoff.
• A detail calibration and validation have done for hydraulic (Run 1 and Run 2)
and water quality (TSS and Nitrate-N) module. A well performed calibration output
were derived having 0.73 and 0.69 NSE value and 0.61 and 0.53 RSR value the calibra-
tion. NSE value of greater than 0.5 and lower RSR value are acceptable for SWMM.
Validation output is more significant as it has the RSR <0.6 (Run 1 is 0.42 and Run 2 is
0.51). on the hand, TSS and Nitrate-N calibration also showed significance perfor-
mance. TSS and Nitrate-N calibration was done for two different monitoring sites
(Site_1 and Site_5 from two different land-use). Nitrate-N calibration performed very
well in terms of NSE (0.78), R2 (0.93) and RSR (0.41) value especially for the Site_1.
• In terms of water quality, performances are evaluated for bioretention, permea-
ble pavement, and vegetative swale in terms of TSS, Nitrate-N, and Orthophosphate-P
removal from the source. The LID removal efficiency reached up to 26% for Nitrate-N
and up to 81% for TSS and up to 27% for Orthophosphate-P for 1-yr rainfall design.
119
The LID removal efficiency for three different controls of the model was independent
of the six different (1-yr, 2-yr, 5-yr, 10-yr, 50-yr, and 100-yr) rainfall intensity and du-
ration. Overall, the developed model performed acceptably for rainfall of up to 120 mm
(10 years rainfall design). Nonetheless, for the return period of more than 10-yrs and for
the rainfall amount of more than 120 mm, the assigned LIDs for the sub-catchment
cannot manage properly.
Finally, lower reduction efficiency showed for the return period for 50-yr and
100-yr for TSS, Nitrate-N, and Orthophosphate-P for all three designed LIDs. We need
more LID to be installed to handle the situation or minimize the impact in this suburban
area.
Reference
1. USEPA (U.S. Environmental Protection Agency). Our built and natural environments:
a technical review of the interactions between land use, transportation, and environmen-
tal quality, 2001, P.4.
2. Davis, A.P., McCuen, R.H. Stormwater Management for Smart Growth. 2005,
Springer, New York, USA.
3. Kayhanian, M., Fruchtman, B.D., Gulliver, J.S., Montanaro, C., Ranieri, E., Wuertz, S.
Review of highway runoff characteristics: comparative analysis and universal implica-
tions. Water Res., 2012, 46 (20), 6609–6624.
4. Barbosa, A.E., Fernandes, J.N., David, L.M. Key issues for sustainable urban storm-
water management. Water Res., 2012, 46 (20), 6787–6798.
5. Harbor, J. A practical method for estimating the impact of land-use change on surface
runoff, groundwater recharge, and wetland hydrology, Journal of the American Plan-
ning Association, 1994, 60(1), 95-108
6. Moscrip, A.L., & Montgomery, D.R. Urbanization, flood frequency and salmon abun-
dance in Puget Lowland streams. Journal of the American Water Resources Associa-
tion, 1997, 33(6), 1289-1297.
7. Eshtawi, T., Evers, M., Tischbein, B., Diekkrüger, B. Integrated hydrologic modeling
as a key for sustainable urban water resources planning. Water Res., 2016, 101, 411–
428.
120
8. USEPA (US Environmental Protection Agency). Cost of urban stormwater control,
2002, Washington, DC: USEPA.
9. Simonovic, S. P., and Peck, A. Updated rainfall intensity duration frequency curves for
the City of London under the changing climate. Water Resources Research Rep. Book
29., 2009, Dept. of Civil and Environmental Engineering, The Univ. of Western On-
tario, London, ON, Canada
10. Elliott, A.H., Trowsdale, S.A. A review of models for low impact urban stormwater
drainage. Environ. Model. Softw, 2007, 22 (3), 394–405. https://doi.org/10.1016/j.en-
vsoft. 2005.12.005.
11. Lee, J.G. and Heaney, J.P. Estimation of urban imperviousness and its impacts on storm
water systems. Journal of Water Resources Planning and Management, 2003, 129 (5),
419–426. doi:10.1061/(ASCE) 0733-9496(2003)129:5(419)
12. Pitt, R., Williamson, D., Voorhees, J., Clark, S. Review of historical street dust and dirt
accumulation and washoff data. In: James, W., Irvine, K., McBean, E., Pitt, R. (Eds.),
Effective Modeling of Urban Water Systems – Monograph 13, 2005, CHI, Toronto,
Ontario, Canada.
13. Huber WC. Deterministic modeling of urban runoff quality. In: Torno HC, Marsalek J,
Desbordes M (eds) NATO advanced research workshop on urban runoff pollution, se-
ries G: ecological sciences, vol 10, 1986, Springer, New York.
14. Pyke, C., et al. “Assessment of low impact development for managing stormwater with
changing precipitation due to climate change, Landscape Urban Plann. 2011, 103(2),
166–173.
15. Hunt,W.; Smith, J.; Jadlocki, S.; Hathaway, J.; Eubanks, P. Pollutant removal and peak
flow mitigation by a bioretention cell in urban Charlotte, NC. J. Environ. Eng. 2008,
134, 403–408.
16. Sieker, H., & Klein, M. Best management practices for stormwater-runoff with alterna-
tive methods in a large urban catchment in Berlin. Germany. Water Science and Tech-
nology, 1998, 38(10), 91–97.
17. U.S. Environmental Protection Agency (USEPA). Stormwater technology fact sheet:
Porous pavement, 1999, EPA 832-F-99-023, Washington, DC.
18. Jia, H., Lu, Y., Yu, S., and Chen, Y. Planning of LID-BMPs for urban runoff control:
The case of Beijing Olympic Village, Sep. Purification Technol, 2012, 84(9), 112-119
19. Benedict, M. E., & McMahon, E. T. Green infrastructure: Linking landscapes and com-
munities, 2006 Washington, DC: Island Press
121
20. NRC. Urban Stormwater Management in the United States. 2008, Washington, DC:
National Academies Press.
21. Ahiablame, L., Engel, B., & Chaubey, I. Representation and evaluation of low impact
development practices with L-THIA-LID: An example for planning. Environment and
Pollution, 2012a, 1(2). doi:10.5539/ep.v1n2p1
22. Baek, S.S., Ligaray, M., Park, J.P., Shin, H.S., Kwon, Y., Brascher, J.T., Cho, K.H.
Developing a hydrological simulation tool to design bioretention in a watershed. Envi-
ron. Modell. Software, 2017.
23. Rossman, L. Storm Water Management Model Reference Manual Volume I–Hydrology
I, 2015.
24. Lee, J.G., Selvakumar, A., Alvi, K., Riverson, J., Zhen, J.X., Shoemaker, L., Lai, F.H.
A watershed-scale design optimization model for stormwater best management prac-
tices. Environ. Modell. Software, 2012 37, 6–18.
25. Beyerlein, D. Low Impact Development Computations—WWHM. In World Environ-
mental and Water Resources Congress 2011: Bearing Knowledge. For Sustainability,
pp. 558–576.
26. Ahiablame, L.M., Engel, B.A., Chaubey, I. Representation and evaluation of low im-
pact development practices with L-THIA-LID: An example for site planning. Environ.
Pollution, 2012b, 1 (2), 1.
27. Ackerman, D., Stein, E.D. Evaluating the effectiveness of best management practices
using dynamic modeling. J. Environ. Eng. 2008, 134, 628–639.
28. Elliot, A.H., Trowsdale, S.A., Wadhwa, S. Effect of aggregation on on-site stormwater
control devices in an urban catchment model. J. Hydrol. Eng., 2009, 14, 975–983.
29. Li, Q., Wang, F., Yu, Y., Huang, Z., Li, M., Guan, Y. Comprehensive performance
evaluation of LID practices for the sponge city construction: A case study in Guangxi,
China. J. Environ. Manage., 2019, 231, 10–20.
30. Jia, H.,Yao, H., Tang, Y., Yu, S.L., Field, R., Tafuri, A.N.., LID-BMPs planning for
urban runoff control and the case study in China, J. Environ. Manag., 2015, 149, pp. 65-
76
31. US Geological Survey. Understanding the Effects of Stormwater Management Prac-
tices on Water Quality and Flow, 2019, https://www.usgs.gov/centers/sa-water/sci-
ence/understanding-effects-stormwater-management-practices-water-quality-and?qt-
science_center_objects=0#qt-science_center_objects
32. Yang, L.; Tian, F.; Niyogi, D. A need to revisit hydrologic responses to urbanization by
incorporating the feedback on spatial rainfall patterns. Urban Clim., 2015, 12, 128–140.
122
33. Jahan K, Pradhanang SM, Bhuiyan MAE. 2021. Surface Runoff Responses to Subur-
ban Growth: An Integration of Remote Sensing, GIS, and Curve Number. Land. 2021;
10(5):452. https://doi.org/10.3390/land10050452
34. Jahan, K.; Pradhanang, S. 2020. Predicting Runoff Chloride Concentrations in Subur-
ban Watersheds Using an Artificial Neural Network (ANN). J. Hydrol. 2020, 7, 80,
doi:10.3390/hydrology7040080.
35. Hunt, W.; Smith, J.; Jadlocki, S.; Hathaway, J.; Eubanks, P. Pollutant removal and peak
flow mitigation by a bioretention cell in urban Charlotte, N.C. J. Environ. Eng. 2008,
134, 403–408.
36. Dietz, M. Low impact development practices: A review of current research and
recommendations for future directions. Water Air Soil Pollut. 2007, 186, 351–363.
37. Jackisch, N.; Weiler, M. The hydrologic outcome of a Low Impact Development (LID)
site including superposition with streamflow peaks. Urban Water J. 2017, 14, 143–159.
38. Wilson, C.; Hunt,W.;Winston, R.; Smith, P. Comparison of runoff quality and quantity
from a commercial low-impact and conventional development in Raleigh, North Caro-
lina. J. Environ. Eng. 2014, 141, 05014005.
39. Bosley, I.; Kern, E. Hydrologic Evaluation of Low Impact Development Using a Con-
tinuous, Spatially-Distributed Model. Master’s Thesis, 2008, Virginia Polytechnic In-
stitute and State University, Blacksburg, VA, USA.
40. Brown, R.; Line, D.; Hunt, W. LID treatment train: Pervious concrete with subsurface
storage in series with bioretention and care with seasonal high-water tables. J. Environ.
Eng. 2011, 138, 689–697.
41. Lenhart, H.A.; Hunt, W.F., III. Evaluating four storm-water performance metrics with
a North Carolina coastal plain storm-water wetland. J. Environ. Eng. 2010, 137, 155–
162.
42. Mayer, A.L.; Shuster,W.D.; Beaulieu, J.J.; Hopton, M.E.; Rhea, L.K.; Roy, A.H.;
Thurston, H.W. Environmental reviews and case studies: Building green infrastructure
via citizen participation: A six-year study in the Shepherd Creek (Ohio). Environ.
Pract., 2012, 14, 57–67.
43. Palla, A.; Gnecco, I. Hydrologic modeling of low impact development systems at the
urban catchment scale. J. Hydrol., 2015, 528, 361–368.
44. Jia, H.;Wang, X.; Ti, C.; Zhai, Y.; Field, R.; Tafuri, A.N.; Cai, H.; Shaw, L.Y. Field
monitoring of a LID-BMP treatment train system in China. Environ. Monit. Assess.,
2015, 187, 373.
45. Jahan, K., Pradhanang, S.M., Boving, T. Modeling Runoff Quantity with different LID
structures in the Suburban Watershed using Stormwater Management, Hydrological
Science journal, 2021, URL: http://mc.manuscriptcentral.com/hsj
123
46. Tsihrintzis.V.A., and Hamid, R. Runoff Quality prediction from small urban catch-
ments using SWMM, Hydrological Processes, 1998, VOL., 12, 311-329 (1998)
47. Di Modugno, M.; Gioia, A.; Gorgoglione, A.; Iacobellis, V.; La Forgia, G.; Piccinni,
A.F.; Ranieri, E. Build-Up/Wash-Off Monitoring and Assessment for Sustainable Man-
agement of First Flush in an Urban Area. Sustainability, 2015, 7, 5050-5070.
https://doi.org/10.3390/su7055050
48. Northeast Regional Climate Center, http://www.nrcc.cornell.edu/wxsta-
tion/pet/pet.html
49. Huber, W.C. and Dickinson, R.E. Storm Water Management Model, Version 4: Users’
Manual, 1988, EPA 600/3-88/001a-Environmental Research Laboratory, EPA, Athens,
Georgia.
50. Mein, R. G., and Larson, C. L. Modeling Infiltration During a Steady Rain, Water Re-
sources Research, 1973, 9(2), 384–394.
51. Rossman, L., & Huber, W. Storm water management model reference manual: Volume
III – Water Quality, 2016, US Environmental Protection Agency, Office of Research
and Development, National Risk Management Laboratory, Cincinnati, OH, 45268.
52. Almedeij, J., Esen, I.I. Modified Green-Ampt infiltration model for steady rainfall. J.
Hydrol. Eng., 2013, 19 (9), 04014011.
53. Warwick, J.J., and Tadepalli, P. Efficacy of SWMM Application, J.Wat. Re-
sour.Plan.Manage., ASCE, 1991, 117, 352-366
54. Zhang, Z., Koren, V., Reed, S., Smith, M., & Moreda, F. Comparison of Simulation
results using SSURGO- based and STATGO-based parameters in a distributed hydro-
logic model, 3rd Federal Interagency Hydrologic Modeling Conference, 2006, Reno,
NV.
55. Mednick, A.C. Does soil data resolution matter? Stae soil geographic database versus
soil survey geographic database in rainfall-runoff modeling across Wisconsin, Journal
of Soil and Water Conservation, 2010, 65, 190-199.
56. Bhaduri, B., Harbor, J., Engle, B., & Grove, M. Assessing watershed-scale, long-term
hydrologic impacts of land-use change using a GIS-NPS model, Environmental Man-
agement, 2000, 26, 643-658.
57. Wiles, J.J., Levine, N.S. A Combined GIS and HEC model for the analysis of the effect
of urbanization on flooding; the Swan Creek Watershed, Ohio Environmental & Engi-
neering Geosciences, 2002, 8, 47-61.
58. Anderson, R.M., Koren, V.I., Reed, S.M. Using SSURGO data to improve Sacramento
Model a priori parameter estimates. Journal of Hydrology, 2006, 320, 103 – 116.
124
59. Reed, S., Schaake, J., Zhang, Z. A distributed hydrologic model and threshold fre-
quency-based method for flash flood forecasting at ungauged locations. Journal of Hy-
drology, 2007, 337, 402-420.
60. Jewell, T.K., D.D. Adrian, and T.J. Nunno. Methodology for Calibrating Stormwater
Models, Journal of the Environmental Engineering Division, 1978, 104 (3): 485–501.
61. James, L.D., and S.J. Burges. Selection, Calibration, and Testing of Hydrologic Models,
1982, In Hydrologic Modeling of Small Watersheds, edited by C.T. Haan, 437–72. St.
Joseph, MI: American Society of Agricultural Engineers.
62. Ahiablame, L.M.; Engel, B.A.; Chaubey, I. E_ectiveness of low impact development
practices in two urbanized watersheds: Retrofitting with rain barrel/cistern and porous
pavement. J. Environ. Manag. 2013, 119, 151–161.
63. Tuomela, C.; Sillanpää, N.; Koivusalo, H. Assessment of stormwater pollutant loads
and source area contributions with stormwater management model (SWMM). J. Envi-
ron. Manag., 2019, 233, 719–727.
64. Kingston Campus Master Plan, 2018. University of Rhode Island, Kingston Campus,
South Kingstown, RI. GORDON R. ARCHIBALD, INC. 200 Main Street Pawtucket,
RI 02860.
65. Stormwater Management Plan (SMP), South Kingstown, Rhode Island. Town of South
Kingstown, July 17, 2020. BL Companies. BL Project Number: 18C6704.
https://www.southkingstownri.com/DocumentCenter/View/5219/Stormwater-Man-
agement-Plan?bidId=
66. Nash, J.E.; Sutcli_e, J.V. River flow forecasting through conceptual models part I—A
discussion of principles. J. Hydrol. 1970, 10, 282–290.
67. Krause, P., D. P. Boyle, and F. Bäse. Comparison of different efficiency criteria for
hydrological model assessment, In: Advances in geosciences, 2005, 5, pp. 89–97.
68. Dongquan, Z., C. et al. GIS-based urban rainfall-runoff modeling using an automatic
catchment-discretization approach: a case study in Macau, In: Environmental Earth Sci-
ences, 2009, 59 (2), p. 465.
69. Devore, J. L. and K. N. Berk. Modern Mathematical Statistics with Applications, 2007,
Thomson Brooks/Cole. 845 pp. ISBN: 9780534404734.
70. Golmohammadi, Golmar; Prasher, Shiv; Madani, Ali; Rudra, Ramesh. Evaluating
Three Hydrological Distributed Watershed Models: MIKE-SHE, APEX, SWAT, Hy-
drology, 2014, 1, no. 1: 20-39. https://doi.org/10.3390/hydrology1010020
71. Nearing, M.A., L.Deer-Ascough, and J.M. Laflen, Sensitivity analysis of the WEPP
hillslope profile erosion model. Trans. ASAE, 1990
125
72. Brezonik, P. L. & Stadelmann, T. H. Analysis and predictive models of stormwater
runoff volumes, loads, and pollutant concentrations from watersheds in the Twin Cities
metropolitan area, Minnesota, USA. Water Research 36, 2002 (7), 1743–1757.
73. El-Hassanin, A. S., Labib, T. M., and Gaber, E. I. Effect of vegetation cover and land
slope on runoff and soil losses from the watersheds of Burundi. Agric. Ecosyst. Envi-
ron., 1993, 43(3–4), 301–308.
74. Huang, J.L.; Du, P.F.; He, W.Q.; Ao, Z.D.; Wang, H.C.; Wang, Z.S. Local sensitivity
analysis for urban rainfall-runoff modeling. China Environ. Sci. 2007, 27, 549–555.
75. Wang, F.; Wang, X.; Zhao, Y.; Yang, Z.F. Long-term periodic structure, and seasonal-
trend decomposition of water level in Lake Baiyangdian, Northern China. Int. J. Envi-
ron. Sci. Technol. 2014, 11, 327–338.
76. Buckman, H. O., and N. C. Brady. The Nature and Properties of Soils. Macmillan, 567
pp, 1960.
126
CHAPTER 4
Manuscript IV
Predictive Uncertainty Estimation of Stormwater Management Model using
Bayesian Markov Chain Monte Carlo approach
prepared for Hydrological Processes Journal, July 2021
Khurshid Jahan1,*, Soni M Pradhanang1,*, and Russell D. Briggs2
1Department of Geosciences, University of Rhode Island, RI, Kingston, USA 2Department of Sustainable Resources Management, ESF SUNY College of Environ-
mental Science and Forestry, NY, USA
*Correspondence: [email protected] (K.J.), [email protected]
(S.M.P)
127
Abstract
The reliability of any hydrological model output depends on the uncertainty analysis of
both input and output. Uncertainty might arise mainly due to errors in data measurement,
data collection, assumptions, and limitations placed. This study analyzes the uncertainty
of Storm Water Management Model (SWMM) data input and output using the Bayesian
Markov Chain Monte Carlo (MCMC) approach due to its flexibile model fitting
capbility with data sets of different complexity and limitation. As the variation of inputs
contributes to output variation differently, the study adopts fully bayesian sensitive
analysis wherein the response variability is examined within the entire input space to
consider the interaction among the inputs. Thus, the quantification of uncertainty
facilitates the generation of more consistent or reliable flow predictions. The study
reveals that rainfall depth, slope, and manning coefficients are the important sensitive
parameters for the Chipuxet River watershed area, Rhode Island. The outcome showed
the Interaction between the model sensitive parameters affects peak flow. Moreover,
Nash-Sutcliffe Efficiency (NSE) values of 0.73, and 0.69, R2 values of 0.75, and 0.64,
and RSR value of 0.53 and 0.61 for runoff quantity module in two different runs prove
model efficiency. For runoff quality module in SWMM, total suspended sediments,
nitrate and nitrite, and ortho phosphate calibrated and measured NSE, R2, and RSR value
which also showed the significant performance. Posterior distribution of the model
parameter reveals that average observed values is within ±20% range from the model
values and most of the observed data are within the 95% uncertainty band.
Keyword: MCMC, SWMM, Bayesian, Uncertainty, Sensitivity, Rhode Island
128
1. Introduction
SWMM (Storm Water Management Model) is a popular and widely used
deterministic rainfall-runoff model for simulating water outflows, inflows, and storages
within a sub-catchment or watershed during a single event or continuous simulations
[1]. SWMM is designed primarily for urban areas [2]. SWMM is capable of estimating
a number of commonly measured runoff pollutants along with runoff quantity. During
SWMM modeling, different parameters are fed into the system. Parameters values can
be default values (established by the manual), measured values (taken as primary data),
estimated values (from the previous literature), and values obtained from the trial-and-
error method. As the watershed areas vary in terms of land-use, soil properties, slope,
etc., multiple interactions occur between features of the drainage system or among
different hydrological processes creating a complex modeling system [3].
