Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Water quality management in small to medium sized distribution networks : optimizing chlorine disinfection… Islam, Nilufar 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2015_November_Islam_Nilufar.pdf [ 5.08MB ]
Metadata
JSON: 24-1.0166702.json
JSON-LD: 24-1.0166702-ld.json
RDF/XML (Pretty): 24-1.0166702-rdf.xml
RDF/JSON: 24-1.0166702-rdf.json
Turtle: 24-1.0166702-turtle.txt
N-Triples: 24-1.0166702-rdf-ntriples.txt
Original Record: 24-1.0166702-source.json
Full Text
24-1.0166702-fulltext.txt
Citation
24-1.0166702.ris

Full Text

WATER QUALITY MANAGEMENT IN SMALL TO MEDIUM SIZED DISTRIBUTION NETWORKS: OPTIMIZING CHLORINE DISINFECTION STRATEGIES   by  Nilufar Islam  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY  in  The College of Graduate Studies (Civil Engineering) THE UNIVERSITY OF BRITISH COLUMBIA (Okanagan) September 2015 © Nilufar Islam, 2015ii  ABSTRACT Main objective of this research is to optimize booster chlorination to ensure high quality water in the distribution network (DN). Focus of this research is primarily small to medium sized DNs that lack continuous monitoring, where chlorination is the predominant disinfection practice to maintain acceptable drinking water quality. In this research, new methods and strategies have been proposed to help in selecting location and dosages for booster chlorination, which protects against microbiological contamination and biofilm growth but also limit formation of harmful disinfectant by-products (DBPs), and chlorine related taste & odors issues.  This research developed index and risk based schemes to optimize water quality in DNs. Three indices have been proposed: 1) non-compliance potential index (NCP index), 2) modified Canadian Council of Ministries of the Environment Water Quality index (Modified CCME WQI), and 3) intrusion risk potential (IRP). NCP index has been developed to evaluate regulatory violations of DBPs using Bayesian Belief Network. The modified CCME WQI, an extension of commonly accepted CCME WQI, has been developed to evaluate Stage 1 and Stage 2 DBP rules. Risk based index (IRP) has been developed to identify potential intrusion points based on pollutant source, water main characteristics, soil properties, operational and land use factors and population served. Three optimization schemes and algorithms have been proposed to determine the number of booster locations and corresponding dosage levels. Modified CCME WQI has been used to select optimal dosage for booster chlorination using response surface optimization.  It uses temporal series data for free residual chlorine (FRC) and converts into an index by maximizing water quality. Later another optimization algorithm has been developed to locate booster stations using FRC and total trihalomethane time series data. This algorithm is called maximum covering location problem, which has been developed using EPANET –MSX programmer’s toolkit integrated with Matlab coding. Third optimization algorithm has been developed to minimize the impacts of contaminant intrusion using booster chlorination. This scheme uses multi-objective genetic algorithm to select both location and dosages for booster chlorination. Proposed methods and strategies have been demonstrated using case studies on City of Kelowna and Quebec City DNs.  iii  PREFACE I, Nilufar Islam, conceived and developed all the contents of this thesis under the supervision of Drs. Rehan Sadiq and Manuel J. Rodriguez. The other co-authors of the articles include Drs. Christelle Legay, Ashraf Farahat, and Mohammad Abdullah M. Al-Zahrani. Both supervisors have reviewed all manuscripts and provided feedbacks to improve the manuscripts and the thesis. Most of the contents of this thesis are published, accepted or submitted for publication in journals and conference proceedings.  A version of Chapter 2 has been accepted for publication in Environmental Reviews Journal with the title “Contaminant intrusion in water distribution networks: review and proposal of an integrated model for decision making” (Islam et al. 2015a). This article has been accepted and under editorial revisions.   A version of Chapter 3 has been submitted for publication in Environmental Monitoring and Assessment Journal with the title “Assessing regulatory violations of disinfection by-products in water distribution networks using a non-compliance potential index” (Islam et al. 2015b).  A version of Chapter 4 has been submitted in Water SA Journal with the title “Assessment of water quality in distribution networks through the lens of DBP Rules” (Islam et al. 2015c).   Contents from Chapters 5 and 8 are under internal review for a possible publication in Water Research Journal with the title “Optimizing booster chlorination to minimize the impacts of contaminant intrusion in a small water distribution network” (Islam et al. 2015d).   A version of Chapter 6 has been published in Environmental Monitoring and Assessment Journal with the title “Optimizing booster chlorination in water distribution networks: a water quality index approach” (Islam et al. 2013).   A version of Chapter 7 is under preparation for a possible publication in Water Research Journal with the title “Locating chlorine booster stations in small water distribution networks: a methodology using optimization and trade-off analysis” (Islam et al. 2015e).  A version of Chapter 6 has been published in The Proceedings of the 15th Canadian national Conference & 6th Policy Forum on Drinking Water (2012) with the title iv  “Locating optimal locations for booster chlorination using a modified CCME water quality index” (Islam et al. 2012a).   A version of Chapter 6 has been published in the 1st International Conference on Advances in Civil Engineering (2012) with the title “Managing water quality in small distribution networks using booster chlorination: a water quality index approach” (Islam et al. 2012b).      v  TABLE OF CONTENTS ABSTRACT .................................................................................................................................... ii PREFACE ...................................................................................................................................... iii TABLE OF CONTENTS ................................................................................................................ v LIST OF TABLES .......................................................................................................................... x LIST OF FIGURES ....................................................................................................................... xi LIST OF ABBREVIATIONS ...................................................................................................... xiv LIST OF SYMBOLS ................................................................................................................. xviii ACKNOWLEDGEMENTS ........................................................................................................ xxii DEDICATION ........................................................................................................................... xxiv Chapter 1 : INTRODUCTION........................................................................................................ 1 1.1 Background and motivation ............................................................................................. 1 1.2 Research objectives .......................................................................................................... 4 1.3 Thesis organization ............................................................................................................... 4 Chapter 2 : LITERATURE REVIEW AND PROPOSED FRAMEWORK .................................. 6 2.1 Literature review ................................................................................................................... 6 2.1.1 Water quality failure in distribution network................................................................. 6 2.1.2 Water quality regulations ............................................................................................... 9 2.1.3 Water quality monitoring ............................................................................................. 11 2.1.4 Water quality assessment ............................................................................................. 14 2.1.5 Water quality management .......................................................................................... 21 2.2 Thesis framework................................................................................................................ 39 2.2.1 Non-compliance potential index .................................................................................. 39 2.2.2 Modified CCME WQI for complying with DBP regulations ...................................... 41 2.2.3 Risk based model for contaminant intrusion ............................................................... 41 vi  2.2.4 Optimization to select booster dosage ......................................................................... 41 2.2.4 Optimization to select booster location........................................................................ 41 2.2.5 Optimization to select both booster location and dosage............................................. 42 Chapter 3 NON-COMPLIANCE POTENTIAL INDEX ............................................................. 43 3.1 Background ......................................................................................................................... 43 3.2 Methodology ....................................................................................................................... 44 3.2.1 Data collection ............................................................................................................. 44 3.2.2 Estimating probability of exceedance .......................................................................... 48 3.2.3 Inference ...................................................................................................................... 50 3.3 Case studies ......................................................................................................................... 54 3.3.1 Study area..................................................................................................................... 54 3.3.2 Seasonal assessment: Québec City .............................................................................. 57 3.3.3 Regression analysis for Quebec City ........................................................................... 58 3.3.4 EPANET results for City of Kelowna ......................................................................... 60 3.3.5 Uncertainty and sensitivity analysis............................................................................. 62 3.4 Discussion and future possibilities...................................................................................... 63 3.5 Summary ............................................................................................................................. 68 Chapter 4 : DEVELOPING MODIFIED CCME WQI TO COMPLY WITH DBP RULES ....... 69 4.1 Background ......................................................................................................................... 69 4.2 Methodology ....................................................................................................................... 70 4.2.1 Parameter selection ...................................................................................................... 70 4.2.2 Conversion to sub-indices ............................................................................................ 70 4.2.3 Assignment of weights ................................................................................................. 72 4.2.4 Aggregation.................................................................................................................. 77 4.3 Water quality assessment: a case study .............................................................................. 79 vii  4.3.1 Study area..................................................................................................................... 79 4.3.2 Sampling and analytical methods ................................................................................ 79 4.3.3 Monthly water quality assessment ............................................................................... 79 4.3.4 Spatial variability of WQI ............................................................................................ 82 4.4.5 Comparison with other models .................................................................................... 83 4.4.6 Sensitivity after adding other DBPs ............................................................................. 83 4.4 Summary ............................................................................................................................. 87 Chapter 5 : IDENTIFYING CONTAMINANT INTRUSION POINTS ...................................... 88 5.1 Background ......................................................................................................................... 88 5.2 Methodology ....................................................................................................................... 88 5.2.1 Sanitary mains .............................................................................................................. 90 5.2.2 Water mains ................................................................................................................. 90 5.2.3 Soil corrosivity ............................................................................................................. 92 5.2.4 Water main pressure .................................................................................................... 93 5.2.5 Consequence factors .................................................................................................... 94 5.3 Case study ........................................................................................................................... 96 5.3.1 Study area..................................................................................................................... 96 5.3.2 Results and discussions ................................................................................................ 97 5.4 Summary ........................................................................................................................... 102 Chapter 6 : OPTIMIZING BOOSTER DOSAGE ...................................................................... 104 6.1 Background ....................................................................................................................... 104 6.2 Methodology ..................................................................................................................... 105 6.2.1 EPANET network analysis ........................................................................................ 107 6.2.2 Temporal and spatial analyses ................................................................................... 109 6.2.3 Critical zone selection ................................................................................................ 111 viii  6.2.4 Factorial analysis and 3D surface optimization ......................................................... 112 6.3 City of Kelowna - A Case Study....................................................................................... 113 6.3.1 Temporal and spatial analyses ................................................................................... 114 6.3.2 Optimizing booster chlorination ................................................................................ 117 6.3.3 Model sensitivity ........................................................................................................ 119 6.4 Summary ........................................................................................................................... 119 Chapter 7 : OPTIMIZING BOOSTER LOCATIONS................................................................ 124 7.1 Background ....................................................................................................................... 124 7.2 Methodology ..................................................................................................................... 125 7.2.1 Reaction kinetics ........................................................................................................ 125 7.2.2 Optimization to locate booster stations ...................................................................... 127 7.2.3 Modelling THM species ............................................................................................ 129 7.2.4 Cancer and non-cancer risk potentials ....................................................................... 131 7.2.5 Life cycle costing for booster chlorination ................................................................ 132 7.3 Scenario Analysis.............................................................................................................. 133 7.4 City of Kelowna- Case Study ........................................................................................... 140 7.4.1 Optimization results ................................................................................................... 143 7.4.2 Trade-off results ......................................................................................................... 150 7.5 Summary ........................................................................................................................... 154 Chapter 8 : MINIMIZING THE IMPACTS OF CONTAMINANT INTRUSION .................... 155 8.1 Background ....................................................................................................................... 155 8.2 Methodology ..................................................................................................................... 156 8.2.1 Intrusion and DBPs modelling ................................................................................... 157 8.2.2 Multi-objective optimization for booster chlorination ............................................... 158 8.3 Case Study ........................................................................................................................ 162 ix  8.3.1 Selecting the “best” solution ...................................................................................... 165 8.3.2 Water quality improvement ....................................................................................... 170 8.4 Summary ........................................................................................................................... 175 Chapter 9 : CONCLUSIONS AND RECOMMENDATIONS .................................................. 176 9.1 Summary and Conclusions ............................................................................................... 176 9.2 Limitations and Recommendations................................................................................... 178 9.2.1 Indices ........................................................................................................................ 178 9.2.2 Optimization schemes and algorithms ....................................................................... 179 9.2.3 Decision support tool ................................................................................................. 181 References ................................................................................................................................... 183 Appendices .................................................................................................................................. 208 Appendix A: Water quality parameter, significance and monitoring locations ...................... 208 Appendix B: Selected water quality parameter measurement details ..................................... 210 Appendix C: Risk analysis approaches used for contaminant intrusion ................................. 212 Appendix D: Geostatistical approaches used in contaminant intrusion studies ..................... 213 Appendix E: Software tools used for predicting contaminant intrusion ................................. 214 Appendix F: Proposed integrated model for contaminant detection and mitigation .............. 215 Appendix G: Life cycle cost estimation for gas chlorination ................................................. 218    x  LIST OF TABLES Table 2-1: Regulations relevant to DN safety ............................................................................... 10 Table 2-2: Water quality index limitations ................................................................................... 17 Table 2-3: Slope factor, reference dose and other hazard information for commonly available DBPs ............................................................................................................................. 19 Table 2-4: Related QMRA studies ................................................................................................ 22 Table 2-5: Summary of the optimization studies from 1992 to 2013 ........................................... 26 Table 2-6: Objective functions used in common optimization studies ......................................... 29 Table 2-7: Algorithm selection criteria for optimization .............................................................. 34 Table 2-8: Water quality optimization in a DN related to booster disinfection............................ 37 Table 3-1: Regression results for predicting TTHM and HAA5 .................................................. 61 Table 3-2: NCP evaluation for SP 6 with TTHM and HAA5 models .......................................... 62 Table 4-1: Regulatory maximum contaminant level (MCL) and maximum contaminant level goals (MCLG) for DBPs and related parameters (USEPA 2010b) .............................. 72 Table 4-2:Sub-index (SI) functions used in this study.................................................................. 74 Table 4-3: Sub-criteria with scores for estimating parameter weights ......................................... 75 Table 4-4: Estimating the parameters weights .............................................................................. 78 Table 5-1: Proposed rating schemes for water main diameter and age ........................................ 91 Table 5-2: Scoring schemes for soil corrosivity for metallic pipe (Sadiq et al. 2010) ................. 93 Table 5-3: Soil corrosivity scoring system for cementitious and plastic pipes (Sadiq et al. 2012) ...................................................................................................................................... 94 Table 5-4: Proposed point scoring method for water main pressure ............................................ 94 Table 6-1: Summary of chlorine decay kinetics ......................................................................... 108 Table 6-2: Temporal and spatial analysis results ........................................................................ 114 Table 7-1: Details for scenario analysis ...................................................................................... 136    xi  LIST OF FIGURES Figure 1-1: Reported water quality optimization studies for small, medium, and large-sized DNs during a period of 1997 to 2012 ................................................................................... 3 Figure 1-2: Thesis structure and organization ................................................................................ 5 Figure 2-1: Common pathways and situations for contaminant intrusions .................................... 8 Figure 2-2: Dose-response for chemical risk assessment ............................................................. 19 Figure 2-3: Sample solution for the set of covering algorithm ..................................................... 30 Figure 2-4: Flow chart of the thesis framework............................................................................ 40 Figure 3-1: Methodology explanation to estimate non-compliance potential (NCP) index ......... 46 Figure 3-2: Estimating probability of exceedance for a) single value, and b) a range ................. 50 Figure 3-3: Explanation of breakpoints in combined probability calculation .............................. 51 Figure 3-4: An explanation of BBN using probability and conditional probability table (CPT) . 52 Figure 3-5: Sampling points in a Quebec City DN ....................................................................... 56 Figure 3-6: Study area in the city of Kelowna DN ....................................................................... 57 Figure 3-7: Seasonal NCP index evaluation ................................................................................. 59 Figure 3-8: NCP evaluation in a small part of the city of Kelowna DN ....................................... 63 Figure 3-9: Uncertainty analysis in NCP evaluation- Quebec City 2007 ..................................... 65 Figure 3-10: Sensitivity analysis, a) TTHM, and b) HAA5 ......................................................... 66 Figure 3-11: Possible NCP index associated with microbiological contamination ...................... 67 Figure 4-1: Methodology to assess water quality using modified CCME WQI for DBP rules.... 71 Figure 4-2: Proposed sub-index function, a) increasing, and b) optimal ...................................... 73 Figure 4-3: Monthly variation of water quality index in seven sampling stations ....................... 81 Figure 4-4: Water quality index spatial variation using Kriging in ArcMap a) Summer 2006, b) Fall 2006, c) Summer 2007, and c) Fall 2007 ............................................................ 84 Figure 4-5: Comparison with other WQIs .................................................................................... 85 Figure 4-6: Sensitivity observation after parameter addition in a) Turbidity, b) FRC, and c) TTHM ........................................................................................................................ 86 Figure 5-1: Hierarchical scheme to identify intrusion points in a DN .......................................... 89 Figure 5-2: City of Kelowna water main system with Thiessen polygons ................................... 98 Figure 5-3: Point scoring results showing pollution source index (PSI) in the City of Kelowna DN .............................................................................................................................. 99 xii  Figure 5-4: Point scoring results showing structural failure index (SFI) in the City of Kelowna DN ............................................................................................................................ 100 Figure 5-5:  Point scoring results showing Soil corrosivity index (SCI) in the City of Kelowna DN ............................................................................................................................ 101 Figure 5-6: Intrusion risk potential (IRP) ranks on the City of Kelowna DN ............................ 103 Figure 6-1: Optimization steps using modified CCME WQI ..................................................... 106 Figure 6-2: Temporal variation in residual chlorine concentration at a given node ................... 108 Figure 6-3: CCME WQI and other possible modified functions ................................................ 109 Figure 6-4: Fuzzy excursion estimation using a modified CCME WQI function: ..................... 111 Figure 6-5: Percentage of various water main materials installed from 1939 to 2010 ............... 113 Figure 6-6: City of Kelowna DN, a) water main network in GIS shape file, b) EPANET      model, and c) zoning system with booster station ................................................... 115 Figure 6-7: Temporal and spatial analysis using modified CCME WQI for a) summer and b) winter ....................................................................................................................... 116 Figure 6-8: Water main system for Zone 1 ................................................................................. 117 Figure 6-9: 3D surface optimization (with dose 0.6 and 1 mg/L) in Zone 1 .............................. 118 Figure 6-10: Optimization with an additional booster station (for dose 0.6 and 1 mg/L) in       Zone 1 .................................................................................................................... 121 Figure 6-11: Water quality analysis with optimization for summer a) present situation and b) after optimization ................................................................................................... 122 Figure 6-12: Sensitivity analysis for a modified CCME WQI ................................................... 123 Figure 7-1: Steps involved in optimization and trade-off analysis ............................................. 126 Figure 7-2: Conversion functions used for a) FRC, and b) TTHM ............................................ 128 Figure 7-3: Network information for scenario analysis, a) network connectivity and dimensions, and b) base demand .................................................................................................. 135 Figure 7-4: Simulated FRC for, a) Scenario 1, b) Scenario 2, c) Scenario 3 and d) Scenario 5 . 137 Figure 7-5: Simulated TTHM for, a) Scenario 1, b) Scenario 2, c) Scenario 3 and d) Scenario      5 ................................................................................................................................ 139 Figure 7-6: Trade-offs between a) water quality index and life cycle cost (LCC), b) Cancer     risk potential (CRP) and non-cancer risk potential (NCRP) .................................... 140 Figure 7-7: Study area on the city of Kelowna DN with scenario details for the case study ..... 142 xiii  Figure 7-8: Water quality change for Scenario 1, a) FRC and b) TTHM concentrations .......... 144 Figure 7-9: Water quality change for Scenario 2, a) FRC and b) TTHM concentrations .......... 145 Figure 7-10: Water quality change for Scenario 3, a) FRC and b) TTHM concentrations ........ 146 Figure 7-11: Water quality change for Scenario 4, a) FRC and b) TTHM concentrations ........ 147 Figure 7-12: Water quality change for Scenario 5, a) FRC and b) TTHM concentrations ........ 148 Figure 7-13: Water quality change for Scenario 6, a) FRC and b) TTHM concentrations ........ 149 Figure 7-14: WQI change after booster station addition ............................................................ 150 Figure 7-15: Trade-offs between water quality index and life cycle cost (LCC) for the case   study ....................................................................................................................... 152 Figure 7-16: Case study results, a) CRP, and b) NCRP variations ............................................. 153 Figure 8-1: Proposed methodology with three components for analysis .................................... 156 Figure 8-2: City of Kelowna – a) water main system with Thiessen polygons, and b) a small   part for performing optimization.............................................................................. 