Moreover, complexity might come from runoff depth, peak flow, lag time,
overflow volume, duration. Hence a unified theory for all kind of watershed and
catchment is difficult to establish at different temporal and spatial scales. As a result,
uncertainty analysis of hydrological model input and output is gaining attention in urban
environments [4, 5]. The many inputs that are used to predict both water quantity
(%imperviousness, depression storage for impervious and pervious, sub-catchment
width, slope, area, Manning’s coefficient, rainfall intensity) and quality (soil
conductivity rate, conduit roughness, conduit length, storage depth) have different levels
of uncertainty [6]. In some cases, especially in the highly urbanized watershed,
uncertainty is higher in water quality prediction [6, 7]. For example, SWMM had an
average prediction error for various pollutant loads of less than 20% [8]. To address
129
these issues numerous studies [9, 10, 11, 12, 13] adopted different uncertainty
techniques like the Bayesian estimation Monte Carlo approach on the Storm Water
Management Model.
Uncertainty of a model implies quantifying the uncertainty in the model output
due to uncertainty in the model input parameters. For example, uncertainty arises in a
hydrological model due to different variability and limitations in its input data, structural
parameter, and calibration. Natural conditions, measurement limitations, data scarcity-
all of those contribute to input data uncertainty [14]. Hence, instead of using a
deterministic model with the fixed input parameters (as in Figure 1a), an uncertainty
model is constructed using the distribution of possible values for each of the parameters
(as in Figure 1b) and outputs a distribution of possible values.
Inability or error in determining effective parameter along with natural
variability, observational limitation & errors in calibration data can lead to parameter
uncertainty [16, 17]. Hence model structures determine the model performance [18]. All
models are mostly approximations to the real-life situation due to the theoretical
constraints, inadequate knowledge, numerical and process simplifications, and lack of
a unifying theory, even with known input parameters [19]. Furthermore, this leads to
structural uncertainty i.e., predictive uncertainty of the output. In some cases, it can
contribute most to the predictive uncertainty [20, 21].
130
Figure 1: (a)- Traditional Model, (b)- Uncertainty Model (adapted and modified from
[15])
Structural uncertainty cannot be wiped out, regardless of error-free input data,
since responses found in the catchment are averaged over space and time [22]; however,
it can be diminished if the predominant processes are sufficiently addressed [23];
Most of the time, calibration of model parameters and evaluation of prediction
take place using observed parameters like streamflow, discharge which might introduce
uncertainty due to measurement error, limitation, natural conditions, or seasonal
variation [24, 25].
All of the above uncertainties mentioned earlier take part in the predictive
uncertainty of model output. Predictive uncertainty is usually heteroscedastic, the
magnitude of uncertainty varying with the extent of the model output (non-Gaussian
residuals). Comparison of predicted data to observed data is necessary to perform a post-
(a)
(b)
131
audit validation of a model [26]. Observed data is needed in order to demonstrate
whether SWMM 5.1 LID controls can predict runoff.
Generalized Likelihood Uncertainty Estimation (GLUE) [27] and the classical
Bayesian Monte Carlo (i.e., Bayesian inference approach) [28] are widely used
uncertainty analysis approaches. As Bayesian methods require a large amount of data,
it is difficult to implement while there is data scarcity. Hence the GLUE method is easier
to implement than the Bayesian method, though it is questioned for its statistical
incoherence and inconsistency [29, 30, 31]. Because of these limitations, Monte Carlo
simulations are proposed for uncertainty quantification by estimating the probability
density function (PDF) of the model parameters and predictions. [32, 33, 34, 35]
Markov-chain Monte Carlo (MCMC) are extended version providing computational
efficiency. In MCMC, probability of next possible events depends on the state of the
existing events, whereas in MC, probability of next possible events is calculated based
on different sets of past events within a range of probability distribution.
Moreover, as some model parameters might not influence individual model
output, the consideration of all parameters in model simulation significantly increases
model convergence time. Therefore, it is essential to apply a parameter selection
methodology for fixing an appropriate set of parameters applicable for the particular
study.
In order to identify the parameters that influence model output, it is necessary to
carry out a sensitivity analysis (SA) during uncertainty quantification. In general, there
are two approaches for SA: (a) local sensitivity analysis where a single parameter is
changed one at a time and the corresponding output is analyzed, and (b) global
132
sensitivity analysis where responses are evaluated considering simultaneous changes of
all the variables within their respective input spaces. Parameters are ranked in order of
sensitivity and denote the effective ones for reducing variance between simulated results
and observed data [36]. Sometimes model sensitivity can be perceived from previous
studies and documentation for similar watersheds, but it might be necessary for a
different kind of watershed to make sensitivity analysis [37]. An overview of the SA
can be found from Saltelli et al., 2000 [38]. Furthermore, when a model is involved with
many inputs, SA is an effective tool for optimizing input space [39].
Numerous studies have cited that runoff volume is most sensitive to the percent
impervious area of a sub-catchment [6, 8, 40]. In addition, some studies concluded that
runoff volume was most sensitive to the impervious depression storage (Dstore-Imperv)
and Manning’s coefficient for impervious areas (N-Imperv) while the shape of the
hydrograph was sensitive to sub-catchment width [41]. The uncertainty of a simplified
urban drainage model developed in previous studies has been evaluated in this context.
Different approaches can be used for uncertainty identification. A Bayesian approach
coupled with Monte Carlo analysis has been used [42] in this study. The Bayesian
approach expresses uncertainties in the model parameters in terms of probability.
Parameter uncertainty is quantified first by introducing a prior probability distribution,
representing historical or expert information before collecting new data.
UA Applications to suburban watersheds are minimal. However, this study is the
first to apply an MCMC scheme that works within a formal Bayesian framework for
UA for suburban watersheds using SWMM. The conducted Bayesian uncertainty
approach applied on the developed SWMM model for a suburban watershed area. Due
133
to suburban growth [43], recently occurred (2010, and 2013) flood events and their
impacts, and having high chloride during the winter season due to overused of road salt
made the authors curious for detail analysis for the suburban areas [44]. It is important
to mention that suburban areas also have started to be affected through the stormwater
runoff like urban areas. We focus here on addressing the sensitive or responsive
parameters of linking the MCMC method to the developed stormwater model. The
linking output will help to update the quantity model parameter distribution. The overall
objective of this research is mentioned below:
• to identify sensitive parameters for the study area using monitored rainfall
events.
• to identify SWMM model output (peak flow) response with the sensitive param-
eters
• to estimate model parameter uncertainty distribution
• to identify predictive uncertainty of SWMM model output (peak flow) using
Bayesian Markov Chain Monte Carlo (MCMC) Method.
2. Study Area, Data & Materials
This research has been carried out in the Chipuxet watershed located in South
Kingstown, Rhode Island. The selected study area is split into 12 different sub-
catchments (Figure 2). The catchment has one rainfall gauge station (NOAA Station ID
54796), one automated stream gauge flow (USGS 01117350) run by USGS, six
stormwater quality monitoring sites and monthly average evapotranspiration data.
Stormwater runoff quality was measured with the samples collected from 42 individual
134
storm events from six different sites. The samples were collected from 2016 to 2019.
The area is a low elevated region having a low to moderate slope. The range of the slope
is (7-10) %.
Stormwater quality monitoring sites are established near the parking lot, agricultural
practice area.
Figure 2: Location Map of the Study Area
The original elevation data, including Digital Topographic Map (DTM) 1:2000
and Remote Sensing Images (RSI) 1:2000, are the required inputs for the SWMM model.
Watershed delineation for the Chipuxet Watershed was done using ArcGIS 10.6.1 using
135
DEM file data. The study area relatively very low elevation area having average
elevation 45 meter.
3. Material & Methodology
3.1 SWMM Model
The study is modeled with SWMM version 5.1 [47]. SWMM is designed with
four different computational routine or building blocks (runoff, storage, transport and
extran). Furthermore, low impact developments (LIDs) modeling like porous
pavements, bioretention cells, infiltration trenches, vegetative swales, and rain barrels
are also possible through SWMM5.1. As the system can capture complex processes and
different responses to different inputs and configurations, SWMM is advised to use
hydrological modeling in urban areas [48]. In this study, runoff routine is used for runoff
quantity simulation by transporting through a drainage network consisting of pipes,
channels, storage/treatment devices, pumps, and regulators [48 - 50].
Runoff is simulated using a dynamic-wave equation. Depression storage is used
for considering water loss through infiltration. The land-use map of the study area
provides the percentage of impervious and pervious portions. The depth of depression
storage on impervious and pervious portions of the sub-catchment, Manning’s
coefficient for overland flow over the impervious and pervious portions of the sub-
catchment, the percent of impervious area without depression storage, and the
infiltration parameters of Green Ampt’s equation [51]- all these are parameters in runoff
block of SWMM were used to calibrate the model. The minimum and maximum values
were obtained from the SWMM user’s manual [47] and relevant literature [52,53]. The
136
model is appropriately calibrated to identify an optimal set of parameters that provide
minimum deviation between simulated and observed results. Topographical and
catchment parameters are formulated using GIS applications and transferred to SWMM.
Runoff volumes produced by the model greatly vary by the types of land-uses like
residential, commercial, agricultural, recreational etc.
3.2 Objective Function
Model parameter sensitivity and the uncertainty analysis was done prior to the
calibration of SWMM. The model was developed for both quantity and quality using 42
individual rainfall events. Stormwater quality data were collected from the six different
site (Figure 2) and calibrated with entire sets of data from 2016 and 2019. Two
individual calibrations run obtained using the daily discharge data from USGS
01117350 Chipuxet river at West Kingston, RI (Figure 2). Calibration was conducted
for two different model set ups (Run 1 and Run 2) in two sub-catchments (sub-
catchment 3 and 6) from two different land types (agricultural and parking lot) where
the monitoring sites are located. For water quality calibration, we considered TSS,
Nitrate-N, and orthophosphate-P for site 1 and site 5. 30 rainfall events out of 42 events
considered for the model calibration and the rest of 12 rainfall events used for model
validation. The adjusted validation parameter is presented in Table 1.
The optimal parameter values for the study area were selected according to the
Nash-Sutcliffe Efficiency (NSE) [54].
NSE = 1 − ∑ (Oi−𝑃𝑖)2n
i=1
∑ (𝑂i−O′)2ni=1
………………………………………………………(1)
137
The coefficient of determination, R2, an established measure in statistics quantifying
the model’s fit, is described as:
R2 = [ ∑ (𝑂𝑖
ni=1 −𝑂′)(𝑃𝑖−𝑃′)
√∑ (𝑂𝑖−𝑂′)2√∑ (𝑃𝑖−𝑃′)2ni=1
𝑛𝑖=1
] 0 ≤ R2 ≤ 1 ……………….………………..(2)
where n is the number of observations in the period under consideration, Oi is the i-th
observed value, 𝑂′ is the mean observed value, 𝑃𝑖 is the i-th model-predicted value, and
𝑃′ is the mean model-predicted value. NSE range is −∞ and 1, where 1 is considered as
the best fit for any model [55]. According to Dongquan et al. 2009 [56] an NSE value
of ≥0.5 is acceptable for SWMM. The acceptable R2 is 1 or close to 1, the larger
proportions of the observed variability can be explained with the simulations of
estimated values [57].
To evaluate the developed model for the goodness of fit test root mean square error
(RMSE), the RMSE-observations standard deviation ratio (RSR) is calculated as the
RMSE and standard deviation ratio of measured data. The RMSE and RSR are
explained as follows
RMSE = √1
𝑛 ∑ (𝑃𝑖 − 𝑂𝑖)2𝑛
𝑖=1 …………………………………………..………….. (3)
RSR = 𝑅𝑀𝑆𝐸
𝑆𝑇𝐷𝐸𝑉𝑜𝑏𝑠 …………………………………………………………………… (4)
where n is the number of observations in the period under consideration, 𝑃𝑖is the i-th
model-predicted value and 𝑂𝑖 is the i-th observed value, 𝑆𝑇𝐷𝐸𝑉𝑜𝑏𝑠 is the standard
deviation of the observed value. For RSR values between 0 and 1 are acceptable when
field-specific data are accessible for calibration. RSR varies from the optimal value of
138
0 to a large positive value. The lower RSR, the lower the RMSE, and the better the
model simulation performance [58].
Calibration is conducted for two different sets of parameters. The adjusted
parameters are given in Table 1.
Table-1: Model Parameter and Ranges used for Calibration and Uncertainty Analysis
Name Description Minimum Maximum
SWMM Parameter Percent adjustment
Width Sub-catchment width (m) -90 120
Slope Sub-catchment Slope -15 10
%Imperv Percentage of impervious area (%) -20 25
N-Imperv Manning n for impervious area -65 85
N-Perv Manning n for pervious area -60 95
Dstore-Imperv Depression storage for impervious area (mm) -90 95
Dstore-Perv Depression storage for pervious area (mm) -95 150
Conduit n Manning’s roughness -55 100
Conduit length Length of the conduit (mm) -60 90
3.3 Bayesian Approach and Error Modeling
Bayesian approach or analysis is a statistical paradigm that allows combining
prior information of the research or project to guide the statistical interpretation process
using probability statements. The “starting level” of knowledge regarding a parameter
can be expressed by a statistical distribution that we call “Prior distribution.” The prior
reflects our level of knowledge concerning the parameter. The model representing the
data generating mechanism for the observed sample is called sampling distribution, or
likelihood. The updated or ultimate distribution of the parameter after observing the
139
sample is called posterior distribution. From the posterior, we can extract inferential
quantities of interest: mean, variance, probability intervals, etc.
For this study, the prior parameters are the sub-catchment characteristics, for
example, area, width, % imperviousness, depression storage impervious, depression
storage pervious, Manning’s roughness, slope, drainage parameters like conduit length,
roughness, storage, and the parameter of three LID controls (Bioretention, Porous
Pavement, and Vegetative Swales). The model outputs (e.g., peak flow, TSS, Nitrite
and Nitrate and ortho phosphate) can be defined as the following function
�̂� = 𝑓(𝑋, 𝜃) ……………………………………………………………………... (5)
where �̂� = vector of predicted (simulated) values = {�̂�1, … . �̂�𝑛}
𝑋 = matrix of input forcing; and
𝜃 = vector of model parameters.
Now if, 𝑌 = vector of observed or measured values = {𝑦1, … . 𝑦𝑛}
the prediction error or residuals is calculated as,
𝐸𝑛(𝜃) = 𝑌 − �̂� ………………………………………………………………… (6)
The traditional calibration method minimizes error prediction by determining a
single set of optimal parameter values. As uncertainty arises from a different perspective,
the model needs to take those into account during simulation. Hence, the Bayesian
approach is adopted in this study. The Bayesian method outputs posterior probability
distribution (PPD) of the parameters to address the associated identity [59].
The PPD is defined as:
𝑃(𝜃 𝑌⁄ ) =𝑃(𝑌 𝜃⁄ )𝑃(𝜃)
∫ 𝑃(𝑌 𝜃⁄ )𝑃(𝜃)𝑑𝜃 …………………………………………………………… (7)
140
where 𝑃(𝜃) = prior distribution or knowledge of the parameters, 𝑃(𝑌 𝜃)⁄ =
conditional probability for the measured data Y given the parameter vector 𝜃,
In many cases, the prior distribution is given a significant variance or a uniform
distribution, reflecting the lack of knowledge about the parameter. A prior distribution
can also be determined from previous studies or expert knowledge. The PPD states that
the current state of the parameters is proportional to the likelihood function multiplied
by the prior density. 𝑃(𝑌 𝜃)⁄ is also called the likelihood function and the critical
parameter in Bayesian analysis. As equation (5) has an integral at the denominator, it is
difficult to estimate by an analytical approach. Hence, Bayesian analysis is coupled with
the Markov Chain Monte Carlo Simulation.
3.4 Markov Chain Monte Carlo (MCMC)
The Monte Carlo method draws samples from the target distribution and
estimates the mean and variance of the drawn samples. When a sample drawing depends
on the prior sample, this is called the Markov Chain property. Hence MCMC allows
random sampling even with the distribution of many variables by narrowing in on the
quantity that is being approximated from the distribution.
The Metropolis-Hastings (MH) algorithm is a widely used flexible variant of the
MCMC algorithm. MH uses a proposed probability distribution for the samples and sets
acceptance criteria for the new sample to be included in the chain [60, 61]. Usually, in
each iteration, current likelihood values are compared with the new likelihood value,
and new PPD is accepted according to the likelihood ratio. Hence MH is helpful where
141
the subsequent state probability distribution cannot be sampled directly. Since model
outputs vary according to the input parameters’ influence, the study conducted a
sensitive analysis before Bayesian Monte Carlo Analysis.
3.5 Sensitivity Analysis (SA)
A model is highly sensitive to the parameter that leads to a significant variation
in model output even there is the slightest change in the parameter value. On the other
hand, if the model output varies insignificantly with the variation in the input parameter,
the model is low sensitive to that parameter. Thus, sensitivity analysis and uncertainty
quantifications are closely related to each other.
In this study, the variability of the response is investigated with respect to a
probability distribution over the entire input space [62], and it adopts the “Sobol”
method [63] where the objective function is the sum of smaller functions with subsets
of the input space. Sobol sensitivity indices denote the quantity of variations in model
output that each uncertainty parameter is responsible for. Parameter with lower Sobol
index has a small impact in variations of the output, whereas higher index denotes the
model output will vary significantly with the change in the parameter.
If model input and output are 𝑥 and 𝑧 respectively, then density 𝑢(𝑥), and the
appropriate marginal densities 𝑢𝑖(𝑥𝑖) defines the uncertainty distribution to the
sensitivity of 𝑧 with respect to change in 𝑥 , that is the input to the SWMM model. So,
the marginal conditional expectation
142
𝐸[𝑧]𝑥𝐽 = {𝑥𝑗: 𝑗 ∈ 𝐽} = ∫ 𝐸[𝑧|𝑥]𝑢(𝑥)𝑑𝑥−𝐽𝑅𝑑−𝑑𝑗 …………………………..……….. (8)
Where 𝐽 = {𝑗1, … … . 𝑗𝑑𝑗}, a subset of input spaces, 𝑥−𝑗 = {𝑥𝑗: 𝑗 ∉ 𝐽}, and the marginal
uncertainty distribution is given by 𝑢𝐽(𝑥𝐽) = ∫ 𝑢(𝑥)𝑑{𝑥𝑖 ∶ 𝑖 ∉ 𝐽}𝑅
𝑑−𝑑𝑗 .
If the inputs are not correlated, then the variability of 𝐸[𝑍/𝑥𝐽] is decomposed as
𝑣𝑎𝑟 (𝐸[𝑧|𝑥]) = ∑ 𝑉𝑗𝑑𝑗=1 + ∑ 𝑉𝑖𝑗1≤𝑖<𝑗≤𝑑 + … . + 𝑉1……..𝑑, …………………………... (9)
where 𝑉𝑗 = 𝑣𝑎𝑟 (𝐸[𝑧|𝑥𝑗]), 𝑉𝑖𝑗 = 𝑣𝑎𝑟(𝐸[𝑧|𝑥𝑖 , 𝑥𝑗]) − 𝑉𝑖𝑉𝑗,
Now, the Sensitivity Indices, 𝑆𝐽 = 𝑉𝐽
𝑣𝑎𝑟(𝑧), will sum to one over all possible 𝐽 and are
bounded to [0,1] and measure the significance of a set 𝐽 of inputs.
In this study, sensitivity for each input is calculated in two ways:
1st order index for 𝑗𝑡ℎ variable, 𝑆𝑗 = 𝑣𝑎𝑟 (𝐸[𝑧|𝑥𝑗])
𝑣𝑎𝑟(𝑧) ………………………………….. (10)
The total sensitivity for input j, 𝑇𝑗 =𝐸[𝑣𝑎𝑟(𝑧|𝑥−𝑗)]
𝑣𝑎𝑟(𝑧) …………………………………. (11)
where 𝑥−𝑖 denotes all uncertain parameters except 𝑥−𝑖.
The 1st order indices measure the contribution variability to the output for each
input variable. In contrast, the total effect indices reflect the portion of variability that
total variation in each input is responsible for. Hence the sum of the total Sobol
sensitivity indices is equal to or greater than one [60]. If no higher-order interactions are
present, the sum of both the first and total order Sobol indices equals one.
Due to the involvement of many integrals, estimation of the sensitivity indices
is not straightforward, and fully Bayesian Markov Chain Monte Carlo simulation is
143
adopted during sensitivity analysis [65]. At each MCMC iteration, the output is
predicted by randomly sampling from the input set according to the Latin Hypercube
Sampling (LHS) scheme. Finally, a single realization of the sensitivity indices is
calculated [62].
In summary –
• Variance-based global sensitivity analysis is carried out using Monte Carlo sim-
ulation to find out influential mode parameters.
• The prior distribution is assumed either from the available knowledge of the past
studies [66] or from experimental measurement. Lack of knowledge of the pa-
rameter often defines prior distribution as uniform distribution in many cases.
• When there is limited data availability, the posterior distribution assumes a sim-
ilar shape to the prior. On the contrary, data sufficiency greatly influences the
shape of the posterior distribution [67]. Posterior distributions are calculated us-
ing the Bayesian MCMC method
• The uncertainty band is calculated using the 95% percentiles of the posterior
distribution of the predicted output. Model is rejected if most of the observed
points fall outside the band. The wider the bands, the higher the uncertainty in
estimation and the lower confidence in model output.
• For Each Bayesian update stage, MCMC methods applied above are run 10000
times for computing posterior distribution
144
4. Findings and Discussions
4.1 Sensitivity Analysis
Identification and selection of significant parameters and their values have a
great impact on the model simulation. The model’s simulations were carried out using
input data on both hydrological and hydraulic parameters in SWMM5.1. Impervious
coefficient, depression storage (D) of both the impervious and pervious areas, width (W)
of the model area, manning coefficient (n) are identified as significant parameters by
some studies [68, 69]. The developed SWMM found rainfall depth, sub-catchment
area, %impervious, and slope as the sensitive or responsive parameters (Figure 3a) for
the hydrology module. In addition, SWMM found conduit roughness, conduit length,
and storage depth as responsive parameters for this suburban catchment (Figure 3b) for
the water quality module. The present studies also reveal that rainfall depth, area,
impervious percentage, slope, and manning coefficients are important sensitive
parameters. Rainfall depth shows linear effects almost, slope behaves like sine waves,
impervious and Manning Roughness shows the inverse relationship (Figure 4a). Here,
Figure 4b shows that the rate of change of the random parameters used in this analysis
generated different absolute relative sensitivity indexes for first-order indices. The
sensitivity index (SI) reveals rainfall depth and slope have higher sensitive indices,
where rainfall depth is more influential than slope (Figure 4b). Impervious percentage
and manning roughness are relatively low first-order indices.
145
Figure 3: a) Model runoff sensitivity analysis of sub-catchments for a) hydrological
parameters; b) the outfall of the catchment concerning hydraulic parameters [In SWMM,
time patterns allow external dry weather flow to vary periodically. They comprise of a
set of adjustment factors applied as a multiplier (SWMM5.1 Manual)]
Full effect sensitive index shows that sensitive index for the total effect of all
these parameters significantly higher than corresponding first-order indices. So the
effect on peak flow manifests largely through an interaction between input parameters
(Figure 4(c)). However, there is a significant variation in the main effects with 90%
central error bars. (Figure 4(d)). Uncertainty distribution of the sensitive parameters is
given in Figure 4(e).
(a) (b)
(c)
146
Figure 4: Sensitive Parameters Response of SWMM Quantity Module (a) Main
Response of the selected parameter (b) first-order SI indices represent the response of
each of the variables (c) Total Effects SI indices measure the portion of variability that
is due to total variation in each input or variable (d) Main effects for each input variables
which obtained as a by-product of the variance analysis with mean 90% error interval
(e) Uncertainty distribution of Sensitive parameters
Sensitivity analysis (Figure 5) was also analyzed for the quality module in this
study, like the hydrology module. The analysis reveals that conduit length, conduit
roughness, and storage depth are the most sensitive parameter. Conduit roughness and
(e)
(d)
147
storage depth behave like sine waves, whereas the conduit length responds inversely
(Figure 5a). Figure 5b indicates that the rate of change of the random parameters used
in this analysis generated different absolute relative sensitivity indexes for first-order
indices. The analysis showed that conduit depth and storage are more sensitive than
conduit roughness. Main effects of the selected parameter (Figure 5c), significant
variation in the main effects with 90% central error bars. (Figure 5d), uncertainty
distribution of the sensitive parameters Figure 5(e) is also analyzed here for the quality
module.
(d)
)
(b)
(a)
(b)
(c)
148
Figure 5: Sensitive Parameters Response of SWMM hydraulic Module (a) Main
Response of the selected parameter (b) first-order SI indices represent the response of
each of the variables (c) Total Effects SI indices measure the portion of variability that
is due to total variation in each input or variable (d) Main effects for each input variables
which obtained as a by-product of the variance analysis with mean 90% error interval
(e) Uncertainty distribution of Sensitive parameters
4.2 Model Efficiency
Based on the above sensitivity analysis and past studies, the different
combinations for model parameters are set for model simulation, and the best result
found according to the parameter settings (in Table-2)
Table-2: Model Parameter Settings
Parameter Adjustment Initial Value Calibrated value (Run 1) Calibrated value (Run 2)
% Impervious x 1.0 X 0.23 X 0.25
Depression Storage (mm) 2.032 1.524 2.2
Sub-catchment Width 1 1 1
(e)
149
Impervious "n" 0.02 0.02 0.017
Pipe "n" 0.013 0.011 0.012
Soil Conductivity (mm/h) 13.97 5.7 4.9
Manning’s n for swale 0.24 0.15 0.14
Suction head (cm) 20.32 18.2 17.5
DStore-Perv 0.25 0.22 0.2
DStore-Imperv 0.02 0.017 0.018
Model efficiency is calculated in terms of NSE, R2, and RSR: Table 3 is
represented the calibration and validation output for the SWMM, corresponding Nash-
Sutcliffe Efficiency (NSE) showed an average value 0.73 which indicates the good fit
of model because NSE values greater than 0.5 are acceptable for SWMM simulation
[70]. R2 values also signify the model performance. In terms of RSR value, the values
between 0 and 1 are acceptable though the ideal value of RSR is 0. The lower RSR, the
lower the RMSE, and the better the model simulation performance [64]. For the
developed SWMM model, the avg RSR value is less than 0.52 that also admit the model
efficiency.
Table 3: Model Calibration and Validation Output
Model
Eval-
uation
Peak Flow Cali-
bration
Peak Flow Vali-
dation
TSS
Calibration
Nitrate and Ni-
trite
Calibration
Ortho Phos-
phate Calibra-
tion
Run 1 Run 2 Run 1 Run 2 Site_1 Site_5 Site_1 Site_5 Site_1 Site_5
NSE 0.73 0.69 0.77 0.70 0.73 0.77 0.78 0.68 0.71 0.68
R2 0.75 0.64 0.8425 0.7991 0.87 0.82 0.93 0.92 0.90 0.88
RSR 0.53 0.61 0.42 0.51 0.58 0.52 0.41 0.55 0.62 0.57
150
4.3 Parameter Uncertainty
Posterior distribution of the model parameter (slope and intercept) indicates
uncertainty between observed and model peak flow. Posterior mean value of the slope
is 0.87 (red circle in Figure 6 (c)) with ranges from 0.81 to 0.97 and mean value of
intercept is 0.12 ((red circle in Figure 7 (a))) with ranges from 0.00 to 0.24. Few negative
values in interception occur because of the algorithm’s effort to minimize error, though
these negative values have no meaning in reality. An average observed value will be
within ±20% range from the model values considering the above values. For example,
if the model assumes the average peak flow is 100 m3/sec, then the actual value will be
from 80m3/sec to 120m3/sec. The green dotted line indicates the true values of the
parameter. The range will shrink and converge to a single value for the greater number
of simulations, indicating a greater confidence in the model parameters.
Figure 6: (a) Posterior distribution of Intercept (b) 10000 Iteration of MCMC Simulation
with the random progression of the intercept of the model (c) Posterior distribution of
151
Slope (d) 10000 Iteration of MCMC Simulation with the random progression of the
slope of the model
Posterior distribution of intercept (Figure 6a) and slope (Figure 6b) extends key
information regarding the comparative significance of the developed SWMM
parameters considered for the analysis. Except for percentage imperviousness (Imperv)
for impervious subareas (Dstore-Imperv), slope and mannings roughness for both
pervious and impervious subareas are not comprehensive. This suggests that the
SWMM parameters that exhibited no wide uncertainty range substantially affect the
Chipuxet Watershed rainfall-runoff characteristics. This also has a practical
consequence in terms of prioritizing resources on data collection. This means the
availability of more accurate or sensitive data may help in improving the accuracy of
runoff quantity and quality simulation for the selected study area.
4.4 Predictive Uncertainty
Runoff predictive uncertainty is estimated by propagating the different samples
of the posterior distribution through the SWMM5.1 model after the posterior
distribution of the model parameters is done, and the runoff and reporting the respective
prediction uncertainty ranges (e.g., 95% confidence interval). However, this prediction
interval represents parameter uncertainty only; it doesn’t consider other sources of error,
including model structural, forcing data, and calibration data uncertainty. Figure 7(a)
represents the bands derived from MCMC simulations of predicted & observed peak
flow. The light grey line indicates the 95% uncertainty band. The black line denotes the
fitted line of the model, which is within 95% uncertainty limits.
152
The uncertainty on model output can be quantified by the mean and the
maximum amplitude of the 95% confidence bounds. While the mean bound amplitude
(MBA) represents the average uncertainty on the computed hydrograph over the entire
event (Peak flow, TSS, Nitrate-N, and Orthophosphate-P), the maximum bound
amplitude (MaxBA) usually occurs for the most significant value. It is, therefore, an
indicator of the uncertainty in the prediction of the peak flow, TSS, Nitrate-N, and
Orthophosphate-P. The MBA is (differences lower than 10%) when the acceptability
threshold is lower than 0.8 (Fig. 7), 0.95 (TSS), TN (0.8), and 0.75 (phosphate).
Figure 7: (a) Predictive Uncertainty Band (b) MCMC simulation of the data
(a)
(b)
153
The red and blue lines indicate the predictive uncertainty that considers both the
parametric uncertainty and the residual error model. Most of the observed points are
within this limit, with some exceptions that might result from uncertainty in
measurement. For example, a wider band in high volume peak flow means higher
uncertainty as the frequency of high volume rainfall is lower than the low and medium
intensity rainfall. Figure 7(b) also shows that more than 70% of data are within Monte
Carlo lower and upper limits.
154
Figure 8: Predictive Uncertainty Band MCMC simulation of a) TSS, b) Nitrate-N , and
c) Orthophosphate-P
5. Conclusion
The study adopted the Bayesian Markov Chain Monte Carlo approach in
demonstrating parameter uncertainty and predictive uncertainty in SWMM model
simulation on the Chipuxet watershed, Rhode Island, USA, by generating 10000 sets of
sensitive parameters identified sensitive analysis. The value of the non-random
parameters has importance in uncertainty quantification. Hence during calibration, most
likelihood values of the nonrandom parameters need to be set up. The numerical
efficiency between observed and simulated hydrographs was computed by measuring
RSR, R2, and NSE. The calibration attempt was in good agreement with the observed
counterparts when evaluated graphically using several goodness-of-fit measures (RSR,
R2, and NSE). The average NSE value is greater than 0.7, R 2 is 0.82, and the average
value of RSR (<0.6) placed the model in good shape.
155
Sensitive analysis suggests that runoff in the study area is sensitive to rainfall
depth, impervious percentage, slope and manning coefficients, conduit length, conduit
roughness, storage depth and combined interaction between these parameters
influencing peak flow. Total predictive analysis with 95% confident interval captures
most of the observed discharge data supporting the uncertainty quantification of the
model.
The prior knowledge of the distribution of the parameter plays a significant
impact in the model uncertainty structure. In this study, parameters are considered to
follow uniform distributions.
Future studies include considering surrounding environment response, spatial
variation, management criteria in uncertainty quantification, and cost-benefit analysis
for accommodating the wide range of peak flow, TSS, Nitrate-N, and Orthophosphate-
P considering the occurring probability and duration.
References
1. Rossman, L. A. Storm Water Management Model user's manual version 5.0. United States
Environmental Protection Agency, Water Supply and Water Resources Division, National
Risk Management Research Laboratory, 2010, Cincinnati, OH.
2. Metcalf and Eddy, University of Florida, Water Resources Engineers. 1971. Storm Water
Management Model, Vol. I. Final Report, 11024DOC07/71 (NTIS PB-203289), U.S. EPA,
Washington, D.C.
3. Hamel, P.; Fletcher, T.D. Modelling the impact of stormwater source control infiltration
techniques on catchment baseflow. Hydrol. Process. 2014, 28, 5817–5831
4. Petrucci, G.; Bonhomme, C. The dilemma of spatial representation for urban hydrology
semi-distributed modelling: Trade-offs among complexity, calibration and geographical
data. J.Hydrol. 2014, 517, 997–1007
5. Vrebos, D., Vansteenkiste, T., Staes, J., Willems, P., Meire, P. Water displacement by sewer
infrastructure in the Grote Nete catchment, Belgium, and its hydrological regime effects.
Hydrol. Earth Syst. Sci. 2014, 18, 1119–1136.
156
6. Jewell, T. K., T. J. Nunno, and D. D. Adrian. Methodology for calibrating stormwater mod-
els. Journal of the Environmental Engineering Division, 1978, 104(3), 485-501.
7. Tsihrintzis, V. A. and R. Hamid. Runoff quality prediction from small urban catchments
using SWMM. Hydrological Processes, 1998, 12(2), 311-329, doi:
10.1002/(SICI)10991085(199802)12:2, 311, AID-HYP579>3.0.CO;2-R
8. Baffaut, C. and J. W. Delleur. Calibration of SWMM runoff quality model with expert sys-
tem. Journal of Water Resources Planning and Management, 1990, 116(2), 247-261, doi:
10.1061/(ASCE)0733-9496(1990)116:2(247).
9. Kanso, A., Gromaire, M. C., Gaume, E., Tassin, B., and Chebbo, G. Bayesian approach for
the calibration of models: Application to an urban stormwater pollution model, Water Sci.
Technol., 2003, 47(4), 77–84.
10. Walker, J. “Statistical techniques for assessing water quality effects of BMPs.” J. Irrig.
Drain Eng., 1994, 120(2), 334–347.10.1061/(ASCE)0733-9437(1994)120:2(334)
11. Mailhot, É. Gaume, J.-P. Villeneuve. Uncertainty analysis of calibrated parameter values of
an urban storm water quality model using metropolis Monte Carlo algorithm. Water Science
and Technology, IWA Publishing, 1997, 36 (5), pp. 141. ⟨10.1016 / S0273-1223 (97) 00468-
X⟩ . ⟨Hal-00779528⟩
12. Gaume, E., Villanueve, J. P., and Desbordes, M. “Uncertainty assessment and analysis of
the calibrated parameter values of an urban storm water quality model.” J. Hydrol. (Amster-
dam), 1998, 210(1–4), 38–50.10.1016/S0022-1694(98)00171-1
13. Arabi, M., Govindaraju, R., and Hantush, M. “A probabilistic approach for analysis of un-
certainty in the evaluation of watershed management practices.” J. Hydrol. (Amsterdam),
2007, 333(2–4), 459–471.10.1016/j.jhydrol.2006.09.012
14. Beck, M.B. Water quality modeling: a review of the analysis of uncertainty, Water Re-
sources Research, 1987, 23 (8), 1393–1442.
15. S. Tennøe, G. Halnes, G. T. Einevoll, Uncertainpy: A Python Toolboxfor Uncertainty
Quantification and Sensitivity Analysis in ComputationalNeuroscience, Frontiers in Neu-
roinformatics 12, 2018.
16. Wagener, T.; Gupta, H.V. Model identification for hydrological forecasting under uncer-
tainty, Stoch. Environ. Res. Risk Assess. 2005, 19, 378–387.
17. Moges, E.; Demissie, Y.; Larsen, L.; Yassin, F. Review: Sources of Hydrological Model
Uncertainties and Advances in Their Analysis. Water, 2021, 13, 28.
https://doi.org/10.3390/w13010028
18. Butts, M.B.; Payne, J.T.; Kristensen, M.; Madsen, H. An evaluation of the impact of model
structure on hydrological modelling uncertainty for streamflow simulation, J. Hydrol.,
2004, 298, 242–266.
19. Parker, W. S. Predicting weather and climate: Uncertainty, ensembles and probability. Stud-
ies in History and Philosophy of Modern Physics, 2010, 41, 263–272.
157
20. Rojas, R.; Feyen, L.; Dassargues, A. Conceptual model uncertainty in groundwater model-
ing: Combining generalized likelihood uncertainty estimation and Bayesian model averag-
ing, Water Resour. Res., 2008, 44, 44.
21. Troin, M.; Arsenault, R.; Martel, J.-L.; Brissette, F. Uncertainty of Hydrological Model
Components in Climate Change Studies over Two Nordic Quebec Catchments, J. Hydro-
meteorol., 2018, 19, 27–46.
22. Kuczera, G.; Renard, B.; Thyer, M.; Kavetski, D. There are no hydrological monsters, just
models and observations with large uncertainties. Hydrol. Sci. J., 2010, 55, 980–991.
23. Crspo, P.J.;Feyen,J.;Buytaert,W.;Bücker,A.;Breuer,L.;Frede,H.G.;Ramírez,M.Identify-
ingcontrolsofthe rainfall–runoff response of small catchments in the tropical Andes, J.Hy-
drol., 2011, 407, 164–174.
24. Kiang, J.E.; Cohn, T.A.; Mason, R.R., Jr. Quantifying Uncertainty in Discharge Measure-
ments: A New Approach. In Proceedings of the World Environmental and Water Resources
Congress, 2009, Kansas City, MO, USA, 17–21 May 2009; American Society of Civil En-
gineers: Reston, VA, USA, 2009; pp. 1–8.
25. McMillan, H.; Krueger, T.; Freer, J. Benchmarking observational uncertainties for hydrol-
ogy: Rainfall, river discharge and water quality, Hydrol. Process., 2012, 26, 4078–4111.
26. Anderson, M.P. and W. W. Woessner. The role of the postaudit in model validation. Ad-
vances in Water Resources, 1992, 15, 167-173, doi: 10.1016/0309-1708(92)90021-S
27. Beven, K. J., and Binley, A. M. The future of distributed models: Model calibration and
uncertainty prediction, Hydrol. Processes, 1992, 6(3), 279–298.
28. Dotto, C.; Kleidorfer, M.; Deletic, A.; Rauch, W.; McCarthy, D.; Fletcher, T. Performance
and sensitivity analysis of stormwater models using a Bayesian approach and long-term
high resolution data. Environ. Model. Softw., 2011, 26, 1225–1239
29. Mantovan, P., and Todini, E. “Hydrological forecasting uncertainty assessment: Incoher-
ence of the GLUE methodology.” J. Hydrol., 2006, 330(1–2), 368–381.
30. Stedinger, J. R., Vogel, R. M., Lee, S. U., and Batchelder, R. “Appraisal of the generalized
likelihood uncertainty estimation (GLUE) method.” Water Resour., 2008, Res., 44(12),
W00B06.
31. Vrugt, J. A., Ter Braak, C. J. F., Gupta, H. V., and Robinson, B. A. Equifinality of formal
(DREAM) and informal (GLUE) Bayesian approaches to hydrologic modeling?, Stoch. En-
viron. Res. Risk Assess., 2009b, 23(7), 1011–1026.
32. Kuczera, G., and Parent, E. Monte Carlo assessment of parameter uncertainty in conceptual
catchment models: The Metropolis algorithm, J. Hydrol., 1998, 211(1–4), 69–85.
33. Schoups, G., and Vrugt, J. A. A formal likelihood function for parameter and predictive
inference of hydrologic models with correlated, heteroscedastic, and non-Gaussian errors,
Water Resour., 2010, Res., 46(10), W10531.
158
34. Thiemann, M., Trosset, M., Gupta, H. V., and Sorooshian, S.“Bayesian recursive parameter
estimation for hydrologic models.” Water Resour. Res.,2001, 37(10), 2521–2535.
35. Vrugt, J. A., Gupta, H. V., Bouten, W., and Sorooshian, S. “A Shuffled Complex Evolution
Metropolis algorithm for optimization and uncertainty assessment of hydrologic model pa-
rameters.” Water Resour. Res., 2003, 39(8), SWC 1-1–SWC 1-14.
36. McCuen, R. H., The role of sensitivity analysis in hydrologic modeling. Journal of Hydrol-
ogy, 1973, 18(1), 37-53, doi: 10.1016/0022-1694(73)90024-3.
37. Engel, B., D. Storm, M. White, J. Arnold, and M. Arabi. A hydrologic/water quality model
application. Journal of the American Water Resources Association (JAWRA), 2007, 43(5),
12231236, doi: 10.1111/j.1752-1688.2007.00105.x.
38. Saltelli A, Chan K, Scott E (eds) Sensitivity analysis. Wiley Series in Probability
and Statistics, Wiley, 2000
39. Matthew A. Taddy, Herbert K. H. Lee, Genetha A. Gray, and Joshua D. Griffin. Bayesian
guided pattern search for robust local optimization, Technometrics, 2009, 51:389– 401.
40. Liong, S., W. Chan, and L. Lum,. Knowledge‐based system for SWMM runoff component
calibration. Journal of Water Resources Planning and Management, 1991,117(5), 507-524,
doi: 10.1061/(ASCE)0733-9496(1991)117:5(507).
41. Zaghloul, N. A. Sensitivity analysis of the SWMM runoff-transport parameters and the ef-
fects of catchment discretisation. Advances in Water Resources, 1983, 6(4), 214-223, doi:
10.1016/0309-1708(83)90059-3.
42. Bayes T, Price R. An Essay Towards Solving a Problem in the Doctrine of Chances. By the
late Rev. Mr. Bayes, communicated by Mr. Price, in a letter to John Canton, MA. and F.R.S,
Philosophical Transactions of the Royal Society of London, 1763, 53, 370-418.
43. Jahan K, Pradhanang SM, Bhuiyan MAE. Surface Runoff Responses to Suburban Growth:
An Integration of Remote Sensing, GIS, and Curve Number. Land, 2021; 10(5):452.
https://doi.org/10.3390/land10050452
44. Jahan, K.; Pradhanang, S. Predicting Runoff Chloride Concentrations in Suburban Water-
sheds Using an Artificial Neural Network (ANN). J. Hydrol., 2020, 7, 80, doi:10.3390/hy-
drology7040080.