163 Figure 8-3: Intrusion risk potential (IRP) ranks on a part of the City of Kelowna DN .............. 164 Figure 8-4: Pareto fronts for average ChRP vs average MRP .................................................... 165 Figure 8-5: Pareto fronts for average ChRP vs life cycle cost (LCC) ........................................ 166 Figure 8-6: Pareto fronts for average MRP vs life cycle cost (LCC) ......................................... 167 Figure 8-7: Proposed and present booster stations in the City of Kelowna ................................ 168 Figure 8-8: Average ChRP, average MRP and LCC comparisons for five selected optimal solutions ................................................................................................................... 169 Figure 8-9: Average ChRP, average MRP and LCC comparisons among initial conditions, present booster stations and proposed solution after optimization .......................... 172 Figure 8-10: Proposed booster effect on Node ET67 for, a) E.coli O157:H7, b) Free residual chlorine, and c) TTHM concentrations .................................................................. 173 Figure 8-11: Proposed booster effect on Node J-6355 for, a) E. coli O157:H7, b) Free residual chlorine, and c) TTHM concentrations .................................................................. 174 Figure 9-1: Conceptual framework for decision support tool ..................................................... 182    xiv  LIST OF ABBREVIATIONS A    Land use- Agricultural AC    Water main made of asbestos cement AHP    Analytical hierarchical method ANN    Artificial neural network AVV    Air vacuum valves BBN    Bayesian belief network BDCM    Bromodichloromethane BIF    Bromide incorporation factor BWAs    Boil water advisories C    Lane use- Commercial CCME WQI Canadian Council of Ministries of the Environment Water Quality Index  CD    Land use- Comprehensive Development CDF    Cumulative distribution function CFU    Colony forming unit ChRP    Chemical risk potential CI    Water main made of Cast iron CIPRA    Cast Iron Pipe Research Association CONC    Water main made of Concrete COP    Water main made of Copper CPCB- WQI   WQI by Central Pollution Control Board CPT    Condition probability table CRP    Cancer risk potential CWQI    Canadian WQI CWWA    Canadian Water and Wastewater Association DA     Dissemination area DBAA    Dibromoacetic Acid DBCM    Dibromochloromethane DBPs     Disinfectant by-products xv  DCAA    Dichloroacetic acid DFI    Driving force index DI    Water main made of Ductile iron DIPRA    Ductile Iron Pipe Research Association DN     Distribution Network DS    Dempster-Shafer E. coli    Escherichia coli EPANET-MSX    EPANET Multi-Species Extension FRC    Free Residual Chlorine GA    Genetic algorithms GC-ECD    Gas chromatograph with electron capture detectors GC–MS    Gas chromatograph with mass spectroscopy GIS    Geographical Information System GWT    Ground water table HAA5    Summation of five species of Haloacetic acid HAAs    Haloacetic acids HD    Health District HDPE     High density polyethylene HPC     Heterotrophic plate counts I     Land  use- Industrial ID    Identification number for water main IESWTR    Interim Enhanced Surface Water Treatment Rule IRA-WDS   Integrated Risk Assessment of Water Distribution Systems IRIS    Integrated Risk Information System IRP    Intrusion Risk Potential  LCC    Life cycle cost LIMS    Laboratory information management system LOAEL    Lowest-observed-adverse-effect level LUCI    Land use consequence index LUW    Land use weight MBAA    Monobromoacetic Acid xvi  MCAA    Monochloroacetic acid MCL    Maximum contaminant level MCLG     Maximum contaminant level goal MCLP    Maximum Covering Location Problem MCWQI   Modified CCME WQI MMCD    Master Municipal Construction Documents MOGA    Multi-Objective Genetic Algorithm MOUSE    Model Of Urban Sewers MRDL    Maximum residual disinfectant level MRP    Microbial risk potential NCP index    Non-Compliance Potential index  NCRP    Non-cancer risk potential NOAEL     No-observed-adverse-effect level NSF-WQI   WQI by National Sanitation Foundation    Nephelometric Turbidity Unit O-WQI    Oregon WQI P/W    Public & Institutional PBA    Particle backtracking algorithm PD    Population density PDCI    Population density consequence index PFU    Plaque-forming units PSI    Pollution source index PVC    Polyvinyl chloride PW- WQI   WQI proposed by Pesce and Wunderlin (2000) QMRA    Qualitative microbial risk assessment RAM-W    Risk Assessment Methodology for Water Utilities RfD    Reference dose RN    Representative Node RR    Land use- Rural Residential RU/RM    Land use-Urban Residential xvii  S- WQI    WQI by Said et al. (2004) SCADA    Supervisory Control and Data Acquisition SCI    Soil corrosivity index SCI-C    Soil corrosivity index for cementitious pipes SCI-M    Soil corrosivity index for metallic pipes SCI-P    Soil corrosivity index for plastic pipes SDWA    Safe Drinking Water Act  SEMS    Security Emergency Management System SF    Slope factor SFI    Structural failure index ST- WQI    WQI by Swamee and Tyagi (2007) Stage 1 D/ DBPR  Stage 1 Disinfectants and Disinfection By-products Rule  STEEL    Water main made of Steel SWTR    Surface Water Treatment Rule  T&O    Taste and odor TBM    Bromoform TCAA     Trichloroacetic acid TCM    Chloroform TCR     Total Coliform Rule  TEVA    Threat Ensemble Vulnerability Assessment  TEVA-SPOT   The Threat Ensemble Vulnerability Assessment and Sensor Placement Optimization Tool THMs    Trihalomethanes TOC    Total organic carbon TTHM    Total Trihalomethane USEPA    US Environmental Protection Agency U-WQI    Universal water quality index VIF    Variance inflation factor VSAT    Vulnerability Self-Assessment Tool WQI    Water Quality Index WQP    Water Quality parameter xviii  LIST OF SYMBOLS A/V Surface area per unit volume within the pipe AF   Exposure factor AR Rating assigned based on water main age and type of material (Table 5-1) AT   Average time (days) BP I Order of the breakpoint BW   Body weight (kg) C FRC concentration (mg/L) CF  Conversion factor Cw   Chemical concentration in water (mg/L) tNC  FRC concentration at node N and at time step t tNChRP  Chemical risk potential for Node N and time step t d Dose (concentration multiplied by the volume of ingestion)= Conc × V(volume) D Water main diameter (mm)  DBPi disinfection by-products under study, e. g, if i=1, it is TTHM and if i=2, it is HAA5 DO   Applied dosage in the booster stations DR Rating assigned based on diameter (Table 5-1) E  Activation energy (J/mol) ED   Exposure duration (year) xix  EF   Exposure frequency (days/year) Exp   Exposure dose (mg/kg/day) F Frequency factor to calculate Kb F1   Scope used in CCME WQI calculation F2   Frequency used in CCME WQI calculation F3   Amplitude used in CCME WQI calculation FE   Fuzzy excursion ( )xF x  Probability of the random variable X (parameter value) being less than, or equal to, x (regulatory threshold) HI   Hazard index IR   Ingestion rate (L/day) Ir Annual interest rate (%) k    Overall chlorine decay co-efficient (hr-1) ks Probability of survival (depends on the organism and test) kb  Bulk reaction rate co-efficient (hr-1) kP  Pathogen kinetic decay constant kw Wall reaction rate co-efficient  L Sanitary main length (m)  LCCNb Estimated life -cycle cost for specific booster dosage and node Nb  LF A linear proportionality constant between TTHM and FRC decay 10 yrsLCC  10years of life cycle cost of booster chlorination xx  MChc Maintenance cost for hypo-chlorination ($/yr) tNMRP  Microbiological risk potential for Node N and time step t  N50 Experimental dose at which 50% of the population is expected to be affected Nb Order of node selected as booster station nse  Normalized sum of excursions nsfe    Normalized sum for fuzzy excursions P E. coli O157:H7 concentration (CFU/L) p(E) The evidence probability p(H) Probability of having the hypothesis true p(H/E) The hypothesis belief of H upon observing evidence E PChc Amount of capital installment payments for hypo-chlorination over y years Pi  Final parameter weight by aggregation in modified CCME WQI in Chapter 4 PI DBP precursor indicator Pin   Response (probability of infection) Pn NCP probability for a certain parameter n Q Flow rate for the node selected for booster station (L/Sec) Qi   Demand or flow rate of Node i QN Base water demand for Node N RfD   Reference dose S i, j Score assigned in each parameter in Chapter 4 xxi  SE   Sum of excursion SF   Slope factor SI Sub-index value NTHM  THM species concentrations with N number of bromide atoms; N = 0, 1, 2, and 3 for Chloroform (CHCl3), bromodichloromethane (CHBrCl2),  dibromochloromethane (CHBr2Cl), and bromoform (CHBr3) respectively V1, V2 Factors proposed to estimate modified CCME WQI in Chapter 4 WBpi Weight assigned for each breakpoint Wj Criteria relative weight in Chapter 4 α Beta-poison distribution parameter αhc Cost of chlorine per litre ($/L) β Concentration of liquid chlorine (%)          xxii  ACKNOWLEDGEMENTS First and foremost I would like to thank my supervisors Dr. Rehan Sadiq and Dr. Manuel J. Rodriguez for their patience, guidance and encouragements throughout my research life. It has been an honor to be Dr. Sadiq’s first Graduate student. I can proudly say that he has taught me everything from scratch. Because of his immense encouragement I have achieved Canada wide exposure in research, received awards, and leadership training, which helped me to believe in myself. I really appreciate all his contributions of dedicated time, and great ideas to make my PhD experience dynamic and exciting. He was always supportive not only academically, but also emotionally and morally, which makes me feel being under the best supervisor. I would like to also express my heartfelt thanks to my committee members Dr. Kasun Hewage, Dr. Kenneth Chau, and Dr. Bahman Naser for their valuable comments and constructive suggestions. This research could not be possible without financial support from the Natural Sciences and Engineering Research Council for my NSERC-CGSD2 scholarship and funding provided by Dr. Rehan Sadiq and Dr. Manuel J. Rodriguez. I am very grateful to Dr. Alex Francisque and Dr. Carolyn Labun who helped me in preparing NSERC- CGSD2 proposal. I would like to thank Dr. Majid Mohseni from RES'EAU-WaterNET for providing me various networking opportunities to improve my research with feedbacks from researchers across Canada. I am grateful to Mr. Fred Schaad from the City of Kelowna for providing their valuable data resources, whenever I requested.  For providing data, I would like to also thank Dr. Christelle Legay.  Especial thanks to Dr. Deborah J. Roberts for her encouragement and motherly guidance throughout my graduate life.  I would like to thank my research group, especially Dr. Bahar Reza and Dr. Anna Scheili for their constant inspiration and friendship. My time in Kelowna was made enjoyable because of many friends, especially due to laughter created by Sanjida Minar, Tammeen Siraj, Shelir Ebrahimi, Hanna Hamid, Afrin Hossain and Riffat Binte Siddique. Thanks to Mahfuza Begum Lona and Farjana Alam Mourumi for their delicious foods and selfless care which alleviated my loneliness. I gratefully acknowledge my best friend Ashikur Rahman for his friendship, guidance, encouragements and continuous help in Matlab.  xxiii  For unconditional love and care, I would like to thank my family members, my father, mother, sisters, paternal uncle Mafiz, nephew, niece, and Rony. Thanks to Rony for his patience on my fairy dreams and teaching me how to drive during a stressful period of his life. Thanks to my sisters Nasima Islam, Nazifa Islam, and Nabila Islam for bearing all discrepancies that they might have felt due to my parents’ extreme care to me. Lastly, thanks to my mother, Nurjahan Begum and my father Engr. A. K. M. Nurul Islam Mondal, who have sacrificed their lives for me and provided me whatever was needed to grow in my life. Thank you.   xxiv    DEDICATION To my father Engr. A. K. M. Nurul Islam Mondal, who eagerly waits for my publications and PhD (the very first PhD from his family). A man, whose only dream is to see his daughters highly educated, well established, and resilient.     1  Chapter 1 : INTRODUCTION 1.1 Background and motivation Unsafe drinking water is the root cause of many health related problems, especially in the case of small communities as they are lacking technical and financial resources to implement a well-protected drinking water supply system. Boil-water advisories (BWAs) have been common in all size of municipalities across Canada, but occurrences in First Nations and small sized communities are particularly high (Health Canada 2007). In the year 2013, 80% boil water advisories came from communities with less than 500 populations in Canada (Environment Canada 2015). Approximately 90,000 illness and 90 deaths/ yr have been reported in Canada due to compromised drinking water (Sierra Legal Defence Fund 2006).  Moreover, after episodes like Walkerton (2000), North Battleford (2001) and the Kashechewan First Nations Reserve (2005), integrated water quality management from source to tap has been emphasized. To ensure safe supply of drinking water, protection of water distribution network (DN) is much more critical as it is closest to the human consumption, and even the best treated water can deteriorate or be contaminated before consumption (Francisque et al. 2009). Water can be compromised within a DN due to contaminant intrusion and deterioration  caused by sudden pump shut-downs, power failures, demand increases, pressure drop inside the network, cross-connection, back-siphonage, breakage, leakage, treatment breakthrough, and even by the biofilm release (Besner et al. 2012; 2011; Betanzo et al. 2008). However, municipalities are more concerned about DN protection against pathogens. One of the main barriers to ensure safe water in a DN is to maintain detectable levels of residual disinfectant through secondary disinfection, which provides a safeguard against microbiological contamination and helps to control and reduce biofilm regrowth (USEPA 2004). Among available disinfectants, chlorine is the most commonly used because it is comparatively cheap, easy to handle, and above all, ensures long-term free residual chlorine (FRC) within a DN before it is supplied to the consumers (Brown et al. 2011). The USEPA surface water treatment rule (USEPA 2004) recommends 0.2 mg/L free residual chlorine in the DN. Besides, the WHO (2004) recommends a FRC within a range of 0.2 to 0.5 mg/L for populations prone to cholera, dysentery etc. However, chlorination also generates 2  undesirable disinfection by-products (DBPs), e.g., trihalomethanes (THMs) and haloacetic acids (HAAs) after reactions between FRC and natural organic matter. Humans can be exposed to DBPs by ingestion, inhalation and dermal contact.  Cooking, bathing, washing dishes and drinking the municipal water are some of the activities related to these exposures. Some DBPs may lead to detrimental health impacts, e.g., bladder cancer, reproductive effects, and child development effects (Hebert et al. 2010). Besides,  customers may reject the water due to chlorine related taste and odor complaints  though the water may be safe to drink (Boccelli et al. 2003; Gibbs et al. 2010; Kang & Lansey 2010; Ostfeld & Salomons 2006). This is especially common in First Nation Communities, where taste & odor issues related to higher chlorine dosages may not be acceptable. Commonly regulated thresholds for TTHMs1 and HAA52 are 80 µg/L and 60µg/L, respectively (USEPA 2004).  Managing these thresholds is essential to safeguard public health from adverse acute and chronic health effects. Trade-off between DBPs and microbial contamination is suggested to minimize both types of risks in the Interim Enhanced Surface Water Treatment Rule (USEPA 2004).  Booster chlorination can be adapted to address these issues efficiently by providing optimum FRC at chlorine depleted nodes in a DN. Studies have suggested this practice to reduce  total chlorine usage, taste & odor complaints, cost for chemical usage, and contact time which, in turn, can help to reduce DBPs (Boccelli et al. 2003; Carrico and Singer 2009). However, placement of a new booster station will add on costs for capital, utility operation & maintenance, and chemical usage. Therefore, booster stations require careful placement with appropriate dosage to minimize costs with higher risk reduction benefits. These contradictory and competing challenges warrant an in-depth study to optimize booster chlorination in DNs.  Many studies in the past have reported results of water quality optimization using booster stations (Cozzolino et al. 2005; Gibbs et al. 2010; Kang and Lansey 2010; Lansey et al. 2007; Ostfeld and Salomons 2006; Parks et al. 2009; Prasad et al. 2004; Propato 2006; Tryby et al.                                                  1 TTHM- total trihalomethanes consisting of the sum of chloroform,bromoform, bromodichloromethane, and dibromochloromethane 2 HAA5 Haloacetic acids consisting of the sum of the following five HAA species: monochloroacetic, dichloroacetic , trichloroacetic, monobromoacetic and dibromoacetic acids.  3  2002). These studies are mainly based on predefined threshold concentrations of FRC that aim at optimizing dosage, number and the location of booster points. Fewer studies have been conducted for smaller or medium-sized DNs in comparison for larger DNs. Figure 1-1 shows the percentage of published articles on the water quality optimization in small (serving population <3, 300) and medium-sized [3300, 50,000] as compared to the large-sized networks (> 50,000] during 1997 to 2012.  This highlights the need for detailed investigation of the techniques and methods required for water quality management in small and medium-sized DNs [3300, 50,000] (USEPA 2007).    Figure 1-1: Reported water quality optimization studies for small, medium, and large-sized DNs during a period of 1997 to 2012 However, it was observed that the studies reported in the literature had certain limitations, which provided motivation for the current research. Index based approaches can be adapted in optimization to integrate number of samples, frequency and specific values of selected parameter(s). It can serve as a mean to evaluate and ensure water quality while maintaining regulatory thresholds.  Also, there have been limited studies focusing on life cycle cost (LCC) for booster chlorination, microbiological and chemical risk effects and related trade-offs. In this research, various regulatory threshold and risk based indices have been proposed to evaluate the water quality. Later, these methods and strategies have been implemented to optimize location 4  and dosages for booster chlorination. The research is mainly focused on small to medium sized DNs, and guides informed decision making under limited resources and expertise.  1.2 Research objectives The main goal of this research is to ensure water quality safety in DN by optimizing secondary chlorination using boosters. As a first step, various indices are developed to evaluate the water quality and potential intrusion risk. Subsequently optimization schemes are proposed using index and risk based objective functions. To accomplish this goal, the specific objectives of this research are to: 1. develop and implement non-compliance potential index for assessing regulatory violations of DBPs, 2. modify and implement CCME WQI for Stage 1 and Stage 2 DBP rules, 3. propose risk based intrusion risk potential model to identify potential contaminant intrusion points in a DN, 4. select optimal dosage for booster chlorination using modified CCME WQI, 5. locate booster stations using optimization and trade-off analysis, and 6. select location and dosage for booster chlorination based on life cycle costing, microbiological and chemical risk potentials specific to contaminant intrusion. 1.3 Thesis organization Thesis contains nine chapters. Figure 1-2 shows thesis structure and organization related to each objective of this study. Chapter 2 covers literature review and develops the framework for this research. Chapters 3 to 8 are based on research gaps and concepts discussed in Chapter 2. Objectives 1, 2, 3, 4, 5, and 6 have been achieved and discussed in Chapters 3, 4, 5, 6, 7, and 8, respectively. Finally, conclusions and recommendations are provided in Chapter 9.    5                       Figure 1-2: Thesis structure and organization  Chapter 1: Introduction Chapter 2: Literature review and proposed framework Chapter 3: Non-compliance potential index Chapter 4: Modifying CCME WQI for DBPs Chapter 5: Risk-based model to identify potential intrusion points Chapter 6: Optimization scheme to select booster dosage Chapter 7: Optimization scheme to locate booster dosage Chapter 8: Optimization scheme for contaminant intrusion Chapter 9: Conclusions and recommendations Objective 1 Objective 2 Objective 3 Objective 4 Objective 5 Objective 6 6  Chapter 2 : LITERATURE REVIEW AND PROPOSED FRAMEWORK Some of the subsections of this chapter have been combined and accepted for publication to Environmental Reviews Journal entitled “Contaminant intrusion in water distribution networks: review and proposal of an integrated model for decision making” by Islam, N., Farahat, A., Al-zahrani, M. A. M., Rodriguez, M. J., and Sadiq, R. (Islam et al. 2015a).  There are two parts in this chapter. The first part reviews state-of the art literature related to this research and identifies weaknesses and limitations of existing literature. In the second part, the thesis framework has been proposed to ensure water quality safety in DNs using various index and risk based optimization schemes. 2.1 Literature review 2.1.1 Water quality failure in distribution network A DN is the last part of a water supply system and provides water for domestic, commercial and industrial uses (Khanal et al. 2005). Water quality failure in a DN is determined by specific microbiological, physio-chemical and aesthetic parameters (Deng et al. 2011). However, researchers are more often concerned about water quality failure at  source and treatment levels (Isovitsch and Vanbriesen 2008) than at DNs. Yet, contamination in DNs is critical because no strong treatment is available at this level except secondary disinfection (Khanal et al. 2005). Monitoring data to evaluate water quality for DNs are very limited.  Moreover, the travel time between the DN to consumers can be quite short, which has potential for community disaster.  Water quality in a DN is vulnerable to three possible means of contamination (NRC 2003): 1) inadequate water treatment at the treatment plant; 2) degradation in the DN; and 3) contaminant intrusion. Inadequate treatment can cause contamination: for example, injured microbes after insufficient treatment may cause contamination if conditions are favorable in the DN. Water quality may also degrade as a result of internal corrosion, nitrification, higher water age (detention), biofilm growth, disinfection by-products, lining by-products, pH stability and scale formation (NRC 2003). However, contaminants can also enter a DN through storage tanks, reservoirs,  pump stations, cross-connections, water mains, and permeation (NRC 2003; Sadiq et al. 2006). Moreover, there have been a number of waterborne outbreaks: for example, in 7  Milwaukee (US)  403,000 people  were affected by Cryptosporidium (Ford and MacKenzie 2000); in Gideon, Missouri, there was an outbreak caused by Salmonella typhimurium (Nilsson et al. 2005); and in Southington, Connecticut, VOCs caused an outbreak (Aral and Maslia 1997). A final barrier to ensure water quality safety is required which can handle these emergency outbreaks. Figure 2-1 provides common pathways and situations for contaminant intrusion in a DN. Unprotected reservoirs,  tanks,  and pump stations can cause contamination as a result of external sources such as the feces of beavers, squirrels and rabbits; airborne microorganisms; bird droppings; organic matter from leaves; and microorganisms at the bottom of the storage tank after flooding or ground water infiltration (Sadiq et al. 2006). Cross-connection is another potential pathway (Besner et al. 2011; Clark et al. 2007; Huang et al. 2009; Lee and Deininger 1992; Propato and Uber 2004a) typically combined with  backflow events, when backflow prevention devices are not installed or not installed properly. Aging of pipes initiates the breakage and results in contaminant intrusion through broken pipes, gaskets, and cross-connections. Ever-present pipe cracks, holes, faulty gaskets and faulty appurtenances can contribute  in reducing the water tightness of a distribution network which may, in turn, create a contaminant pathway (Cristo and Leopardi 2007, 2008; Deng et al. 2011; Sadiq et al. 2006). Kirmeyer et al. (2001) rank different pathogen entry routes in qualitative scales of low, medium and high based on the opinions of an expert panel. Cross-connections and water main repairs/breaks (intrusion) are ranked high; and uncovered storage facilities ranked medium.  Sources of intrusions through water mains can originate from adjacent broken sewers, contaminated ground water near a high water table and contaminated soil (Besner et al. 2011; Khanal et al. 2005; Kim et al. 2012, 2008; Lee and Deininger 1992; Yan et al. 2007) (Figure 2-1).  Sadiq et al. (2006) list two types of contaminant sources - chemical and microbiological.  Chemical contaminants from pesticides, petroleum products, fertilizers, solvents, detergents, pharmaceuticals, etc., are common , while Karim et al. (2003) report total coliform, fecal coliform (E. coli), clostridium, bacillus, Enteroviruses, Norwalk and Hepatitis A viruses near water mains.  8  Figure 2-1: Common pathways and situations for contaminant intrusions9  DNs can be vulnerable to various sudden imposed threats, such as intentional contamination by terrorist attacks (Ebacher et al. 2012; Huang et al. 2007, 2009; Khanal et al. 2005; Nilsson et al. 2004) that include physical attacks, cyber disruption and bio-chemical contamination.  Contaminant intrusion also requires a driving force , e.g., a negative pressure or a pressure less than 20 psi (Besner et al. 2011). Besner et al. (2011) describe two types of events caused by pressure in DNs; 1) transient events, and 2) sustained low/negative pressure events. Transient events, water hammer and surge events are generally initiated by rapid changes in water velocity, lasting from a few milliseconds to few minutes. These events are caused by pumps turning on and off, unplanned power outages, sudden changes in demand when opening and closing fire hydrants, and water main rupture (Figure 2-1). Conversely, sustained low/negative pressure events are of long duration and can last from minutes to hours. These are caused by natural disasters such as earthquakes. Mitigation steps applied during these events cause interruption and water outages which ultimately result in low/ negative pressure. Increased demand caused by flushing after repairs changes the hydraulic conditions and causes sustained low/negative pressure. Therefore, intrusion can take place at various locations depending on three basic conditions: 1) a pathway, 2) a source, and 3) a negative/low pressure inside a water main. These conditions require careful management using proper regulations, monitoring/sampling data, and decision support tools.  2.1.2 Water quality regulations There are various regulations that apply to water quality in DNs. Table 2-1 outlines regulations such as the Surface Water Treatment Rule (SWTR), Stage 1 Disinfectants and Disinfection By-products Rule (Stage 1 D/ DBPR), and Interim Enhanced Surface Water Treatment Rule (IESWTR). The table provides regulatory thresholds for possible parameters indicating contaminant intrusion. There are regulatory thresholds including thresholds enforced by SWTR and Stage 1 D/DBPR, IESWTR, (Kirmeyer et al. 2000), and guidelines suggested by USEPA (http://www.epa.gov/), WHO (WHO 2008) and Health Canada (HealthCanada 2012). Generally, residual disinfectants are maintained to protect the system from microbiological contamination. Therefore, tracking the loss of free residual chlorine is commonly suggested by regulations. The Total Coliform Rule (TCR) is used to protect DNs from adverse health effects caused by disease-10  causing pathogens. The TCR requires the sampling of water from DNs for total coliforms. It requires fecal coliform or E. coli testing if the test for total coliforms is positive.  The SWTR applies to water supply systems influenced by surface water. Generally, the SWTR serves to offset pathogen exposure caused by Giardia lamblia and Legionella viruses. The SWTR requires a minimal residual chlorine concentration of 0.2mg/L at the point of entry in a DN. The Stage 1 D/DBPR requires to maintain maximum residual disinfectant levels (MRDL) (Table 2-1). However, residual chlorine cannot protect DNs from microbes like Cryptosporidium and Giardia because of their resistant nature to chlorination. Additionally, Table 2-1 provides threshold values for some parameters representing contamination.  Table 2-1: Regulations relevant to DN safety Water quality parameter Regulatory limit Reference Disinfectant residual At the entry point: > 0.2 mg/L on continuous basis; distribution network: MRDL chlorine 4.0 mg/L, MRDL chloramine 4.0 mg/L, annual average SWTR; Stage 1 D/DBPR  Turbidity  Uof the time IESWTR  Umaximum Disinfectant residual or HPC bacteria count Detectable level of disinfectant residual or HPC bacteria ≤ 500 CFU/mL in 95% of the samples collected each month for any 2 consecutive months SWTR Total coliform  <5% positive (large systems) TCR Copper MCLG= 1.3 mg/L USEPA  2mg/L WHO Chlordane 0.0002mg/L WHO Nitrite MCLG= 1.3 mg/L; MCL=1mg/L USEPA  3mg/L WHO Ethylene glycol 9.5mg/L Health Canada Fluoride MCLG= 4 mg/L; MCL=4mg/L USEPA  1.5mg/L WHO Lead MCLG= 0mg/L;  USEPA  0.01mg/L WHO Chromium MCLG= 0.1mg/L; MCL=0.1mg/L USEPA  0.05mg/L WHO SWTR- Surface water treatment rule; Stage 1 D/DBPR – Stage 1 Disinfectant by-products rule; IESWTR- Interim Enhanced Surface Water Treatment Rule; MRDL - Maximum Residual Disinfectant Level; U - Nephelometric Turbidity Unit; CFU- Colony forming unit; TCR- Total Coliform Rule; MCLG – maximum contaminant level goal; MCL – maximum contaminant level; HPC- Heterotrophic Plate Count The Safe Drinking Water Act (SDWA) covers compliance of water quality standards (Khanal et al. 2005). The “Bioterrorism Act of 2002” requires vulnerability assessment reports submitted to 11  USEPA for all DNs serving more than 3,300 people (Khanal et al. 2005). In the wake of September 11, 2001, the Homeland Security Presidential Directive/HSPD-9  (2006) (HSPD-9 2006) requires federal agencies to develop contaminant intrusion-related monitoring systems. Therefore, DN safety greatly depends on effective monitoring and decision-making tools. 2.1.3 Water quality monitoring Most communities have water quality monitoring programs at source and at the treatment plant. However, water quality can change in the DN as a result of contaminant intrusion, biofilm growth and other reactions. Water quality monitoring/sampling is required in the DN to track changes and ensure safe drinking water for consumers. In addition, monitoring fulfils regulatory requirements. It increases customer confidence, minimizes the use of chemicals at treatment, and deals with emerging water quality issues. Sampling data contribute to important decision making through modelling and analysis.  