45. Wood-Pawcatuck, 2000. Wood-Pawcatuck wild and scenic river, Chipuxet River,
http://wpwildrivers.org/the-rivers/chipuxet-river/
46. RI Water Resource Board, 2000. Chipuxet_Specific watershed.
http://www.wrb.ri.gov/documents/Chipuxet_specific%20watershed.pdf
47. Rossman, L.A. Storm Water Management Model User’s Manual, 5th ed.; Environment Pro-
tection Agency: Cincinnati, OH, USA, 2005; pp. 125–137
48. Zoppou, C. Review of urban storm water models. Environ. Modell. Softw., 2001, 16, 195–
231.
159
49. Gironas, J.; Roesner, L.A.; Rossman, L.A.; Davis, J. A new applications manual for the
Storm Water Management Model (SWMM). Environ. Modell. Softw., 2010, 25, 813–814
50. Huber W.C. and Dickinson R.E. Storm water management model, version 4: User’s manual,
1988, EPA- 600/3- 88/001a, U.S. EPA, Georgia.
51. Horton, R.E. An approach toward a physical interpretation of infiltration capacity. Soil Sci.
Soc. Am. 1940, 5, 399–417.
52. Zhao, G.; Pang, B.; Xu, Z.; Du, L.; Zhong, Y. Simulation of urban storm an Dahongmen
drainage area by SWMM. J. Beijing Normal Univ. Nat. Sci. 2014, 50, 452–455.
53. Shi, R.; Zhao, G.; Pang, B.; Jinag, Q.; Zhen, T. Uncertainty Analysis of SWMM Model
Parameters Based on GLUE Method. J. China Hydrol. 2016, 36, 1–6.
54. Nash, J.E.; Sutcli_e, J.V. River flow forecasting through conceptual models part I—A dis-
cussion of principles. J. Hydrol., 1970, 10, 282–290.
55. Krause, P., D. P. Boyle, and F. Bäse. Comparison of different efficiency criteria for hydro-
logical model assessment. In: Advances in geosciences, 2005, 5, pp. 89–97.
56. Dongquan, Z., C. et al. “GISbased urban rainfall-runoff modeling using an automatic catch-
ment-discretization approach: a case study in Macau”. In: Environmental Earth Sciences,
2009, 59 (2), p. 465.
57. Devore, J. L. and K. N. Berk. Modern Mathematical Statistics with Applications, 2007,
Thomson Brooks/Cole. 845 pp. ISBN: 9780534404734.
58. Golmohammadi, Golmar; Prasher, Shiv; Madani, Ali; Rudra, Ramesh. 2014. "Evaluating
Three Hydrological Distributed Watershed Models: MIKE-SHE, APEX, SWAT" Hydrol-
ogy 1, no. 1: 20-39. https://doi.org/10.3390/hydrology1010020Moriasi, D.N.; Gitau, M.W.;
Pai, N.; Daggupati, P. Hydrologic and water quality models: Performance measures and
evaluation criteria. T. Asabe 2015, 58, 1763–1785.
59. Devore, J. L. and K. N. Berk. Modern Mathematical Statistics with Applications, 2007,
Thomson Brooks/Cole. 845 pp. ISBN: 9780534404734.
60. Hastings, W. K. Monte Carlo Sampling Methods using Markov Chains and their Applica-
tions. Biometrika, 1970, 57, 97–109.
61. Smith, A. F. M., and G. O. Roberts. Bayesian Computation Via the Gibbs Sampler and
Related Markov Chain Monte Carlo Methods, Journal of the Royal Statistical Society. Se-
ries B (Methodological), vol. 55, no. 1, 1993, pp. 3–23. JSTOR, www.jstor.org/sta-
ble/2346063. Accessed 3 June 2021.
62. Saltelli A., Making best use of model evaluations to compute sensitivity indices, Computer
Physics Communications, 2002, 145, 280-297.
63. Sobol,I.M., Global sensitivity indices for nonlinear mathematical models and their Monte
Carlo estimates. Math Comput Simulat, 2001, 55(1–3),271-280, doi:10.1016/S0378-
4754(00)00270-6
160
64. Glen G, Isaacs K. Estimating Sobol sensitivity indices using correlations. Environ. Model.
Software, 2012, 37, 157e166.
65. Gramacy, R. and Matt Taddy. Categorical Inputs, Sensitivity Analysis, Optimization and
Importance Tempering with tgp Version 2, an R Package for Treed Gaussian Process Mod-
els, Journal of Statistical Software , 2010, 33, 1-48.
66. Jaynes, E.T. Prior PROBABILITIES. IEEE Transactions on Systems Science and Cyber-
netics, 1968, 4 (3), 227–241.
67. Marshall, L., Nott, D. J. & Sharma, A. A comparative study of Markov chain Monte Carlo
methods for conceptual rainfall-runoff modeling. Water Resour. Res., 2004, 40,
10.1029/2001WR002501
68. Lee, G.Fred and Jones-Lee, Anne. Impact of Municipal and Industrial Non-Hazardous
Waste Landfills on Public Health and the Environment: An Overview. Prepared for Cali-
fornia EPA Comparative Risk Project, Sacramento, CA. 1994. Available at:
http://www.gfredlee.com/cal_risk.htm
69. Elliot, A.H., Trowsdale, S.A. A review of models for low impact urban stormwater drain-
age, Environ. Modell. Softw., 2007, 22.394-405
70. Dongquan, Z., C. Jining, W. Haozheng, T. Qingyuan, C. Shangbing, and S. Zheng.
“GISbased urban rainfall-runoff modeling using an automatic catchment-discretization ap-
proach: a case study in Macau”. In: Environmental Earth Sciences, 2009, 59 (2), p. 465.
161
CHAPTER 5
Manuscript V
Predicting Runoff Chloride Concentrations in Suburban Watersheds Using an
Artificial Neural Network (ANN)
Published in Hydrology 2020, 7, 80
Khurshid Jahan 1 and Soni M Pradhanang 2,*
1 Khurshid Jahan1,*, Soni M Pradhanang1,*, and Russell D. Briggs2
1Department of Geosciences, University of Rhode Island, RI, Kingston, USA 2Department of Sustainable Resources Management, ESF SUNY College of Environ-
mental Science and Forestry, NY, USA
*Correspondence: [email protected] (S.M.P)
162
Abstract: Road salts in stormwater runoff, from both urban and suburban areas, are of
concern to many. Chloride-based deicers [i.e., sodium chloride (NaCl), magnesium
chloride (MgCl2), and calcium chloride (CaCl2)], dissolve in runoff, travel downstream
in the aqueous phase, percolate into soils, and leach into groundwater. In this study, data
obtained from stormwater runoff events were used to predict chloride concentrations
and seasonal impacts at different sites within a suburban watershed. Water quality data
for 42 rainfall events (2016-2019) greater than 12.7 mm (0.5 inches) were used. An
artificial neural network (ANN) model was developed, using measured rainfall volume,
turbidity, total suspended solids (TSS), dissolved organic carbon (DOC), sodium, chlo-
ride, and total nitrate concentrations. Water quality data were trained using the Leven-
berg-Marquardt back-propagation algorithm. The model was then applied to six differ-
ent sites. The new ANN model proved accurate in predicting values. This study illus-
trates that road salt and deicers are the prime cause of high chloride concentrations in
runoff during winter and spring, threatening the aquatic environment.
Keywords: stormwater; artificial neural network (ANN); water quality; road salts;
deicers
163
1. Introduction
Urban areas require the construction of buildings, roads, and parking areas, yet
such urban development causes hydrologic impacts and pollution as pervious surfaces
are made impervious [1]. For safety, given abundant snowfall during the winter season,
most communities in New England use salt or deicing on roads and parking areas. Road
salts or deicing during the winter season are the primary factors for increasing salinity
in surface soils, surface water, groundwater, and runoff. In the USA, an average of 24
million metric tons of road salt is applied each year to roads [2]. It is well-established
that the application of road salts leads to the accumulation of sodium and chloride in
soils and surface waters [3–5], with adverse impacts on downstream aquatic ecosystems
[6]. In fact, when impervious surface areas increase, the areas that need to be deiced
also increase.
Chloride-based deicers [i.e., sodium chloride (NaCl), magnesium chloride
(MgCl2), and calcium chloride (CaCl2)] dissolve in runoff, percolate into soils, and leach
into groundwater. Chloride from chloride-based deicers does not efficiently precipitate
or biodegrade but is absorbed by mineral/soil surfaces [7]. Although winter road deicing
is an essential service for urban areas in USA (especially in the upper Midwest and
Northeast), it contributes to a significant increase in chloride concentration [8]. The
United States Geological Survey (USGS) (2014) conducted a temporal, seasonal, and
environmental analysis of chloride concentrations in urban areas and assessed effects
on water quality and the environment, especially on aquatic organisms across the USA
164
[8]. This study concluded that there is an increasing trend of high chloride concentra-
tions in urban areas due to expansion of impervious cover that requires deicing.
An increasing trend in chloride concentrations in some US rivers is shown in
Figure 1 and attributed to increased usage of road salt; the trend is positive in New
England (USGS, 2014). The increasing salinity not only threatens aquatic ecosystems,
but also contributes to corrosion in water distribution systems. Salinity can increase
even with little snowfall due to efficient transfer of chloride to wastewater discharge or
septic systems [8]. Areas that have no snow, such as Florida, also showed increasing
salinity. In the case of Florida, less than normal rainfall since 1990 combined with
groundwater level decline due to over pumping explain the observed increasing salinity
[9–11]. The characteristics and degradation of urban runoff quality and its impact on
the environment largely depend on the urban land-use practices, site geology, and hy-
drogeology. The large quantity of road salt that is applied every year for snow removal
in the USA is one of the major contributors to declining stormwater quality.
165
Figure 1. Chloride concentration trends from 1992 to 2012 in the USA show regional
differences (modified from United States Geological Survey Report, 2014)
Recently, several studies have applied artificial neural network (ANN) methods to
predict resulting water quality based upon input variables [12]. Since 1990, ANN has
been applied in many fields, including environmental sciences, ecological sciences, and
water engineering [13]. According to Haykin (1999) [14], ANN is highly capable in
modeling nonlinear system estimation and is highly adaptable. ANN allows precise pre-
dictions of the target parameter for specific materials or stages [14, 15].
In this study, an ANN model is developed with a back-propagation algorithm. The
back-propagation algorithm incorporates highly nonlinear relationships [15]. The ANN
model was developed for rapid calculation and prediction of selected water quality var-
iables at any location of interest. Within the model, unknown parameter weights are
166
adjusted to obtain the best correlation between appropriate input parameters or a histor-
ical set of model inputs and the corresponding outputs [16]. This study provides the
ANN modeling method needed to simulate and forecast chloride concentrations in run-
off. The aim of the study is to (i) develop an ANN model of the system trained using a
small data set, (ii) obtain the best-fit models for predicting chloride concentrations using
data from monitoring sites, (iii) evaluate the ANN model performance using 3 years
(2016-2019) of observed data versus predicted data from the model, and (iv) determine
the accuracy of the ANN model performance. The model also assesses the impact of
road salt applications through assessment of a spatial density distribution focused on
probable high chloride concentration in an area.
2. Materials and Methods
A three-year study on the effectiveness of the stormwater best management prac-
tices (BMPs) was conducted in the Chipuxet watershed of South Kingstown, Rhode
Island, USA (Figure 2).
Figure 2. The map illustrates the location of the study area in Rhode Island, USA. Red
bounding box shows the exact location of the study site shown above (Google Earth
screenshot- on the right).
167
An overview of the chloride concentration for three years (2016–2019) from six
sites represented in Figure 3. Based on the analysis of the stormwater runoff quality
data, an apparent seasonal variation of chloride concentration is observed (Figure 3).
The higher concentration of chloride was found during winter and early spring season
(at the tail end of winter). Our study results are highly consistent with the study
conducted by the USGS (2014) [7]. The USGS study showed the increasing trend of
chloride concentration in the New England zone, and our data also provided the same
impression. As stated above, high chloride concentration (0.8–197.9) mg/L was seen on
the winter samples for all the sites. Site 5 and 6 are in close proximity to the parking lot;
parking lots are highly impermeable, and exhibited higher chloride concentration (0.9–
197.9) mg/L than the remaining four sites that are close to the agricultural field (Figure
3). The chloride concentration data were then used to investigate the future scenario
through the ANN model. The steps used to develop the model include the choice of
model performance criteria, preprocessing of available data, the selection of appropriate
model inputs, and network structure.
Figure 3. Variability in observed chloride concentrations in runoff are shown, from
2016 to 2019.
168
2.1. Artificial Neural Network (ANN)
The ANN concept was first introduced by McCulloch and Pits in 1943, and ANN
applications in research areas started with the back-propagation algorithm for feed-for-
ward ANN in 1986 [17,18]. ANNs consist of multiple layers; basic layers are common
to all models (i.e., input layer, output layer), and several hidden layers may be needed
(located between the input and output of the algorithm) [19]. Each of the layers in an
ANN consists of a parameterizable number of neurons. Neurons are activation functions
of adjustable weight based on a priori and domain knowledge [20]. In this study, an
ANN with three different learning approaches, such as back-propagation neural network
(Levenberg–Marquardt), curve fitting, and density distribution, were considered and
adapted to develop the final model for predicting and validating chloride concentration
in the runoff. The overall objective of the ANN model was to reduce model error, E,
defined as
𝐸 =1
𝑝∑ 𝐸𝑝
∞
𝑛=1
(1)
where p = total number of training patterns and Ep = error for the training pattern p.
Ep is calculated with the following equation:
𝑬𝒑 =𝟏
𝟐∑(𝒐𝒌 − 𝒕𝒌)𝟐
𝑵
𝒌=𝟏
(2)
where N = total number of output nodes; ok = network output at the kth output node;
and tk = target output at the kth output node. Additional details on the mechanics of this
study are described below.
169
2.2. Back Propagation (BP) Algorithm
Back propagation (BP) is the most widely used method for training multiplier feed-
forward networks. Before BP, almost all of the networks used non-identifiable complex
binary nonlinear methods to self-test, such as step functions, statistical time series mod-
els, auto regressive integrated moving average (ARIMA), and moving average (MA)
[21,22]. Layered networks from BP algorithm are useful for nontrivial calculations with
the different attractive features such as fast response, fault tolerance, the ability to ob-
servation from input parameters, and the capability to generalize beyond the training
data. A set of input variables is needed to train the network to match desired outputs,
with a function that measures the “value” of differences between network outputs and
desired values [22]. The most straightforward implementation of the standard BP algo-
rithm adjusts the network weights and biases in the target direction, and this adjustment
helps to achieve the model accuracy.
The back-propagation neural network structure consists of two or more layers of
neurons, and network weights connect all the neurons [22,23]. The final output is cap-
tured by the developed system, when input data pass through the hidden layers to the
output layer. This process is shown in Equation 3.
𝑢𝑗 = ∑ 𝑤𝑗𝑖𝑥𝑖
𝑝
𝑖=1
(3)
In Equation 3, Wji represents the weights that connect two neurons i and j, and every
neuron calculates its output based on the number of stimulations it obtains from the
given input vector xi, where xi is the input of neuron i. The “net input” of a neuron is
170
measured as the weighted sum of total number of input variables, and the output of the
neuron is based on the active function (active function indicates the magnitude of the
“net input” [24]). BP is a training algorithm consisting of two steps: first, values are fed-
forward, and second, error is calculated and propagated back to the earlier layers.
2.2. Curve Fitting Algorithm
Polynomial models for curves are given by equation (4)
𝑦 = ∑ 𝑝𝑖 𝑋𝑛+1−𝑖
𝑛+1
𝑖=1
(4)
where n+1 and n represent the order of the polynomial and the degree of the polynomial,
respectively, and the range of n is 1 ≤ n ≤ 9.
A third-degree (cubic) polynomial equation (5) is also pertinent
𝛾 = 𝑝1 𝑥3 + 𝑝2 𝑥
2 + 𝑝3𝑥 + 𝑝4 (5)
Polynomials (as in Equation 5) are frequently used when a simple experiential
model is required, or when a model needs interpolation or extrapolation. The main ad-
vantages of polynomial fitting comprise cognitive flexibility for the most complex and
large data sets [25]. The polynomial curve fitting process is simple and linear [25].
2.4. Density Distribution algorithm
Distribution fitting applies to model the probability distribution of a single variable.
The normal distribution is the most applied statistical distribution approach in research.
171
In this study, we calculated the probability density function (PDF) of the predicted chlo-
ride concentration. The following equation (6) is used in this study to specify the prob-
ability of the predicted output [25].
𝑓(𝑥) = 1
√2𝜋𝜎exp(
−(𝑥 − µ)2
2𝜎2, > 0 (6)
Where 𝜎 is standard deviation, 𝜎2 is variance, and µ is mean.
2.5. Model Structure
In recent years, neural network technology has been adopted in water quality
prediction, in which the back-propagation network is commonly used [26,27]. The
model created in this study is a BP neural network model with a single hidden layer
(Figure 4). In this ANN, the input layer is R, the hidden layer is a1, the output layer is
a2, the weight matrix of the input layer is IW1.1, and the weight matrix from the hidden
layer to the output layer is LW2.1. The threshold values of the hidden and output layers
are b1 and b2, respectively. f1 and f2 are the neuron transfer functions of the hidden and
output layers, respectively.
As theoretically verified, the BP model as shown in Figure 4 can handle any
nonlinear function with minimum interruptions at any accuracy as long as there are a
sufficient number of neurons in the hidden layer of the model and the number of neurons
are determined based on a priori and domain knowledge [28]. The proposed ANN model
(Figure 5) has two hidden layers of sigmoid neurons that are followed by an output layer
of linear neurons. Sigmoid neuron makes the output smoother, and a small change in
172
the input only causes a little variation in the output [29]. This network system can be
utilized as a general function approximator. It can estimate any function with a finite
number of discontinuities given sufficient neurons in the hidden layer [30,31].
Figure 4. This diagram illustrates the structure of a conventional feed-forward back-
propagation neural network model
In the developed model structure (Figure 5), the input and output variables are
established for the evaluation of water quality. Multiple layers of neurons of the devel-
oped ANN structure with nonlinear transfer functions let the network assess nonlinear
and linear relationships that underlay input and output vectors. The final output layer
is linear, allowing the network to produce values outside the range -1 to +1.
Figure 5. This diagram illustrates the structure of the artificial neural network (ANN)
model developed in this work.
173
2.5.1. ANN Parameter Selection: Hidden Layers and Nodes
The number of hidden layers in ANN model is usually determined by trial and error.
The number of training set samples should be higher than the number of synaptic
weights, a rule of thumb for defining the number of hidden nodes [31,32]. Most ANN
modelers usually consider a one-hidden-layer network (i.e., the number of hidden nodes
is between input nodes and (2*(input nodes) + 1) [30]). However, hidden nodes should
not be less than the maximum of one third of input nodes and the number of output
nodes. The optimum value of hidden nodes is fixed by trial and error. Networks with
minimum number of hidden nodes are usually preferred due to better generalization
capabilities and fewer overfitting problems. For this study, a trial and error procedure
for the number of hidden node selection was carried out by gradually changing the num-
ber of hidden layer nodes.
2.5.2. ANN Parameter Selection: Learning Rate and Momentum
The functions of the learning rate and momentum parameters are to enhance model
training and ensure that error is reduced. There is no precise rule for the selection of
values for these parameters. Here, the learning rate was controlled by internal valida-
tion: after the end of each epoch, the weights were updated. The number of epochs with
the smallest internal validation error indicates which weights to select [33]. In this study,
the learning rate for the weights connecting input layer and the hidden layer was set at
double the size of the learning rate for the weights connecting the hidden layer to the
output layer, to increase the rate of network convergence. The momentum was initially
fixed at a value of 0.015, with the number of hidden nodes initially estimated as number
of input nodes +1, similar to the study conducted by Maier and Dandy (1996) [34].
174
2.5.3. ANN Parameter Selection: Initial Weights
When the weights of a network is trained by BP, it is always better to initialize from
small, non-zero random values, although ANN modelers can start over with a different
set of initial weights [22,23]. In this study, the amplitude of a connection between two
nodes (synaptic weights) of the proposed ANN networks was adjusted using the nor-
mally distributed random numbers having the range from-1 to 1.
2.5.4. ANN Parameter Selection: Selection of Input Variables
In an ANN, one of the main tasks is to determine the model input variables that
significantly affect the output variable(s). The selection of input variables is usually
related to a priori knowledge of output variables, inspections of time series plots, and
statistical analysis of potential inputs and outputs. In this study, the input variables for
the present neural network modeling were selected based on a statistical correlation
analysis of the runoff quality data, the prediction accuracy of water quality variables,
and domain knowledge. Domain knowledge is the specific field knowledge that sup-
ports interpretation of data when applying machine learning algorithms like regression,
stepwise approach, and classification to predict some test data [35]. In a stepwise ap-
proach, separate networks are trained for each input variable [36]. We experimented
with the water quality variables included in the parameters listed above in several mod-
els to both identify the optimal predictive model and reduce the monitoring cost by in-
cluding fewer input parameters. After selecting the appropriate input variables, the next
step involved determining appropriate lags for each of these variables. The selected ap-
175
propriate input variables are rainfall amount, duration of rainfall, intensity, runoff coef-
ficient, runoff depth, peak discharge, turbidity, total suspended solids (TSS), dissolved
organic carbon (DOC), sodium, chloride, and total nitrate concentrations that were used
to develop the ANN model. Appropriate lags are needed for complex problems, where
the numbers of potential inputs are significant, and no a priori knowledge is available.