Although, data can be generated by hydraulic and water quality simulations sampling data are still required for basic inputs.  NRC (2004) defines five steps for a proactive monitoring program: 1) parameter selection, 2) monitoring locations/stations selection, 3) determination of monitoring frequency, 4) sampling method selection, and 5) data management and report. 2.1.3.1 Parameters, locations and frequency Municipalities select parameters based on regulatory requirements, type of source water and water main materials, age, maintenance and conditions. NRC (2004) has categorized parameters, which has been further modified for contaminant intrusion to include: 1) key regulatory parameters: E. coli, FRC, heterotrophic plate counts (HPC); 2) operational parameters: ammonia, nitrate, nitrite, turbidity, flow, pressure; 3) routine parameters: temperature, pH, alkalinity, conductivity, color, soluble metals (lead and iron); 4) event-based parameter: iron, manganese, taste, odor, hydrocarbons. Monitoring locations are selected based on regulatory requirements, historical data, system characteristics, type of infrastructure, and water age. Most regulations have specific instructions regarding the selection of monitoring location. For example, NRC (2004) mentions regulations in Quebec (Canada) where they suggest that at least 50 percent of samples be collected from the outermost boundary of a distribution network. Historical data can indicate a possible location where previous contamination took place and suggest a monitoring location. Water main age and 12  breakage records can be tracked to select monitoring locations. Sensitive areas such as those supplying water to hospitals or places with a high population density require monitoring stations. Reservoirs pump stations and fire hydrants need frequent sampling as these are potential locations for contamination. Distribution areas near dead-end zones and poor hydraulics with higher water age also require sampling points. Tracer studies can help in locating these areas. Appendix A indicates commonly monitored parameters, for monitoring purposes and suggested sampling locations based on AWWARF (2003) and NRC (2004). FRC is possibly the only parameter to detect microbiological water quality degradation, along with other indirect parameters such as turbidity, and conductivity. However, it depends on the type of contaminant. For example, contamination by petroleum can be detected using hydrocarbons. Additionally, some municipalities use non-regulated parameters to better understand what is affecting the water quality in their DNs.  Monitoring frequency is also a regulatory requirement to monitor contamination in a timely manner. It identifies acute changes in water quality and helps in understanding system vulnerability and variability. There are two types of frequency: routine monitoring and non-routine monitoring. Routine monitoring is performed in a regular fashion, e.g., hourly, daily, weekly, monthly, quarterly and annually. Continuous monitoring of free residual chlorine and turbidity at the point of entry is common for most municipalities. Online monitoring (explained later) is an example of a routine monitoring practice. Non-routine monitoring is a practice generally performed after a specific event, e.g., breakage, repair and flushing, etc.  2.1.3.2 Sampling methods Three types of sampling methods are available to monitor water quality in DNs: 1) online instruments, 2) automatic samplers and 3) manual samples. Online instruments are installed permanently in the DN and do not require operator intervention, except for maintenance and calibration. This sampling method can sample, analyze and generate reports in a regular fashion even on a per-second basis. Because of the high cost of online instruments, most municipalities use them only at the point of entry. The use of online monitoring of free chlorine, pressure and flow is very common. However, it can be challenging to operate these instruments in colder climates. Automatic samplers collect samples without recourse to an operator. However, technical personnel are needed to retrieve and analyze the samples in a laboratory. For manual samples, water sample collection and calibration should be carried out by an operator or 13  technical personnel. Manual samples are easy to collect, but require trained technical personnel.  However, handling of sampling manually may result in contamination. Monitoring methods are selected on the basis of regulatory requirements, frequency, monitoring locations, costs, operation and maintenance of related equipment, availability of techniques, laboratory facilities, storage means, transportation facilities and technical staff. Measurement methods associated with each technique can be manual or sensor-based. For example, 24-hour incubation is common to sample E. coli in water. Appendix B indicates available measurement methods for commonly used parameters in contaminant intrusion. The names of various sensors are also mentioned in this table. However, the details of these measurement sensors are beyond the scope of this study. Appendix B also emphasizes that there is a need for rapid and effective sensor measurements for microbiological contaminants. Turbidity is not a direct measure of microbes. This metric is typically considered a surrogate for microbial contamination as E. coli and other microbes may require multiple days of incubation time. A catastrophic event is quite likely to take place during this incubation time.  2.1.3.3 Managing and reporting data Monitored data must be managed properly to identify problematic areas, control treatments and generate regulatory reports, statistical-trend analysis, water quality modelling and decision making. Common ways to manage data include the creation of standard databases, spreadsheet-based software, laboratory information management systems and commercial data management. Additionally, reports need to be generated to address consumer inquiries and to create awareness on water main breaks, health concerns in an area, flushing, construction, firefighting and cleaning, major changes in water usage and floods or any extreme weather event. One common approach is to develop a Supervisory Control and Data Acquisition (SCADA) system to generate the necessary warning during any contamination event.  Regulations have been introduced for adhering to certain parameters related to sources of contaminants and operational conditions, but there is a serious lack of specific asset management regulations.  For example, regulatory guidance towards Master Municipal Construction Documents (MMCD) standards (MMCD 2015) might be a future possibility to regularly store and transfer municipal data (e.g., condition assessment using breakage, installation year,  and performance data). Basically, standards like MMCD ensure regular monitoring of water main 14  conditions that can be incorporated into various modeling approaches, e.g., modeling related to water quality assessment and management. 2.1.4 Water quality assessment Water quality is assessed based on index and risk based approaches. The index evaluates the water quality and the risk assesses the potential risk due to water quality failure.  2.1.4.1 Index based approaches Water quality assessment can be carried out by using a unit less index, called water quality index (WQI), which generally consists of selected water quality parameters. The index value may reflect the number, frequency and magnitude, whereby regulatory standards (for various water quality parameters) are not maintained over a specific period of time. Generally, the WQI formulation includes the following three steps: (1) selecting potential water quality parameters, (2) converting non-commensurable water quality parameter (WQP) measurements into a monotonic quality scale to obtain sub-indices (unit less measure), and (3) aggregating sub-indices into a unit less number index (WQI).  The units of water quality parameters are non-commensurate. In addition, some parameters positively impact water quality whereas others are negatively correlated with water quality. Parameters like FRC can even have double impact on water quality such as increasing FRC can improve microbial water quality while decreasing chemical water quality. Therefore, before aggregation is performed, a transformation (using appropriate functions) is required to convert the observed values of each WQP into a monotonic quality scale   [0, 1]. The conversion can be done by using two types of functions; quality function, where 1 represents the best quality and 0 represents the worst quality or with failure function, where 1 represents complete failure and 0 means complete maintenance of the water quality. The transformed values of a water quality parameter are generally referred to as sub-index, where each sub-index   [0, 1] regardless of the original units.  After transformation, the final step is an aggregation of sub-indices. There are four common types of aggregation formulations that include additive, multiplicative, logical or based on water quality guidelines. A few common formulations used to develop water quality indices for mostly surface waters such as National Sanitation Foundation (NSF-WQI) (Brown et al. 1970), Oregon 15  (O-WQI) (Dunnette 1979) , PW- WQI (Pesce and Wunderlin 2000), Central Pollution Control Board (CPCB- WQI) (Sarkar and Abbasi 2006), Universal water quality index (U-WQI) (Boyacioglu 2007), S- WQI (Said et al. 2004), Canadian Council of Ministers of the Environment Water Quality Index (CCME WQI ) (Khan et al. 2003) and ST- WQI (Swamee and Tyagi 2007).  Aggregation formulations commonly encounter problems as a result of the abstraction of data. Table 2-2  provides some of those important limitations, which includes ambiguity, eclipsing, and rigidity. For example, weighted arithmetic mean has an eclipsing problem (i.e., one or more sub-indices show poor quality but the overall index does not reflect it), whereas root sum power addition suffers from an ambiguity problem (i.e., all sub-indices show acceptable quality but the overall index shows unacceptable quality). Minimum and maximum operators are free from ambiguity and eclipsing, but they fail to reflect the change in any sub-index other than the lowest (or highest) sub-index value in the group. Many aggregation formulations are expert dependent by using expert defined weights (Brown et al. 1970) and fail to show the effect of regulatory failure with their formulation. Canadian Council of Ministries of the Environment WQI (CCME WQI) or Canadian WQI (CWQI) came out as the WQI with comparatively less limitation such as less complexity, and less eclipsing  besides being able to handle regulatory effects without experts’ opinion (Table 2-2).  CCME WQI is determined based on three factors: 1) scope (F1), 2) frequency (F2), and 3) amplitude (F3) (Khan et al. 2003):   2 2 21 2 3100 ( )1.732F F FCCMEWQI          [2-1] Scope (F1) is the percentage of failed variables to the total number of variables. Frequency (F2) is the percentage of regulatory non-compliance incidence (‘failed values’) for all parameters under consideration. Amplitude (F3) refers to the amount by which the non-compliance values do not meet their regulatory guideline. The value 1.732 is a normalization factor to keep the final CCME WQI from 0 to 100 (Khan et al. 2003).  16  However, the formulation has some basic limitations and can be modified. The formulation can be more sensitive if water quality parameter relative weights are incorporated. Assigning weights will make the formulation expert dependant as often experts judgements are used in this regard (Espejo et al. 2012). Hurley et al. (2012) have assigned relative weights among the factors F1, F2, F3 instead of weights  to the water quality parameters. Espejo et al. (2012) have proposed a scheme to assign weights among the parameters by placing the parameters in different levels. However, the study was not specified for DBP rules.   2.1.4.2 Risk based approaches Chemical risk assessment Chemical risk assessment framework was first developed by National Research Council (NRC 1983), which contains four basic steps: 1) hazard identification, 2) dose-response assessment, 3) exposure assessment, and 4) risk characterization.  Hazard identification defines the potential chemical substances that can cause health problems. It can be chronic or acute depending on the animal or human studies. Microbial risk assessment Microbial risk assessment refers to the steps to understand, reduce, and prevent risks caused by hazardous microbes, whether natural or anthropogenic, intentional or unintended. Though microbial risk assessment evolves from chemical risk assessment, there are fundamental differences between them; e. g. the exposure of chemical is mainly chronic while microbial exposure is acute. Moreover, quantification of microbial risk is harder compared to the chemical one as it depends on many factors such as microbial growth, inactivation by disinfection and treatments. Quantitative microbial risk assessment (QMRA) is a popular way to estimate the risk of infection by specific microbes (Austin et al. 2012; Besner et al. 2011). This approach is performed by the office of water to support SWTR and the LTESWTR (USEPA 2010b). The standard QMRA consists of four basic steps (Ahmed et al. 2010):  (1) Hazard Identification: Hazard identification is to define the potential microbes (e.g., protozoan, virus, and bacteria) which are harmful to the system and also investigates the effect of the microbes. Most drinking water related QMRA address enteric viruses (Hepatitis A, Norwalk 17  virus, rotavirus), parasites (Ascaris, Eimeria, Cyclospora, Toxoplasma), protozoa (Cryptosporidium and Giardia) and bacteria (E. coli, Shigella, Salmonella, Vibrio cholera) (Haas et al. 1999). Unlike microbial water quality assessment, E. coli O157:H7 can be considered for simplicity to investigate the effect of chlorination with QMRA in DN. Table 2-3 states the commonly available DBPs and their group identification by IRIS (Integrated Risk Information System) (USEPA 2014).  Dose-response relationship examines the effect of the applied dosage to animal or human host body. This step is dependent upon the studies already conducted for particular chemical. The effect on the animal or human host helps to calculate terms such as no-observed-adverse-effect level (NOAEL), lowest-observed-adverse-effect level (LOAEL), reference dose (RfD), and slope factor (SF) (Figure 2-2). The highest level of exposure or dose at which there is no significant increase in response is called NOAEL. The term will be called LOAEL if there is significant response. RfD estimates a daily oral exposure to the human population, which has no appreciable risk of deleterious especially during lifetime. Table 2-2: Water quality index limitations Reference WQI name Eclipsing Ambiguity Rigidity Complexity in formulations Unable to handle regulatory effect Expert dependent Brown et al. (1970) NSF-WQI       Dunnette (1979) O-WQI       Pesce and Wunderlin (2000) PW-WQI       Sarkar and Abbasi (2006) CPCB-WQI       Boyacioglu (2007) U-WQI       Said et al. (2004) S-WQI       Swamee and Tyagi (2007) ST- WQI       Khan et al. (2003) CWQI         Nil  Low  Medium  High  18  RfD is provided so that chemicals do not exceed this level and to ensure that there are no adverse health effect. The slope of the line drawn from NOAEL to the origin with zero dosage and zero response is called SF.  Exposure assessment estimates the amount of contact to create adverse effect. The exposure route can be through oral ingestion, dermal adsorption, and inhalation for DBPs while cooking, washing, bathing, laundering, cleaning, and showering (Wang et al. 2007). The exposure of individual DBP or combinations of DBPs can be estimated by using a common equation (Wang et al. 2007): wC IR EF AF EDExpBW AT            [2-2] where Exp = exposure dose (mg/kg/day); Cw= Chemical concentration in water (mg/L); IR = ingestion rate (L/day); EF = exposure frequency (days/year); AF = exposure factor (assumed 1); ED = Exposure duration (year); BW = Body weight (kg); and AT = the average time (days). Finally, risk characterization is the combination of the dose-response assessment and the exposure assessment. It is conducted to determine cancer or non-cancer effects. For example, the slope factor helps to estimate the risk of cancer as: CancerRisk Exp SF                      [2-3] For non-cancer effects, the hazard index (HI) is a popular way adapted, for example adapted in Wang et al. (2007): ExpHIRfD           [2-4] Microbial risk assessment Microbial risk assessment refers to the steps to understand, reduce, and prevent risks caused by hazardous microbes, whether natural or anthropogenic, intentional or unintended. Though microbial risk assessment evolves from chemical risk assessment, there are fundamental differences between them; e. g. the exposure of chemical is mainly chronic while microbial exposure is acute. Moreover, quantification of microbial risk is harder compared to the chemical 19  one as it depends on many factors such as microbial growth, inactivation by disinfection and treatments. Quantitative microbial risk assessment (QMRA) is a popular way to estimate the risk of infection by specific microbes (Austin et al. 2012; Besner et al. 2011). This approach is performed by the office of water to support SWTR and the LTESWTR (USEPA 2010b). The standard QMRA consists of four basic steps (Ahmed et al. 2010):  (1) Hazard Identification: Hazard identification is to define the potential microbes (e.g., protozoan, virus, and bacteria) which are harmful to the system and also investigates the effect of the microbes. Most drinking water related QMRA address enteric viruses (Hepatitis A, Norwalk virus, rotavirus), parasites (Ascaris, Eimeria, Cyclospora, Toxoplasma), protozoa (Cryptosporidium and Giardia) and bacteria (E. coli, Shigella, Salmonella, Vibrio cholera) (Haas et al. 1999). Unlike microbial water quality assessment, E. coli O157:H7 can be considered for simplicity to investigate the effect of chlorination with QMRA in DN. Table 2-3  provides the required SF, RfD for commonly available DBPs in drinking water.   Figure 2-2: Dose-response for chemical risk assessment Microbial risk assessment Microbial risk assessment refers to the steps to understand, reduce, and prevent risks caused by hazardous microbes, whether natural or anthropogenic, intentional or unintended. Though 20  microbial risk assessment evolves from chemical risk assessment, there are fundamental differences between them; e. g. the exposure of chemical is mainly chronic while microbial exposure is acute. Moreover, quantification of microbial risk is harder compared to the chemical one as it depends on many factors such as microbial growth, inactivation by disinfection and treatments. Quantitative microbial risk assessment (QMRA) is a popular way to estimate the risk of infection by specific microbes (Austin et al. 2012; Besner et al. 2011). This approach is performed by the office of water to support SWTR and the LTESWTR (USEPA 2010b). The standard QMRA consists of four basic steps (Ahmed et al. 2010):  (1) Hazard Identification: Hazard identification is to define the potential microbes (e.g., protozoan, virus, and bacteria) which are harmful to the system and also investigates the effect of the microbes. Most drinking water related QMRA address enteric viruses (Hepatitis A, Norwalk virus, rotavirus), parasites (Ascaris, Eimeria, Cyclospora, Toxoplasma), protozoa (Cryptosporidium and Giardia) and bacteria (E. coli, Shigella, Salmonella, Vibrio cholera) (Haas et al. 1999). Unlike microbial water quality assessment, E. coli O157:H7 can be considered for simplicity to investigate the effect of chlorination with QMRA in DN. Table 2-3: Slope factor, reference dose and other hazard information for commonly available DBPs Name of the chemical Last updated* Cancer group Oral slope factor, SF ((mg/kg/d)-1) Reference dose, RfD (mg/kg/d) Chloroform (TCM) 10/19/2001 Probable human carcinogen(B2) NA 1 x10-2 Dibromochloromethane (DBCM) 03/01/1991 Possible human carcinogen (C) 8.4 x10-2 2 x10-2 Bromodichloromethane (BDCM) 03/01/1991 Probable human carcinogen(B2) 6.2 x10-2 2 x10-2 Bromoform (TBM) 03/01/1991 Probable human carcinogen(B2) 7.9 x10-3 2 x10-2 Monochloroacetic acid (MCAA)  Not classified as to human carcinogenicity (D) NA 4 x10-3 Dichloroacetic acid (DCAA) 09/11/2003 Likely to be carcinogenic to humans 5 x10-2 4 x10-3 Trichloroacetic acid (TCAA) 09/30/2011 Suggestive evidence of carcinogenic potential 7 x10-2 2 x10-2 *for oral exposure only; NA-not available 21   (2) Exposure Assessment: It generally identifies the exposure routes such as inhalation, ingestion, and dermal absorption, the time period for the exposure, and the amount or the level of the exposure. The main focus of the exposure assessment is the amount of oral consumption per day, which depends on the place and the type of consumer. There is much discrepancy about the daily water consumption. An average 1L/day can be assumed based on previous study conducted by Besner et al. (2012).  . (3) Dose-Response Assessment: Dose-response assessment can be conducted by using appropriate model to define the response as the probability of infection. Microbial concentration is required in colony forming units for bacteria, plaque forming units for viruses and cysts or oocyst counts for protozoa. Numerous studies have been conducted on different hosts to observe the response over the dose and aggregated in a QMRA data base (Haas 2012). The data base assumes the response as either beta-poison or exponential model with the following equations: 1 k dinP e             [2-5] 1502 11 [1 ]inP dN               [2-6] where Pin= Response (probability of infection); d= dose (concentration multiplied by the volume of ingestion)=Conc×V(volume); ks= Probability of survival (depends on the organism and test); N50= the experimental dose at which 50% of the population is expected to be affected; α= Beta-poison distribution parameters. Table 2-4 states distribution parameter values for commonly available microbes. Based on the above three steps, the final step would be risk characterization for risk management. 2.1.5 Water quality management  2.1.5.1 Common management strategies Various management strategies are recommended by NRC (2003) to ensure safety in DNs. The strategies are site-specific and applicable with available technology and resources.  22  One common strategy is to use disinfection, such as chlorination in the DN to kill microorganisms in contaminated water. Residual chlorine can be monitored on-line and covers many organisms, except Cryptosporidium.  Alternative disinfectants such as ozone and ultraviolet light are recommended for DNs prone to Cryptosporidium. Most regulations focus on maintaining a minimum 0.2mg/L of residual chlorine. Increasing chlorine dosage can create harmful DBPs which are also prohibited by regulations. Many municipalities use booster chlorination, a mean of secondary disinfection applied in chlorine depleted areas. If managed properly, booster disinfection ensures minimum chlorination with maximum benefits.  Maintaining positive pressure is necessary, since negative or low pressure is one of the prime reasons for contaminant intrusion. Keeping water mains in good condition with fewer leaks can help in maintaining positive pressure. Elevated storage construction, pressure control equipment (control valves, air chambers, and surge relief valves) installation and prevention of the submergence of air valves can also help in this respect. Likewise, back-up pumps and standby pumps in the event of sudden power outages and slowed-down fire hydrant operations can help maintain adequate pressure in water mains. Table 2-4: Related QMRA studies Organisms Reference Best fit model Parameters for the best fitted model Host type & route Dose units Response k α N50 Salmonella McCullough and Elsele (1951) Beta-poisson  3.18×10-01 3.71 ×1004 Huma, Oral CFU positive stool culture Shigella DuPont et al. (1972) Beta-poisson  2.65×10-01 1.48×1003 Huma, Oral CFU positive stool culture isolation Virrio Cholerae Hornick et al. (1971) Beta-poisson  2.50×10-01 2.43×1002 Human, Oral CFU Infection E. Coli O157:H7 DuPont et al. (1971) Beta-poisson  1.55×10-01 2.11×1006 Human, Oral CFU positive stool isolation Legionella pneumophila Muller et al. (1983) Exponential 5.9×10-02   Guinea pig, Inhalation CFU Infection Cryptosporidium parvum Messner et al. (2001) Exponential 5.7×10-02   Human, Oral Oocysts infection Giardia duodenalis Rendtorff (1954) Exponential 1.99×10-02   Human, Oral Cysts infection CFU- Colony forming unit;  23  Backflow can be detected based on customer complaints and drop in pressure.  Decreases in residual chlorine, pH and color change, reverse water meter running and coliform detection can also confirm backflow. Backflow can be prevented by separating potable and non-potable water supplies, installing backflow prevention devices, implementing backflow-cross-connection control programs, maintaining positive pressure, and through disinfectant residuals. A survey is recommended ahead of the time when building a water main system. Once contamination is detected, flushing is recommended. Unidirectional flushing is more effective than traditional flushing. Unidirectional flushing flows one way from larger diameter to smaller diameter water mains. Swabbing and pigging are not recommended because they require special arrangements. They may be recommended for pipes with larger water mains for which larger volumes of water for flushing are required. Control of valves and fire hydrants is recommended. Studies related to blending water sources should be carried out ahead of time to understand the effects of blending in water mains. Water main systems should be designed to allow fewer dead end zones, detention time and more positive pressure. When not lined, water main systems may also be subjected to external corrosion by aggressive soil and internal corrosion. Replacement is uneconomical unless the water main has a high breakage rate and insufficient hydraulic capacity.  Training and certification are required for operators owing to the increased complexity of distribution systems. The Association of Boards of Certification for examination and certification of operators is mandatory for most jurisdictions in Canada. Associations like the Canadian Water and Wastewater Association (CWWA) are developing organized programs for operators. Effective customer care and meetings with stakeholders can provide some oral barriers to contamination intrusion.  Municipalities should have a comprehensive water quality monitoring program so as to ensure safe water before it arrives at the consumer’s tap. Municipalities should have cost-benefit analysis for sampling methods and important parameters should be monitored within limitations. Monitored data should be linked to a GIS, laboratory information management system (LIMS), and supervisory control and data acquisition (SCADA). Further modelling based on monitored data can contribute important decision making. 24  Apart from various management strategies, this research will focus on contaminant intrusion and booster chlorine management to ensure water quality safety in DN.  2.1.5.2 Contaminant intrusion modelling  In the literature, various approaches have been proposed to model contaminant intrusion. Many of these approaches consider laboratory data in conjunction with simulation results. Typically, optimization studies are used to identify the contaminant source, place and design sensors, and select monitoring stations. Traditional microbial risk assessment, kinetic modelling and negative exponential modelling are some of the common ways to address risk analysis. Soft-computing techniques such as the Dempster-Shafer, Fuzzy logic and Bayesian network are also adapted to risk analysis. Geographical and statistical modelling can be performed for topography and connectivity analysis. Moreover, commercial software and tools can be used to model risk, vulnerability and condition assessment.  Optimization Optimization generally consists of variables, objective functions, constraints, and modelling assumptions (Hart and Murray 2010). Table 2-5 summarizes up-to-date optimization studies related to contaminant intrusion. The table provides the overall purpose, details of objective functions and the algorithms used in each study. Commonly used problem formulations, hydraulic and water quality analyses and optimization schemes are discussed hereafter.    The steps required to perform an optimization study related to contaminant intrusion include the following: 1) select variables, 2) define objective functions, 3) define constrains, 4) state the problem mathematically, 5) perform hydraulic and water quality simulations, 6) select appropriate algorithm / algorithms, and 7) perform the optimization. Table 2-5  provides possible variables inside objective functions. Variables are selected based on the overall objective of optimization. For example, most studies are for sensor placement or contaminant source identification.  Variables such as contaminated water volume, detection time, demand coverage, detection likelihood, and affected population are common in objective functions. The detection time is important if a DN is large: contamination can take a number of hours or even days to be detected. Water demand is also an important variable as it is an indirect measure of the affected population. However, it can vary significantly with water withdrawal from one node to another 25  and the variation of diameters and flow. Affected population sometimes represents water demand which expresses the importance of the node to be selected. Consumed contaminated water can be another variable which is the total volumetric water that exceeds a predefined threshold for a selected water quality parameter. Demand coverage is the total demand percentage of the nodes, also used as a variable. Detection likelihood, the probability of contaminant detection by at least one sensor, is another variable to be considered.  The objective functions using these variables can be to: 1) minimize the detection time (z1), 2) minimize the affected population before the detection ( z2), 3) minimize water contamination demand before detection (z3), 4) maximize detection likelihood (z4), 5) maximize demand coverage (z5), and 6) minimize  simulated and measured data difference (z6). Objective functions z1, z2, z4, and z6 are presented in detail in Table 2-6. Algorithm-based objective functions are also used for z3, and z5. Generally, objective functions are stated with constrains in the optimization studies. Maintaining pressure, or dosage or chlorine thresholds are some common constraints for contaminant intrusion. Hydraulic and water quality analysis are necessary in most optimization studies to predict the final outcome. The analysis can predict contaminant spatial and temporal variations throughout the DN. The steady-state simulation tool, EPANET, is commonly used as a hydraulic and water quality simulator in pressurized pipe networks. The model can predict flow and pressure at various points in a DN using an extended period simulation from one day to one week. Additionally, concentrations for specific contaminants can be predicted with EPANET using transport and chemical reactions (Hart and Murray 2010). For example, Chick-Watson first order kinetics was stated in Betanzo et al. (2008) to model microbial growth. 