Lags allow the model to establish significant connection or bonding between the output
and the input variables. By doing so, the best network performance is retained, and the
effect of adding each of the remaining inputs in turn is assessed. The correlations be-
tween the input variables and output variables are computed separately for each lagged
input variable [37]. In this study, optimal networks for each of these combinations were
obtained with these time-lagged variables, and the results were compared with the target
dataset.
2.5.5. ANN Parameter Selection: Data Partition
It is essential to divide the data set in such a way that both training and overfitting
test data sets are statistically comparable. The test set should be approximately 10–30%
of the size of the training set of data [38]. In this study, the water quality data were
divided into three partitions: the first set contained 70% of the records used as a training
set, the second test contained 15% of the records and was used as an overfitting test set,
and the rest of the data (15%) were used as the validation set. This process is necessary,
because the efficiency of the developed neural network model is highly dependent on
the quantity and quality of the data as stated by Palani et al., 2008 [18].
176
2.5.6. ANN Parameter Selection: Model Performance Evaluation
The model's efficiency was evaluated using the root mean square error (RMSE, see
Equation 7), the mean absolute error (MAE, see Equation 8), and R2 (see Equation 9)
[39]. Scatter plots and time series plots were used for visual comparison of the observed
and predicted chloride concentrations values.
R2 values of zero indicate that the observed mean is as good a predictor as the
model, R2 value of one represents a perfect fit, and a negative R2 value reflects a better
predictor than the model [40]. Depending on the sensitivity of water quality parameters
and any mismatch between the forecasted and measured water quality variables, one
can decide whether the predictive power of the ANN model is accurate enough to inform
crucial decisions regarding data usage.
RMSE = √1
𝑁∑(𝑋𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 − 𝑋𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑)2 (7)
MAE = 1
𝑁 ∑ |𝑋𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 − 𝑋𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑| (8)
𝑅2 = 1 − 𝐹
𝐹0 (9)
F and F0 could be described using following two equations.
F = ∑(𝑋𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 − 𝑋𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑)2 (10)
F0 = ∑(𝑋𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 − 𝑋𝑀𝑒𝑎𝑛 𝑂𝑏𝑠𝑒𝑟𝑣𝑒𝑑)2 (11)
Where N is the total number of observations.
177
The other primary criterion used to select the optimum ANN model was the sum of
square error (SSE), determined from the following empirical equation:
𝑆𝑆𝐸 = ∑ 𝜔𝑖
𝑛
𝑖=1
(𝑦𝑖 − ŷ𝑖)2 (12)
where wi are the weights and yi and ŷi are the observed response value and the fitted
response value, respectively.
The weights determine how much each response value influences the final pa-
rameter estimates. A high-quality data point influences the fit more than a low-quality
data point. Weighting data is recommended if the absolute weights are known, or if
there is good cause for weighting data differently.
3. Results and Discussion
3.1. Model Output
The BP ANN architecture was applied to five hidden layers with different acti-
vation functions and initial weights of 0.3, optimum learning rate (0.1), and momentum
(0.015), as described in Sections 2.5.2 and 2.5.3. The proposed ANN model was de-
signed considering 11 input variables from 42 storm events. The sensitivities of the in-
put parameters for the chloride concentration prediction are smaller than those used for
the validation dataset. An individual ANN model was run for each of the sites consid-
ering the same ANN model structure shown in Figure 5 and the input parameters. The
parameters that produce the ‘‘best results” for all sites (Table 1) were then used as the
final chloride concentration prediction. The model output or the performance of the
ANN model was evaluated based on the R2 values for training, validation, and testing.
178
R2 values for each of the sites were similar for the three data partitions, as indicated in
Table 1. The weights are methodically changed by the learning algorithms such that for
a given input, the difference between the ANN output, and the actual output was small.
The developed ANN model with nine hidden nodes was considered optimal here, con-
sidering the output (Table 1 and 2). The optimum network parameters associated with
the model output are presented in Tables 1 and 2. Validation errors were calculated after
the optimization of the network parameters and the topology. Error was calculated in
two ways. First, the cross validation was applied. In this method, data were separated
into three parts: training (70%), validation (15%), and testing (15%). The output of the
first technique is shown in Table 1. Secondly, the ANN predicted outputs were validated
using curve fitting technique. Curve fitting analysis showed a good fit between the tar-
geted or observed and the predicted value (Figure 6). The model outputs were consid-
ered acceptable based on the R2 values for training, validation, and testing. In general,
the accuracy of the model can be improved by adding data to the validation step or to
input variables.
Table 1. ANN model output for all the sites.
Location Performance (Epoch) Training (%) Validation (%) Testing (%)
Site_1 9 95 87 88
Site_2 5 100 72 95
Site_3 5 100 52 93
Site_4 3 99 97 93
Site_5 3 99 78 91
Site_6 2 97 83 88
179
3.2. Curve Fitting Analysis
Curve fitting analysis examines the relationship between target output and the
model output. The fit between the target and predicted values were represented for all
six sites (Figure 6). Except for site number 1, all the other sites had the best fitting
between the two datasets (target and model output). This fulfilled the aims of applying
the polynomial bi-square fitting for the presence of concentrated chloride. Polynomial
fitting with a high-order polynomial uses the large predictor values as the basis for a
matrix, which often creates scaling problems [25]. In this study, most of the analyzed
chloride concentration range is from 1 to 20 mg/l, but during the winter these ranges are
from 25 to 200 mg/l. No axis range modifications were made here; axes ranges were
kept as appropriate for the chloride concentration range. A reasonably good match be-
tween the output from the developed ANN model and the curve fitting output was ob-
tained for the sites. To illustrate this, a prediction boundary is provided in Figure 6 for
every site with a 95% confidence interval.
180
Figure 6. Curve fitting assessment between ANN output and target data of six different
sites. (a) Site 1, (b) Site 2, (c) Site 3, (d) Site 4, (e) Site 5, and (f) Site 6. Site 1, 3, 4 and
5 showed the similar trend and data were not scattered. On the other hand, site 2 and 6
has scattered data point but 80% of dataset were in the confidence interval.
181
Table 2. Summary of the curve fitting assessment.
File Name SSE R-Square Adj R -sq RMSE
Site_1 1.96 0.99 0.99 0.22
Site_2 171.58 0.93 0.93 2.07
Site_3 7.9 0.99 0.99 0.44
Site_4 231.3 0.99 0.99 2.4
Site_5 473.6 0.98 0.98 3.4
Site_6 4051 0.93 0.92 10.06
Curve-fitting information regarding the developed ANN model output is pre-
sented in Table 2; note the SSE, R-Square, and RMSE values are robust. All the ANN
models constructed using nine nodes in the hidden layer produced the lowest SSE.
Therefore, the site 3 ANN model showed the lowest value of SSE, and the model for
Site 6 showed the highest SEE value having the lowest R-square value. In summary, all
the ANN models developed here for six sites showed an acceptable range for all the
model justification factors. The predicted values are reasonable for all the sites. Curve
fitting assessment is a cross-validation approach, proving the accuracy of the developed
ANN model. No significant difference in the R2 values can be seen in the Table 2.
3.3. Density Distribution of the Predicted Chloride Concentration
The density distribution was applied to show the spatial distribution of the pre-
dicted chloride concentration values (Figure 7). Usually, the continuous data values tend
to cluster around the mean in a normal distribution, and the farther a value is from the
182
mean, the more uncertain it is. The tails are asymptotic, which implies that they ap-
proach but never meet the X-axis. In this study, density distribution curve fitting resulted
in a 95% confidence interval. This 95% confidence interval means that 95% of values
fall within two standard deviations from the mean.
Considering the six study sites, four sites are close to an agricultural field and
two are close to a parking lot; the spatial distribution range of the agricultural field sites
is smaller than those close to the parking lot. For sites 5 and 6, more than 70% of pre-
dicted chloride concentrations are clustered at the mean value, and the peak is wide. On
the other hand, the density distribution for sites 1, 2, 3, and 4 cover less than 50% pre-
dicted chloride concentration values. Sites 5 and 6 have higher chloride concentrations,
because they received salt/chloride from both sides (from the road and the parking lot).
The highest spatial range was observed for sites 5 and 6. These results are consistent
with the fact that sites 5 and 6 received the chloride from both sides.
3.4. Cross Validation Based on Snow and Precipitation Events
Predicted data were cross validated by taking advantage of temporally close
snow and precipitation events. Chloride concentrations were high in storm events that
followed severe snow events. Here, we analyzed data from US climate data repositories
with three years (2017, 2018, and 2019) of snow and daily precipitation [41] (Figure 8).
183
Figure 7. Density distributions (ANN outputs) for chloride concentrations at all sites (a) Site 1, (b)
Site 2, (c) Site 3, (d) Site 4, (e) Site 5, and (f) Site 6 reveal shifting means. Spatial distribution covered
<50% for site 1, 2, 3 and 4 whereas site 5 and 6 covered >70% of predicted chloride.
184
Figure 8. Precipitation and snow events (a) 2017, (b) 2018, and (c) 2019 in South Kings-
town, RI. (a) total four significant snow events (>10 cm) and low to moderate precipi-
tation events appeared during winter in 2017; (b) three significant snow (>10 cm) events
(among them one was the worst >25 cm) and low to moderate with high frequency
(c)
185
precipitation occurred during winter in 2018; and (c) only one significant snow and low
to moderate precipitation happened during the winter in 2019.
Figure 9. Predicted concentrations for three sites reflect heavy Cl inventory in Winter
2018.
Rhode Island receives approximately 94 cm of snow every year, but snowfall
totals can vary significantly from town to town, even though the state is relatively small
and the terrain is flat [42]. Moreover, the number of snow events vary widely from year
to year. Both salt and sand are used on roads during snow events. Given additional chlo-
ride derived from snow removal deicers, Rhode Island collects a considerable amount
of chloride in its surface water and groundwater, and the salts accumulate in the soils
and later percolate into the groundwater. The groundwater becomes saltier every year,
since chloride is a dissolved phase and cannot be removed naturally from the water [42–
44]. Seventy percent of the salt applied to roads stays within the region’s watershed
[45]. In this study, chloride data for rain events occurring immediately after snow events
in 2017, 2018, and 2019 showed that chloride concentrations increased.
186
Runoff pollutant concentrations also depend on the size and duration of precip-
itation event. Both longer duration storms and storms of high intensity impact chloride
concentrations. Longer period storms and high intensity storms can dilute the pollutant
concentration. For example, on January 4th, 2018, a 220 mm snow event preceded 830
mm of rainfall on January 13th, 2018, in a storm of 4 hours’ duration. Chloride concen-
trations were highest after the January 13th storm events. On the other hand, a 147 mm
snow event occurred on March 10th, 2017, followed by March 17th, 2017 storm of 7
hours’ duration. The detected chloride concentration from March 17th, 2017 storm
events were not significant relative to 2018 storm events. In the ANN model predicting
chloride concentration, runoff volume and duration of the rainfall are considered as a
positive sensitivity parameter. The March 2017 storm pair, which did not lead to ele-
vated chloride, could reveal the counteracting impacts of street density, street width,
and location of the street. Furthermore, the accumulated chloride could be attributed
primarily to the amount of salt application, which varies from event to event.
Chloride concentrations greater than 600 ppt (1 mg/l = 1ppt) are considered
harmful for freshwater aquatic life and for the groundwater in general [46]. The devel-
oped ANN model prediction (Figure 9) and probability density output (Figure 7) indi-
cated that aquatic habitats at sites 5 and 6 are at risk. State planners need to take neces-
sary action regarding the implications of road salting and snow removal.
4. Conclusion
In this study, a new ANN model is developed to predict elevated chloride con-
centrations due to road salt and deicer applications in a suburban watershed, based on
187
three years of data (2016-2019) collected at six study sites. Study sites are close to ag-
ricultural land (Sites 1, 2, 3, and 4) and an impervious parking lot (Sites 5 and 6). Sea-
sonal variation is evident in the three years of collected data. For the ANN model, input
variables were derived from the hydrometeorological database, stormwater runoff qual-
ity, key network parameters, and network topology. Preliminary ANN models were con-
structed using a subset of all data (for 42 storm events from 2016 to 2019) where it
covered all four seasons (15 winter events, 6 spring events, 7 summer events, and 15
fall events). A series of sensitivity analyses were considered to determine the relative
significance of input variables used in the ANN models. Applying the BP algorithm,
developed ANN models showed a good fit between observed and predicted data (about
91%). Model accuracy was initially optimized using a cross-validation approach, and
the developed model offers an appropriate and time-efficient approach to constraining
the target water quality parameter. The curve fitting assessment resulted in a 95% con-
fidence interval, used here as cross validation of ANN outputs, and provided an opti-
mum summary for every site. The predicted ANN outcome could be more significant
or could be trained better if the study duration were longer than three years and/or in-
volved more frequent events. This study focused on the winter season because of the
amount of road salt applied to the impervious surfaces, generating high concentrations
of chloride in runoff water. The presence of chloride in non-winter season data is neg-
ligible compared to the winter season, but the detection of chloride could be due to the
use of fertilizer in the agricultural zone. According to the best-fit results, chloride in the
study area is mostly affected by the rain and snow that occurred during the winter sea-
188
son, and chloride concentration depends on storm duration, intensity, and runoff vol-
ume. We propose neural network modeling as an effective tool for water quality param-
eter prediction. Finally, density distribution analysis revealed the spatial distribution
(the amount of clustered data value around the mean) of the chloride concentration.
Density distributions showed about 70% of clustered value is detected for Sites 5 and 6,
due to the nearby parking lot. The other four sites, which are close to the agricultural
zone, cover less than 50%. These findings again point to used road salt as the main agent
of chloride delivery to the groundwater. ANN modeling of environmental data has great
potential in future work on improved prediction of chloride and other pollutant concen-
trations, and provides a useful tool for water resource and environmental managers.
Author Contributions: KJ and SMP, conceptualization; KJ, data collection, analysis,
and writing the original draft preparation; SMP, writing, review, and editing, SMP,
funding acquisition. All authors have read and agreed to the published version of the
manuscript.
Funding: The research study is funded by the Rhode Island Department of Transporta-
tion, SPR-234-2362, and the University of Rhode Island.
Acknowledgments: Authors would like to show their gratitude to the Rhode Island
Department of Transportation (RIDOT) for their financial support. In addition, we also
thank Dr. Gavino Puggioni and Dr. Nisa Khan for their valuable comments and recom-
mendations. We are also thankful to Dr. Thomas Boving for site preparation help. Eng-
lish editing assistance was received through Dr. Dawn Cardace.
189
Conflicts of Interest: The authors declare no conflicts of interest. The funders had no
role in the design of the study; in the collection, analyses, or interpretation of data; in
the writing of the manuscript; or in the decision to publish the results.
References
1. Li, D.; Wan, J.; Ma, Y.; Wang, Y.; Huang, M.; Chen, Y. Stormwater Runoff Pollutant
Loading Distributions and Their Correlation with Rainfall and Catchment Characteristics
in a Rapidly Industrialized City. PLoS ONE 2015, doi:10.1371/journal.pone.0118776.
2. Koleskar, R.K.; Mattson, N.C.; Peterson, K.P.; May, W.N.; Prendergast, K.R.; Pratt, A.K.
Increase in wintertime PM2.5 sodium and chloride linked to snowfall and road salt appli-
cation. Atmos. Environ. 2018, 177, 195–202.
3. Dugan, H.A.; Bartlett, S.L.; Burke, S.M.; Doubek, J.P.; Krivak-Tetley, F.E.; Skaff, N.K.;
Summers, J.C.; Farrell, K.J.; McCullough, I.M.; Morales-Williams, A.M.; et al. Salting our
freshwater lakes. Proc. Natl. Acad. Sci. USA 2017, 114, 4453–4458,
doi:10.1073/pnas.1620211114.
4. Jackson, R.B.; Jobbagy, E.G. From icy roads to salty streams. Proc. Natl. Acad. Sci. USA
2005, 102, 14487–14488, doi:10.1073/pnas.0507389102.
5. Kelting, D.L.; Laxson, C.L.; Yerger, E.C. Regional analysis of the effect of paved roads on
sodium and chloride in lakes. Water Res. 2012, 46, 2749–2758, doi:10.1016/j.wa-
tres.2012.02.032.
6. Findlay, S.E.G.; Kelly, V.R. Emerging indirect and long-term road salt effects on ecosys-
tems. Year Ecol. Conserv. Biol. 2011, 1223, 58–68, doi:10.1111/j.1749-
6632.2010.05942.x.
7. Levelton Consultants Limited. Guidelines for the Selection of Snow and Ice Control Mate-
rials to Mitigate Environmental Impacts; National Cooperative Highway Research Pro-
gram, American Association of State Highway, and Transportation Officials; Transporta-
tion Research Board: Washington, DC, USA, 2008; Volume 577.
8. USGS. Evaluating Chloride Trends Due to Road-Salt Use and Its Impacts on Water Qual-
ity and Aquatic Organisms; USGS: Reston, VA, USA, 2014.
9. Salinity Network. FWRMC Salinity Network Workgroup, Watershed Monitoring Section
Quick Links. 2019. Available online: https://floridadep.gov/dear/watershed-monitoring-
section/content/salinity-network#:~:text=Over%20the%20past%20several%20dec-
ades,of%20groundwater%20from%20our%20aquifers (May 10, 2019).
10. Florida Salinity Network Workgroup. Groundwater Level Conditions for the Upper Flori-
dan Aquifer Based on Percentile Ranks Pilot Study for May, 2010. 2015. Available online:
http://publicfiles.dep.state.fl.us/DEAR/DEARweb/FWRMC/FWRMC%20Salnet_docu-
ment_SNWGWLPRIPilotFinal_cv%20ada%203-27-15.pdf (September 2015).
190
11. Status and Trends Report, SFNRC Technical Series 2012:1, Salinity and Hydrology of
Florida Bay, Status and Trends 1990–2009. South Florida Natural Resources Center, Ev-
erglades National Park: Homestead, FL, USA. Available online:
https://www.nps.gov/ever/learn/nature/upload/SecureSFNRC2012-1LoRes.pdf (January
2012).
12. Kumar, A.; Sharma, M.P. Assessment of water quality of Ganga river stretch near Ko-
teshwar hydropower station, Uttarakhand, India. Int. J. Mech. Prod. Eng. 2015, 3, 82–85.
13. Juair, H.; Sharifuddin, M.Z. Sensitivity analysis for water quality index (WQI) prediction
for Kinta River. World Appl. Sci. J. 2011, 14, 60–65.
14. Haykin, S. Neural Networks: A Comprehensive Foundation; Prentice Hall: Upper Saddle
River, NJ, USA, 1999.
15. Vasanthi, S.; Kumar, A. Application of artificial neural network techniques for predicting
the water quality index in the parakai lake, Tamil Nadu, India. Appl. Ecol. Environ. Res.
2018, 17, 1947–1958.
16. Najah, A.; El-Shafie, A.; Karim Amr, O.A. Application of artificial neural networks for
water quality prediction. Neural Comput. Appl. 2013, 22, 187–201.
17. Rumelhart, D.E.; Hinton, G.E.; Williams, R.J. Learning internal representations by error
propagation. In Parallel Distributed Processing; Rumelhart, D.E., McClelland, J.L., Eds.;
MIT Press: Cambridge, MA, USA, 1986.
18. Palani, S.; Liong, S.; Tkalich, P. An ANN application for water quality forecasting. Mar.
Pollut. Bull. 2008, 56, 1586–1597.
19. Deep Artificial Intelligence, Hidden Layer. Available online: https://deepai.org/machine-
learning-glossary-and-terms/hidden-layer-machine-learning (December 2012).
20. Obuchowski, A. Understanding Neural Networks 1: The Concept of Neurons. 2019. Avail-
able online: https://becominghuman.ai/understanding-neural-networks-1-the-concept-of-
neurons-287be36d40f (March 23, 2019).
21. Sing, K.; Basant, A.; Malik, A.; Jain, G. Artificial neural network modeling of the river
water quality—A case study. Ecol. Model. 2008, 220, 888–895,
doi:10.1016/j.ecolmodel.2009.01.004.
22. Gavindaraju, R.S. Artificial neural network in hydrology, II: Hydrologic application.
ASCE task committee application of artificial neural networks in hydrology. J. Hydrol.
Eng. 2000, 5, 124–137.
23. Wang, Q.H. Improvement on BP algorithm in artificial neural network. J. Qinghai Univ.
2004, 22, 82–84.
24. Kuo, J.T.; Hseih, M.H.; Lung, W.S.; She, N. Using artificial neural network for reservoir
eutrophication prediction. Ecol. Model. 2006, 200, 171–177.
191
25. Mathworks. Matlab User’s Guide (for Use with MATLAB); MathWorks, Inc.: Natick, MA,
USA, 2001.
26. Lee, T.L.; Jeng, D.S. Application of artificial neural network for long-term tidal predic-
tions. Ocean Eng. 2002, 29, 1003–1022.
27. Moon, S.K.; Woo, N.C.; Lee, K.S. Statistical analysis of hydrographs and water table fluc-
tuation to estimate groundwater recharge. J. Hydrol. 2004, 292, 198–204.
28. Wu, Z.S.; Xu, B.; Yokoyama, K. Decentralized parametric damage detection based on neu-
ral networks. Comput. Aided Civ. Infrastruct. Eng. 2002, 17, 175–184.
29. Kumar, N. Sigmoid Neuron—Building Block of Deep Neural Networks, Towards Data
Science. 2019. Available online: https://towardsdatascience.com/sigmoid-neuron-deep-
neural-networks-a4cd35b629d7 (March 7, 2019).
30. Xu, B.; Wu, Z.S.; Yokoyama, K. Response time series-based structural parametric assess-
ment approach with neural networks. In Proceedings of the 1st International Conference
on Structural Health Monitoring and Intelligent Infrastructure (SHMII-1 2003), Tokyo, Ja-
pan, 13–15 November 2003; Swets & Zeitlinger: Lisse, The Netherlands, 2003; pp. 601–
609.
31. Tarassenko, L. A Guide to Neural Computing Applications; Arnold Publishers: London,
UK, 1998.
32. Hechit-Neilsen, R. Kolmogorov’s mapping neural network existence theorem. In Proceed-
ings of 1st IEEE International joint Conference of Neural Network 1987; Institute of Elec-
trical and Electronics Engineers: New York, NY, USA, 1987.
33. Tsoukalas, L.H.; Uhrig, R.E. Fuzzy amd Neural Approaches in Engineering; Wiley Inter-
science: New York, NY, USA, 1997; 587p.
34. Maier, H.R.; Dandy, G.C. The use of artificial neural networks for the predictionof water
quality parameters. Water Resour. Res. 1996, 32, 1013–1022.
35. Anand, S. Why Domain Knowledge Is Important in Data Science, medium.com. 2019.
Available online: https://medium.com/@anand0427/why-domain-knowledge-is-im-
portant-in-data-science-anand0427-3002c659c0a5 (March 18, 2019).
36. Yang, X.; Gandomi, A.; Talatahri, S.; Alavi, A.H. Metaheuristics in Water, Geotechnical
and Transport Engineering; Zali, M.A., Retnam, A., Eds.; Elevier Inc.: Amsterdam, The
Netherlands, 2013; ISBN: 978-0-12-398296-4.
37. Masters, T. Advanced Algorithms for Neural Networks. A C++ Sourcebook; John Wiley
and Sons, Inc.: New York, NY, USA, 1993.
38. Neuroshell 2TM. Neuroshell Tutorial; Ward Systems Group, Inc.: Frederick, MD, USA,
2000.
39. Nash, J.E.; Sutcliffe, J.V. River flow forecasting through conceptual models; part 1—A
discussion of principles. J. Hydrol. 1970, 10, 282–290.
192
40. Wilcox, B.P.; Rawls, W.J.; Brakensiek, D.I.; Wight, J.R. Predicting runoff from rangeland
catchment: A comparison of two models. Water Resour. Res. 1990, 26, 2401–2410.
41. Rhode Island Division of Planning 2014. Road salt/sand application in Rhode Island,
Statewide planning technical paper number: # 163 (March 31, 2014)
42. Hinsdale, J. How Road Salt Harms the Environment, State of the Planet. 2018. Available
online: https://blogs.ei.columbia.edu/2018/12/11/road-salt-harms-environment/ (Decem-
ber 11, 2018).
43. Elgin, E. Salt Runoff Can Impair Lakes. Salting Roads, Parking Lots, and Sidewalks Can
Turn Our “Fresh” Water Salty. 2018. Available online:
https://www.canr.msu.edu/news/salt_runoff_can_impair_lakes (February 22, 2018).
44. Kelting, D.; Laxson, C. Review of Effects and Costs of Roads De-Icing with Recommen-
dations for Winter Road Management in the Adirondack Park. Adirondack Watershed In-
stitute Report # AWI2010-01. 2010. Available online: http://www.protectadks.org/wp-
content/uploads/2010/12/Road_Deicing-1.pdf (February 2010).
45. Rastogi, N. Salting the Earth: Does Road Salt Harm the Environment? The Green Lantern.
2010. Available online: https://slate.com/technology/2010/02/does-road-salt-harm-the-en-
vironment.html (February 16, 2010).
46. Nagpal. N. K.; Levy. D.A.; and MacDonald. D.D 2004. Ambient Water Quality Guidelines
for Chloride 2. National Library of Canada Cataloguing in Publication. Environmental
Management Act, 1981. http://wlapwww.gov.bc.ca/wat/wq/BCguidelines/chloride.html (3
of 22) [1/15/2004 3:20:27 PM]
193
CHAPTER 6
CONCLUSION
The novelty of the study is assessed the stormwater BMPs’ performances on
runoff volume reduction, assessed their performance on peak flow attenuation, and the
effectiveness of BMPs on runoff pollutant reduction in a suburban watershed. A four-
pronged analysis based on 1) RS-GIS based SCS-CN model developed; 2) EPA SWMM
5.1 for hydrologic and hydraulic simulation; 3) parameter estimation using the Bayesian
approach, and 4) predictive modeling through Artificial Neural Network.
An integrated study based on RS-GIS and SCS-CN model was conducted to
assess the suburban growth and its impacts on surface runoff over the study area in
South Kingstown, Rhode Island, USA. First, land-use pattern changes were detected
through remote sensing application using the supervised classification method on four
different Landsat images from 1994, 2004, 2014, and 2020. The analysis revealed 68.6%
urban land expansion from 1994 to 2020 in this suburban area for 36 years. Then surface
runoff estimated using a GIS-based SCS-CN model that showed an increasing spatial
pattern of 2% to 10% of surface runoff from 1994 to 2020. This study also told that
suburban growth began after 2000 in this area, and within 16 years, this land-use change
started to show its substantial impact on surface runoff.
Suburban areas are becoming equally important as urban areas to investigate
stormwater management. After ensuring the suburban growth and its impact on surface
194
runoff EPA SWMM5.1 is developed in simulating stormwater runoff quantity in this
study area. The conducted study investigated the comparative performance in terms of
runoff depth and peak flow of different low impact development (LID) structures under
various rainfall designs (1-year, 2-year, 5-year, 10-year, 25-year, 50-year, and 100-year
return periods). The study is carried out to test the performance of 4 different scenarios
(absence of LID, permeable pavement, bioretention, and vegetative) controlling runoff
depth and peak flow. The developed model calibration and validation were conducted
for three setups, i.e., run1, run 2, and run 3, in three different sub-catchments. The
respective NSE values of run1, 2, and 3 were 0.73, 0.59, and 0.64, with an average value
of 0.65. The average R2 value was 0.76. Environmental benefit also analyzed, and the
outcome suggests that LID structures (bioretention, permeable pavement, and
vegetative swale) are effective for suburban areas. In summary, the reduction rates of
LID facilities for different rainfall intensities per unit area (sq. km) were ranked as
follows: BR > PP > VS. Bioretention cell showed steady effectiveness in every rainfall
intensity even for the 100 years return period (31% ± 5%).
The water quality module of LID in the stormwater management model
(SWMM) was used through the RUNOFF block to develop the SWMM-LID model
using the buildup and wash-off method in this study. The developed SWMM-LID
applied to assess the effectiveness of the LID structures (bioretention, permeable
pavement, and vegetative swale) in controlling the runoff pollutants (total suspended
sediments, Nitrate-N, and Orthophosphate-P) based on 42 individual rainfall events
from 2016 – 2019 under various designs storms (1-year, 2-year, 5-year, 10-year, 25-
195
year, 50-year, and 100-year return periods). We used 30 rainfall events for model
calibration, and the rest of the 12 rainfall events were considered for model validation.
Moreover, Nash-Sutcliffe Efficiency (NSE) values of 0.73, and 0.69, R2 values of 0.75
and 0.64, and RSR values of 0.53 and 0.61 for runoff quantity module in two different
runs prove model efficiency. The selected LID structures were effective to removing
runoff pollutants up to 27% to 83% for the TSS, 11% to 28% for Nitrate-N, and 14% to
27% for Orthophosphate-P. Also, a detailed uncertainty of SWMM-LID model data
input and output analysis was conducted applying the Bayesian Markov Chain Monte
Carlo (MCMC) approach. Sensitivity analysis suggests that runoff in the study area is
sensitive to rainfall depth, impervious percentage, slope and manning coefficients,
conduit length, conduit roughness, storage depth, and combined interaction between
these parameters runoff and water quality (Total suspended solid, Nitrate-N, and
Orthophosphate-P). The prior knowledge of the parameter distribution plays a
significant impact in the model uncertainty structure, whereas parameters are considered
to follow uniform distributions in this study. Total predictive uncertainty for peak flow,
TSS, Nitrate-N, and Orthophosphate-P analysis with 95% confident interval captures
most of the observed data supporting the uncertainty quantification of the model.
The predictive model is developed using Back Propagation Neural Network (BPNN)
method for chloride from road salt in the study area. The innovation of the conducted
study were to develop an ANN model of the system trained using a small data set, to
obtain the best-fit models for predicting chloride concentrations using data from
monitoring sites, to evaluate the ANN model performance using 3 years (2016–2019)
196
of observed data versus predicted data from the model, and to determine the accuracy
of the ANN model performance. The model also assesses the impact of road salt
applications by assessing a spatial density distribution focused on probable high
chloride concentration in an area. The conducted study suggests that ANN modeling off
environmental data has great potential in future work on improved prediction of chloride
and other pollutant concentrations and provides a helpful tool for the water resource and
environmental managers.
Scope of the work
Even though much progress has been made in the science of LID to understand
these practices' performance, there are still many aspects and challenges that must be
assessed and addressed to support widespread LID adoption. These needs are discussed
hereafter in more detail. They include, among others:
• There is a need for continued data collection to evaluate LID systems over different
spatial and temporal scales and climatic conditions.
• Need for assessing the removal of emerging and difficult-to-measure contaminants by
LID practices.
• Need to rank the performance of LID practices from lot scales to watershed and re-
gional scales.
• Development of easy-to-use decision support tools incorporating LID practices.
197
CHAPTER 7
Supplementary
Data Presentation and Analysis
How effective are the bioswales in improving water quality? A case study of
a sub-urban catchment
An overview of field and laboratory analysis data along with statistical analysis.
Khurshid Jahan1,*, Soni M Pradhanang1,*, and Rebecca Brown2
1Department of Geosciences, University of Rhode Island, RI, Kingston, USA 2Department of Plant Science and Entomology, University of Rhode Island, Kingston,
RI, USA
*Correspondence: [email protected] (K.J.), [email protected]
(S.M.P)
198
Abstract
Best management practices (BMPs) are popular approaches to improve hydrology and
stormwater quality in urban areas. This study applied the BMPs such as bioretention
and vegetative filter strips to assess its performance in removing the non-point pollutants
of stormwater runoff from a suburban area. Sedimentation and filtration within the grass
layer are the primary mechanisms of pollutant treatment. The results compared the
reduction efficiency between control and treatment sites. The treatment sites were
treated with organic soil mix with native grass and Rhode Island Department of
Transportation (RIDOT) park mix. The entire analysis was done for 42 individual
rainfall events from 2016 to 2019 applying the ANOVA where three different variables
were considered (season, rainfall category, and the site type) for nitrate, ortho-phosphate,
and total suspended sediments. The study found that pollutants significantly responded
with seasons and differed in the control and treatment sites. Furthermore, the treatment
sites are capable of reducing pollutants. All three pollutants are consistently removed
during the post-treatment period. For the total suspended solids, the applied bioretention
and vegetative swale reduce from 20% to 90%, nitrate and nitrite reduced up to 60%,
and it also reduces ~75% ortho-phosphate. Statistical analysis also found that both
treatment mechanisms were equally effective for the selected area.
Keywords: Best Management Practices, Suburban, Analysis of Variance,
Effectiveness.
199
1. Introduction
The US Environmental Protection Agency [1] considers non-point sources as
the leading threat to water quality in the Nation. Urban stormwater runoff is a non-point
pollution source and contributes to excess sediments and nutrients to the receiving water
[1-4]. Excess nutrients can interrupt the delicate balance of the aquatic ecosystem.
Nutrients come from the natural breakdown of human and animal wastes and fertilizers
applied to residential, agricultural, recreational, and commercial landscapes. Best
management practices are considered to counteract the non-point pollutant sources and
reduce their impact [5]. Roadside best management practices (BMPs) are techniques or
methods that aim to prevent or reduce the overall negative impacts of stormwater and
improve the quality of stormwater runoff cost-effectively [6]. These BMPs can be
characterized into three types based on their performance: (i) source control, (ii) flow
control, and (iii) runoff treatment [6]. A source control BMP is effective at preventing
and/or redirecting pollutants before entering the storm sewer system. It can be a
structural component of a planned site or a procedural BMP. Only a few permanent
source control BMPs (such as street sweeping, deicing, and spill control) can be
frequently used for a roadway. Therefore, source control BMPs are used more
commonly during construction and for the permanent portion of non-roadway projects
such as rest areas and park and ride lots. Stormwater flow control BMPs are designed
to control the flow rate or the amount of runoff leaving a site after development. The
primary mechanisms used to manage flow control include dispersion, infiltration, and
detention. Increased flows can cause downstream damage due to flooding, erosion, and
scour, and degradation of water quality and in-stream habitat because of the channel and
200
streambank erosion. Runoff treatment BMPs are designed to remove runoff pollutants
using various mechanisms, including sedimentation, filtration, plant uptake, ion
exchange, adsorption, precipitation, and bacterial decomposition. Different types of
structural BMPs such as infiltration, dispersion, bio-infiltration, wetpool, oil control
BMPs are widely applied as runoff treatment.
Structural stormwater treatment practices apply primarily in Rhode Island for
taking care of water quality, and it includes 1) wet vegetated treatment systems, 2)
infiltration practices, 3) filtering systems, 4) green roofs, and 5) open channel practices
[7]. An open channel practice (vegetative swale) is applied to this study. Vegetative
swales are usually shallow open channels with gentle side slopes, filled with erosion
and flood-resistant vegetation, designed to convey, control, and improve stormwater
through infiltration, sedimentation, and filtration [8,9]. Vegetative swales are mainly
used to slow runoff velocity and improve water quality. The use of vegetated filter strips
and swales in the highway and adjacent roadside environment has been extensively
studied in field trials in Texas [10,11], North Carolina [12], Kansas [13], Washington
[14], Maryland [15], and internationally [16]. Vegetated filter strips have been shown
to remove 50 to 98% of TSS, 23 to 50% of nitrate, and 33 to 80% of phosphate [15, 17],
while infiltration rates are lower than for bioretention BMPs, standard vegetated front
slopes, and swales are just as effective at treating stormwater [18].
Organic amendments of roadside soils combined with plantings of native
perennially grass noticeably improve the persistence of perennial vegetation by
201
increasing soil fertility and water retention [19, 20, 21] and enhancing stormwater
infiltration treatment. The benefits of the soil amendments are expected to last for years
after the initial application. The ideal soil organic matter mix has sufficient nitrogen and
phosphorus to support vegetation and sufficient carbon to support a healthy soil
microbial and invertebrate community [22]. In addition, many native kinds of grass have
deeper root systems than the turfgrass species typically seeded on roadsides [23]; such
deep-rooted species can anchor slopes and absorb nitrate and phosphate from deep
within the soil profile, preventing leaching to groundwater. Therefore, the proposed
BMP applied organic soil amendments and vegetative grasses to control the runoff
quality.
This study analyses the effectiveness of vegetative swales and bioretention for
the removal of stormwater pollutants. Vegetative swale effectively removes many
pollutants (nitrate and phosphate) from urban runoff depending on structural design and
composition [24 – 27]. The standard bioretention design and the soil mix constituents
and amounts are presented in Figure 1 that generally applied in Rhode Island for
stormwater management. Rhode Island faced nitrate, phosphate, and TSS problem
through the stormwater runoff. Effectiveness is measured based on the removal capacity
of nitrate, phosphate, and TSS. According to Hunt et al. (2006) [25], the phosphorus
removal efficiency increased with decreasing phosphorus content of the soil mixture,
and Dietz and Clausen (2005) [27] explained that the phosphorus removal efficiency
worked better after adding the soil mixture. Nitrogen removal is also effective through
the soil media of bioretention.
202
2. Site Description
Two sites are considered as control sites, and four are vegetative swales BMP
treatment sites (Figure 1). The size of each site depends on the width of the bioretention
swales. Unlike the BMP sites, the control sites received no soil amendment (compost and
biosolids) and vegetated with the original grass mix. For a treatment site, approximately
3 lbs of nitrogen per 1000 sq. ft was used to prepare compost and biosolids. The carbon-
to-nitrogen ratio of this mix is about 12:1. The mixtures were added to the treatment sites.
After soil amendments have been placed, two kinds of grass mixtures were seeded; one
was the standard RIDOT grass mix, which includes Lolium Perenne, Festuca Rubra, and
Poa Pratensisand the other was a native grass mix containing Schizachyrum Scoparium,
Agrostis Capillaries, Eragrostis Spectabilis, and Festuca Rubra. Both RIDOT and native
grass mixes represent common grass species found in Rhode Island. The plots were
seeded by hand, followed by hydro-mulching to prevent erosion. The soil amendments
and vegetative strips worked together as a filter for the stormwater runoff. Total sus-
pended solids (TSS), Nitrate (NO3--), Nitrite (NO2-), Phosphate (PO43-), Chloride (Cl-),
Dissolved Organic Carbon (DOC), and Iron (Fe) were tested in the Hydro-systems and
Water Quality Laboratory at the URI. Instrumental analysis methods are provided in Ta-
ble 3.
203
Figure 1: Roadside Best Management Practices Sites within the URI Campus
The results will compare the filtration effectiveness on the control versus the
treated sites over the three years. All data is statistically analyzed and visualized to
compare the stormwater runoff quality between control and treatment sites.
Table 1: Monitoring Strategies
Sites Location
ID
Activity Plan Sample collection
Treatment
Sites
Site 1 Sites are seeded
RIDOT mix and
native grass mix.
Stormwater runoff samples
were collected during
every storm event (>0.5
inches) from 2016 to 2019.
Site 3
Site 4
Site 6
Control
Sites
Site 2 Control sites
kept as it is. Site 5
204
3. Materials and methods
3.1 Study Design
In this study, two control sites and four treatment sites were monitored during
the study period from August 2016 to July 2019. Each of these sites are equal length of
50ft and approximately equal width of ~8ft. Furthermore, the pre-treatment (calibration
period) started from August 2016, whereas the post-treatment started from June 2017.
The relationship between control and treatment sites during the treatment period
(sample collection period for control sites) is described by a simple linear regression
between the paired observation, taking the form:
𝑋𝑖 = 𝑏0 + 𝑏1(𝑌𝑖) + 𝑒
where 𝑋𝑖 and 𝑌𝑖 represent water quality concentration 𝑏0 and 𝑏1 are the regression
coefficients representing the regression intercept and slope, respectively, e is the
residual error. The significance of the connection between paired observations is tested
using the analysis of variance (ANOVA) between treatment and control sites. So, the
entire analysis is designed to evaluate sediments and nutrient removal capabilities. We
considered here three treatments:
a) Treatment 0: 2 control sites (Site 2 and Site 5)
b) Treatment 1: 2 treatment sites (added organic soil with native grass mix) (Site_1
and Site_4)
c) Treatment 2: 2 treatment sites (added organic soil with RIDOT park mix) (Site_3
and Site_6)
Site 1 to Site 4 is close to agricultural land and Sites 5 and 6 are situated close to the
parking lot (Figure 1).
205
3.2 Rainfall pattern and size
This study is investigated using forty-two rainfall events from August 2016 to
July 2019. rainfall eventsThe actual concentration of a pollutant in stormwater runoff
varies with several factors land use, the time between runoff events, duration and
intensity of rainfall and runoff, characteristics of the pollutant, season, extent of exposed
soils, such as with construction sites, and extent of connected impervious surfaces [28 -
30]. Therefore, in this study, we considered rainfall intensity and season to assess the
stormwater pollutant characteristics.
Each of the rainfall events is categorized based on the rate of rainfall. According
to Table 2 and the characteristics of the rainfall events, we found four categories of
rainfall (light-moderate, moderate-heavy, heavy, and heavy-torrential).
Table 2: Types of rainfall based on rainfall intensity
Rainfall Categories Rate of the rainfall
Light rain 0 - 4 mm/day
Light-Moderate 4 – 16 mm/day
Moderate - Heavy rain 16 – 32 mm/day
Heavy rain 32 - 64 mm/day
Heavy- Torrential rain 64 – 128 mm/day
Torrential rain >128 mm/day
Source: Alpart et al., 2002 [31]
3.3 Monitoring Data Collection and Analysis
In this study, stormwater runoff samples were collected from 6 different sites
from 2016 – 2019, covering about 42 rainfall events. All those rainfall events were
selected based on rainfall—only those events were greater than 12.7 mm (0.5 inch).
From all the sites total of 6 stormwater runoff samples were taken during each event,
206
and pH, temperature, and turbidity analysis were conducted immediately after collecting
the samples. All samples were refrigerated (<40C) until testing for parameters including
seven anions (Fluoride, chloride, nitrite, nitrate, bromide, sulfate, and phosphate) and
six cations (lithium, sodium, ammonium, potassium, magnesium, and calcium) along
with total suspended sediments (TSS), Dissolved Organic Carbon (DOC), and Iron. The
water quality parameters, testing methods, and their relative analytical method detection
limits (MDL) are summarized in Table 3. All the analytical data were gathered to
conduct robust statistical analysis.