26  Table 2-5: Summary of the optimization studies from 1992 to 2013 Reference Overall purpose/purposes Objective functions Algorithm/ method used Detection time Affected population Consumed contaminated water volume Detection/ contamination likelihood Demand coverage Difference=simulated-measured data Baranowski and LeBoeuf 2008 Minimize contaminant concentration     √  Genetic algorithm and cost function Berry et al. 2006 Placing sensors    √   Mixed integer linear programming Couboul and Ghanem 2013 Placing sensors    √   Greedy algorithm, Monte Carlo simulations Cristo and Leopardi 2008 Locating contaminant source      √ Pollution matrix Eliades and Polycarpou 2006 Placing sensors √ √ √ √ √  Solved using multi-objective Pareto solutions Guan et al. 2006 Locating contaminant source      √ Optimal predictor corrector algorithm Huang et al. 2007 Placing sensors √ √  √   Genetic algorithm  Ostfeld et al. 2008 Placing sensors √ √ √ √   Expert opinion, Monte Carlo simulation, pollution matrix Isovitsch and Vanbriesen 2008 Placing sensors  √ √    Spatial analysis and ANN Kim et al. 2008 Locating contaminant source      √ ANN Kim et al. 2012 Placing sensors √   √   Multi-objective genetic algorithm. Kumar et al. 1997 Selecting monitoring stations   √    Integer programming method Laird et al. 2005 Locating contaminant source      √ Nonlinear programming and backtracking algorithm  27  Table 2-5: Summary of the optimization studies from 1992 to 2013 Lee and Deininger 1992 Selecting monitoring stations     √  Integer programming method Ostfeld and Salomons 2004 Selecting monitoring stations   √    Pollution matrix, and genetic algorithm Preis and Ostfeld 2006a Locating contaminant source      √ A hybrid model trees - linear programming Preis and Ostfeld 2006b Placing sensors      √ Genetic algorithm  Preis and Ostfeld 2008 Designing sensor network √   √   Genetic algorithm II and detection redundancy Preis et al. 2008 Quantifying uncertainly in contaminate source      √ Genetic algorithm Sanctis et al. 2010 Locating contaminant source      √ Back tracking algorithm Shastri and Diwekar 2006 Sensor placement using uncertainty in demands  √  √ √  Linear programming Watson et al. 2004 Placing sensors √ √ √ √(failed detection)   Mixed integer linear programming, extent of contamination Weickgenannt et al. 2010 Placing sensors    √ √  Genetic algorithm II Woo et al. 2001 Selecting monitoring stations     √  Integer programming method using a set of covering algorithm Xu et al. 2008 Placing sensors   √    Graph theory, greedy heuristic algorithm ANN- Artificial Neural Network28  Additionally, the Monod equation can be adapted to model bacterial growth within a DN (Zhang et al. 2004). EPANET-MSX is an extension of EPANET and capable of modelling the dynamics of the reactions for multiple interacting species. Studies (Austin et al. 2012; Betanzo et al. 2008; Grayman et al. 2007) have modelled microbial growth using EPANET-MSX in DNs. However, EPANET assumes perfect mixing behavior, further explored by researchers at the University of Arizona. The AZRED 1.0 program was created in this regard and hydraulic simulation for predicting contaminant intrusion was carried out assuming imperfect mixing (Kim et al. 2012). EPANET Multi-Species Extension (EPANET-MSX) can be used in future to model both microbial and chemical (free residual chlorine) concentrations. The final steps in optimization include the selection of algorithms and performing the analysis. Preis and Ostfeld (2008) divided optimization studies into single-objective and multi-objective studies. Both single- and multi-objective studies can be solved by the following methodology or algorithm: 1) mathematical optimization, 2) pollution matrix, 3) backtracking algorithm, 4) set covering algorithm, 5) stochastic approach, 6) soft-computing approach, and 7) other approaches such as geographical and statistical approaches. Additionally, multi-objective optimization can be solved using Pareto-solutions.  Mathematical optimizations such as mixed integer linear programming (Berry et al. 2006; Watson et al. 2004), integer programming (Kumar et al. 1997; Lee and Deininger 1992; Woo et al. 2001), non-linear programming (Laird et al. 2005), and even linear programming (Shastri and Diwekar 2006) have been used in previous studies. Any of these methods can be selected based on the variables and assumptions required for optimization. Most are single-objective, rather than multi-objective optimizations. Simple linear programming is also used in some studies if there are other complications, such as uncertainty, need to be considered.  Pollution matrix is used in studies (Cristo and Leopardi 2008; Kessler et al. 1998; Ostfeld and Salomons 2004) to represent the domain of contaminant detection in the DN. The concept is used to assign coefficients in a matrix to represent the state of contamination, or not being contaminated, for each node in a DN. A level of service for the detection system is set by defining the allowable consumed volume of contaminated water prior to detection. A 1000cm3 level of service means that the detection system design is capable of detecting contamination before a total consumption of 1000cm3 of contaminated water. The matrix can be represented as N x N matrix with 0-1 co-efficient values for each node, where N= number of nodes; co-29  efficient=1, when there is contamination and 0 when the node is not contaminated (Kessler et al. 1998). The jth row lists contaminated nodes due to contamination in ith node. The volume of consumed contaminated water is estimated before arrival from i to j using the shortest path (minimum travel path). If the volume exceeds the maximum allowable level of service, the simulation will stop; otherwise, it will assign 1 to each node.  Table 2-6: Objective functions used in common optimization studies Variable used Brief definition Reference and overall objective Mathematical objective functions  (denoted as z) in literature Detection time The time elapsed from the contamination intrusion to the very first detection of the pollutant.  Kim et al. (2012)- stated for sensor placement; Preis and Ostfeld (2008)- Sensor design  1min( )min ( )d idt tz E t where, td= detection time; ti= detection time for the ith sensor; E(td) is the mathematical expression of the detection time td.  Affected population The number of population exposed before detection. Watson et al. (2004)- Placing sensor; Shastri and Diwekar (2006)- Sensor placement with uncertainity 21minniiz W P x    Where, n= node; W= weight assigned for each detection; P=population exposed; x= contamination incident  and 1 for positive ones; Detection likelihood The probability of detection. Kim et al. (2012)- stated for sensor placement; Preis and ostfeld (2008)- Sensor design 411max( )Srrz dS    where, dr =1 if contamination r  is detected; dr= 0when contamination is not detected; S= total number of detection. Simulated and measured data difference Simulated data gathered from software or by fitness function and measured data from laboratory studies. Preis et al. (2008)-quantifying uncertainty in contamination source 261min [ ( ) ( )]rdtSM Sii ii tz D t D t   Where, S= monitored stations; td= first detected time; tr= response time; D= data either monitored or simulated; M= monitored data; Si= simulated data; t=time The backtracking algorithm or particle backtracking algorithm (PBA) is a special type of inverse problem. The inverse problem refers to simulations where parameter values or boundary conditions are measured by the state of the system. Contaminant source identification or sensor designs are generally based on a backtracking algorithm where researchers measure the state of the system and boundary conditions for the contaminant source. Based on this, various algorithms are incorporated along with the PBA concept to identify the source of contamination. Shang et al. (2002) have an input-output model to define the relationship between selected water 30  quality parameters at the inlet and outlet. The model is based on tracking water parcel movement by identifying water flow paths. The algorithm has been expanded further by Laird et al. (2005) using mathematical optimization such as  nonlinear programming. Sanctis et al. (2010) also extended the algorithm by incorporating data uncertainties. Preis and Ostfeld (2006a) developed an algorithm using the same concept, except for the incorporation of a mathematical optimization. In addition to tracking water parcel movement, the simulation-optimization approach is also incorporated as PBA. The simulation-optimization is an advanced level trial-and-error approach unless it fulfills the objectives. For example, Guan et al. (2006) present a simulation-optimization where the algorithm is a continuous optimal predictor algorithm.  The set of covering algorithm is a special form of graph theory usually combined with a pollution matrix algorithm. It finds an optimal combination of solutions to cover the system (Kessler et al. 1998). A cost comparison is carried out to effectively compare possible solutions. The cost is generally compared with a total number of sensors and monitoring locations. A solution that covers most of the areas is selected for this algorithm. For example, Figure 2-3 states an application of the set covering algorithm that suggests 3 monitoring stations for contaminant intrusion event. The solution is also selected based on the assigned level of service.   Figure 2-3: Sample solution for the set of covering algorithm 31  DNs possess many uncertainties that can be incorporated into optimization. For example, water demand fluctuates tremendously from day to day (Comboul and Ghanem 2013). Pollutant concentration can be different based on the algorithm used. Therefore, it is always a good idea to assess the robustness of analysis by performing a stochastic analysis in optimization. Randomly generated data using the Monte-Carlo simulation is a common approach for stochastic optimization (Comboul and Ghanem 2013; Cristo and Leopardi 2007; Preis and Ostfeld 2008b). Ostfeld and Salomons (2004) first introduced possible uncertainty into demand related to contaminant events. They expanded their work in Ostfeld and Salomons (2005) and other researchers started incorporating this into optimization studies. Population uncertainty was considered in Berry et al. (2005) and optimization uncertainty was considered in Carr et al. (2006). Shastri and Diwekar (2006) improved integer programming with sensor costs and incorporating uncertainty.  Sanctis et al. (2010) incorporated a fuzzy term “state data” into uncertainty analysis. Cristo and Leopardi (2007) proposed a pathway analysis using time varying concentrations to incorporate measurement and demand uncertainty.  Zadeh (1984) defines soft-computing as a method for handling imperious, incomplete and uncertain information. In optimization, soft-computing methods are employed to obtain near optimal solutions. It can be any heuristic technique such as fuzzy logic, evidential reasoning, artificial neural network (ANN), data mining and genetic algorithms (GA). Genetic algorithm (Al-Zhrani and Moied 2003; Baranowski and LeBoeuf 2008; Kim et al. 2008; Preis and Ostfeld 2006a, 2008; Weickgenannt et al. 2010) and ANN (Isovitsch and Vanbriesen 2008; Kim et al. 2008) are the most frequently adapted optimization algorithms in studies related to contaminant intrusion (Table 2-5).  GA is inspired by Charles Darwin’s theory of evolution, which adapts the process of natural selection to solve optimization. GA has proven to be a powerful tool to solve optimization problems. It follows the analogy of nature’s evolution and is quite successful in solving complex combinational and organizational problems. A new set of artificial creatures called strings is created from the last step of information and so on. The technique uses past information to speculate the analysis and improve performance (Goldberg 1989). ANN is a computational model capable of pattern generation inspired by central nervous systems. It is a powerful non-linear statistical data modelling tool to model complex relationships between input and output to generate the pattern.  32  Hart and Murray (2010) state that an algorithm or optimization solver selection depends on three criteria: 1) guarantee to find the real solution, 2) computer memory, and 3) runtime. Other optimization solvers can be superior in terms of obtaining the solution, but soft computing techniques such as GA can find near-optimal solutions quickly. Indeed, a DN made up of many nodes requires a solver that can perform with low memory and less computational time. Therefore, GA is used frequently to solve optimization studies related to contaminant intrusion.  Ostfeld and Salomons (2004) proposed a genetic algorithm (GA) with a previously described pollution matrix to select the best monitoring stations. Preis and Ostfeld (2006b) also performed GA for optimization by minimizing the simulated and monitored data. Huang et al. (2008) proposed GA with a data mining approach for locating sensors in DNs. Later, they selected GA as the best solution.  Kim et al. (2008) applied ANN along with a computational flow model MOUSE (Model Of Urban Sewers) to identify pathogen release locations. Baranowski and LeBoeuf (2008) proposed GA to minimize both contaminant concentrations and the cost of demand alternation.  Preis and Ostfeld (2008) applied a multi-objective genetic algorithm to place sensors.  Preis et al. (2008) suggested GA linked with EPANET to identify the source of contamination. Al-Zahrani and Moied (2003) used GA to optimize monitoring stations to identify proper locations for monitoring stations that can cover a DN. There can also be other advanced GA such as Genetic Algorithm II to solve the optimization problem (Weickgenannt et al. 2010). In cases of multi-objective optimization, Pareto solutions, based on a trade-off analysis of objective functions, are frequently used  (Huang et al. 2007; Kim et al. 2008; Preis and Ostfeld 2008; Weickgenannt et al. 2010) . There are other, less popular algorithms such as the greedy algorithm (Comboul and Ghanem 2013; Xu et al. 2008)  and the optimal predictor corrector algorithm (Guan et al. 2006). Table 2-7 compares common optimization algorithms, where each algorithm is evaluated as low, medium, and high under specific selection criteria.  Here, the selection criteria are simplifying optimization problem, self-sufficiency, handling multi-objectives, representing uncertainty, efficiency in obtaining solutions, less memory and computational time requirements. Table 2-7 concludes that GA can be regarded as the most effective algorithm in handing most of the criteria and is a strong candidate for multi-objective optimization as well as decision making. 33  Risk analysis  Risk analysis due to contaminant intrusion in a DN is a difficult task because it depends on many factors such as microbiological growth, inactivation by disinfection and treatment technologies. Appendix C indicates some modelling approaches used in risk analysis. The analysis includes vulnerability assessment, soft-computing based likelihood estimation, exponential models, and QMRAs. These studies are either generic (Perelman et al. 2013; Propato and Uber 2004b; Sadiq et al. 2006, 2007) or carried out for microbiological contamination for norovirus, rotavirus, Cryptosporidium (Austin et al. 2012; Ebacher et al. 2012; Teunis et al. 2010). Propato and Uber (2004b) first developed a dynamic vulnerability assessment in a DN using EPANET- MSX and kinetic modelling. The method identified the interaction of free residual chlorine and microbes in a DN. Ebacher et al. (2012) proposed a methodology for assessing public health risks associated with Cryptosporidium using negative exponential models.  This method investigates primarily negative pressure events in DNs for two possible pathways: leakage orifices and submerged air vacuum valves (AVVs). Austin et al. (2012) developed a methodology encompassing QMRA, EPANET and Monte Carlo simulations to perform exposure assessments for Cryptosporidium. Teunis et al. (2010) performed hydraulic and Monte Carlo simulations for virus attack. Appendix C briefly describes these studies.  It is observed that previously described QMRA is a commonly adapted procedure.  Previously discussed soft-computing techniques for optimization can be also adapted for risk analysis, source detection and water quality failure studies. Sadiq et al. (2006) propose risk of contamination based on the Dempster-Shafer (DS) theory of evidence. DS theory is based on evidential reasoning proposed by Dempster (1967) and Shafer (1976) and is derived from the Bayesian theory.  This theory is powerful enough to combine probability theory with rule-based systems. Sadiq et al. (2007) upgraded the DS theory using AHP (analytical hierarchical method) and the Dempster-Shafer rule of combinations to predict water quality failure for contaminant intrusion. Perelman et al. (2013) developed a Bayesian believe network with cluster analysis to identify sources of contamination. In this method, an inference among the sensors using spatial analysis is considered to obtain the solution. 34  Table 2-7: Algorithm selection criteria for optimization Algorithm name Advantage /selection preference Simplify problem Self-sufficient Handles multi-objectives Represents uncertainty Efficient in obtaining solutions Requires less memory Requires less computational time General mathematical programming H M L N M M M Using pollution matrix N N N N L M M Backtracking algorithm N N N N L M M Set of covering algorithm N N N N L M M Stochastic approach N M M H M M M Soft computing -fuzzy logic & evidential reasoning M M M H M M M Soft computing -ANN L H H H M M M Soft computing -GA L H H H H H H  Low L Medium  M High H Nil N 35  Geographical and statistical methods Appendix D indicates some methods related to geographical and statistical concepts: regression analysis, GIS (Geographical Information System) based analysis, clustering, and statistical moment considerations. Helbling and VanBriesen (2009) adapted nonlinear regression to estimate chlorine demand signal. The analysis also includes microbiological interactions. Sample data are incorporated to minimize observed and predicted values for chlorine demand. A GIS based software tool known as GIS-CSI has been adapted by Shen and Mcbean (2012), where  main information related to water such as aging and land use importance are integrated to make decisions regarding source identification. Butera et al. (2013) propose a geostatistical approach with hydraulic modelling to assess uncertainty in the system. Perelman and Ostfeld (2011) performed cluster analysis to assess the topography and connectivity analysis for sensor placements and contaminant isolation. The clusters were defined according to the flow directions in pipes. Cluster analysis is the process of classifying a set of objects into groups of similar criteria. Typically, it involves statistical analysis with pattern recognition. The clusters ultimately create connectivity relationships and define system interconnections.  Commercial software Appendix E provides a brief description of some commercial software available for contaminant intrusion in DNs. Most of these software were developed by USEPA (US Environmental Protection Agency), the AWWA Research Foundation (AwwaRF) and Sandia National Laboratories. These software either predicts the effect of contamination through risk assessment, vulnerability assessment and condition assessments (AWWARF 2003; Hart and McKenna 2009; Jaeger et al. 2010; Khanal et al. 2005; USEPA 2010b; Yan et al. 2007) or provides responses in the form of digital warnings (Berry and Hart 2008; USEPA 2010a). Basically, hydraulic and water quality simulations are integrated with risk assessment and condition assessment tools.  For example, Yan et al. (2007) developed a software known as integrated risk assessment of water distribution systems (IRA-WDS). The software has three modules, 1) contaminant ingress model - the location of the pollutant from nearby sewer pipes, and open drains, 2) a contaminant transport model, and 3) a pipe condition assessment model- combination from a number of physical, environment and operational parameters.   36  The interrelationship among various intrusion analysis compiled with common management strategies can be observed in Appendix F. Appendix F describes a conceptual model specific for intrusion and  its management.  2.1.5.3 Optimization for booster chlorination Apart from intrusion related optimization, Table 2-8 provides few recent studies related to the optimization of booster chlorination. As can be noticed from the table that minimization of booster locations, dosage, mass, injection rate, and costs are commonly the objective functions, whereas maintaining FRC levels are used as constrains.  Other microbial and chemical water quality parameters can be added into the objective function or as constraints to solve a multi-objective function. Multi-objective optimization using genetic algorithms, mixed integer linear programming and linear least-squares coupled with EPANET hydraulics analysis are common approaches for optimizing booster chlorination. Optimization based on an overall “human health risk” or “water quality index” can be a new approach which demands further exploration in the concept of water quality management in DNs. For example, instead of using objective function of minimizing mass in DNs, maximizing WQI or minimizing risks can be the new objective functions.   37   Table 2-8: Water quality optimization in a DN related to booster disinfection Reference Method used General objective Objective function/functions Constrains Boccelli et al.  (1998)  Linear superposition,  Finite time linear programming-simplex method.  Dynamic scheduling of booster chlorination Minimizing mass using mass injection rate times time elapsed. Maximum and minimum FRC  Kansal & Arora (2004) Optimization using software LINDO Reliability analysis of booster chlorination  Min number of booster stations,  Min total mass flow rate, and  Reliability in terms of normalization with time step. Maximum and minimum FRC Prasad et al. (2004) Multi objective genetic algorithm Optimizing location and injection scheduling   Minimizing dosage, and   Maximizing demand volume NS Propato and Uber ( 2010) Linear least-squares formulation Ensure FRC uniformity in spatial and temporal variations. Minimizing injection rates  Maximum (4mg/L) and minimum (0.2mg/L) FRC Cozzolino et al. (2005) Hydraulic and water quality simulation with:  complex mass-balance equation,  Monte Carlo simulation, and  With predefined index. Locating optimal booster stations and dosage. Minimizing the overall dosage.  Maximum (1mg/L) and minimum (0.2mg/L) FRC;   Minimum injection rate (0.1 mg/s) and maximum injection rate (30 mg/s).   Taylor et al. (2006) Genetic algorithm (GA) Optimal scheduling of pumping and booster injection  Minimizing pumping costs,   Minimizing operational costs, and   Minimizing design costs.  Maximum (4mg/L) and minimum (0.2mg/L) FRC;  Max and min pressure constrains, and  Normalized tank storage function. Lansey et al. (2007)  Genetic algorithm (GA), and   Mixed integer linear programming Locating booster stations. Minimizing the total mass by multiplying the mass injection rate by flow rate and time.  Maximum (4mg/L) and minimum (0.5mg/L) FRC;  Rico-Ramirez et al. (2007) Two stage stochastic approach to incorporate uncertainty. Minimize the number of booster stations. Minimize the number of booster stations with uncertainty as function. NS Parks  et al. (2009) Analysis using EPANET 2 and EPANET-MSX. Optimizing sensor network for protecting contaminant intrusion. NS Maximum (4mg/L) and minimum (0.5mg/L) FRC; Carrico and Singer (2009) EPANET-MSX analysis. Predicting chlorine decay and THM formations NS NS      38  Table 2-8: Water quality optimization in a DN related to booster disinfection Kang and Lansey (2010) Genetic algorithm (GA) with EPANET. Optimize valve-operation to reduce reaction time.  Minimize mass flow rate multiplied by the flow rate and the FRC concentration, and  Minimize the total mass of injected chlorine.   Maximum (4mg/L) and minimum (0.2mg/L) FRC;  Minimum (40 psi) and maximum (100 psi) pressure. Hongxiang et al. (2010) Particle swarm optimization with genetic algorithm Locating booster stations. Maximizing a function consists of a number of booster stations and multiplication of level of coverage and partial influence. Maximum and minimum FRC. Gibbs et al. (2010) Various parameter settling methods in Genetic algorithm (GA), e.g., parameter less, convergence due to drift, typical parameter values, and self-adaptive parameters. Investigate different approach of GA to optimize chlorine injection. Minimizing injection chlorine rate.   Maximum (4mg/L) and minimum (0.2mg/L) FRC;  Gibbs et al. (2010) Genetic algorithm (GA) Optimizing pumping schedule to save costs.  Minimizing total dose, and   Maximizing water volume NS Meng et al. (2011) Particle Backtracking Algorithm and the GA. Minimizing the amount of booster dosage Minimizing the total mass.  Maximum (0.8mg/L) and minimum (0.02mg/L) FRC; Behzadan et al. (2012) Multi objective genetic algorithm Optimal location and scheduling  Minimizing  mass,   Maximizing volume with FRC constrains,  Maximizing volume with THM constrains.  Maximum and minimum FRC and THM threshold. NS-not specified 39  2.2 Thesis framework To overcome limitations in previous studies, various models have been proposed. A flowchart (Figure 2-4) illustrates the thesis framework. The flowchart also relates the thesis framework with thesis objectives. The study has been divided into water quality assessment and water quality management. Water quality assessment has been classified into index based and risk based models. Index based models have been proposed to evaluate regulatory violations related to chlorination. Regulatory violation may occur if there is: 1. over chlorination resulting higher levels of DBPs and taste and odor complaints, or 2. insufficient chlorination to ensure optimal or detectable FRC concentrations.  Various indices have been proposed including non-compliance potential index(NCP index) (Objective 1), modified CCME WQI specified to DBP rules (Objective 2),  modified CCME WQI using FRC concentrations, and modified CCME WQI using FRC & TTHM concentrations. A risk based method has been also proposed to evaluate contaminant intrusion potential at DNs (Objective 3). Later, various optimization schemes have been proposed to ensure water quality management in DNs. Previously, developed indices have been used in two optimization schemes to select dosage (Objective 4) and location for booster stations (Objective 5) respectively. Lastly, an optimization scheme has been developed to locate and select dosage simultaneously (Objective 6). This optimization scheme combines contaminant intrusion model, life cycle cost (LCC) analysis for booster chlorination and chemical as well as microbiological risk assessments. Finally, conclusions and recommendations have been proposed based on the indices and optimization schemes.  2.2.1 Non-compliance potential index Non-compliance potential index (NCP index) has been developed as a part of water quality assessment in DNs (Chapter 3). The index can evaluate regulatory violations by DBPs. TTHM, HAA5 and FRC have been selected as the variables to predict the index. A methodology has been proposed to implement the index using either monitored data, empirical model results (e.g., multiple linear regression) and disinfectant kinetics through EPANET simulations. 40    Figure 2-4: Flow chart of the thesis framework 41  2.2.2 Modified CCME WQI for complying with DBP regulations A modified version of CCME WQI has been proposed to assess the water quality. The study has developed a scheme to make this index suitable for DBP rules. A scoring method is proposed based on Analytical Hierarchical Process (AHP) and DBP rules. Later, CCME WQI has been modified with these weights to perform water quality assessment in DNs. The spatial water quality variations have been presented with this index using Kriging- a geostatistical method. This identifies regions with relatively poor water quality and highlights the potential locations for re-chlorination.  2.2.3 Risk based model for contaminant intrusion A point scoring scheme has been developed to identify potential intrusion points within a DN. The methodology has considered factors such as pollutant source (sanitary main characteristics), pollution pathway (pipe diameter, length, age, and surrounding soil properties), consequence (population, and land use), and operational factors (water pressure). The proposed methodology combines different sets of Geographical Information System (GIS) data using advanced ArcMap 10 operations. 2.2.4 Optimization to select booster dosage An index based on regulatory chlorine thresholds for microbial, chemical and aesthetics criteria has been proposed to help engineers make intelligent decisions. An innovative scheme for maintaining adequate residual chlorine with optimal chlorine dosages has been established. Another modified version of popular CCME WQI has been used as objective function. Preliminary temporal and spatial analyses identified critical zones (relatively poor water quality) in the DN. Temporal FRC concentrations predicted with EPANET software has been used to estimate the index, later coupled along with the optimization scheme to select booster dosage for the critical zones.  2.2.4 Optimization to select booster location An index based approach has been proposed to locate booster stations using optimization. The approach uses EPANET-MSX programmers’ toolkit to predict FRC and total trihalomethane (TTHM) and later converts to water quality index (WQI). Maximum covering location problem 42  (MCLP), a heuristic algorithm, has been used for optimization for maximizing WQI. TTHM has been converted into THM species using quadratic optimization and later to cancer and non-cancer risk potentials. Finally, the required number of booster stations has been decided after a risk-cost trade-off analysis using risk potentials, WQI, and life cycle cost (LCC) of booster chlorination.  2.2.5 Optimization to select both booster location and dosage In this scheme, intrusion has been modelled for a microbiological indicator E. coli O156: H7 using EPANET-MSX programmer’s toolkit. Previously selected intrusion points have been selected as the intrusion entry points in a DN. Location and dosage for booster chlorination have been selected based on Multi-Objective Genetic Algorithm (MOGA). Quantitative microbial risk assessment (QMRA) and chemical risk assessment frameworks have been adapted to estimate risk potential variability. As objective functions, risk potentials and LCC for booster chlorination have been minimized during optimization.    43  Chapter 3 NON-COMPLIANCE POTENTIAL INDEX A version of this chapter has been submitted for publication to the Environment Monitoring and Assessment Journal with the title “Assessing regulatory violations of disinfection by-products in water distribution networks using a non-compliance potential index” by Islam, N., Sadiq, R., Rodriguez, M. J., and Legay, C. (Islam et al. 2015b).  