Table 3: Summary of the parameters and analytical method used including detection
limits(MDL)
Parameters Reference Method MDL*
(mg/l)
Determination of Turbidity Turbidity EPA method 180.1 -
Total Suspended Sediments TSS Calculation -
Dissolved Organic Carbon DOC NDIR detection method -
Nitrite NO2− Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.02
Nitrate N03- Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.02
Sulfate SO42- Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.05
207
Phosphate PO43- Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.005
Potassium K+ Thermo Scientific™ Dionex™ CERS™ 500
Cation Electrolytically Regenerated Suppressor
(2 mm): P/N 082543
0.01
Chloride Cl- Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.05
Iron (Total) Fe 1,10 phenanthroline method. USEPA 8008 0.02
Bromide Br- Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.05
Sodium Na+ Thermo Scientific™ Dionex™ CERS™ 500
Cation Electrolytically Regenerated Suppressor
(2 mm): P/N 082543
0.01
Fluoride F- Thermo Scientific™ Dionex™ AERS™ 500
Anion Electrolytically Regenerated Suppressor
(2 mm); P/N 082541
0.05
*MDL: Method detection limit collected from Western Washington Phase II Municipal
Stormwater Permit report [32].
3.4 Turbidity
Turbidity is a principal physical and optical property of water that usually
measure the clarity. It is known to be the quickest and cost-effective parameter to assess
the amount of eroded sediment. Generally, turbidity is measured in both nephelometric
208
turbidity units (NTUs) and Jackson turbidity units (JTUs). In this study, NTU are
applied here for turbidity measurement. It therefore can be an unintended indicator of
potential health risks associated with the outflow water from stormwater runoff [33].
Turbidity of the stormwater runoff is highly dependent on rainfall intensity and the
duration or the substrate materials [34]. In this study, turbidity was measured for six
different monitoring sites from 2016 – 2019 (Figure 2). The average range of the
turbidity is 9 NTU excluding 5 high range turbidity (60.7, 64, 184, 49, and 102 NTU).
Those 5 extreme values were represented in the figure in separately with value. Among
5 extreme values 3 values were measures in summer season and rest of the two values
found during the spring season. Turbidity showed a variation in different types of
rainfall intensity and season. It showed that the maximum Turbidity (NTU) in the
control site 5 in heavy rainfall. and minimum turbidity found for all sites during
moderate-heavy rainfall. It showed almost same turbidity in the heavy torrential because
of size of sample (sample size=2). On the other hand, maximum turbidity concentration
(3.5 mg/l) was detected during winter season for site 5.
Figure 2: An overview of Turbidity in the Six different monitoring sites
209
3.5 Chloride
Chloride concentrations are represented in Figure 3 from six different
monitoring sites for three years (2016–2019). Seasonal variation of chloride
concentration is observed in this analysis and the higher concentration of chloride was
found during winter and early spring season in most of the sites having the higher
median than the other two season (summer and fall). According to the study, the highest
range chloride concentration (0.8-197.9) mg/l was measured from winter runoff samples.
A detail study about the chloride prediction applying artificial neural network (ANN)
[35] and its impact is presented in Chapter Five. In the figure, four extreme values were
represent separately including their values.
Figure 3: An overview of Chloride concentration in the Six different monitoring sites
In this study, a detail statistical analysis is conducted for Nitrate-N,
Orthophosphate-P, and total suspended sediments (TSS).
210
3.5 Statistical Analysis
The significance of the relationship between paired observations was tested
using analysis of variance (ANOVA). Analysis of variance (ANOVA) is a statistical
degree analysis method that separates an observed cumulative variability into two sets,
i.e., systematic, and random [36, 37]. Usually, ANOVA test is applied to determine the
influence of independent variables on the dependent variables. A model was developed
to test the stormwater quality of the samples collected from the six different sites. The
test assumes that the regression residuals: usually are distributed, have equal variances
between treatments, and are independent [36, 37]. Statistical analyses are done in two
steps. For the first step, analysis of variance is applied and in the other step, stormwater
pollutant reduction rates are analyzed using the reduction rate (RR) equation. Statistical
analyses were applied using the R software. For comparing differences between samples
in two different types of sites (control and treatment) taken simultaneously, both one-
way and two-way ANOVA was applied in this study. One-way ANOVA helped to
assess the significance of each of the considered variables on the stormwater pollutants.
In contrast, two-way ANOVA allowed assessing the interaction between
continuous quantitative output and categorical variables and assessing the significant
impact of every experimental variable. Two-way ANOVA is used here for water quality
assessment on turbidity, TSS, DOC, chloride, nitrate-N, and orthophosphate-P, which
shows the difference of variation in both types of sites and explained the seasonal and
rainfall intensity impact on impact on the stormwater pollutant. The methodology of
one and two-way ANOVA are described below:
211
3.5.1 One-way ANOVA
One-way ANOVA analyze and compare the means between the selected groups
to determine the statistical significance between them. The ANOVA produces an F-
statistic, the ratio of the variance calculated among the means to the variance within the
samples [36, 37].
F = variation between sample means/variation within the samples
F- critical (0.05) = 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 1
𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 2
R2 = 1 − 𝑆𝑆𝐸
𝑆𝑆𝑇
R2 (adj) = 1 - 𝑀𝑆𝐸
𝑀𝑆𝑇
P-value = 2*P (TS≥|ts||Ho is true) = 2* (1-cdf(|ts|))
where SSE is the sum of square error, SST is the total sum of the square, MSE is mean
square error, MST is the mean square of the variable.
3.5.2 Two-way ANOVA
Two-way ANOVA is a statistical analysis method for a study with a continuous
quantitative consequence and two or more categorical independent variables [36, 37].
The usual expectations of normality, equal variance, and independent errors apply. A
two-way ANOVA was employed to examine whether temporal changes in stormwater
quality occurred over the study period. Post Hoc Least Significant Difference and
Multiple Comparison tests were again used to determine where the differences between
means were located. The first one-way ANOVA was applied to determine the statistical
differences between each season and the rainfall categories in this study. Furthermore,
in this study, a two-way ANOVA was employed to determine the temporal changes in
212
the individual control and treatment sites during the period of this study. Homogeneity
is also tested for the categorical variables before conducted a two-way ANOVA analysis.
Table 4: Analysis of variance for linear regression
Source Degrees of
Freedom
Sun of Squares Mean squares F
Regression
Residual
Total
1
n-2
n-1
(𝑆𝑥𝑦)2/𝑆𝑥2
𝑆𝑦2 − (𝑆𝑥𝑦)2/𝑆𝑥
2
𝑆𝑦2
(𝑆𝑥𝑦)2/𝑆𝑥2
𝑆𝑦𝑥2
[(𝑆𝑥𝑦)2/𝑆𝑥2]/ 𝑆𝑦𝑥
2
The value for Table 2 is calculated from
𝑆𝑦2 = ∑ 𝑌𝑖
2 − (∑ 𝑌𝑖)2
𝑛
𝑆𝑥2 = ∑ 𝑋𝑖
2 − (∑ 𝑋𝑖)
2
𝑛
𝑆𝑥𝑦 = ∑ 𝑋𝑖𝑌𝑖 − ∑ 𝑋𝑖 ∑ 𝑌𝑖
𝑛
𝑆𝑦𝑥2 =
𝑆𝑦2 − (𝑆𝑥𝑦)2)/𝑆𝑥
2
𝑛 − 2
Also, the regression and coefficient of determination are determined from:
𝑏1 = 𝑆𝑥𝑦
𝑆𝑥2
𝑏0 = 𝑌′ − 𝑏1𝑋′
213
𝑟2 =(𝑆𝑥𝑦)2/𝑆𝑥
2
𝑆𝑦2
The resulting F statistic would indicate that the regression relationship adequately
explains a significant amount (p<0.001) of the variation in paired quality data.
Table 5: Categorical variables
Categorical Variables Parameters
Season Fall
Spring
Winter
Summer
Rainfall types based on rainfall intensity Light
Light-moderate
Moderate-Heavy
Heavy
Heavy-Torrential
Torrential
Site Types Treatment 0 (Control)
Treatment 1 (amended soil with new vegetation)
Treatment 2 (amended soil with RIDOT park
mix)
The null hypothesis that we considered for this study are described as follows:
• There is no difference in average TSS, nitrate and nitrite, and ortho-phosphate
concentration for any treatment type.
• There is no difference in average TSS, nitrate and nitrite, and ortho-phosphate
concentration for any rainfall type.
214
• There is no interaction effect between treatment type and rainfall type on the
average TSS, nitrate and nitrite, and ortho-phosphate concentration.
3.6 Reduction Rate
To assess the effectiveness of the roadside BMPs, we compared pollutant
concentrations between the control and treatment sites for every events. Due to shallow
relief, event mean concentration couldn’t be generated in this site. Flow data and
pollutant flux could not be achieved even after installed shallow v-notch weirs. Grab
sample data are available for each of the events for six different sites. We compared
treatment sites 1, 3, and 4 with the control site 2 as all these four sites are close to the
agricultural field. The other two sites, Site 5 (control) and Site 6 (treatment), are close
to the parking lot. Site 5 and site 6 are compared to each other. Pollutant concentration
reductions (in percent) were determined applying the following formula:
ΔC =( 𝑃𝑐−𝑃𝑡
𝑃𝑐) x 100
where ΔC= Pollutant concentration reduction (%), Pc = concentration of stormwater
runoff pollutant for control site, Pt = concentration of stormwater runoff pollutant for
the treatment site.
5. Result and Discussion
In the 2.1 section, we described the location of the site and the land types. Sites
1 to 4 are close to agricultural sites, and sites 5 and 6 are located at the edge of the
parking lot. So, here we compared the reduction rate and the average concentrations
215
based on land types. Control sites 2 are compared for treatment sites 1, 3, and 4. And
sites 5 and 6 are compared to each other due to having the same land types.
5.1 Nitrate-N
5.1.1 analysis on one-way and two-way ANOVA
Nitrate-N concentration was compared with control and treatment sites and
efficacy of each site were evaluated. The control sites' nitrogen composition closely
matched the parallel studies of urban runoff [33-35], with nitrate representing 45% of
the total nitrogen concentration and the rest of the remaining nitrite making up the
average. Therefore, nitrate was measured as an indicator of the nutrient content of the
runoff. Nitrate concentrations showed a variation in different types of rainfall (Figure
4), where it showed that the nitrate concentration was maximum (15.7 mg/l) in the
control site 2 for heavy rainfall (Table 2) and minimum nitrate concentration is detected
in the site 5 and 6 (parking lot sites) for moderate-heavy and heavy rainfall. Figure 5
shows seasonal differences for site 2-4. The maximum nitrate concentration (15.7 mg/l)
is detected during the spring season for site 2 followed by site 3 (13.5 mg/l) and 4 (11.3
mg/l). The fall and summer seasons showed a similar nitrate concentration for site 1 to
site 4. The lowest median observed in the winter season.
216
Figure 4: Statistical parameter of the different types of rainfall
Figure 5: Statistical parameter of the different types of seasons
To assess the overall interaction between three variables, a two-way model was
developed for this study, and AIC (1139.9) showed the lowest value for the treatment
type, rainfall and seasonal variables interaction (Table 6. The model revealed that for
this area and the analysis is more interactive with season and treatment rather than
rainfall category. The reason for having lower interaction on rainfall category is the low
elevation and almost flat slope of the study area. The residual and Normal Q-Q plots
(Figure 6) showed the efficiency of the model.
217
Table 6: Model selection-based AIC for Nitrate-N, Orthophosphate-P and TSS
Model Nitrate-N Orthophosphate-P TSS
A 1139.9 914.7 293.3
B 1141.3 921.5 296.1
C 1145.8 922.8 299.5
D 1157.9 930.6 300.7
Model A: Treatment type + Rainfall + Season
Model B: Treatment Type + Rainfall
Model C: Treatment Type + Season
Model D: Rainfall+ Season
The diagnostic plots (Figure 6) showed the variance (residuals) across the range
of the observed data. The red lines represent the mean of the residuals and since it lies
closer to the zero lines horizontally means no significant outliers and no high difference
in variance. In this figure, the mean of the residual lied horizontally on the zero lines.
The normal Q-Q showed an inverted s shape. Approximately from the values (-3, -2.2),
the sample grows lower than the standard normal distribution; therefore, it takes longer
for sample quantiles to increase. From the values (-2.2, 2), the sample seems to grow at
approximately the same pace as the standard normal distribution.
218
Figure 6: Residuals and Normal Q-Q plot of the developed model
There was a statistically significant difference between the NO3-N level of
control and treatment runoff samples as indicated by two-way ANOVA (F(3, 50)=15.07,
p < 0.01). Significant variance was observed among the treatments with F(3, 59)=11.11,
p<0.01 with control for rainfall intensity and season. A post-hoc multiple comparison
test method was applied, and it also showed significant differences (p<0.01) among the
nitrate levels of intensive treatment compared to the control site.
The study found a significant response of nitrate for treatment variables
(treatment 0, treatment 1, and treatment 2) (Figure 7). Figure 7 showed the mean of
three different treatment and the result are presented in the table 7
Table 7: Nitrate-N response on treatment type
Treatment Type Group No of the Sites Mean (mg/l)
Treatment 0 A Site_2 5.43
219
B Site_5 1.89
Treatment 1 C Site_1 4.21
D Site_4 4.22
Treatment 2 E Site_3 4.43
F Site_6 1.4
Figure 7: Nitrate and Nitrite response to treatment types in each of the sites
Due to a significant interaction between treatment 2 and treatment 0 (P< 0.05),
it was clearly proved that the treatment 2 is the more effective in reducing nitrate
concentration. The interaction between treatment 1 and treatment 0 (P > 0.05) found
treatment site 1 less effective than treatment 2. However, the mean of the two treatment
sites (Table 6) (treatment 1 and treatment 2) is very close. So, the very close variance
of treatments 1 and 2 proved that they both are equally effective.
220
5.1.2 Reduction rate of the Nitrate-N
As shown in figure 9, the pre-treatment and post-treatment periods are separated
with a line. The first 9 rainfall events were under the pre-treatment period for all sites,
and the rest of the 33 events were under the post-treatment. Reduction of nitrate is
represented in figure 8 where it showed the overall reduction between control and
treatment sites (Site 2 (control) vs. Site 1 (treatment). Out of total 42 rainfall events,
NO3-N reduction was seen for 29 rainfall events, and the range of reduction (%) was (0
to 55)% whereas for the site 3 NO3-N reduction was seen for 34 rainfall events, and the
range was (0 to 52)%. For site 4, reduction range was up to 60% for 37 rainfall events.
Considering the other sites, site 5 and site 6 had a lower amount of nitrate concentration
than the other 4 sites. However, the comparison between site 5 and site 6 showed that a
significant amount of nitrate was removed (average reduction rate 45%) through the
implemented organic soil layer.
For most of the storm events, especially for the treatment sites with a small
number of events showing high export of nitrogen, mainly during the summer season.
The transport of NO3-N during summer rainfall events is likely due to the organic nature
of the swales, perhaps tied to unnecessary sources of nutrients or other organic debris
[36]. This study also showed Nitrate-N removal through the soil media, and the
comparison reflected the positive influence of the BMPs for this area.
222
Figure 8: Comparison of Nitrate-N concentration reduction between control and
treatment site. Blue colored symbol represented the reduction (%), and the red color
symbol represented the negative reduction for the treatment site. The dot lines separate
Pre-treatment and post-treatment periods.
5.2 Orthophosphate-P (PO4-P)
Orthophosphate-P is one of the primary nutrients for the metabolic responses in
plants and animals. However, as soon as it arrives on surface, the groundwater or a
wetland making it provisionally inaccessible to active organisms by developing bonds
to soil particles [37].
5.2.1 analysis on one-way and two-way ANOVA
Orthophosphate-P (PO4-P) concentrations showed a variation for different types
of rainfall (Figure 9) intensity. The PO4-P concentration was maximum (8.4 mg/l) in the
control site 2 for heavy and moderate-heavy (8.3 mg/l) rainfall, and minimum
orthophosphate-P concentration is detected in the site 5 and 6 (parking lot sites) as well
for moderate-heavy and heavy rainfall. On the other hand, figure 10 showed that
223
maximum PO4-P concentration (8.4 mg/l and 8.3 mg/l) was detected during the fall and
winter season for site 2. Figure 9 also showed a variation in different types of rainfall
intensity, where it showed that the phosphate concentration was maximum (8.4 mg/l) in
the control site 2 for heavy rainfall and moderate-heavy (8.3 mg/l) and minimum
phosphate concentration is detected in the site 5 and 6 (parking lot sites) as well for
moderate-heavy and heavy rainfall. On the other hand, figure 11 described that
maximum phosphate concentration (8.4 mg/l and 8.3 mg/l) was detected during the fall
and winter season for site 2.
Figure 9: Statistical parameter of the different types of rainfall
224
Figure 10: Statistical parameter of the different types of seasons
The results of the two-way ANOVA were F(5,21) = 8.4, p < 0.05, which showed
significant differences among the phosphate values. A two-way ANOVA was employed
to understand the effects of rainfall intensity and season on orthophosphate-P variation.
The results found a significant main effect of heavy rainfall (F(1.02, 8.5) = 8.3, p <
0.05), and a significant main effect of season (F(4, 9) = Fall, p b< 0.05, ηp 2 = 0.44).
The post-hoc multiple comparison test disclosed significant differences (pb<0.05)
between control and treatment orthophosphate-P concentrations. Furthermore, the PO4-
P concentrations were higher in the control sites than in treatment sites. Additionally,
lower phosphate concentration was observed in the parking lot sites rather than those
close to the agricultural field.
To assess the overall interaction between three variables, a two-way model was
conducted for this study, and AIC (914.7) showed the lowest value for the treatment and
seasonal variables interaction like nitrate and nitrite (Table 6). The model revealed that
225
for this area and the analysis is more interactive with season and treatment rather than
rainfall category. The reason for having lower interaction on rainfall category is the low
elevation and almost flat slope of the study area. The residual and Normal Q-Q plots
(Figure 11) showed the efficiency of the model.
The diagnostic plots (Figure 11) showed the variance (residuals) across the range
of the observed data. The red lines represented the mean of the residuals. The more it
lies on the zero lines horizontally means no significant outliers and no high difference
on variance. In this figure, the mean of the residual lied horizontally on the zero lines.
The Normal Q-Q showed an inverted s shape. Approximately from the values (-3, -1.8),
the sample grows lower than the standard normal distribution; therefore, it takes longer
for sample quantiles to increase. From the values (-1.8, 2.3), the sample seems to grow
at approximately the same pace as the standard normal distribution.
Figure 11: Residuals and Normal Q-Q plot of the developed model
226
The study found a significant response of orthophosphate-P for treatment variables
(treatment 0, treatment 1, and treatment 2) are shown in the figure 12 and the result of
the mean of each treatment types are represented in the Table 8.
Table 8: Orthophosphate-P response on treatment type
Treatment Type Group No of the Sites Mean (mg/l)
Treatment 0 A Site_2 4.04
B Site_5 1.69
Treatment 1 C Site_1 2.85
D Site_4 2.59
Treatment 2 E Site_3 2.54
F Site_6 1.32
Figure 12: Orthophosphate-P response to treatment types in each of the sites
Due to a significant interaction between treatment 2 and treatment 0 (P< 0.05) it
was clearly proved that the treatment 2 is the more affective in terms of reducing nitrate
concentration. The interaction between treatment 1 and treatment 0 (P > 0.05) found
227
treatment site 1 less effective than treatment 2. However, the mean of the two treatment
sites (Table 7) (treatment 1 and treatment 2) is very close. So, the very close variance
of treatments 1 and 2 proved that they both are equally effective.
5.2.2 Reduction rate of the Orthophosphate-P (PO4-P)
The reduction of PO4-P concentration is presented in figure 13 where it showed
the overall reduction between control and treatment sites (Site 2 (control) vs Site 1
(treatment) A removal was observed here. Out of the total 42 rainfall events, PO4-P
concentration decreased for 35 rainfall events, and the range of reduction (%) was (0 to
80) % whereas for the site 3 removal range of PO4-P concentration was rainfall events
(0 to 78)% considering 30 rainfall events. For site 4, the removal rate was up to 80% for
32 rainfall events, Considering the other sites, site 5 and site 6 had a lower amount of
PO4-P concentration than the other four sites. However, the comparison between site 5
and site 6 showed 28 rainfall events removal through the implemented organic soil
media.
228
Figure 13: Comparison of Orthophosphate-P concentration reduction between control
and treatment site. Blue colored symbol represented the reduction (%), and the red color
symbol represented the negative reduction for the treatment site.
229
5.3 Total Suspended Sediments (TSS)
Total suspended sediments (TSS) are defined as the particles in water that will
not pass through a 0.45-μm filter.) The TSS include clay and silt particles, fine organic
debris, and other particulate matters [38]. Elevated concentrations of sediments affect
the clarity of the water [38]. Higher concentrations result in less light passing through
water, which reduces the photosynthesis of aquatic plants and can lead to rapid heating
of water that might adversely affect the aquatic life that has been adapted to a lower
temperature [38]. In addition, suspended sediments can serve as carriers of toxins such
as pesticides that readily cling to the particles’ surfaces [38]. In the following sections,
one and two-way analysis of various for TSS are presented.
5.2.1 Analysis of One-way and Two-way ANOVA
TSS showed a variation in different types of rainfall (Figure 14) intensity and
season (Figure 15). This results however is different from that is NO3-N and PO4-P. It
showed that the maximum TSS (3.5 mg/l) in the control site 5 in heavy rainfall and
moderate-heavy (2.5 mg/l) and minimum TSS concentration is detected in site 1 through
site 4 during heavy and moderate-heavy rainfall. Figure 16 also shows that maximum
TSS concentration (3.5 mg/l) was seen during the winter season for site 5.