3.1 Background An index based approach describing regulatory violations can be helpful in decision making as it - 1) is simple, direct and easy to understand because of the multiple variables conversion to a single index, 2) can describe complex data series such as seasonal variations, 3) contains less uncertainty as only water quality variables are required unlike risk assessment methods, 4) can help to predict regulatory violations by maintaining thresholds for multiple variables, and 5) can also evaluate microbiological water quality as well as chemical water quality by incorporating FRC as variable. Previously, index based formulations were popular to evaluate surface and ground water quality (Boyacioglu 2007; Brown et al. 1970; Dunnette 1979; Espejo et al. 2012; Hurley et al. 2012; Liou et al. 2004; Pesce and Wunderlin 2000; Said et al. 2004; Sarkar and Abbasi 2006; Swamee and Tyagi 2007). Most indexes cannot handle data spare conditions. Moreover, the aggregation formulations were more on integrating water quality variables rather than probability of regulatory violations. There were also limited applications to evaluate water quality in DNs (Islam et al. 2013) and to evaluate regulatory violations by DBPs (Sadiq and Rodriguez 2004). This chapter proposes a mean to estimate regulatory violations using a non-compliance potential index (NCP index) specific to DBPs. The NCP index predicts the probability of regulatory violations based on time series data. The Bayesian belief network (BBN) is used as a modelling tool to calculate the index. The BBN has the ability to deal with subjective and qualitative data and complex causal relationships while providing simple inference mechanism (Liu et al. 2009; Ren et al. 2009). It can also incorporate expert judgements and missing information. In the field of water quality, Joseph et al. (2010) propose a BBN model for compliance with water quality regulations. However, their application was site-specific and not focused on DBPs.  44  The evaluation with the NCP index requires temporal series data for selected variables. This may not be practical in many cases, especially for small communities. Therefore, the study proposes a complete scheme to evaluate a DN using NCP index under three possible conditions: 1) with DBP and FRC monitored data; 2) with predicted levels of DBPs using empirical (e.g., regression-based) models based on monitored DBP precursor indicators (Golfinopoulos and Arhonditsis 2002; Rodriguez and Sérodes 2005; Uyak et al. 2007; Zhang et al. 2011); and 3) with predicted levels of DBPs based on variable kinetics using a simulator such as EPANET and its ad-on. Moreover, important regulatory decisions can be made based on the index value. For instance, one can tell how many variables have violated the regulations or on the eve of violation for a specific period of time. 3.2 Methodology The NCP index for DBPs involves measuring the degree of regulatory violations for selected DBPs. Recent developments in analytical instrumentation and the ability to measure very low levels of chemicals have allowed researchers to measure and report many DBPs (Richardson and Ternes 2011).  Most require expensive analytical methods, such as gas chromatography and mass spectrometry (Richardson et al. 1996). These analytical methods demand careful operation and appropriate interpretation by technical staff, something that may not be possible in many municipalities. Therefore, the study proposes a flexibility methodology to estimate NCP with or without information about DBPs. Figure 3-1 states the detailed methodology, where the basic steps are, Step 1: Data collection, Step 2: Converting time series data to probability of exceedance, and Step 3: Inference the probability of exceedance using BBN. 3.2.1 Data collection  Regulations focus mainly on compliance with TTHM, and HAA5 thresholds. Therefore, these parameters were selected in this study to estimate the NCP index. FRC was also considered as it may represent an indication of both microbiological and chemical water quality (with DBPs). Moreover, data for FRC is commonly available in many municipalities. However, frequent TTHM and HAA5 data (i.e., with a high sampling frequency and spatially covering the study area) may not be available. As a result, three possible means, or paths, are proposed to obtain 45  desired water quality parameter values (Figure 3-1). Path 1: This path requires temporal series data for TTHM, HAA5, and FRC. A comprehensive monitoring/sampling program is needed in this respect. Laboratory experiments can generate data when a close simulation of full-scale systems (Sadiq and Rodriguez 2004) is possible. However, the data may not be suitable for non-compliance evaluation.   Path 2: Path 2 consists of using TTHM and HAA5 data estimates from empirical models. TTHM and HAA5 modeling requires information about factors responsible for their formation. In previous studies, FRC, temperature (T), pH and UV254 have been measured as precursors for DBPs (Golfinopoulos and Arhonditsis 2002; Zhang et al. 2011). These were also used in this study. Multiple linear regression was suited to model TTHM and HAA5. Typically, power function models are popular for modeling TTHM and HAA5 (Golfinopoulos and Arhonditsis 2002; Zhang et al. 2011). These models are appropriate for a situation where the reactions among the parameters are barely comprehensible. In this case, a logarithmic transformation was adapted to derive coefficients. For example, the input (DBP precursor indicators) and the outputs (TTHM, HAA5) are transformed into logarithm values such as: 1 2( )( )...............( )i i ia b mi i nDBP K PI PI PI            10 10 10 1 10 2 10log log log ( ) log ( )............... log ( )i i i i i nDBP K a PI b PI m PI     [3-1] where DBP= disinfection by-products under study, e. g, if i=1, it is TTHM and if i=2, it is HAA5; PI=DBP precursor indicators; n= order of DBP precursor indicators; K, a, b….m= regression co-efficient. Equation [3-1] was converted to a linear equation after log10 transformations and used for multiple linear regression. Commercial software SPSS (IBM SPSS statistics 21) was used for stepwise regression. Compared to standard regression, stepwise regression is more applicable because it selects the best parameters (here DBP precursor indicators) to predict the model.   46    Figure 3-1: Methodology explanation to estimate non-compliance potential (NCP) index47  Path 3: This path consists of TTHM and FRC data obtained from EPANET simulation. Established chemical kinetics can be used to predict the levels of desired DBPs. EPANET MSX is an adapted and commonly used tool in many recent studies (Brown et al. 2011; Clark 1998) (Figure 3-1). Kinetics for FRC and TTHM are well established in comparison to HAA5. Therefore, in Path 3, HAA5 was not included in the calculations of NCP index. Numerous kinetic models were commonly used to predict chlorine decay over time in a distribution network (Carrico and Singer 2009; Huang and McBean 2007). The first-order decay model is more popular and commonly used compared to others, because it is simple and can be easily used with the principle of linear superposition (Behzaidan et al. 2012).  Chlorine reacts with various chemicals in water such as organic matter, ammonia and metallic compounds. Chlorine also reacts with wall material and attached biofilms, and tubercles. Therefore, chlorine decay kinetics can be divided in two groups: bulk decay and wall decay (Al-Jasser 2007; Courtis et al. 2009; Fisher et al. 2011; Hallam et al. 2002; Hallam et al. 2003). The chlorine reaction in the bulk flow can be expressed as follows: bdCk Cdt             [3-2] where Kb=bulk reaction rate co-efficient (hr-1); C= FRC concentration (mg/L). Bulk reaction rate co-efficient (Kb) can be estimated using the Arrhenius equation (Hallam et al. 2003): [ ]( 273)ER Tbk Fe          [3-3] where F= Frequency factor (depends on order of reaction and experiment) =3×10-9 (assumed from Hallam et al. (2003)); E = activation energy (J/mol); R = ideal gas constant (8.31 J/mol°C); E/R = 6,616 °C. The reaction of chlorine at or near the wall can be related to bulk reaction and can be expressed as follows (Rossman 2000): ( )( )WdC Ak Cdt V             [3-4] 48  where kw=wall reaction rate co-efficient,  can be assigned based on pipe material, condition (basically repair condition) and age (Al-Jasser 2007; Hallam et al. 2002); A/V= the surface area per unit volume within the pipe. EPANET software models the FRC based on the above equations, where kb and kw have to be assigned for DN pipes.  Clark and Sivaganesan (1998) developed kinetic equations to predict TTHM formation based on chlorine consumption. The equations need regression analysis based on explanatory variables such as TOC, temperature, FRC and pH data, which make them more complex and unsuitable for areas where these data are unavailable. Clark (1998), and Trussel and Umphres (1978) provided a much simpler approach compared to Clark and Sivaganesan (1998), where TTHM formation is predicted based on consumed chlorine. A linear differential relationship is used for this method between TTHM and the chlorine consumption (Brown et al. 2011; Elshorbagy 2000):  ( )d TTHM dCLFdt dt            [3-5] where TTHM = TTHM concentration (mg/L); LF = a linear proportionality constant between TTHM and FRC decay. EPANET-MSX can model FRC and TTHM using Equation [3-2] to [3-5] and can predict temporal variations for different DN locations.  Note that Figure 3-1 shows an order of preference for three paths: Path 1 is more preferable than Paths 2 and Path 3; Path 2 is more preferable than Path 3. 3.2.2 Estimating probability of exceedance Probability distributions represent the probability that a random value X will be less than, or equal to, a value, x. For example, the cumulative distribution function (CDF) can be written as: ( ) ( )xF x P X x            [3-6] where the right side of the equation represents the probability of the random variable X (parameter value) being less than, or equal to, x (regulatory threshold). CDF for TTHM, HAA5 or FRC, can be plotted by using a series of monitored or modeled data. Figure 3-2 represents qualitative plots of CDF function in which the maximum value of the function is 1, representing largest value of probability. This figure shows how the real value of a water quality parameter x 49  can be converted to probability Px and represents both situations, e.g., for one value (Figure 3-2a), and a range (Figure 3-2b). NCP is about regulatory non-compliance; therefore the function for not being equal to, or less than, the threshold value can be expressed as: ( ) 1 ( ) 1 ( )x x xF x F x P X x              [3-7] Equation [3-7] is applicable for TTHM and HAA5 as regulations suggest thresholds of 80µg/L, and 60µg/L respectively. Similarly, the probability of not being a range can be expressed as (Figure 3-2b): 1 2 1 2 1 2 2 1( ) 1 ( ) 1 ( ) 1 ( ( ) ( ))x x x x x x xF x F x P x X x F x F x              [3-8] Equation [3-8] is applicable for FRC as regulations suggest a range of 0.2-4mg/L. Sensitivity was increased by using several breakpoints between 0 and the threshold value. For example, the threshold value for TTHM is 80 µg/L (xBP5). Intermediate breakpoints such as xBP1 = 0µg/L, xBP2 = 20µg/L, xBP3 = 40µg/L, and xBP4 = 60 µg/L can be assigned by simple multiplication of the threshold with 0, 0.25, 0.5, and 0.75, respectively. Figure 3-3 states details regarding the breakpoints for FRC, TTHM, and HAA5. The probability for each breakpoint is estimated using Equations [3-7] and [3-8]. Finally, the combined probability can be estimated using a weighted average of the breakpoints: 5050( )i iin BP BPinBPiF x WPW         [3-9] where Pn= NCP probability for a certain parameter n; BP i= order of the breakpoint; WBPi= weight assigned for each breakpoint. The assigned weights are 0, 0.25, 0.5, 0.75 and 1, same as the breakpoint multiplication factors.   50    Figure 3-2: Estimating probability of exceedance for a) single value, and b) a range 3.2.3 Inference The Bayesian belief network (BBN) is a graphical model to inference probabilistic relationships among variables (Pearl 1988). It represents a directed acyclic graph where the elements/nodes represent variables (TTHM, FRC, etc.) and the links represent causal dependencies among variables. For example, Figure 3-4 represents a BBN consisting: 1. a set of variables (such as FRC, TTHM, and NCP index) and a set of directed links among the variables, a) b) 51  2. a set of mutually exclusive states for variables such as positive (+Ve) and negative (-Ve) NCP, which means state of violating regulations or not, and 3. a condition probability for each variable with parent variables. TTHM and FRC are the parent variables for NCP index (NCP index is the child of TTHM and FRC) as the link goes from TTHM/ FRC to NCP index (Figure 3-4). Condition probabilities are used to quantify dependencies for each node from the parents in the BBN. Condition probability table (CPT) is assigned for each variable based on the parent variables. A variable with no parent is considered as unconditional probability (e.g., FRC and TTHM in Figure 3-4). In this study, probability conversion (described in Section 3.2.2) from the temporal data was used in this respect.   Parameter Unit XBP1 XBP2 XBP3 XBP4 XBP5 FRC mg/L 0 - 4 0.05 - 4 0.1 - 4 0.15 - 4 0.2 - 4 TTHM µg/L 0 20 40 60 80 HAA5 µg/L 0 15 30 45 60 Assigned weights  0 0.25 0.5 0.75 1 Figure 3-3: Explanation of breakpoints in combined probability calculation 52  The main idea behind BBN depends on the use of Bayes’ theorem to manage uncertainty by the conditional probability dependencies among variables (Pearl 1988). In a traditional BBN analysis with n number of mutually exclusive hypotheses Hk, k= 1, . . . , n, and a given evidence E, the probability can be expressed as:    1( / ) ( )P Hj / E( / ) ( )j jnk kkp E H p Hp E H p H        [3-10]  Figure 3-4: An explanation of BBN using probability and conditional probability table (CPT) where p(H/E)- The hypothesis belief of H upon observing evidence E, p(E/H)-likelihood of E to be observed if H is true, p(H)- probability of having the hypothesis true, and p(E)- the evidence probability.  53  The combined probability for a child is the joint probability of all the parent variables, which can be expressed as the product of all the parameter probabilities and the conditional probabilities gathered from the CPT. The basic BBN can be modified with addition and multiplication operators denoted as “” and “” respectively. If P is the probability distribution over n with discrete random variables {X1 ;X2 ; ... ;Xn}, then the modified BBN can be defined as follows. Conditional independence: 1 2 31( , , ,...., ) ( / ( ))nn l llP X X X X p X Parents X       [3-11] Joint probability: ( , ) ( ) ( / )m l l m lP Y y X x P X x P Y y X x            [3-12] Marginalization rule: ( ) ( ) ( / )m l m llP Y y P X x P Y y X x            [3-13] where X is the parent variable of order l; Y is the child variable of order m . Figure 3-4 describes conditional probabilities for NCP index using FRC and TTHM, which will be used in Equation [3-13] to estimate NCP index. The condition probabilities can be obtained through expert judgement or training from data (Joseph et al. 2010; Nadkarni and Shenoy 2001). However, expert opinion and extensive literature review was adapted in this study. Figure 3-4 represents CPT for NCP index using Path 3 (Using TTHM and FRC), while similar approach was adapted for Path 1, and Path 2 (Using TTHM, HAA5, and FRC). As NCP is dependent on 3 (Path 1 and 2) or 2 parameters (Path 3), the possible combination are 8 (23) or 4(22) (Figure 3-4). Finally, important regulatory decisions can be made based on the index value such as: 1. [0<NCP≤0.2]: Very low (VL). The water is safe from the perspective of regulatory violation. Variables are very low or negative for TTHM and HAA5 and within optimal range for FRC. 2. [0.2<NCP≤0.4]: Low (L). The value can be in this range, when some or all of the parameters follow an increasing trend or are on the way to violate the regulatory guidelines. 54  3.  [0.4<NCP≤0.6]: Medium (M). One parameter has violated the regulation. 4. [0.6<NCP≤0.8]: High (H). Some parameters have violated the regulations. 5. [0.8<NCP≤1]: Very high (VH). All parameters have violated the regulations. 3.3 Case studies In order to illustrate the methodology to estimate the NCP index, case studies were conducted in Québec City and the city of Kelowna (Provinces of Quebec and British Columbia, Canada, respectively) DN. The NCP index was calculated for the Québec City DN using monitored and regression model data (Path 1 and Path 2 stated in Section 3. 2.1). For the city of Kelowna DN, data obtained from the EPANET simulation was used to estimate the NCP index (Path 3 stated in Section 3. 2.1). It should be noted that the case studies are general applications to show how we can use monitored, regression and EPANET simulation data (Path 1, 2, and 3) using NCP index, rather than using the best available data at the respective locations.   3.3.1 Study area  Québec City A DN serving 240,000 inhabitants in Québec City (Province of Quebec, Canada) was considered as a case study. The St. Charles River situated inside the study boundary (Figure 3-5) is the water source refined by a treatment plant that delivers water to the DN (Figure 3-5). Sampling campaigns were conducted for the years 2005, 2006, 2007, and 2008. Legay et al. (2011) reported sieving, flocculation, sedimentation, filtration, intermediate ozonation, and chlorination as the treatment processes. During these campaigns, seven sites were sampled, denoted as SP 1, SP 2, SP 3, SP 4, SP 5, SP 6, and SP 7 (Figure 3-5). Re-chlorination was applied in sampling points denoted as SP1, SP2 and SP3. Factors affecting THM and HAA formation, such as FRC, T, pH, and UV254 were measured from the samples collected for 2005, 2006, 2007, and 2008. TTHM and HAA5 were also analyzed but only for 2006, 2007, and 2008. In this study, samples for THM and HAA measurement were collected in 40mL vials from the faucets of restrooms. Samples were collected after a 5min flow to get the water from the DN and not from the building pipes. Ammonium chloride was added as a dechlorinating agent prior to 55  collecting the water. Samples for UV254 and pH measurements were collected in 125mL plastic bottles. Temperature and FRC parameters were measured in situ during the sample collection. All samples were conserved at 4°C in a cooler and transported to the laboratory. EPA methods 524.2 (USEPA 1995a) and 552.2 (USEPA 1995b) were adapted to analyze THMs and HAAs using gas chromatographs with mass spectroscopy (GC–MS) and electron capture detectors (GC-ECD), respectively. Blanks and internal standards were made for proper quality assurance. Four THM species were analyzed (chloroform, BDCM, DBCM and bromoform). The sum of these four compounds represents TTHM. The analytical protocol assured detection limits of 0.3, 0.3, 0.4, 0.5 μg/L for chloroform, BDCM, DBCM, and bromoform, respectively. Five HAA species were analyzed with detection limits of 1.3, 0.9, 0.4, 1.0 and 0.7 μg/L for Monochloroacetic Acid (MCAA), Dichloroacetic Acid (DCAA), Trichloroacetic Acid (TCAA), Monobromoacetic Acid (MBAA) and Dibromoacetic Acid (DBAA), respectively. The sum of these five HAA species represents HAA5. FRC was measured using a Hach colorimeter (model DR-820) and a titrimetric method (Standard Methods 4500-Cl G). UV254 measurement was carried out using a spectrophotometer (Pharmacia, model 80-2097-62) at 254 nm and with a 50mm optical path quartz cell. First, the NCP index was estimated using 2006 and 2007 sampling data. The overall time period of one year was divided into two seasons: summer (May to October), and winter (November to April). Average values for each parameter were calculated for two seasons. Later, TTHM and HAA5 regression models were developed using data collected from 2006, 2007, 2008 samplings. The NCP index for SP 6 was evaluated using these models and data collected for 2005. It should be noted that only SP 6 could be evaluated as data were available for this point only. Finally, an example of uncertainty analysis was carried out using 2007 data. City of Kelowna Figure 3-6 shows a small part of the city of Kelowna DN, where we estimated the NCP index using EPANET simulation. There were 298 nodes, 3 tanks, and one reservoir. An EPANET model was collected from the city and necessary assumptions were made for the small part. The study area had 3 booster stations (Figure 3-6), where 1 mg/L of chlorine dosage was applied. Water supply with a constant chlorine dosage of 0.7 mg/L was assumed for the reservoir.  56    Figure 3-5: Sampling points in a Quebec City DN It should be noted that chlorine dosage at the reservoir and the booster stations were made based on the information collected from the city. Only bulk decay co-efficient (kb) was considered 57  using the Arrhenius equation and temperature value stated in Islam et al. (2013). The analysis was carried out for the summer (May to October) only, as summer produces the worse water quality according to Islam et al. (2013). The EPANET –MSX and Matlab toolkit was used to predict TTHM, and FRC. Finally, highest NCP indexes were reported after a comparison among the nodes. 3.3.2 Seasonal assessment: Québec City The seasonal NCP index was evaluated for the sampling points of the Québec City DN (Figure 3-7) using TTHM, FRC and HAA5 monitoring data (Path 1). The NCP index was higher in summer than in winter for both 2006, and 2007. FRC values were comparatively low, while TTHM and HAA5 concentrations were higher in summer. This is probably associated with the higher temperature which increases reaction rates. Except for SP 1, SP 2 and SP 3, the sampling points were in the very low range of NCP (0<NCP ≤0.2). SP 1 and SP 2 were sometimes in the low range (0.2<NCP ≤0.4) (Figure 3-7).  Figure 3-6: Study area in the city of Kelowna DN 58  These results indicate that the system is quite safe or that some of the parameters are on the threshold of regulatory violation. The NCP was medium (0.4<NCP≤0.6) for SP 3 in the summers of 2006 and 2007, indicating that one of the parameters may have violated the regulation. Detectable FRC is advisable in most provinces rather than maintaining a range for FRC. Therefore, some of the indexes were in the medium range. In winter, the NCP index for SP3 was higher compared to other sampling points. SP 1, 2 and 3 were identified as the points supplied by re-chlorinated water, which may result in more chlorine application and longer water residence times in the system involving higher TTHM, and HAA5 concentrations. Moreover, SP 2 and 3 were closely located (Figure 3-5) re-chlorination points and their superposition may also result in higher TTHM and HAA5 concentrations. Better management of the system using optimization studies on dosage and location could help in limiting the NCP values. It should be noted that the index was based on USEPA guidelines, also adapted in Québec City. 3.3.3 Regression analysis for Quebec City DBP precursor indicators (FRC, Temp., pH, and UV254) from the 2006, 2007 and 2008 sampling campaigns were used to develop models for TTHM and HAA5. Table 3-1 states the regression results for the seven sampling points. The stepwise procedure was applied for the multiple linear regression in SPSS software (IBM SPSS statistics 21). The method only selects significant parameters (p<0.10), therefore regression coefficients were declared “ns” (non-significant) for many parameters (Table 3-1). Previously, models were considered well performed when the R2>0.5 (Golfinopoulos and Arhonditsis 2002; Uyak et al. 2007; Zhang et al. 2011). In this study, most of the models performed well (0.725>R2>0.589) for TTHM except for SP 1(R2=0.484). HAA5 models also had performed well (0.831>R2>0.568) for most of the sampling points except SP 4 (R2=0.403). Durbin –Watson values were in the desired range (1.5 to 2.5) for most of the models, except for SP 1 (for both TTHM and HAA5 models) and SP 6 (TTHM model). In general, Temp (T), pH, and UV254 emerged as the significant parameters (p<0.05). More especially, T emerged as the most significant parameter. Variance inflation factors (VIF) were in the desired range (VIF<5) and showed less multicollinearity among the parameters. 59    Figure 3-7: Seasonal NCP index evaluation00.050.10.150.20.250.30.350.40.450.5Summer 2006 Winter 2006 Summer 2007 Winter 2007Non-complinace potential (NCP) Index SP 1SP 2SP 3SP 4SP 5SP 6SP 760  For example, the predicted models for SP 6 were: 1.53 0.47 4.55 1.0225410 ( )( )( )TTHM T pH UV          [3-14] 0.62 0.81 0.84475 10 ( )( )HAA T FRC         [3-15] DBP precursor indicators from 2005 were available to evaluate SP 6 only. However, other predicted models can be useful for future use. Models stated in Equation [3-14], and [3-15] were adapted to predict TTHM and HAA5 (Table 3-2) in SP6. Monthly data were available for this purpose and mean and standard deviations were estimated for summer and winter. Finally, the NCP index was estimated using predicted TTHM, and HAA5 values. NCP index values were in the medium range (0.4<NCP ≤0.6), indicating at least one parameter with a regulatory violation. Table 3-2 states that for both summer and winter, TTHM concentrations were higher than the regulatory standard (80µg/L<TTHM). For the summer of 2005 in particular, the predicted TTHM concentration was quite high (i. e. 146µg/L) and different from the measured TTHM concentration previously used to estimate NCP values. A similar situation was observed for winter. However, more sampling data will improve the model in future. 3.3.4 EPANET results for City of Kelowna The summer NCP index was estimated for 298 nodes in the city of Kelowna DN. FRC and TTHM values were obtained after the EPANET MSX toolkit using a Matlab coding. Assumptions were made from a previous study (Elshorbagy 2000) and some adjustments were made to make it suitable for the study area. Finally, inference was carried out using the CPT table (Figure 3-4). Figure 3-8 shows the evaluation results for 298 nodes. Twelve nodes emerged in the medium level (0.4<NCP ≤0.6), while 16 were low (0.2<NCP ≤0.4). The rest of the nodes (270 in number) came out to as very low NCP level (0<NCP≤0.2). The result supports the evaluation result reported in Islam et al. (2013), where the same area was evaluated to supply “very good” water quality. However, Islam et al. (2013) only used FRC, while in this study, TTHM was incorporated. Better management with optimization could improve this situation. It should be noted that the result might be more realistic if constants, e.g., linear proportionality constants, were applied from laboratory studies on the city of Kelowna’s water supply.  61  Table 3-1: Regression results for predicting TTHM and HAA5 TTHM model prediction Sampling location  SP 1 SP 2 SP 3 SP 4 SP 5 SP 6 SP 7 R2 0.484 0.703 0.695 0.675 0.589 0.724 0.673 n 23 24 22 23 22 22 22 Durbin-Watson 1.447 1.959 2.163 2.057 1.581 1.269 2.373  β p VIF β p VIF β p VIF β p VIF β p VIF β p VIF β p VIF α 1.13 0  -1.78 0.394  -1.93 0.383  1.60 9E-05  0.76 4E-07  -1.53 0.362  2.17 0.001  Log10(T) 0.39 2E-04 1.00 0.35 1E-05 1.00 0.32 0 1.07 0.44 8E-06 1.00 0.58 3E-05 1.00 0.47 0 1.00 0.33 0 1.06 Log10(pH)  ns  7.12 0.017 1.50 6.44 0.026 1.30  ns   ns  4.55 0.042 1.34  ns  Log10(UV254)  ns  2.05 0.001 1.51 1.64 0.001 1.24 0.50 0.024 1.00  ns  1.02 0.004 1.34 0.77 0.046 1.06 Log10(FRC)  ns   ns   ns   ns   ns   ns   ns  HAA5 model prediction Sampling location  SP 1 SP 2 SP 3 SP 4 SP 5 SP 6 SP 7 R2 0.568 0.651 0.831 0.403 0.681 0.677 0.588 n 23 24 22 23 22 22 22 Durbin-Watson 1.209 2.285 2.28 1.67 1.825 1.857 2.034  β p VIF β p VIF β p VIF β p VIF β p VIF β p VIF β p VIF α -2.97 0.098  -3.91 0.073  -0.79 0.565  0.64 7E-06  0.50 2E-04  0.62 0  2.81 0.006  Log10(T) 0.31 2E-04 1.00 0.28 2E-04 1.00 0.35 2E-07 1.07 0.40 0.001 1.00 0.75 2E-06 1.00 0.81 0 1.77 0.43 0 1.06 Log10(pH) 4.59 0.029 1.00 5.81 0.022 1.04 4.19 0.021 1.30  ns   ns   ns   ns  Log10(UV254)  ns   ns  0.98 7E-04 1.24  ns   ns   ns  1.31 0.033 1.06 Log10(FRC)  ns  0.49 0.029 1.04  ns   ns   ns  0.8447 6E-04 1.77  ns  R2: Statistical coefficient of determination; n: number of observations for model development; β: partial slope coefficient; p: level of significance; α: intercept; ns: non-significant; VIF: Variance inflation factor; T: temperature; FRC: free residual chlorine62  Table 3-2: NCP evaluation for SP 6 with TTHM and HAA5 models Season Parameter Unit Source Mean, µ Standard deviation, σ NCP  Index Summer FRC mg/L Sampling data 0.28 0.08 0.60 TTHM µg/L Model predicted 146 22 HAA5 µg/L Model predicted 7 2 Winter FRC mg/L Sampling data 0.44 0.08 0.43 TTHM µg/L Model predicted 79 25 HAA5 µg/L Model predicted 11 2 3.3.5 Uncertainty and sensitivity analysis An uncertainty analysis was performed to incorporate reality. Distributions were assigned in the CPT table rather than deterministic values. All the distributions were assumed to be normal for simplicity. Oracle crystal ball software was used to perform the uncertainty analysis after 10,000 iterations (Oracle Crystal Ball 11.1). The uncertainty analysis was applied to the Québec City DN for 2007 only. Figure 3-9 states the uncertainty analysis results for each sampling point in summer and winter. For example, SP 7 results can be seen from the figure showing an uncertainty ranges for summer and winter.  The NCP index value ranged from 0.09 to 0.13 for summer and from 0.009 to 0.011 for winter. Similarly, uncertainty results for other sampling points can be observed in Figure 3-9. For most cases, the results were at the same level as stated in Figure 3-7, except for SP 2: between low to medium; SP 3: between low to medium; and SP 5: from low to very low in the summer.  A sensitivity analysis was also performed to observe variable importance for NCP index estimation. One thousand random values were assigned to each variable and the NCP estimated for them. Figure 3-10 shows the NCP variation for randomly changing TTHM and HAA5, where both of them clearly show an increasing trend. The sharp trends in TTHM and HAA5 also show their importance in estimating NCP index. 63  3.4 Discussion and future possibilities In this study, NCP index is proposed which can evaluate the DN for possible regulatory violations for DBPs. Important regulatory decisions can be made based on the results using simple language.  Figure 3-8: NCP evaluation in a small part of the city of Kelowna DN Low 64  The index value can interpret how many variables have violated regulations or on the way to violation.  Moreover, the study proposes a scheme that might be applied to different situations i.e., with extensive sampling data, limited precursor indicator data or hydraulic simulations. This ultimately makes the scheme suitable even for smaller municipalities, where limited sampling data is available.  Among the proposed methods, a regression model using available precursor indicators (Path 2) could be applied to estimate the NCP index. However, other precursor indicators such as water age (water residence time) and dosage can be incorporated into the model to obtain more realistic results. The sampling size for the regression analysis was small (n=22-24) and a better predictive model could possibly be obtained if more sampling data were available. For the method using EPANET simulation (Path 3) to predict FRC and TTHM concentrations, better results might be obtained if laboratory-based studies were available for the study area.  