230
Figure 14: Statistical parameter of the different types of rainfall
Overall, the TSS range in all six sites was low (0.074 – 3.5) mg/l. The results of
the two-way ANOVA were F(3,12) = 3.5, p < 0.05, which showed significant
differences among the TSS values among the sites. A two-way ANOVA was employed
to understand the effects of rainfall intensity and season on TSS variation. The results
showed that there was a significant main effect of heavy rainfall (F(2.4, 1.5) = 3.5, p <
0.05) and a significant main effect of season (F(3, 1.1) = winter, p< 0.05, ηp 2 = 0.14).
The post-hoc multiple comparison test disclosed significant differences (p<0.05)
between control and treatment TSS concentrations.
To assess the overall interaction between three variables, two-way model
conducted for this study and AIC (293) (Table 6) showed the lowest value for treatment
and seasonal variables interaction. The model revealed that for this area and the analysis
is more interactive with season and treatment rather rainfall category. The reason for
having lower interaction on rainfall category is the low elevation and almost flat slope
231
of the study area. The residual and Normal Q-Q plot (Figure 16) showed the efficiency
of the model.
Figure15: Statistical parameter of the different types of seasons
The diagnostic plots (Figure 16) showed the variance (residuals) across the range
of the observed data. The red lines represented the mean of the residuals. The more it
lies on the zero lines horizontally means no significant outliers and no high difference
on variance. In this figure, the mean of the residual lied horizontally on the zero lines.
The Normal Q-Q showed an inverted s shape. Approximately from the values (-3, -2.4),
the sample grows lower than the standard normal distribution; therefore, it takes longer
for sample quantiles to increase. From the values (-2.4, 1.9) the sample seems to grow
at approximately the same pace as the standard normal distribution.
232
Figure 16: Residuals and Normal Q-Q plot of the developed model
The study found a significant response of ortho-phosphate for treatment
variables (treatment 0, treatment 1, and treatment 2) are shown in the figure 17 and the
result of the mean of each treatment types are represented in the Table 9.
Table 9: TSS response on treatment type
Treatment Type Group No of the Sites Mean (mg/l)
Treatment 0 A Site_2 0.0874
B Site_5 1.02
Treatment 1 C Site_1 0.457
D Site_4 0.507
Treatment 2 E Site_3 0.569
F Site_6 0.550
233
Figure 17: TSS response to treatment types in each of the sites
Due to a significant interaction between treatment 2 and treatment 0 (P< 0.05) it
was clearly proved that the treatment 2 is the more affective in terms of reducing nitrate
concentration. The interaction between treatment 1 and treatment 0 (P > 0.05) found
treatment site 1 less effective than treatment 2. However, the mean of the two treatment
sites (Table 9) (treatment 1 and treatment 2) is very close. So, the very close variance
of treatments 1 and 2 proved that they both are equally effective.
5.2.2 Reduction rate of the Total Suspended Sediments (TSS)
Reduction of TSS is represented in figure 18 where it showed the overall
reduction between control and treatment sites (Site 2 (control) vs Site 1 (treatment).
234
Again, result showed that TSS also reduced through the treatment sites was observed
very clearly from this analysis.
235
Figure 18: Comparison of TSS concentration reduction between control and treatment
site. Blue colored symbol represented the reduction (%), and the red color symbol
represented the negative reduction for the treatment site.
Out of the total 42 rainfall events, TSS reduction was seen for 37 rainfall events,
and the range of reduction (%) was (0 to 100)% in site 1, whereas for the site 3 TSS
reduction for 31 rainfall events ranged from 0% to 988% For site 4, PO4-P concentration
reduction was calculated. For 34 storms TSS reduction range was up to almost 95%.
Site 6 also showed TSS reduction for 33 rainfall events for the implemented soil media,
and the removal range was from (0 to 94)%.
6. Conclusion
The roadside best management practices (BMPs) performance is evaluated to
reduce or improve the quality of stormwater runoff. The grass swales are evaluated as a
simple and effective stormwater control measure for this suburban pollutant treatment.
The grass swales significantly reduced pollutant and mean concentrations for nitrate and
236
nitrite, ortho-phosphate, and TSS. Nutrient treatment was variable and effective for
most rainfall events, and there is a seasonal effect and usually occurring during the
summer months. The grass swales consistently remove all three nutrients except for a
few numbers of rainfall events. Generally, Nitrate-N, orthophosphate-P, and total
suspended sediments are the common pollutant problem for Rhode Island areas from
the stormwater runoff. In addition, non-point source (NPS) pollution is a significant
management concern in Rhode Island’s waters – both surface waters and groundwaters.
The performance of the roadside best management practices is evaluated in this
study. The grass swales significantly reduced pollutant concentration for nitrate,
phosphate, and total suspended sediments. Statistical analysis showed that these soil
media are capable of reducing about (20 – 45)% of Nitrate-N, (20-60)% of
Orthophosphate-P, and (20-90)% of TSS through implemented organic soil media.
This research suggests that the grass swales generally improve the suburban
runoff water quality and can be employed to treat non-point source pollution.
Therefore, they should be considered to reduce the environmental impacts due to
stormwater runoff.
References
1. US EPA. National Water Quality Inventory, 2002 Report. Washing DC: US Environ-
mental Protection Agency, EPA-841-R-02-001, 2002.
2. Park, D.; Song, Y.-I.; Roesner, L.A. Effect of the Seasonal Rainfall Distribution on
Storm-Water Quality Capture Volume Estimation. J. Water Resour. Plan. Manag.,
2013, 139, 45–52.
3. Shrestha, P.; Hurley, S.E.; Wemple, B.C. Effects of different soil media, vegetation,
and hydrologic treatments on nutrient and sediment removal in roadside bioretention
systems. Ecol. Eng. 2018, 112, 116–131.R-02-001, 2002.
237
4. US EPA. National Water Quality Inventory, 2002 Report. Washing DC: US Environ-
mental Protection Agency, EPA-841-R-02-001, 2002.
5. Department of Environmental Resources of Prince George’s County. Low Impact De-
velopment Design Strategies: An Integrated Design Approach. Largo, MD: Develop-
ment of Environmental Resources of Prince George’s County, 1999
6. WSDOT, Stormwater Best Management Practices- Highway Runoff Manual M31-16-
04, April 2014.
7. RISDISM. Rhode Island Stormwater Design and Installation Standards Manual, 2010,
Rhode Island Department of Environmental Management and Coastal Resources Man-
agement Council.
8. USEPA (US Environmental Protection AgencyStormwater technology fact sheet. Veg-
etative Swales, 1999c, Washington D.C: Office of Water, EPA 832-F-99-006
9. Kirby, J.T., Durrans, S R., Pitt, R., & Johnson, P.D. Hydraulic resistance in grass swales
designed for small flow conveyance. Journal of Hydrologic Engineering, 2005, 131 (1),
65-68.
10. Barrett M., Kearfott P., Malina Jr J., Landphair H., Li M.-H., Olivera F., Rammohan P.
PollutantRemoval on Vegetated Highway Shoulders, 2006.
11. Barrett M.E., Walsh P.M., Jr J.F.M., Charbeneau R.J. Performance of vegetative con-
trols for treating highway runoff. Journal of environmental engineering, 1998,
124:1121-1128.
12. Winston R., Bouchard N., Hunt W. (2013) Carbon Sequestration by Roadside Filter
Strips and Swales: A Field Study, Green Streets, Highways, and Development, Advanc-
ing the Practice, ASCE, 2013, pp. 355-367.
13. Ebihara T., Young C.B., Tiwari V., Agee L.M. Treatment of contaminated roadway
runoff using vegetated filter strips, 2009, Kansas Department of Transportation.
14. Yonge D. Contaminant Detention in Highway Grass Filter Strips, 2000, Washington
State University Pullman. pp. 73.
15. Stagge J.H., Davis A.P., Jamil E., Kim H. Performance of grass swales for improving
water quality from highway runoff, Water research, 2012, 46:6731-6742.
16. Deletic A., Fletcher T.D. Performance of grass filters used for stormwater treatment—
a field and modelling study. Journal of Hydrology, 2006, 317:261-275.
17. Mitchell G.F., Riefler R.G., Russ A. Vegetated Biofilter for Post Construction Storm
Water Management for Linear Transportation Projects–Dormant Grass Test Supple-
ment, 2010b, USFHWA/OH- 2010/7.
238
18. Storey B.J., Li M.-H., McFalls J.A., Yi Y.-J. Stormwater Treatmtne with Vegetated
Buffers.Texas Transportation Institute; Cambridge Systematics, Incorporated, 2009,
108p http://trid.trb.org/view/1266327.
19. Brown. R. N.,Gorres. G. H.. The Use of Soil Amendments to Improve Survival of Road-
side Grasses, HORTSCIENCE 46(10):1404–1410. 2011. Retrieved from
http://hortsci.ashspublications.org/content/46/10/1404.full.pdf
20. Block, D. Controlling erosion from highway projects. Biocycle, 2000, 41:59.
21. EPA. Innovative uses of compost. 1997, U.S.Environmental Protection Agency.
22. Karlen, D.L.; Stott, D.E. A framework for evaluating physical and chemical indicators
of Soil quality.1994, In Defining Soil Quality for a Sustainable Environment;
23. Doran, J.W.,Coleman, D.C.,Bezdicek, D.F., Stewart, B.A., Eds.; Soil Science Society
of America: Madison, WI, USA, 1994;pp. 53–72
24. Brown R.N., Maynard B. Evaluation of Native Grasses for Highway Slope Stabilization
& Salt Tolerance, 2010.
25. Davis, A.P., Shokouhian, M., Sharma, H., Minami, C. Water quality improvement
through bioretention media: nitrogen and phosphorus removal. Water Environment Re-
search, 2006, 78 (3), 284–293.
26. Hunt, W.F., Jarrett, A.R., Smith, J.T., Sharkey, L.J. Evaluating bioretention hydrology
and nutrient removal at three field sites in North Carolina. Journal of Irrigation and
Drainage Engineering-ASCE, 2006, 132 (6), 600–608.
27. Kim, H.H., Seagren, E.A., Davis, A.P. Engineered bioretention for removal of nitrate
from stormwater runoff. Water Environment Research, 2003, 75 (4), 355–367.
28. Dietz, M.E., Clausen, J.C. Saturation to improve pollutant retention in a rain garden.
Environmental Science and Technology, 2006 40 (4), 1335–1340.
29. MSM (2010). Minnesota Stormwater Manual 2010
30. Danuta Barałkiewicz, Maria Chudzińska, Barbara Szpakowska, Dariusz Świerk,
Ryszard Gołdyn, Renata Dondajewska Environ Monit Assess. 2014; 186(10): 6789 -
6803. Published online 2014 Jul 2. doi: 10.1007/s10661-014-3889-0
PMCID: PMC4149883
31. P. Alpert, T. Ben-Gai, A. Baharad, Y. Benjamini, D. Yekutieli, M. Colacino, L. Dio-
dato, C. Ramis, V. Homar, R. Romero, S. Michaelides, A. Manes, The paradoxical in-
crease of Mediterranean extreme daily rainfall in spite of decrease in total values, Ge-
ophys. Res. Lett., 2002, 29, 1–31
32. Western Washington Phase II Municipal Stormwater Permit report 2010
239
33. Collins, K.A., Lawrence, T.J., Stander, E.K., Jontos, R.J., Kaushal, S.S., Newcoer, T.A.,
Grimm, N.B., Cole Ekberg, M.L. Opportunities and challenges for managing nitrogen
in urban stormwater: a review and synthesis. Ecol. Eng., 2010, 36,1507e1519.
34. Kruzic, A.P., Schroeder, E.D., 1990. Nitrogen removal in the overland flow wastewater
treatment process e removal mechanisms. Res. J. Water Pollu. Control, 1990, Fed. 62,
867e875.
35. Jahan, K.; Pradhanang, S.M. Predicting Runoff Chloride Concentrations in Suburban
Watersheds Using an Artificial Neural Network (ANN). Hydrology 2020, 7, 80.
https://doi.org/10.3390/hydrology7040080
36. Hirsch, R. Seizing the light: A history of photography, 2000, Boston: McGraw-Hill.
37. Larson. M.G. Analysis of Variance, 2008, https://doi.org/10.1161/CIRCULA-
TIONAHA.107.654335.https://www.ahajournals.org/doi/full/10.1161/circula-
tionaha.107.654335
38. Vaze, J., Chiew, H.S. Nutrient loads associated with different sediment sizes in urban
stormwater and surface pollutants. J. Envir. Eng. ASCE, 2004, 130, 391e396.
39. Taylor, G.D., Fletcher, T.D., Wong, T.H.F., Breen, P.F., Duncan, H.P., 2005.
40. Nitrogen composition in urban runoff implications for stormwater management. Water
Res. 39, 1982e1989.
41. Collins, K.A., Lawrence, T.J., Stander, E.K., Jontos, R.J., Kaushal, S.S., Newcomer,
T.A., Grimm, N.B., Cole Ekberg, M.L. Opportunities and challenges for managing ni-
trogen in urban stormwater: a review and synthesis. Ecol. Eng. 2010, 36,1507e1519.
42. Kruzic, A.P., Schroeder, E.D. Nitrogen removal in the overland flow wastewater treat-
ment process e removal mechanisms. Res. J. Water Pollu. Control, 1990, Fed. 62,
867e875.
43. Razzaghmanesh. M., Beecham. S. Kazemi. F. 2014. Impact of green roofs on storm-
water quality in a South Australian urban environment. Science of the Total Environ-
ment 470–471 (2014) 651–659. http://dx.doi.org/10.1016/j.scitotenv.2013.10.047
240
Appendix
Table: Statistical Parameters value for Nitrate-N of Rainfall category [ Instrument: Ion
Chromatography, Standard: DionexTM combined seven Anion standard, Standard
concentration unit: mg/l, Method: EPA 300.1 (A), Detection Limit: 0.02mg/l] Types of
Rainfall
Paramet
er
Site_1
(Treatme
nt)
Site_2
(Control)
Site_3
(Treatment
)
Site_4
(Treatment)
Site_5
(Control)
Site_6
(Treatment)
Moderate-
Heavy
Min 2 2.1 1.5 1.4 0.54 0.3
Median 4.55 5.2 4.7 4 1.8 1.25
Max 6.1 8.1 6.1 6.8 3.45 2.6
Light-Moderate Min 2.6 3.6 3.18 3.3 0.6 0.9
Median 5.4 5.05 4 4.3 1.9 1.3
Max 7.1 7.5 7.9 4.8 3.1 2.6
Heavy Min 2.1 2.3 1.3 1.4 0.4 0.23
Median 3.7 5.6 3.2 4.1 1.8 1.3
Max 7.6 15.7 13.5 11.3 4.1 2.1
Heavy-
Torrential
Min 3.6 5.4 4.7 5.2 2.2 1.7
Median 5.4 6.1 5.2 5.7 2.5 1.7
Max 7.3 6.8 5.7 6.3 2.9 1.7
241
Table: Statistical Parameters value for Nitrate-N of Rainfall category [Instrument: Ion
Chromatography, Standard: DionexTM combined seven Anion standard, Standard
concentration unit: mg/l, Method: EPA 300.1 (A), Detection Limit: 0.02mg/l] Season Param
eter
Site_1
(Treatm
ent)
Site_2
(Control)
Site_3
(Treatme
nt)
Site_4
(Treatmen
t)
Site_5
(Control)
Site_6
(Treatmen
t)
Fall Min 2 2.3 1.3 1.4 0.4 0.23
Median 3.6 5 3.7 4.1 1.4 0.9
Max 6.2 7.5 7.9 6.3 4.1 2.6
Spring Min 2.1 2.7 2.2 2.8 1.8 1.2
Median 4.4 6.7 5.8 4.7 2.2 1.4
Max 6.1 15.7 13.5 11.3 2.9 2.1
Summer Min 3.1 4.4 2.4 3.3 1.1 0.6
Median 5.4 7.2 4.7 5.2 1.7 1.4
Max 7.3 8.4 6.3 6.8 2.9 2.1
Winter Min 2.1 2.1 1.5 1.4 0.4 0.3
Median 4.1 4.7 3.2 3.8 1.9 1.4
Max 7.6 8.2 11.1 6.2 3.7 2.6
242
Table: Statistical Parameters value for Orthophosphate-P of Rainfall category
[ Instrument: Ion Chromatography, Standard: DionexTM combined seven Anion
standard, Standard concentration unit: mg/l, Method: EPA 300.1 (A), Detection Limit:
0.005 mg/l] Types of
Rainfall
Param
eter
Site_1
(Treatm
ent)
Site_2
(Control)
Site_3
(Treatme
nt)
Site_4
(Treatmen
t)
Site_5
(Control)
Site_6
(Treatmen
t)
Moderate-
Heavy
Min 0.5 0.98 1.2 0.8 0.2 0.2
Median 2.05 4.64 2.8 2.25 2.5 1.3
Max 3.6 8.3 4.4 3.7 4.8 2.4
Light-
Moderate
Min 1.8 1.93 1.9 1.1 0.8 0.4
Median 2.5 4.365 2.7 2.15 1.55 1.1
Max 3.2 6.8 3.5 3.2 2.3 1.8
Heavy Min 1.0 1.8 0.8 0.9 0.1 0.1
Median 2.5 4.5 2.5 2.6 1.5 1.3
Max 7.8 8.4 5.5 5.7 3.0 2.5
Heavy-
Torrential
Min 3.1 3.9 1.2 2.1 1.4 1.4
Median 3.1 4.05 3.25 3.1 2.55 1.75
Max 3.1 4.2 5.3 4.1 3.7 2.1
243
Table: Statistical Parameters value for Orthophosphate of Season [ Instrument: Ion
Chromatography, Standard: DionexTM combined seven Anion standard, Standard
concentration unit: mg/l, Method: EPA 300.1 (A), Detection Limit: 0.005 mg/l] Season Param
eter
Site_1
(Treatm
ent)
Site_2
(Control)
Site_3
(Treatme
nt)
Site_4
(Treatmen
t)
Site_5
(Control)
Site_6
(Treatmen
t)
Fall Min 0.98 1.9 1.08 0.78 0.23 0.21
Median 2.1 4.1 2.8 2.6 1.32 0.98
Max 4.2 8.3 4.9 4.9 3.1 2.12
Spring Min 2.1 2.9 1.3 1.8 0.9 0.6
Median 4.5 4.5 2.1 2.3 1.8 2.1
Max 7.8 6.3 2.7 3.6 2.1 2.5
Summer Min 2.1 3.1 1.7 1.4 1.12 1.2
Median 3.1 4.2 3.2 2.1 2.8 1.5
Max 3.8 5.98 5.3 4.2 4.8 2.4
Winter Min 0.5 1.0 0.8 0.9 0.1 0.1
Median 2.5 3.1 2.5 2.1 1.6 1.3
Max 4.3 8.4 5.5 5.7 2.4 2.4
244
Table: Statistical Parameters value for TSS of Rainfall category [Instrument: Filter
paper, oven, weight scale, Method: 2540D] Types of
Rainfall
Parame
ter
Site_1
(Treatme
nt)
Site_2
(Control)
Site_3
(Treatmen
t)
Site_4
(Treatment
)
Site_5
(Control)
Site_6
(Treatment
)
Moderate-
Heavy
Min 0.06 0.10 0.02 0.06 0.09 0.10
Median 0.53 0.87 0.62 0.46 1.00 0.30
Max 1.10 1.80 1.10 1.60 2.50 1.40
Light-Moderate Min 0.11 0.06 0.07 0.08 0.43 0.50
Median 0.70 0.89 0.47 0.54 0.98 0.65
Max 1.10 0.98 1.02 0.78 1.90 1.50
Heavy Min 0.07 0.12 0.01 0.03 0.12 0.19
Median 0.40 0.98 0.76 0.50 0.98 0.70
Max 0.96 1.60 1.40 1.10 3.50 1.40
Heavy-
Torrential
Min 0.34 1.08 0.76 1.08 0.91 0.2
Median 0.57 1.19 0.98 1.24 1.205 0.65
Max 0.8 1.3 1.2 1.4 1.5 1.1
245
Table: Statistical Parameters value for TSS of Rainfall category [Instrument: Filter
paper, oven, weight scale, Method: 2540D] Season Param
eter
Site_1
(Treatm
ent)
Site_2
(Control)
Site_3
(Treatme
nt)
Site_4
(Treatme
nt)
Site_5
(Control
)
Site_6
(Treatme
nt)
Fall Min 0.07 0.118 0.007 0.028 0.09 0.2
Median 0.56 0.9 0.67 0.65 0.54 0.7
Max 1.1 1.7 1.2 1.4 2.1 1.5
Spring Min 0.074 0.6 0.015 0.102 0.54 0.1
Median 0.26 0.94 0.62 0.45 1.09 0.496
Max 0.7 1.4 0.76 1.2 2.4 1.1
Summer Min 0.06 0.10 0.12 0.13 0.81 0.10
Median 0.45 1.02 0.72 0.89 1.31 0.45
Max 0.80 1.80 1.10 1.60 2.50 0.80
Winter Min 0.11 0.06 0.021 0.08 0.69 0.2
Median 0.45 0.9 0.78 0.4 1.1 0.56
Max 1.1 1.6 1.1 1.1 3.5 1.4