Regulations suggest trade-offs between the DBPs and microbiological water quality. The proposed model has FRC, which partially represents microbiological pollution. However, a specific model is needed for microbiological contamination. Therefore, Figure 3-11 proposes a possible model for evaluating the microbiological NCP index. In a DN, microbiological contamination can take place if external intrusion occurs or through biofilm growth. Biofilm formation generally accelerates the microbiological regrowth of injured microbes in the presence of organic matter (can be measured with UV254, turbidity or other parameters), appropriate temperatures and longer residence times (water age) (Chang et al. 2010). External intrusion can take place owing to the condition of pipes, and with a driving force. The condition of a pipe can be represented by average pipe age and breakage rate. Aged pipes with higher breakage rates tend to have higher contaminant intrusion possibilities, while negative pressure can be the driving force (Sadiq et al. 2007). Parameter values such as FRC, water age and pipe pressure can be obtained by EPANET simulation, while real data can be used for other parameters such as pipe age, breakage rate, HPC, UV254 and temperature. Step-by-step BBN can be applied for NCP evaluation for bio-film growth and external intrusion occurrence.   65   Figure 3-9: Uncertainty analysis in NCP evaluation- Quebec City 200766    Figure 3-10: Sensitivity analysis, a) TTHM, and b) HAA5  Finally, a microbiological NCP index can be obtained from biofilm growth and external intrusion (Figure 3-11). It should be noted that modeling a microbiological NCP index is quite difficult as the relationship among the parameters is complex to understand and requires more exploration for future modeling. However, decision making using trade-offs among the DBP and microbiological NCP indexes might be a future approach for decision makers. 00.20.40.60.810 20 40 60 80 100 120 140 160NCP TTHM (µg/L) a) 00.20.40.60.810 20 40 60 80 100 120NCP HAA5 (µg/L)  b) 67    Figure 3-11: Possible NCP index associated with microbiological contamination68  3.5 Summary There are many regulations related to the control of DBPs and microbiological quality in DNs. A non-compliance potential (NCP) index was proposed for evaluating regulatory violations by DBPs, e.g., TTHM and HAA5. The index can simplify complex situations by combining temporal series data to a single value, and moreover possess less uncertainty compared to traditional DBP risk assessments. BBN was also utilized to combine probabilistic concepts to estimate the NCP index. Moreover, the model is able to interpret how many water quality parameters have violated the regulations or about to violate under no preventive conditions. The methodology also shows various application of the index such as with monitoring data or with predicted data from empirical models or EPANET simulations. In this chapter, case studies were performed in a Québec City DN (Province of Quebec, Canada) using sampling and regression model data. Monthly data series were collected for seven sampling points at the Québec City DN and were measured with standard methods. TTHM, HAA5, FRC and various DBP precursor parameters were measured for the Québec City DN. For both sampled and modeled data, the summer season emerged as the time with greater NCP index values. Another case study was conducted with data obtained by EPANET simulation on a small part of the city of Kelowna DN (Province of British Columbia, Canada). Nodes with higher NCPs were identified for the system, which can be helpful for decision makers to observe areas with possible regulatory violations. An uncertainty analysis was performed by assigning probabilistic inputs to the conditional probability table for the Québec City DN only. The results showed a slight variation in results for some of the sampling points. A sensitivity analysis was performed to observe the possible effect of the parameters. Both TTHM and HAA5 emerged as significant parameters.  Better management is recommended with optimization for booster dosage and locations. Additionally, a conceptual model for assessing microbiological NCP index was proposed, which might help to evaluate regulations using trade-off analysis to relate microbiological and DBP NCP indexes. The proposed model is suitable for any situation, from an area with vast sampling data to one with limited data such as small municipalities.69   Chapter 4 : DEVELOPING MODIFIED CCME WQI TO COMPLY WITH DBP RULES A version of this chapter has been submitted in the Water SA journal with the title “Assessment of water quality in distribution networks through the lens of DBP Rules” by Islam, N., Sadiq, R., Rodriguez, M. J., and Legay, C. (Islam et al. 2014c). 4.1 Background Stage 1 and Stage 2 DBP rules were implemented in 1998 and 2006, respectively (USEPA 2004). These regulations are focused on the control of DBPs exposure by providing various parameter thresholds such as maximum contaminant level (MCL) and maximum contaminant level goals (MCLG). Water quality assessment needs to be done in DNs based on these thresholds from the perspective of: 1. microbiological water quality that can be degraded by inadequate chlorination, 2. chemical water quality that can be compromised by DBPs, and 3. aesthetic water quality (Taste and odour (T & O)) that can be unacceptable due to high levels of chlorination. Many water quality assessment formulations have been proposed to assess surface water (Liou et al. 2004; Pesce and Wunderlin 2000; Said et al. 2004; Swamee and Tyagi 2007) (Section 2.1.4). Those studies have used an index based on water quality parameters for specific usage. CCME WQI is globally accepted index to represent complex water quality data from regulatory perspective. It has been applied to source water and needs to be applied for DNs (Khan et al. 2003). Previously, some limitations were identified in the CCME WQI formulation and modifications were performed (Islam et al. 2013). However, the index is too generic and requires further modification to make it suitable for DBP rules. For example, careful parameter selection is needed for DBP rules. There should be also weights among the parameters, which is missing in CCME WQI. Suitable water quality assessment formulation based on these requirements can 70  help in intelligent decision making with appropriate management strategies. The objective of this chapter is to modify CCME WQI to assess water quality within DNs in the context of DBP rules.  4.2 Methodology The methodology is stated in Figure 4-1, which contains seven steps: 1) water quality parameter selection, 2) converting parameters to sub-indices, 3) generating parameter relative weights, 4) aggregating sub-indices to get the modified CCME WQI, 5) temporal and spatial analyses of a DN, 6) comparing with other WQI formulations, and 7) analysing sensitivity by increasing the number of parameters. It should be noted that in Figure 4-1 the diamond shape states the applied approach to perform a particular step.  4.2.1 Parameter selection DBP rules highlight some parameters related to DBPs (Table 4-1). On this basis, TTHM, HAA5, and FRC, are parameters that can represent the chemical water quality associated with DBPs. Other DBPs can be considered based on the data availability. In order to represent the microbial water quality, turbidity can be incorporated (USEPA 2009). Additionally, total organic carbon (TOC) can be considered based on the availability. DBP rules have mentioned about TOC as precursors for generating DBPs. All of these parameters can contribute towards chemical and microbial water quality from DBP rules’ perspective. It should be noted that the most significant parameters should be selected to avoid redundancy. In small municipalities, TTHM, FRC, and turbidity might come as the only available parameters, which are thus considered for this study. However, in this study additional DBPs (BDCM, DBCM, and DCAA) are added later to see the effect of sensitivity with parameter addition. 4.2.2 Conversion to sub-indices Based on the nature of the parameter selected, two types of sub-index functions are assigned: 1) increasing, and 2) optimal (Figure 4-2). The original CCME WQI has linear normalization, while calculating F3 factor (CCME 2001). Incorporating more breakpoints (denoted as points in Figure 4-2) inside the transformation function instead of using linear function can increase the sensitivity.  71  Increasing sub-index function is applicable for parameters (TTHM, and Turbidity), when increasing value is undesirable. The transformation function can contain 2, 3, or even 4 break points (Figure 4-2a) based on decision makers own choice. For TTHM and Turbidity, 4 point functions are used for this study.  Optimal sub-index is valid for parameter (FRC) for which optimal range of values is desirable to represent the water quality. Figure 4-2b states different possibility of optimal sub-index function.   Figure 4-1: Methodology to assess water quality using modified CCME WQI for DBP rules It can be made of 2-2, 2-3, 3-3, or 3-3 break points. However, 3-3 break point function is used for FRC in this study. 72  The sub-index functions used for this study are stated in Table 4-2 for more details. The functions are carefully made based on the MCL stated in Table 4-1. However, the transformation functions are flexible for the decision makers and can be modified by changing the break points. 4.2.3 Assignment of weights  In this study, a new methodology has been proposed to assign parameter weights based on DBP rules. Firstly, four criteria are selected named as type of parameter, strength posed by regulation, Stage 1 / 2 DBP threshold related , and risk related. There are sub-criteria under each criterion except Criteria 4.  Table 4-1: Regulatory maximum contaminant level (MCL) and maximum contaminant level goals (MCLG) for DBPs and related parameters (USEPA 2010b) Parameter name Unit MCL (mg/L) MCLG (mg/L) in Stage 1 DBP rule MCLG (mg/L) in Stage 2 DBP rule Turbidity NTU 1 or 5*   Free residual chlorine (FRC) mg/L 4**   Total trihalomethanes (TTHM) mg/L 0.08   Chloroform  mg/L  - 0.07 Bromodichloromethane (BDCM) mg/L  0 0 Dibromochloromethane (DBCM) mg/L  0.06 0.06 Bromoform mg/L  0 0 HAA5 mg/L 0.06   Monochloroacetic acid (MCAA) mg/L  - 0.07 Dichloroacetic acid (DCAA) mg/L  0 0 Trichloroacetic acid (TCAA) mg/L  0.3 0.2 Monobromoacetic acid (MBAA) mg/L  - - Dibromoacetic acid (DBAA) mg/L  - - *USEPA (2009a) states  1 NTU when the system uses conventional or direct filtration and 5NTU when the system uses filtration other than the conventional or direct filtration. ** The maximum residual disinfection levels (MRDLs)  A score between 0-10 is suggested under each sub-criterion. Table 4-3 summarizes all of these sub-criteria with scores such as for Criterion 1 (type of parameter) sub-criteria are: 73   Microbes (M) related: A score of 10 will be assigned when the parameter is a direct or indirect measure of microbes,  DBP related: Any DBP species will get a score of 10.  Partially M & DBP related: Any operational parameter related to M and DBP will receive a score of 8.  Precursor: DBP related precursors will get a score of 5.   Figure 4-2: Proposed sub-index function, a) increasing, and b) optimal74  Table 4-2:Sub-index (SI) functions used in this study Parameter name Function name Mathematical representation  Graphical representation Turbidity  Increasing If Turb=0, SITurb=0   If 0< Turb<0.1NTU, SITurb=0.5Turb  If 0.1 NTU < Turb< 0.5NTU, SITurb= 0.05+ 17(𝑇𝑢𝑟𝑏−0.1)8  If 0.5 NTU < Turb< 1 NTU, SITurb= 0.9+ 2(Turb-0.5)  If Turb> 1NTU, SITurb=1   TTHM Increasing If TTHM=0, SITTHM=0   If 0< TTHM<20µg/L, SITTHM=0.005TTHM  If 20 µg/L < TTHM< 50µg/L, SITTHM= 0.1+ (𝑇𝑇𝐻𝑀−20)60  If 50 µg/L < TTHM< 80 µg/L, SITTHM= 0.6+ (𝑇𝑇𝐻𝑀−50)75  If TTHM> 80 µg/L, SITTHM=1   FRC Optimal If FRC=0, SIFRC=0   If 0< FRC<0.15mg/L, SIFRC=1-100𝐹𝑅𝐶12 If 0.15mg/L < FRC< 0.2mg/L, SIFRC= 0.2- 4(FRC-0.15) If 0.2mg/L < FRC< 0.8 mg/L, SIFRC= 0 If 0.8mg/L < FRC< 0.9 mg/L, SIFRC= 8(FRC-0.8) If 0.9mg/L < FRC< 1 mg/L, SIFRC= 0.8+2(FRC-0.9) If FRC> 1 mg/L, SIFRC=1   00.511.50 0.5 1 1.5Sub-index Turb. (SI Turb) Turb. NTU 00.511.50 50 100Sub-index TTHM (SI TTHM) TTHM, µg/L 00.511.5-0.3 0.2 0.7 1.2Sub-index FRC (SIFRC) FRC, mg/L 75  Table 4-3: Sub-criteria with scores for estimating parameter weights Criteria 1: Type of parameter Sub-criteria Brief explanation Score Microbes (M) related Directly or indirectly related to microbes. 10 DBP related Directly measured DBPs 10 Partially M & DBP related Microbial and DBP related operational parameter 8 Precursor DBP related precursors 5 Criteria 2: Strength posed by regulation Sub-criteria Brief explanation Score MCL based Has direct MCL 10 MCLG based Has MCLG to maintain MCLs for others 6 Mentioned in DBP rules Mentioned in DBP rules but without any specific MCLG or MCL 2 Not mentioned  Not mentioned in DBP rules 0 Criteria 3: Stage 1/2 DBP threshold related  Sub-criteria Brief explanation Score Thresholds are available only in Stage 2 DBP rule New MCLG available in Stage 2 DBP rule 6 Thresholds are mentioned both in Stage 1 & 2 DBP rules MCLG available in Stage 1 and stage 2 DBPs 10 No threshold is available in DBP rules   0 Criteria 4: Risk related (cancer) Parameter name Scoring method adapted from Sadiq & Rodriguez (2004) with updated value from IRIS data base Score* Chloroform  0.000 2 BDCM 0.230 8 DBCM 0.307 10 Bromoform  0.029 1 DCAA 0.179 6 TCAA 0.255 9 TTHM 0.566 10 HAA5 0.434 8 Other DBPs  2 Not a DBP  0 *Oral ingestion only. There was no data available for some of the DBPs, therefore, score value 2 is assumed. 76  Scoring for Criterion 2 is done based on the strength posed by the regulations. The parameters that have MCLs are given more score than the parameters with MCLGs, and parameters mentioned without any thresholds (Table 4-3). Criterion 3 is based on the frequency of mention in Stage 1 and 2 rules. For example, parameters have MCL/ MCLG mentioned in both Stage 1 and Stage 2 rules got higher score than the parameters newly mentioned in Stage 2 rule. As controlling risk exposure is one of the main targets of Stage 2 DBP rule, fourth criterion has been considered based on the possible cancer risk due to oral and inhalation exposure. An exposure study (Sadiq and Rodriguez 2004b) has stated an approach for assigning relative weights among the DBPs. The study has utilized IRIS data base to gather the unit risk value for oral and inhalation exposure.  Later, the unit risk values have been normalized to assign relative weights among the DBPs. The same approach has been used but with oral ingestion only using updated value from IRIS (USEPA 2014). However, the weights are again normalized between 0-10 to cope with other criteria. Based on the above mentioned methodology, a score is assigned for each of the fourteen parameters stated in Table 4-1. Si,j can be the denotation of the score where i represents the order of the parameter and j is the criteria order. The summary of these scores can be seen in Table 4-4. Later, four criteria are compared with each other using Analytical Hierarchical Process (AHP), a mathematical technique (Saaty 1977) providing the weights to each parameter related to an issue. Pair-wise comparison is the basis of AHP and estimates the priority vector of the contributing parameters. The summation of the weights in a particular family is equal to 1. Interested readers are referred to other studies for details (Chung and Lee 2009; Tesfamariam and Sadiq 2006). Let us denote the criteria relative weights by Wj , where j again stands for the criteria order. Wj (denoted as W1, W2, W3, and W4 for four criteria) are obtained using AHP and are summarized in Table 4-4.   Finally, the parameter weights (Pi) can be estimated using the scores (S i, j) and the criteria relative weights (Wj) by using following formulation: Pi=∑ 𝑆𝑖,𝑗𝑊𝑗𝑗=𝑗𝑗=1𝑀𝑎𝑥 (∑ 𝑆𝑖,𝑗𝑊𝑗𝑗=𝑗𝑗=1)          [4-1] 77  4.2.4 Aggregation  Finally, the modified CCME WQI aggregation is done by using the following modification: Scope, V1: The percentage of failed parameters is modified using the relative weights of the parameters. V1=∑ 𝑷𝒊×𝑪∑ 𝑷𝒊           [4-2] where C is a count factor 1 or 0. If one parameter fails C=1, otherwise C=0.  Frequency and amplitude, V2: Islam et al. (2013) have combined frequency and amplitude. Another modification will be done in it by combining the relative weights such as for normalized sum of excursions (nse), ,11ni k iiniiSI PnseP           [4-3] where SI= sub-index value; i= order of parameter; k= order for each time step; P= relative weight. V2=0.005 0.005nsense         [4-4] Modified CCME WQI= 100-√V12+V221.414       [4-5] The value should be from 0 to 100, where we can define the water quality such as [0-20]: poor; [21-50]: marginal; [51-70]: fair; [71-80]: good; [81-90]: very good; and [91-100]: excellent water quality (Islam et al. 2013).    78  Table 4-4: Estimating the parameters weights Criteria weights* W1= 0.09 W2= 0.48 W3= 0.27 W4= 0.15 Name Criteria 1: Type of parameter  Criteria 2: Strength posed by regulation Criteria 3: Stage 1/2 DBP threshold related  Criteria 4: Risk related Turbidity S1,1=10 S1,2=10 S1,3=0 S1,4=0 FRC S2,1=8 S2,2=6 S2,3=0 S2,4=0 TTHM S3,1=10 S3,2=10 S3,3=10 S3,4=10 Chloroform  S4,1=10 S4,2=6 S4,3=6 S4,4=2 BDCM  S5,1=10 S5,2=6 S5,3=10 S5,4=8 DBCM  S6,1=10 S6,2=6 S6,3=10 S6,4=10 Bromoform S7,1=10 S7,2=6 S7,3=10 S7,4=1 HAA5 S8,1=10 S8,2=10 S8,3=10 S8,4=8 MCAA  S9,1=10 S9,2=6 S9,3=6 S9,4=2 DCAA S10,1=10 S10,2=6 S10,3=10 S10,4=6 TCAA S11,1=10 S11,2=6 S11,3=10 S11,4=9 MBAA S12,1=10 S12,2=2 S12,3=0 S12,4=2 DBAA S13,1=10 S13,2=2 S13,3=0 S13,4=2 TOC S14,1=5 S14,2=0 S14,3=0 S14,4=0  *Estimated using AHP (Analytic hierarchy process); S= scores assigned from Table 4-3  00.10.20.30.40.50.60.70.80.91Parameter Weights 1 1 0.57 0.36 79  4.3 Water quality assessment: a case study 4.3.1 Study area Water quality is assessed for a DN in Quebec City, province of Quebec, Canada. The study area is the same (Figure 3-5) selected for the NCP index evaluation in Chapter 3. The sampling program went from April 2006 to April 2008 monthly in seven sampling points (Figure 3-5) named as SP 1, SP 2, SP 3, SP 4, SP 5, SP 6, and SP 7, respectively. Most of the sampling points are designated for re-chlorination except SP 4, SP 5, SP 6, and SP 7. 4.3.2 Sampling and analytical methods Samples were collected around 9.00am to 11.00am. The tap water was allowed to run for 5mins before taking the samples. Collected samples were analyzed for measuring FRC and turbidity. Several DBPs are also analyzed, however, only TTHM (summation of chloroform, BDCM, DBCM, and bromoform), is considered for this study. Later, DCAA was added to observe the effect of additional parameter inside the index. Samples were collected in 40mL vials for TTHM, and DCAA. Dechlorinating agent, ammonium chloride, was added on these vials before collecting the samples. For turbidity and FRC, samples were collected in 125mL plastic bottles. Samples were stored in the laboratory at 4°C as soon as the collection was done.  FRC was measured in situ by using a Hach colorimeter (model DR-820) and a titrimetric method (Standard Methods 4500-Cl G). A turbidimeter (Hach, model 2100N) was used for turbidity measurement. Gas chromatography with mass spectroscopy detection (GC–MS) (Varian chromatograph, model 3900) was used to measure TTHM. US EPA method 524.2 (USEPA 1995a) was adapted to conduct the analysis. 0.3, 0.3, 0.4, 0.5µg/L were the detection limits for TTHM species of chloroform, BDCM, DBCM, and bromoform respectively. A gas chromatography with electron capture detector (GC-ECD) (Perkin Elmer chromatograph, model AutoSystem XL) was used to analyze DCAA using USEPA method 552.2 (USEPA 1995b). The detection limit for DCAA was 0.9 µg/L. 4.3.3 Monthly water quality assessment Previously explained scoring method with AHP was carried out to assign weights among the parameters. TTHM and HAA5 came out as the parameters with relatively higher weights (Table 4-4). They were ranked high as they are regulated strongly with MCLs, have higher exposure 80  risk, and are strongly included in the regulations. The proposed WQI is formulated based on firstly chemical (considering only DBPs) and then microbial water quality, therefore, a microbial parameter such as turbidity is weighted a moderate value of 0.57 (out of 1). FRC is one of the common parameters, however indirectly related to microbial and chemical water quality and got a weightage of 0.36. The rest of the parameter weights can also be seen in Table 4-4. The proposed water quality assessment was conducted using only turbidity, FRC and TTHM. These are also commonly available parameters in smaller municipalities. Figure 4-3 shows the monthly water quality index variation in seven sampling points. Sampling points SP 4, SP 6, and SP 7 show relatively unchanged water quality index compared to others. Significant monthly variation can be seen for SP 2, and SP 3.  The water quality index is relatively better in winter (January to April) and starts degrading during summer (May to August). The temperature during summer is high, which increases the reaction rate of chlorine and forms more DBPs. FRC is also expected to degrade as it reacts with organic and inorganic compounds. Mostly, the water quality reached to the lowest levels in the fall season (September to December) and starts improving from the winter. Sampling point SP 2, supplied by re-chlorinated water from a tank located in the distribution system, shows comparatively lower water quality apart from other points. Higher level of FRC (due to re-chlorination in tank) can be one of the main reasons for the relatively lower water quality in sampling point SP 2. The higher residence time of water in the system for this point, which is associated to residence in tank and to its location in the system (in extremity), can also explain higher TTHM levels resulting in lower water quality. 81    Figure 4-3: Monthly variation of water quality index in seven sampling stations82  4.3.4 Spatial variability of WQI Water quality assessment is not feasible to perform at each node of a DN. It requires enormous resources and labor to obtain monitored data. Especially for small municipalities, the availability of data is even limited. In this situation spatial interpolation can be a useful tool to observe spatial variability. Spatial interpolation measures a constituent value at a place where it is not measured but is interpolated based on measurement taken at other places. This can be valuable to map WQI for a DN, especially for small municipalities. Kriging is an advanced Geo-statistical tool and can be used for spatial interpolation (Bayraktar and Turalioglu 2005). It works after investigating the interaction spatial behavior using z values. Basically, it can generate a map for an area using some known values. The details of kriging method can be seen elsewhere (Murphy et al. 2010). There are a number of studies related to water quality at source (Garreta et al. 2010; Murphy et al. 2010), and ground water (Baalousha 2010; Spruill and Candela 1990) using kriging.  There are also studies to interpolate pressure and flow at DNs (Bogdan and Studzinski 2007).  Therefore, kriging will be applied in this study to observe spatial variability of water quality using ArcMap 10.0. As previously observed,  the water quality index is generally lower during summer and fall, therefore, only results for these seasons will be analyzed to study the spatial variation of the index. The monthly results from May to August are combined using simplistic average to represent the summer season, while September to December are combined to represent the fall. Figure 4-4 presents the spatial variation for summer-2006, fall- 2006, summer -2007, and fall-2007. The water quality is between marginal (WQI value: 21-50) to fair (WQI value: 51-70). Much spatial variation can be seen for fall 2007 as water quality varies from marginal to fair. Fall 2006 shows less spatial variation (WQI: 55.85-55.86) as the water quality was all fair throughout the space. The water quality is better in summer 2007 compared to summer 2006. For most cases, the middle region of spatial analysis (Figure 4-4) is showing better water quality compared to the side regions.  The water quality in distribution depends on the flow rate, flow direction, and stagnation time. Higher flow rate can result in more dilution and less contaminant concentrations. It can also result is higher water residence time favoring FRC decay and DBP generation. Stagnation can 83  also result in low FRC and high DBP levels.  Avoiding stagnation zones can thus contribute to improve water quality (Coelho et al. 2003). Introducing re-chlorination or booster points and a minimum chlorine dosage application can be a solution to improve water quality. It will maintain minimum free residual chlorine which will make the system safe for further contamination and with less DBPs. For example, installing new re-chlorination stations near SP 2 and SP 3, and optimizing the chlorine dosages in these stations can improve the water quality, especially, during fall and summer season. Further optimization study is recommended to obtain a solution in this regard. 4.4.5 Comparison with other models Most of the WQI formulations encountered in the literature are rigid and not applicable using other parameters. Therefore, average, weighted average, and conventional CCME WQI are compared with the modified CCME WQI proposed herein. Figure 4-5 stated the comparison between these assessment formulations. Modified CCME WQI proves to be comparable with average and weighted average formula with higher R2 values. Conventional CCME WQI shows less sensitivity as most of the results are showing the same index value. The modified CCME WQI delivers the basic message of the original CCME WQI, but is more specific for DBP rules. Adding other parameters can be done to check the sensitivity. 4.4.6 Sensitivity after adding other DBPs Parameters are added in the modified CCME WQI to observe the variation in sensitivity after adding additional DBPs.  BDCM, DBCM, and DCAA are added step by step in the modified aggregation function and the WQIs are, respectively, called modified CCME WQI level 2, level 3, and level 4.  For example, in modified CCME WQI level 2, turbidity, FRC, TTHM, and BDCM are considered. DBCM is added with turbidity, FRC, TTHM, and BDCM for modified CCME WQI level 3. Level 4 has DCAA with addition with the parameters considered in level 3. It should be noted that additional parameters are selected randomly just to show the effect of sensitivity with parameter addition. Simplistic sub-index functions are used in this regard. MCLGs for BDCM and DCAA are 0. Complete failure (SI value = 1) is assumed if there is any detection of those parameters. For DBCM, simplistic linear functions are assumed with the MCLG of 6µg/L (USEPA 2006).  84     Figure 4-4: Water quality index spatial variation using Kriging in ArcMap a) Summer 2006, b) Fall 2006, c) Summer 2007, and c) Fall 200785   Figure 4-5: Comparison with other WQIs Figure 4-6a states the variation of turbidity for modified CCME WQI, modified CCME WQI level 2, level 3, and level 4. Water quality is showing decreasing trend while increasing turbidity. Level 2 and 3 are showing almost same results.  The water quality showed more degraded value in level 4. The parameter weights are 0.78, 0.81, and 0.74 for BDCM, DBCM, and DCAA respectively. The figure says adding parameter decreases the sensitivity of the formulation. Water quality change while changing FRC (Figure 4-6b) is showing quite different graphical trend compared to turbidity. Water quality is higher, when FRC is from 0.3 to 0.8mg/L. The sensitivity is same as the turbidity means water quality shows less variability with addition of a new parameter. Figure 4-6c presents decreasing water quality while increasing TTHM. The sensitivity of TTHM is much higher compared to turbidity as TTHM has higher calculated weight compared to turbidity. The results for level 2, 3, and 4 are showing similar results with similar sensitivity 0204060801001200 20 40 60 80 100WQI Modified CCME WQI Modified CCME WQI vs AverageModified CCME WQI vs WeightedAverageModified CCME WQI vs CCMEWQI86  observation as FRC and turbidity. The sensitivity analysis concludes that unnecessary parameters addition decreases the sensitivity. Therefore, careful parameter selection is recommended.   Figure 4-6: Sensitivity observation after parameter addition in a) Turbidity, b) FRC, and c) TTHM 87  4.4 Summary USEPA has established Stage 1 and Stage 2 regulations to control DBP exposure to human health. An index to assess the water quality for DBP rules has been proposed in this chapter. The index is based on CCME WQI, however with modification to assign parameter weights after DBP rules. Parameters are selected based on Stage 1 and Stage 2 regulations with one additional parameter that is turbidity, to indirectly represent microbial water quality. A scoring method with AHP is proposed to assign parameter weights based on Stage 1 and 2 regulations. The proposed index has flexibilities such as using flexible parameter conversion functions or any number of parameters. The flexibilities make it suitable for small communities where we do have resources limitations. Intelligent decision making is possible with this tool as it can show the spatial variation by adding other Geo-statistical method. A case study was performed in a DN in Quebec, Canada. The DN is not a small DN, however used as a case study as a large data set was found to assess the water quality. Considering the constraints of smaller municipalities, only turbidity, FRC, and TTHM are considered to assess the water quality. The monthly water quality assessment shows best water quality from January to April. The water quality starts degrading from May and reaches the worst from September to October. Higher temperature increases the reactions between FRC and organic matters, which results less FRC and more DBPs. Finally, water quality starts improving from November and shows the best in winter from January to April. Spatial analysis is done using Geo-statistical method kriging to estimate water quality in regions without sampling stations. The spatial analysis can identify the locations in the DN with relatively lower water quality, where decision makers need to concentrate. Assigning booster stations with adequate dosage can be one of the possible options to deal with this situation. Further studies on optimization are required to improve the water quality through booster stations.   88  Chapter 5 : IDENTIFYING CONTAMINANT INTRUSION POINTS A version of this chapter has been combined with Chapter 8 and under preparation for possible publication in Water Research Journal with the title “Optimizing booster chlorination to minimize the impacts of contaminant intrusion in a small water distribution network” by Islam, N., Rodriguez, M. J., Farahat, A., and Sadiq, R. (Islam et al. 2015d). 5.1 Background Previously in the literature review, three necessary conditions have been identified for a contaminant intrusion (henceforth will refer as intrusion): 1) source of contamination 2) intrusion pathway and 3) a pressure gradient (Lindley 2001). The structural deterioration of water mains can result in leaks which lead to breakage and can serve as an intrusion pathway. Contamination can be microbiological (e.g., viruses, bacteria, protozoa, fungi) and physico-chemical (e.g., debris, soil, pesticides, fertilizers, petroleum, and detergent). Wastewater from adjacent broken sewers are generally the main microbiological pollutant source (Sadiq et al. 2010). Driving force such as low or negative pressures inside the water main will allow contaminants to enter inside the DN. Unprotected DNs can create catastrophic intrusion such as microbiological outbreaks in small municipalities, which demands multi-barrier approach including careful management of DN (Sierra Legal Defence Fund 2006). Therefore, the study proposes a risk based methodology to evaluate intrusion risk potential (IRP) to ensure effective DN management. The model requires common municipal inventory data instead of breakage, leakage and pipe condition information, however contains every aspects of intrusion in a DN. 5.2 Methodology Risk is a two dimensional number that combines likelihood and related consequences of failure (Lawrence 1976). Factors contributing to likelihood and consequences of intrusion are organized hierarchically (Figure 5-1). These factors, non-commensurate in nature, can serve as a surrogate for likelihood and consequences for intrusion risk potential (IRP).  89   Figure 5-1: Hierarchical scheme to identify intrusion points in a DN90  Municipal inventory data such as sanitary main diameter and length, water main diameter, age, and length, soil corrosivity related environmental data, and DN operational parameters such as pressure are considered in the analysis. Moreover, various scales have been proposed to transform non-commensurate data into indices for pollution source, structural failure, soil corrosivity, driving force, land use, and population density (Figure 5-1). Finally, these indices can be aggregated to rank the DN nodes from highest to the lowest values. The top ranked DNs were selected as the potential intrusion points for intrusion modelling. The details of the factors are described below. 5.2.1 Sanitary mains Sanitary sewers around a water main represent potential pollution source for microbiological contamination. Commonly available municipal sanitary GIS map can be used and pollution source index (PSI) can be calculated as follows:  1i ii Ni iiL DPSIL D           [5-1] where L= sanitary main length (m); D= sanitary main diameter (mm); i= order of area affected by a DN node; N= total number of nodes in the DN. A spatial analysis, e, g., analysis with Thiessen polygon, is required to select the area boundary affected by a specific node.  The PSI can be normalized from 0 to 10, where 10 represents the highest and 0 means the lowest possible pollution source. 5.2.2 Water mains  Municipal inventory generally includes water main information such as material type, diameters, length and installation years. These represent structural failure affected by their basic characteristics (Berardi et al. 2008; Hu and Hubble 2007; Kleiner and Rajani 2002). Breakage, leakage and pipe condition information are direct measures for structural failure, but require tedious record management. Moreover, factors and processes affecting pipe performance are not completely understood. Thus, factors such as water main length, diameter and age provide a simplistic way to estimate the structural failure index (SFI).  91  Table 5-1 provides proposed scales for water main diameter and age. The table suggests a number from 0 to 10 for each water main based on diameter and age separately. Table 5-1: Proposed rating schemes for water main diameter and age Granularity for age Age*  range(yr) Age rating (AR) Metallic Cementitious Plastic G 1 0-10 0 0 0 G 2 11-20 1 2 3 G 3 21-30 2 4 5 G 4 31-40 3 5 6 G 5 41-50 4 6 8 G 6 51-60 5 7 10 G 7 61-70 6 9 10 G 8 71-80 7 10 10 G 9 81-90 8 10 10 G 10 90-100 9 10 10 G 11 100->100 10 10 10 Granularity for diameter Diameter range (mm) Diameter rating (DR) G 1 0-40 10 G 2 41-50 9 G 3 51-100 8 G 4 101-200 7 G 5 201-300 6 G 6 301-400 5 G 7 401-500 4 G 8 501-600 3 G 9 601-700 2 G 10 701-inf 1 *- Age=Current year- Installation year; G- Granularity 92  The higher the rating number, there will be more likelihood of deterioration. Different materials are expected to show different deterioration modes with age. Therefore, three alternative rating schemes are proposed for age. Finally, the SFI can be estimated using the proposed equation: Structural failure index (SFI) = 1 11 12Z Zj j j jj jZ Zj jj jL DR L ARL L         [5-2] where DR= Rating assigned based on diameter (Table 5-1); AR= Rating assigned based on water main age and type of material (Table 5-1); j= order of water main under an area representing a DN node; z= total number of water mains inside the area.  5.2.3 Soil corrosivity Previously, point scoring methods (e.g. 10-point, 12-point, 25-point scoring method) were adapted to predict the soil corrosivity (Sadiq et al. 2010) for water mains. Most commonly, 10- point scoring method was applied for metallic pipes (cast iron and ductile iron), which was introduced by the Cast Iron Pipe Research Association (CIPRA) (the former name of Ductile Iron Pipe Research Association/DIPRA) (Sadiq et al. 2005) and later recommended by the American Water Work Association (AWWA 1999). The method uses soil properties like soil resistivity, pH, redox potential, sulphide and moisture content. The suggested score for each soil property is provided in Table 5-2. A point was assigned to each soil property, and soil corrosivity index for metallic pipes (SCI-M) was estimated with mathematical summation. Similarly, soil corrosivity index for cementitious, and plastic pipes are denoted as SCI-C, and SCI-P respectively. A modification was proposed in a previous study (Francisque et al. 2014) is adapted in this present research (Table 5-3).  Table 5-1 specifies selecting one (“or” gate) SCI based on the material of a water main, such as either SCI-M, SCI-C, or SCI-P will be estimated based on material type .  Finally, the SCI for each area under a specific node can be estimated as follows: 93  11Nk kkNNkL SCISCIL          [5-3] where L is the length of a water main (m); k is the order of water main under an area representing a DN node; and N is the total number of water mains inside the area.  Table 5-2: Scoring schemes for soil corrosivity for metallic pipe (Sadiq et al. 2010) Parameter Unit Values  Points Soil resistivity Ω-cm <1,500 10 ≥1,500-1,800 8 >1800-2,100 5 >2,100-2,500 2 >2,500-3,000 1 >3,000 0 Soil pH  0-2 5 2-4 3 4-6.5 0 6.5-7.5 0 7.5-8.5 0 >8.5 3 Redox potential  mV > +100 0 (+) 50 to (+)100 3.5 0 to +50 4 Soil sulphide content  Positive 3.5 Trace 2 Negative 0 Moisture content  Poor drainage(continually wet) 2 Fair drainage (generally moist) 1 Good drainage (generally dry) 0  5.2.4 Water main pressure A piped water supply system is devoted for supplying water at adequate pressure and flow to the consumer. The losses of pressure at a level that the water cannot be supplied with an adequate flow can initiate contaminant intrusions. Various software for hydraulic simulation have been in use to model pressure levels in DNs for over 70 years (Sadiq et al. 2010). However, EPANET is the most widely applied software and is also adapted in this study. A driving force index (DFI) can be assigned for each node after obtaining the pressure values from EPANET analysis and a rating scheme presented in Table 5-4.  94  Table 5-3: Soil corrosivity scoring system for cementitious and plastic pipes (Sadiq et al. 2012) Parameter Values Points  Cementitious pipes Plastic pipes Soil pH < 5.5 2 NA 5.5 - 6.5 3 ≥ 7 3 % of clay ≥ 35 3 3 30.0 - 35.0 2 2 20.0  -30.0 1 1 < 20 0 0 Soil sulphide content Positive 0 NA Trace 1 Negative 3 % of gravel > 30.0 NA 5 15.0 - 30.0 3 8.0 - 14.0 2 < 8.0 0 NA- not applicable Table 5-4: Proposed point scoring method for water main pressure  Granularity (G) no Values* Points G 1 <0 10 G 2 0-10 9 G 3 10-15 8 G 4 15-20 7 G 5 20-25 6 G 6 25-30 5 G 7 30-35 4 G 8 35-40 3 G 9 40-50 2 G 10 >50 1 *Pressure (m) obtained after EPANET 2.0 hydraulic analysis 5.2.5 Consequence factors Unintentional contaminant intrusions will result in potential consequences on consumers, business and other socio-economic activities. In case of failure of a water main, the consequences will depend on various factors among those we will consider the land use around the water main and the population that might be affected.  95  The impact of a water main failure will vary according the usage of the territory that might be affected. Indeed, the impact of a failure in zones used for hospital or schools will be more important than the impact of the same failure on an agricultural land, for example. To consider this aspect, each land use received a weight, called land use weight (LUW) expressing its relative importance. A relative weight has been attributed to each land use using the analytic hierarchy process developed by Saaty (1988). The analytic hierarchy process (AHP) uses pairwise comparisons to estimate the preference weights (wi) of each factor (Saaty 1988). Saaty (1988) provides a scale to assign relative importance to different factors in a group. These preference weights (wi) were normalized to 1, as described by the equation below for n contributory factors: W=∑ 𝑤𝑖𝑛𝑖=1 = 1;  0 ≤ wi ≤ 1         [5-4] A relative weight (wi) for each land use was estimated using AHP, however, the details the method calculations can be seen in Saaty (1988).  Finally, a combined land use consequence index (LUCI) can be estimated for each DN node: 11Nk kkNNkL LUWLUCIL         [5-5] The impact of an intrusion will vary according to the amount of the population of a given area. The impact on an area for which the density of population is very high, for example the downtown, will be more important than the impact of the same water main failure on a rural residential zone with a scattered small amount of persons. To consider this aspect, the density of population for each dissemination area was calculated as follows: Population density, PD (Persons/Km2) = 2        ( )Total populationinadisseminationareaDisseminationarea size Km [5-6] where a dissemination area (DA) is the smallest standard geographic area with an uniform population size from 400 to 700 persons (Statistics Canada 2015).  Lastly, the population density consequence index (PDCI) can be estimated as follows: 96  11Nk kkNNkPDCPIL DL          [5-7] The PSI, SFI, SCI, DFI, PDCI, and LUCI can be aggregated with mathematical summation and DN nodes can be ordered from the highest to the lowest score values. Finally, the top ranked nodes can be identified as the potential intrusion points. 5.3 Case study 5.3.1 Study area City of Kelowna (BC, Canada) inventory includes numerous water mains of different material, diameters, and of highly variable age and conditions. There were 2,586 water mains (Record provided in 2012). Two types of data based on Geographic Information System (GIS) were collected related to water mains. One of them provides the most reliable information about the city water main network, while the other one is most likely the GIS version of the EPANET model. Asbestos cement (AC), cast iron (CI), concrete (CONC), ductile iron (DI), copper (COP), high density polyethylene (HDPE), poly vinyl chloride (PVC), and steel (STEEL) were water main materials. PVC (1430 pipes for a total length of 187,387 m) and AC (849 pipes for 254,258 m of total length) pipes were the most frequent water mains. Many of the CI water mains were installed from 1939 to 1970, while PVC water mains were mostly newly installed. The water main identification (ID) for the same water main was different in those GIS data bases. A separate coding was created in Matlab to identify each water main in both databases. GIS database was also collected for 9,063 sanitary mains. Information on soil properties (e.g., soil resistivity, redox potential, soil sulphide content, pH, % fines) were collected from Pacific Agri-Food Research Centre (Agricultural and Agri-food Canada, Summerland, BC), Ministry of Agriculture (Abbotsford, BC), and Ministry of Environment (Kelowna, BC). GIS shape files were collected on soil properties from the Ministry of Environment, and Pacific Agri-Food Research Centre.  Additionally, soil information e.g., soil type, color, ground water table (GWT) were collected for about 50 bore holes (AGRA 1998) and a zonal analysis was performed for the study area. Assumptions and zonal analysis details can be found in (Sadiq et al. 2012). CI, DI, 97  COP, and STEEL pipes were considered as metallic pipes; while AC, CONC were cementitious pipes; PVC and HDPE were plastic pipes. Land use and population information were collected from the City’s GIS coordination department. There are 96 land use codes, which were described in Bylaws (2012).  These land uses are grouped as Agricultural (A), Rural Residential(RR), Urban Residential(RU/RM), Commercial (C ), Industrial(I), Public & Institutional (P/W), Health District (HD), and Comprehensive Development (CD) groups. Their weights were estimated using AHP (Sadiq et al. 2012). A population distribution map named “Census2006_DisseminationArea” was collected from the City, which shows the total population distribution in 2006 for various DAs. Various operations such as functions in GIS (e.g., clip, intersect, Thiessen polygon) and MS Excel (e.g., VLOOKUP) were used to assign factor values on DN pipes. For example, polygons were created using ArcMap 10 operation Thiessen polygon and were considered as representing areas for DN nodes (Figure 5-2). Later, GIS shape files for sanitary mains, water mains were superimposed to get the intersected sewer mains and water mains under each polygon. For example, separate water main information, e. g., length, diameter, and installation year under each polygon were collected. Similarly, soil corrosivity, land use and population distribution information were also extracted for each polygon. An EPANET model was also collected from the city, which had three reservoirs and twenty tanks. Hydraulic analysis was performed on the model to obtain the average nodal pressure (m) assuming a time period of ten days with 1 hour of time step. 5.3.2 Results and discussions Step by step methods were applied to estimate PSI, SFI, SCI, DFI, LUCI, and PDCI for 2093 DN nodes, and results are presented in ArcMap 10 maps. Index values were incorporated inside the polygons to represent the results for each polygon. Indexes for pollution source, structural failure, and soil corrosivity are presented in Figure 5-3, Figure 5-4, and Figure 5-5 respectively. In this approach, comparisons among the nodes were more important rather than the individual index values. Therefore, these figures represent a graphical comparison among the nodes in four grades (such as Grade 1, Grade 2, Grade 3 and Grade 4), where higher grade values represent more risk potentials.  98   Figure 5-2: City of Kelowna water main system with Thiessen polygons 99   Figure 5-3: Point scoring results showing pollution source index (PSI) in the City of Kelowna DN 100    Figure 5-4: Point scoring results showing structural failure index (SFI) in the City of Kelowna DN 101    Figure 5-5:  Point scoring results showing Soil corrosivity index (SCI) in the City of Kelowna DN102  A polygon with more sanitary mains generally presents nodes with highest potential pollutant source (Figure 5-3). Higher diameter also represents greater volume of pollutants and therefore was considered as an important factor to represent pollutant source. Figure 5-4 represents the SFI on 2093 polygons. Again, results were represented in four grades and the highest ranked DN nodes were observed. The water main age, material, diameter and length were the priority in estimating SFI, however, equally considered.  Similarly, Figure 5-5 represents the soil corrosivity and a comparison among polygons can be observed in four grades. Similar graphical representations were expected for DFI, LUCI and PDCI. Finally, Figure 5-6 states the IRP ranking numbers for almost 2000 nodes. There were some missing factor in estimating SCI, and SFI, therefore were ranked as less possible intrusion points in Figure 5-6 (marked in Blue color).  The ranking numbers were divided in ten classes and were categorized from Class1; Rank 1-200; Class 2: Rank 200-400; Class 3: Rank 400-600; Class 4: Rank 600-800; Class 5: Rank 800-1000; Class 6: Rank 1000-1200; Class 7: Rank 1200-1400; Class 8: Rank 1400-1600; Class 9: Rank 1600-1800; and  Class 10: Rank 1800-2000.  5.4 Summary Microbiological contaminant may enter in a DN if there is pathogen source, path way, and driving force. This can be catastrophic if not properly protected by booster chlorination. This chapter proposed a methodology, which identifies DN nodes with highest possible intrusion risks in a DN. A risk based framework was proposed for estimating intrusion risk potentials (IRPs, IRP for single) using water main, sanitary main, soil properties, land use, population density, and operational factors. Many point scoring methods and assumptions were proposed for these factors including advanced GIS functions using common municipal inventory databases. The methodology was implemented on the City of Kelowna (British Columbia, Canada) DN for IRP estimation. The proposed scheme is expected to be beneficial for the small and first Nations communities, where better management is required by identifying intrusion risk in DNs.  103   Figure 5-6: Intrusion risk potential (IRP) ranks on the City of Kelowna DN 104  Chapter 6 : OPTIMIZING BOOSTER DOSAGE  A version of this chapter has been published in the Environmental Monitoring and Assessment Journal with the title “Optimizing Booster Chlorination in Water Distribution Networks: A Water Quality Index Approach” by Islam, N., Sadiq, R., and Rodriguez, M., J. (Islam at al. 2013). Some other versions have been published as conference proceedings named “Locating optimal locations for booster chlorination using a modified CCME water quality index” in the 15th Canadian national Conference & 6th Policy Forum on Drinking Water (2012) by Islam, N., Sadiq, R., and Rodriguez, M., J. (Islam et al. 2012a) and “Managing water quality in small distribution networks using booster chlorination: a water quality index approach” in the 1st International Conference on Advances in Civil Engineering (2012) by Islam, N., Sadiq, R., and Rodriguez, M., J. (Islam et al. 2012b).  6.1 Background Maintaining optimal FRC concentration to maintain acceptable microbial, chemical and aesthetical water quality is a difficult task that requires optimization study for booster chlorination.  Previously, many studies have been discussed on the optimization of water quality in a DN (Cozzolino et al. 2005; Gibbs et al. 2010; Kang and Lansey 2010; Lansey et al. 2007; Ostfeld and Salomons 2006; Parks et al. 2009; Prasad et al. 2004; Propato 2006; Tryby et al. 2002). These studies are based on predefined thresholds of FRC that aim at optimizing dosage, booster numbers and the location of booster stations. Table 2-5 also stated a list of optimization studies. The typical objective functions were also discussed in Table 2-6. Maximum, minimum or both maximum-minimum FRC thresholds were used as constraints. 0.2 mg/L is generally defined as the lowest level to ensure microbiological control (USEPA 2004). There is uncertainty in the maximum level selection based on objective functions and the real situation. The range of available FRC in Canadian DN ranges from 0.04 to 4 mg/L (Sadiq and Rodriguez 2004) because it is very difficult to maintain even the minimum level of 0.2 mg/L, especially at the extremities. The situation is more critical for the point nearest to the booster station: comparatively high residual chlorine may cause T&O complaints and issues related to DBP formation. A maximum level of FRC concentration to avoid T&O complains is difficult to establish as it depends on customers’ personal choices and sensitivity (NHMRC 2004). For example, Australian Guidelines 105  suggest a threshold level of 0.6 mg/L (Health Canada 2009), but WHO (2004) suggests a range of maximum FRC concentration of 0.2 to 4 mg/L. Fixing a maximum level of 0.6 mg/L FRC may be somewhat strict and unreasonable for real DNs, since it requires many booster points with required dosages which might prove expensive. Therefore, an optimal concentration range of FRC such as 0.2 to 0.8 mg/L could be reasonable.  A DN node can possess a series of FRC concentrations for a certain period of time (can be obtained from EPANET analysis), which can then be converted into a WQI using a transformation function with threshold values of 0.2 to 0.8mg/L. It should be also noted that various FRC thresholds are generally defined to ensure acceptable microbial, chemical and aesthetic water quality. In this study an index is proposed using only FRC concentration unlike other water quality indices that require a number of water quality parameters. This type of index can be appropriate for small or rural communities where FRC is the major parameter defining water quality.  CCME WQI is a popular way of representing complex water quality monitoring data, especially for water quality at source (Khan et al. 2003; Lumb et al. 2006). The CCME WQI formulation is simple, making its acceptability higher among various stakeholders. However, it is associated with some fundamental mathematical problems. In this study, a modified CCME WQI was used to overcome these issues. Later, an optimization scheme is proposed using this index to select dosages for booster chlorination. The scheme might prove helpful to optimize chlorine dosage and potentially useful for water quality management in DNs.  6.2 Methodology Optimization using modified CCME WQI was proposed and applied to obtain minimum booster chlorine dosage in a DN. Figure 6-1 presents the proposed optimization steps such as: Step 1: EPANET network analysis, Step 2: Temporal and spatial analyses, .106    Figure 6-1: Optimization steps using modified CCME WQIModified CCME WQI DN 107  Step 3: Critical zone selection, and  Step 4: Factorial analysis and 3D surface optimization.  The details of these steps are stated below 6.2.1 EPANET network analysis Necessary information about the DN was collected to create the EPANET model. Now, chlorine decay model is required to predict FRC concentrations in a DN. Various chlorine decay models have been used to define chlorine decay in a water distribution network, such as first-order decay, power-law decay (nth order), first-order decay with stable components, power-law decay with stable components (nth order) and parallel first-order decay models (Carrico & Singer 2010; Huang & McBean 2007). However, overall chlorine decay can be broadly divided into two groups: 1) bulk decay-reaction with water constituents and 2) wall decay-reaction with biofilm, wall materials and tubercles. First-order chlorine decay for both bulk and wall decay is more popular because of its simplicity and the availability of analytic solutions (Courtis et al. 2009; Powell et al. 2010; Tamminen & Ramos 2008) adapted to studies. Table 6-1 provides a summary of chlorine decay kinetics adapted to this study. The table describes an Arrhenius equation to estimate the bulk decay coefficient (kb) and various wall decay co-efficients (kw) based on pipe material, condition (basically repair condition) and age (Al-Jasser 2007; Hallam et al. 2002). EPANET can model chlorine decay based on these predefined kinetics and predict the change in residual concentration for a time series in a water distribution network. The EPANET chlorine decay model also requires bulk and wall decay coefficients for each pipe estimated from the information provided in Table 6-1. The EPANET water quality analysis provides a time series of residual chlorine concentrations for a DN node (Figure 6-2).      108  Table 6-1: Summary of chlorine decay kinetics  Reference Equation Notation Overall kinetics Hallam et al. (2003)  Loureiro (2002) Powell et al. (2000a)  Tamminen and Ramos (2008) C(t) = Coe-kt where, k = kb + kw  t = the time period under consideration (hr) C = chlorine concentration after time t (mg/L) Co = initial chlorine concentration (mg/L) k = the overall decay co-efficient (hr-1) kb = bulk decay co-efficient (hr-1) kw = wall decay co-efficient (hr-1) Bulk decay co-efficient (kb) Hallam et al. (2003) [ ]( 273)ER Tbk Fe  F= Frequency factor depending on order of reaction and experiment (assume a value of 3x10-9 from Hallam et al. (2003) E=activation energy (J/mol) R = ideal gas constant (8.31 J/mol°C). E/R= 6,616 oC Wall decay co-efficient (kw) Hallam et al. (2003) Based on material and repair condition Material With repair /without repair Cast iron (CI) 0.12/ 0.48 Ductile Iron (DI) 0.13/0.52 Polyvinyl chloride (PVC) 0.09 Al-Jasser (2007)  Asbestos cement (AC)   0.71* *- based on diameter (200mm) and age (over 50 yrs)   Figure 6-2: Temporal variation in residual chlorine concentration at a given node     TV Time (hr)  0.18 mg/l Chlorine (mg/L) 109  6.2.2 Temporal and spatial analyses The network was divided into a number of zones for spatial/zonal analysis. Representative nodes (RNs) were selected for each zone. For example, arbitrary one middle and two extreme points (will be denoted as RN1, RN2, and RN3) were selected as RNs for each zone in this study. A ten-day EPANET analysis was run for a specific season using a one-hour time step and temporal residual chlorine data were exported from EPANET to a MS Excel sheet for further calculation. It should be noted that the extraction was carried out for each RN separately. Finally, temporal residual chlorine was converted using the modified CCME WQI formulation, which is described below. Previously CCME WQI has been described using Equation [2-1]. The index uses a uniform function, where any FRC concentration lower than 0.2 mg/L and higher than 0.8 mg/L will give a function value of 1. If the value is between 0.2 to 0.8 mg/L, the transformed value will be zero (Figure 6-3). There is no gradual change of the function for which a modification is recommended. Figure 6-3 shows the function used for CCME WQI and some possible modified functions for the modified CCME WQI.   Figure 6-3: CCME WQI and other possible modified functions Figure 6-4 shows the modified CCME WQI function adapted to this study. There are two additional points other than the maximum and minimum threshold points, which are denoted as ‘a’, and ‘d’. Decision makers may define these points based on their personal choice to maintain CCME WQI function FRC 110  water quality. For example, a=0.1 mg/L and d=1 mg/L were used in this study. Excursion points called as fuzzy excursion (denoted as FE and not related to fuzzy logic) were calculated (Figure 6-4) using the proposed function different from CCME WQI. This means that a value lower than 0.1 mg/L or higher than 1mg/L is a complete failure (i.e., function values “1” (Figure 6-4)).  The normalized sum for fuzzy excursions (nsfe) was calculated as follows: 1#niFEnsfefailedtest          [6-1] where FE= Fuzzy excursion Finally, the modified CCME WQI was estimated after calculating a factor denoted as V: V=0.005 0.005nsfensfe          [6-2] Modified CCME WQI= 100-V        [6-3] The value is expected to be between 0 to 100, where various range of  WQI will represent different states of water quality such as [0-20]: poor; [21-50]: marginal; [51-70]: fair; [71-80]: good; [81-90]: very good; and [91-100]: excellent water quality.  The modified CCME WQI has been calculated for the time series plot in Figure 6-2. There were 103 points below 0.2 mg/L from 49 to 240 hrs (Figure 6-2); FE values were calculated for these points. For example, Figure 6-4 shows a FE estimation of 0.2 for a point with 0.18 mg/L residual concentration. Finally, Equation [6-1], [6-2] and [6-3] were used to calculate nsfe, factor V and modified CCME WQI respectively. 111   Figure 6-4: Fuzzy excursion estimation using a modified CCME WQI function:  nsfe =0.2 0.3 ............ 0.4103  =0.175 V=0.17529.80.005*0.175 0.005  Modified CCME WQI=100-29.8=70.2 Therefore, the WQI of 70.2 represents the overall water quality for that particular time series of residual chlorine concentrations. 6.2.3 Critical zone selection Average water quality for each zone was estimated based on the mathematical average of the RNs’ water quality using modified CCME WQI (denoted MCWQI). Average 1 2 33n n nRN RN RNnMCWQI MCWQI MCWQIMCWQI      [6-4] where n = number of zones (i. e 1, 2, 3,…, n) RN1,2,3= Representative nodes for two extreme points and one middle point. a=0.1 0.2  d=1 M N 1 Chlorine, mg/L 0.18 0.2 FE= 112  Relative comparisons between zonal water quality indices will help to select the critical zones with relatively lower water quality (comparison between MCWQI1, MCWQI2, MCWQI3,… MCWQIn) 6.2.4 Factorial analysis and 3D surface optimization Factorial analysis was completed on the critical zones with various chlorine dose levels. A two level factorial analysis was performed using Design Expert software. For example, if there is m number of booster stations, 2m number of combination results will be generated using Steps 1-3 (Figure 6-1). A 3D-response surface optimization was carried out using those 2m number of combinations, where booster dosage minimization is the main constraint and water quality maximization is the objective function such as: Objective function: Maximize average  1 2, ,........ nMCWQI MCWQI MCWQI  Subject to: 0.2 mg/L< DO < 1 mg/L         [6-5] where DO= applied dosage in the booster stations; n= number of zones (for optimization, n refers to critical zones only) (i. e 1, 2, 3,…, n) Factorial analysis with higher level of booster dosage was performed to observe water quality improvement (objective function: maximizing average nMCWQI ); otherwise an extra booster station with a minimum booster dosage was applied (Figure 6-1). It should be noted that the location of a new booster station should be selected based on an arbitrary point in the area with comparatively fewer or no booster stations. However, detailed factorial analysis can be seen with an example in case study section. Steps 1, 2, 3, and 4 (Figure 6-1) are integrated step by step based on results obtained from EPANET analysis, MS Excel calculation (WQI part), and Design Expert software (optimization part).  113  6.3 City of Kelowna - A Case Study The water main system of the City of Kelowna consists of some 2586 water mains made of mainly asbestos cement (AC), cast iron (CI), concrete (CONC), and ductile iron (DI), and polyvinyl chloride (PVC). Basic information regarding the water mains is described in Figure 6-5 and includes the percentage of material installed with various installation dates. Material type  Number of water mains  Total length (m)  Diameter (mm)  AC  849  151, 430  100 to 500  CI  126  19, 541  100 to 400  CONC  33  14, 044  500 to 900  DI  130  26, 354  50 to 750  PVC  1448  187, 387  50 to 750    Figure 6-5: Percentage of various water main materials installed from 1939 to 2010 CI pipes are the oldest installed water mains and the kw value was assigned to consider the age effect for CI water mains only. There was no information gathered on repairs to the water mains. Therefore, the kw value for other materials was assigned the assumption of “without repair” (Table 6-1).The network has five reservoir sources and twenty tanks (Figure 6-6) (based on the EPANET model provided by the City). There are already 16 available booster stations (Figure 0%10%20%30%40%50%60%70%80%90%100%1939-1950 1951-1960 1961-1970 1971-1980 1981-1990 1991-2010PVCDICICONCAC114  6-6c). The general range of booster chlorine dosage is from 0.6 to 1 mg/L (information obtained from the City of Kelowna). The five reservoirs supply water with a constant chlorine dosage of 0.7 mg/L. Average monthly water temperatures were collected for the sampling stations. An Arrhenius equation was used to calculate a kb value using sampling point temperatures for summer (May to October) and winter (November to April), respectively. The kw values from Table 6-1and calculated kb value were incorporated into EPANET network and an analysis performed for a duration of 10 days (0 to 240 hrs). The network was divided into eight zones denoted as Z1, Z2, Z3, Z4, Z5, Z6, Z7, and Z8 (Figure 6-6c) for spatial analysis. 6.3.1 Temporal and spatial analyses Preliminary spatial (for the eight zones) and temporal (for summer, and winter) analysis results using average CCME WQI (using the original CCME WQI equation) and average MCWQI (Equation [6-4]) are provided in Table 6-2. The results are quite similar for these two indices, resulting in a higher R2 value of ~0.92. Figure 6-7 represents the spatial and temporal results using average MCWQI, which shows comparatively worse water quality in summer compared to winter. The higher reaction and related chlorine depletion during summer (due to high temperatures) is the main reason for this observation. Finally, Zone 1 and 2 emerged as the critical zones because they demonstrated comparatively worse water quality in summer (Z1-fair; Z2-marginal). Therefore, optimization steps described in methodology section were applied in these two zones only. Table 6-2: Temporal and spatial analysis results Zone  Summer Winter Average CCME WQI Average  MCWQIn Average CCME WQI Average  MCWQIn Z1 66.8 70.5 80.2 76.2 Z2 32.5 31.2 34.3 32.2 Z3 96.0 86.9 98.6 92.9 Z4 73.0 60.2 81.7 60.4 Z5 96.2 93.5 99.1 96.9 Z6 86.8 68.2 97.0 91.9 Z7 71.0 55.4 90.1 73.8 Z8 99.2 97.2 97.8 94.5 CWQI- Canadian water quality index; MCWQI-Modified Canadian water quality index; n- number of zone115   Figure 6-6: City of Kelowna DN, a) water main network in GIS shape file, b) EPANET model, and c) zoning system with booster station 116     Figure 6-7: Temporal and spatial analysis using modified CCME WQI for a) summer and b) winter Modified CCME WQI Poor (0-20) Fair (51-70) Marginal (21-50) Good (71-80) Very Good (81-90) Excellent (91-100) a) b) 117  6.3.2 Optimizing booster chlorination Figure 6-8 shows the water main system for Zone 1 with already available booster stations, water main pipes, and RNs (RN1, RN2, and RN3). There are three booster stations, e. g., R & E, Clifton and Rio. A number of combinations of chlorine dosages at these booster stations were compiled using factorial combination for dosage levels of 0.6 and 1 mg/L. There were eight combinations (23) referred to C1, C2, …, C8 and EPANET simulation was performed for those combinations. The average MCWQI values for each combination are reported in Figure 6-9.  Figure 6-8: Water main system for Zone 1 Response-surface optimization was performed using 3D-surface plotting in Design Expert software to select an optimal solution based on the combined results provided in Figure 6-9. Equation [6-5] was used to add the objective function and constraint. A decision was made based Representative node Booster location Water main pipes RN RN2 RN3 118  on a factor known as desirability, which ranges from 0 to 1. Here, 0 refers the worst situation and 1 to the best situation.  Combination no Booster concentration (mg/L) MCWQI Avg. MCWQI1 R&E Clifton Rio RN1 RN2 RN3 C1 0.6 0.6 0.6 100 0 100 66.67 C2 1 0.6 0.6 100 0 100 66.67 C3 0.6 1 0.6 100 0 100 66.67 C4 1 1 0.6 100 0 100 66.67 C5 0.6 0.6 1 100 0 100 66.67 C6 1 0.6 1 100 0 100 66.67 C7 0.6 1 1 100 0 100 66.67 C8 1 1 1 100 0.26 100 66.75  Figure 6-9: 3D surface optimization (with dose 0.6 and 1 mg/L) in Zone 1 The optimization function serves to maximize the water quality for each RN (will create higher desirability factor) and minimize booster dosage. Figure 6-9 also shows the optimal combination selection (dosage of 0.8, 0.8, and 0.8 mg/L for the three booster points) with a desirability factor of 0.354, which is much less than the highest desirability value of 1. However, the 3D surface plot shown for only Clifton and R&E booster points as similar types of 3D plots was observed for the other combinations.  0.354 Desirability  119  Booster dosage levels were increased to obtain better water quality. Similar types of factorial and 3D surface optimization were performed and the optimal solution came with a comparatively higher desirability factor of 0.8. But, the average water quality did not improve much from the previous situation of 66.6 (the new value was 66.75). RN2 showed  poor water quality for all the above cases because the booster points were located far from RN2. A new booster station is proposed to increase water quality. Figure 6-10 presents the location of the new booster station and the 3D surface optimization solution related to the situation. The solution (0.6 mg/L on the previous booster stations and 1 mg/L on the new station) is much improved compared to previous situations, with an average improved modified CWQI value of 99.98 (from 66.6). Similar analysis and optimization steps were performed and three new booster stations proposed for Zone 2. Finally, Figure 6-11 shows the improved water quality after optimization (for summer only), and significant improvements to water quality in Zone 1 and 2. 6.3.3 Model sensitivity Uncertainty can be incorporated into the modified CCME WQI formulation by changing the ‘a’ and ‘d’ points in the proposed function (Figure 6-3). This particular analysis situation (a=0.1, d=1) incorporates a higher R2 value of 0.92 for all summer and winter points (modified CCME WQI vs. CCME WQI plot in Figure 6-12). However, in some special cases, the modified CCME WQI proved to be more logical than the CCME WQI.  For example, a particular point had a time series of residual chlorine concentration ranging from 0.17 to 0.22. The estimated modified CCME WQI was 70.2, while CCME WQI was 48. Therefore, it can be concluded that the modified CCME WQI is more logical, in the sense that the residual concentrations do not deviate much from the regulatory threshold value of 0.2 mg/L. Various R2 values were estimated for different ‘a’ and ‘d’ combinations and indicated in Figure 6-12. The highest R2 value obtained was a=0.15 and d=0.9 for the combination, while the lowest value was a=0.15, d=2 (Figure 6-12).  6.4 Summary  Various residual chlorine concentration ranges can handle microbial, chemical and aesthetic water quality separately. An index (WQI) to measure water quality based on regulatory threshold 120  points for microbial, chemical and aesthetic water quality can help manage overall water quality, especially for small or rural communities where free residual chlorine is possibly the only monitored water quality parameter. A modified CCME WQI was proposed to express overall water quality using the free residual chlorine concentrations. Optimization steps to ensure maximum water quality using modified CCME WQI were also proposed to select optimal doses and locations for booster stations. The City of Kelowna (BC, Canada) water main system was used as a case study to prove the concept. Necessary data and information were collected to build the EPANET water quality model. The model simulation provided residual chlorine concentrations for a certain period of time considering the first-order decay coefficient for both bulk and wall decay. The DN was divided into a number of zones and representative nodes selected to express the overall water quality. The preliminary temporal and spatial analysis showed comparatively worse water quality in summer and critical zones were selected. 3D surface optimization was performed for critical zones and additional booster stations and improved chlorine dosages suggested. The scheme can be extended to ensure fewer DBPs by incorporating other commonly regulated DBPs along with residual chlorine.121    Figure 6-10: Optimization with an additional booster station (for dose 0.6 and 1 mg/L) in Zone 1   New booster a) 0.8 Desirability  b) 122   Figure 6-11: Water quality analysis with optimization for summer a) present situation and b) after optimization Modified CCME WQI Poor (0-20) Fair (51-70) Marginal (21-50) Good (71-80) Very Good (81-90) Excellent (91-100) a) b) 123   Figure 6-12: Sensitivity analysis for a modified CCME WQI  Predicted (modified CCME WQI)  CCME WQI 124  Chapter 7 : OPTIMIZING BOOSTER LOCATIONS  A version of this chapter is under preparation for a possible publication to Water Research Journal with the title “Locating chlorine booster stations in small water distribution networks: a methodology using optimization and trade-off analysis” by Islam, N., Sadiq, R., and Rodriguez, M. J. (Islam at al. 2015e). 7.1 Background Many studies have been reported on optimization of booster chlorination and some important ones are previously summarized in Table 2-5. Boccelli et al. (1998) investigated an optimization problem for booster injection scheduling. Minimization of total dosage was used as an objective function. The study was extended by Tryby et al. (2002), where they treated booster locations as variable in the optimization formulation. It can be noticed that the minimization of booster dosage, mass, and injection rates are commonly used as objective functions (Table 2-6), whereas maintaining FRC is used as constraint in optimization studies. However, the studies reported in the literature have certain limitations, which provided motivation for the current study. Number of samples, frequency and specific values of selected parameter(s) can be included in an optimization problem, especially from the perspective of maintaining regulatory thresholds. Most of the studies have been focused on the best method selection for optimization (e.g., genetic algorithms, mixed integer linear programming etc.) rather than the effect on overall water quality. Moreover, there have not been many studies related to small to medium sized DNs. Limiting booster numbers because of higher costs can be a major concern for small communities. However, they have to abide by the regulatory thresholds. There have been limited studies focussed on cost and specific DBP component, e.g., TTHM. Locating booster stations with optimization after meeting regulatory thresholds, and performing trade-off analysis using health risk, and cost can be some an innovative way of decision making at DNs.  Previously, a methodology to optimize dosage for booster chlorination has been presented in this study. This chapter will be focussing on selecting the locations for the booster chlorination. Another modification has been made on CCME WQI after considering FRC and TTHM concentrations.  The index will be incorporated as the objective function for optimization.  Later, 125  the required number of booster stations will be decided with trade-off analysis using chemical risk, water quality and life cycle Cost (LCC) for booster chlorination.  7.2 Methodology The proposed methodology helps in selecting booster locations and the number of booster stations required in a DN. First, optimization has been performed to locate booster stations, and in the next stage, the number of booster stations has been minimized through a trade-off analysis.  Figure 7-1 provides the required steps to perform the analysis. Steps 1-3 have been performed for optimization, while Steps 4-7 have been conducted for trade-off analysis. In Step 1, EPANET model has been created for hydraulic and water quality analyses. The model generates water quality parameter values for optimization (Lin and Yeh 2005; Ostfeld and Salomons 2006a). Analysis requires specific reaction kinetics for the selected parameters (Step 2). FRC and commonly regulated DBPs, such as total trihalomethane (TTHM3)(USEPA 2005a) have been selected as water quality parameters. In Step 3, a Matlab code has been developed to perform optimization using EPANET 2.0 and EPANET-MSX functions. An index, named modified Canadian water quality index (modified CCME WQI) has been used inside the optimization objective function (Islam et al. 2013). Temporal FRC and TTHM values have been converted into modified CCME WQI. THM species have been modelled using quadratic optimization in Step 4. Cancer and non-cancer risk potentials have been estimated using the THM species in Step 5. In Step 6, LCC has been estimated for booster chlorination. Finally in Step 7, the trade-off analysis has been performed using risk potentials, modified CCME WQI and LCC values 7.2.1 Reaction kinetics  Previously, chlorine and TTHM reaction kinetics using Equation [3-2] and [3-5] have been stated. Same kinetics has been adapted in this study using EPANET multi-species extention mode.                                                 3 TTHM: Summation of THM species such as chloroform (CHCl3), bromodichloromethane (CHBrCl2),  dibromochloromethane (CHBr2Cl), and bromoform (CHBr3) 126    Figure 7-1: Steps involved in optimization and trade-off analysis127  Baiscally, EPANET-MSX generates a time series of FRC and TTHM concentrations using these equations. A Matlab coding has been created in this respect, which also performs optimization. 7.2.2 Optimization to locate booster stations Simulated FRC and TTHM concentrations have been converted into an index.  Previously, Islam et al. (2013) (Chapter 6) have proposed a modified CCME WQI; however the index was based on FRC values only. This requires modification to incorporate both TTHM and FRC levels. Therefore, the proposed modified CCME WQI consists of two factors: 1) scope (V1), and 2) integrated excursions (V2). V1= Total number of failed variables 2× 𝟏𝟎𝟎       [7-1] “Failure” is counted for a variable when the concentration exceeds the regulatory threshold. For instance, TTHM will be considered as “failed” if the concentration is above regulatory level of 0.8mg/L (USEPA 2010b). Recommended levels of FRC range between 0.2 of 4 mg/L (USEPA 2006). Maintaining a minimum level of 0.2mg/L may not be possible for all DN nodes, especially at the extremities of the DN.  A highest level of 4mg/L may also create taste & odour complaints. As a result, water quality failure is assumed when the FRC is either above 0.8mg/L or below 0.05mg/L. CCME WQI defined a variable failure when one sample is failed among a series of samples (CCME 2001). In this study, failure has been assumed if 10% of the samples have been failed for a specific variable.  Parameter values have been converted into unit-less excursions using conversion functions shown in Figure 7-2. Functions can be changed based on the required situation or decision maker’s decision. Excursion values have been converted into sum of excursions (SE) such as: 1 1n nFRC TTHMi iE ESEn            [7-2] where EFRC = excursion for FRC by using conversion function stated in Figure 7-2a; ETTHM= excursion for TTHM by using conversion function stated in Figure 7-2b; n = total number of data points. It can be noticed that there are excursion functions defined for FRC; 1) one for a small network and 2) another one is for medium DN. Later, SE has been normalized as follows: 128  20.01 0.01SEVSE          [7-3] Modified CCME WQI has been calculated using the following equation: Modified CCME WQI=100-2 21 21.414V V       [7-4] where 1.414 is a normalization factor that converts the index between 0 and 100. Zero represents the worst and 100 represent the best water quality.   Figure 7-2: Conversion functions used for a) FRC, and b) TTHM 00.20.40.60.811.20 0.5 1 1.5Excursion FRC, mg/L Conversion func.FRC- medium DNConversion func.FRC- small DNa) 00.10.20.30.40.50.60.70.80.910 0.2 0.4 0.6 0.8 1Excursion TTHM (mg/L) Conversion Function TTHMb) 129  Finally, the optimization has been performed using the following objective function: 1( ....... )i iiQ Modified CCMEWQIMaxMax Q Q            [7-5] where Q = demand or flow rate; i = the order of a DN node. The proposed optimization has been solved as an unconstrained optimization using a heuristic method of maximum covering location problem (MCLP). This is an iterative method and selects a node that covers the largest area that is uncovered before (Subramaniam 1996). The algorithm will keep going unless the whole DN area is covered with adequate water quality. A “greedy algorithm” is behind this method, and was adapted previously for locating booster stations (Subramaniam 1996). A Matlab code has been written for this algorithm, which selects first the booster point to ensure the highest water quality for maximum number of nodes.  In next run, the code picks the next booster station among the remaining ones. It finds the best solution with the previously selected booster stations. The process continues unless the desired numbers of boosters have been selected. The coding requires long simulation time, however is reasonable for smaller DNs. 7.2.3 Modelling THM species  To estimate risk potentials, concentrations of individual THM species are required. Reaction kinetics are needed to model Chloroform (CHCl3), bromodichloromethane (CHBrCl2), dibromochloromethane (CHBr2Cl), and bromoform (CHBr3). Lasdon & Warren (1982) proposed a method to model THM species using optimization, which was later used in other studies (Elshorbagy 2000; Lin and Yeh 2005). The method is site specific and is based on an average bromide incorporation factor (BIF) (Gould et al. 1981): 3030[ ][ ]NNNNNNN THMBIFTHM         [7-6] 130  where N is the number of bromide atoms in a THM compound; N = 0, 1, 2, and 3 for Chloroform (CHCl3), bromodichloromethane (CHBrCl2),  dibromochloromethane (CHBr2Cl), and bromoform (CHBr3), respectively.   Using this concept, the bromide distribution factor has been defined as follows(Lin and Yeh 2005):  [ ]NNTHMSTTHM , where N=0, 1, 2, or 3       [7-7] The specific mass balance of THM species can be defined by the molar concentrations such as: [THM0]+ [THM1]+ [THM2]+ [THM3]= [TTHM]      [7-8] where [THM0], [THM1], [THM2], and [THM3] are the molar concentrations of CHCl3, CHBrCl2, CHBr2Cl, and CHBr3 respectively. Equation [7-6] and [7-8] can be simplified by Equation [7-7]: S0+S1+S2+S3=1          [7-9] BIF= S1+2S2+3S3          [7-10] Elshorbagy (2000) defined following constraints: S0≥S1, S2≥S3           [7-11] This study has assumed [THM3] negligible, and therefore, zero has been assigned to bromide concentration. Concentrations of other species have been modelled using quadratic optimization with the following objective function: Min [S1+2S2+3S3- BIF]2         [7-12] Subject to constraints defined in Equation [7-9] and [7-11]. 131  7.2.4 Cancer and non-cancer risk potentials In this step, THM species concentrations have been converted into cancer and non-cancer risk potentials denoted as CRP and NCRP respectively. Reference doses (RfD) and slope factors (SF) have been collected from Integrated Risk Information System (IRIS) database to estimate NCRP and CRP respectively (USEPA 2014) (Microbial risk assessment Microbial risk assessment refers to the steps to understand, reduce, and prevent risks caused by hazardous microbes, whether natural or anthropogenic, intentional or unintended. Though microbial risk assessment evolves from chemical risk assessment, there are fundamental differences between them; e. g. the exposure of chemical is mainly chronic while microbial exposure is acute. Moreover, quantification of microbial risk is harder compared to the chemical one as it depends on many factors such as microbial growth, inactivation by disinfection and treatments. Quantitative microbial risk assessment (QMRA) is a popular way to estimate the risk of infection by specific microbes (Austin et al. 2012; Besner et al. 2011). This approach is performed by the office of water to support SWTR and the LTESWTR (USEPA 2010b). The standard QMRA consists of four basic steps (Ahmed et al. 2010):  (1) Hazard Identification: Hazard identification is to define the potential microbes (e.g., protozoan, virus, and bacteria) which are harmful to the system and also investigates the effect of the microbes. Most drinking water related QMRA address enteric viruses (Hepatitis A, Norwalk virus, rotavirus), parasites (Ascaris, Eimeria, Cyclospora, Toxoplasma), protozoa (Cryptosporidium and Giardia) and bacteria (E. coli, Shigella, Salmonella, Vibrio cholera) (Haas et al. 1999). Unlike microbial water quality assessment, E. coli O157:H7 can be considered for simplicity to investigate the effect of chlorination with QMRA in DN. Table 2-3). The chronic exposure of individual THM species has been estimated for entire lifetime by using the Equation [2-2], where  IR=2L/day, EF=365days/year, ED=70yrs, BW=70kg, and AT=70yrs are assumed from previous studies (Hamidin et al. 2008). It can be noted that acute exposure is same as chronic exposure, as AT value is equal to EF×ED for chronic exposure.   CRP and  NCRP have been estimated using the following equations (Hamidin et al. 2008): 132  N N NTHM THM THMCRP Exp SF           [7-13] NNNTHMTHMTHMExpNCRPRfD          [7-14] The exposure route can be in the form of ingestion, dermal adsorption,  and inhalation for DBPs during cooking, washing, bathing, laundering, cleaning, and showering (Hamidin et al. 2008). Only oral ingestion has been considered in this study. Finally, CRP and NCRP for TTHM have been estimated as follows: 30NNTTHM THMNCRP CRP          [7-15] 30NNTTHM THMNNCRP NCRP          [7-16] 7.2.5 Life cycle costing for booster chlorination Life cycle cost (LCC) analysis is a common way of selecting a cost-effective option after considering various alternatives and their cost throughout their life span. LCC for booster chlorination requires cost components like capital, and operation and maintenance (O& M) costs. Costs for buying a piece land, and later building a simple house on it and getting initial utilities can be the elements for capital cost of booster chlorination. Costs due to chemical, operator, and utilities are some common parts in O & M cost. Generally, gas and hypo-chlorination are two common ways of chlorination in DNs, but only hypo-chlorination is considered in this study as this is commonly used in small municipalities (Card et al. 2015). Hypo-chlorination has less capital but higher O & M costs. It requires frequent visits by operators due to various routine tasks such as mixing the chemicals, and cleaning the chemical pump lines and diaphragms.  A generalized LCC function has been proposed based on a study by Card et al. (2015): (1 ) 3153.6(1 ) 1yhc Ohc hcyir ir Q DLCC PC MCir            [7-17]       133  where LCC= Annual life cycle cost  for hypo-chlorination ($/yr);  PChc= Amount of capital installment payments for hypo-chlorination over y years; ir= annual interest rate (%); αhc= cost of chlorine per litre ($/L); Q= Flow rate for the node selected for booster station (L/Sec); DO= Dosage for the booster station (mg/L); β= Concentration of liquid chlorine (%); MChc= Maintenance cost for hypo-chlorination ($/yr) Finally, a cumulative LCC have been estimated for an overall booster solution:   101nyrs kkLCC LCC          [7-18] where LCC10yrs= the cumulative 10 years LCC after adding cost for each booster station; k= order of booster station; n= total number of booster stations. Card et al. (2015) suggested capital, chlorine, and maintenance costs of $104,000, $1.24 (per liter), and 2,559$ respectively.  Additionally, 10.30% has been assumed as the concentration of liquid chlorine. Finally, a 10years LCC analysis has been performed for booster stations assuming an annual interest rate of 10%.   Equation [7-17] described LCC function for hypo-chlorination. Another LCC function has been described in Appendix G for gas chlorination. 7.3 Scenario Analysis Proposed methodology is implemented for an “example” DN that consists of 28 nodes including one reservoir (Figure 7-3a), which is a modified version of a network discussed in Islam et al. (2011). An EPANET network was created using the basic information provided in Islam et al. (2011). There were 42 pipes with a total length of 21.6Km. The base demand varied from 50 to 133.33 LPS (L/Sec) in nodes (Figure 7-3b). In each node, the base demand was multiplied with a factor ranging from 0.38 for 5am to 1.53 at 9am. The details of the network such as the length, diameter, and connectivity are presented in Figure 7-3a. Water was assumed to be supplied by gravity from the reservoir with 95m total head. As the network is small, a 0.2mg/L chlorine dosage was applied at each booster station. Additionally, a constant 0.2 mg/L chlorine concentration was maintained at the reservoir. Bulk-coefficient (kb) and linear proportionate constant (F) were assumed to be 0.0331/hr and 0.651 from Elshorbagy (2000). A chlorine 134  conversion function for small DN was adapted to make the analysis suitable for small network (Figure 7-2a).  The proposed scheme located four booster stations (Figure 7-3). The optimization selected Node 14, 3, 12, and 10 as the preferred order of booster stations. Five scenarios were created for effect analysis when there was no booster and consecutive addition of boosters. Details of the scenarios can be observed in Table 7-1.  The suggested booster stations were fairly distributed (Figure 7-3a), which ensures certain levels of FRC. Figure 7-4 represents FRC levels in Nodes 11, Node 17 and Node 24.    135   Figure 7-3: Network information for scenario analysis, a) network connectivity and dimensions, and b) base demand   136  Table 7-1: Details for scenario analysis Scenario no Number of boosters  Chlorination and Booster points Reservoir Node 14 Node 3 Node 12 Node 10 1 0 √     2 1 √ √    3 2 √ √ √   4 3 √ √ √ √  5 4 √ √ √ √ √  Node 11 is one of the closed nodes to the first booster station (Node 14). Node 17 and Node 24 represent nodes situated in the middle and farthest points in the DN. These nodes were selected to observe the results as they have shown typical changes after adding booster stations in each scenario.  The analysis was performed for 24 hrs, however results represent variations from 2 to 24 hr of simulation. Only reservoir was supplying chlorinated water for Scenario 1 (Figure 7-4a), hence lowest FRC levels were available in this scenario compared to others. The farthest node (Node 24) showed less and delay in receiving delectable FRC, while the nearest node (for example, Node 11) had detectable FRC from fourth hour of simulation. Node 14 was selected as the booster station in Scenario 2 (Figure 7-4b), therefore only Node 24 showed increasing FRC concentrations.  The water flow direction from Node 14 to node 24 was the main reason behind this change. Figure 7-4c showed simulated FRC after adding another booster station in Node 3, which showed increasing FRC in Node 11, and Node 17. Similar results can be expected for Scenario 4, hence only Scenario 5 has been presented in Figure 7-4d. In this scenario, four booster stations with 0.2mg/L dosage each have been used. The results showed the highest FRC increase in these nodes after adding four booster stations. However, these nodes showed FRC concentration greater than 0.4mg/L which was assumed as the highest threshold for small DN (Figure 7-2a). Similar results were expected for TTHM, and can be observed in Figure 7-5. Highest TTHM concentrations were approximately 7µg/L, 8µg/L, 12µµg/L and 13g/L for Scenario 1, Scenario 2, Scenario 3, and Scenario 5 respectively, which were below the recommended threshold of 80µg/L.    137       Figure 7-4: Simulated FRC for, a) Scenario 1, b) Scenario 2, c) Scenario 3 and d) Scenario 5  00.20.40.62 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24FRC, mg/L Time step, hr Node 11Node 17Node 2400.20.40.62 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24FRC, mg/L Time step, hr Node 11Node 17Node 2400.20.40.62 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24FRC, mg/L Time step, hr Node 11Node 17Node 2400.10.20.30.40.50.62 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24FRC, mg/L Time step, hr Node 11Node 17Node 24d) a) b) c) 138  Figure 7-6 presents the trade-offs between CRP, NCRP, and LCC in Node 11, 17, and 24. LCC shows increasing trend with addition of booster stations in each scenario.  30% nodes showed improvement in water quality after adding the first booster station.  Rest of the nodes remained unchanged in water quality. Only 18% nodes showed improvement in water quality after adding the second booster station and the remaining 82% showed degradation in water quality. Degradation in water quality was observed in each node after adding the third and the fourth booster station. For example, a sharp improvement in water quality was observed in Node 24 after the first booster station addition. This is one of the farthest nodes and therefore required booster station to supply further chlorine. Node 11 and 17 showed slightest increase after adding the first booster station. The water quality kept improving but remaining nodes showed degradation after adding the second booster. Figure 7-6b shows CRP and NCRP for Node 11, 17, and 24.  The figure shows only the maximum CRP and NCRP values between 0-24hrs of simulation. The CRP and NCRPs were increased in small amounts after adding the first booster station. The second booster station addition showed tremendous increase in both CRP and NCRP (Figure 7-6b).  However, there was no significant difference after adding the third and the fourth booster stations. After observing these effects, it can be concluded that one booster station is enough for ensuring water quality in this DN. It is always challenging to decide number of boosters after observing combined nodal effects. For example, some nodes may show increasing trend while others decreasing trend in water quality. Considering the important nodes with higher water demand or using the combined effect (using weighted average) for selected nodes might be a good solution.   139      Figure 7-5: Simulated TTHM for, a) Scenario 1, b) Scenario 2, c) Scenario 3 and d) Scenario 5  0510152 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24TTHM, µg/L Time step, hr Node 11Node 17Node 24a) 0510152 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24TTHM, µg/L Time step, hr Node 11Node 17Node 24b) 0510152 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24TTHM, µg/L Time step, hr Node 11Node 17Node 24c) 0510152 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24TTHM, µg/L Time step, hr Node 11Node 17Node 24d) 140    Figure 7-6: Trade-offs between a) water quality index and life cycle cost (LCC), b) Cancer risk potential (CRP) and non-cancer risk potential (NCRP) 7.4 City of Kelowna- Case Study The methodology was implemented on a part of the City of Kelowna (province of BC, Canada) DN. An EPANET network was collected from the city, and a small part was selected to apply the methodology. Previously, scenario analysis was performed to apply the methodology on a small DN. Next, this small part of the city of Kelowna DN was selected to represent a small to medium sized DN (Figure 7-7). There are one reservoir, three tanks, four pumps, and 340 water mains in this part. The reservoir was assumed to supply water with 0.6mg/L concentration and 341m 3040506070809010000.20.40.60.811.2Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5WQI Life cycle cost (LCC) x 100000 LCC ($/10yr)WQI (Node 11)WQI (Node 17)WQI (Node 24)a) 0.00E+002.00E-034.00E-036.00E-038.00E-031.00E-021.20E-021.40E-021.60E-021.80E-022.00E-020.00E+005.00E-071.00E-061.50E-062.00E-062.50E-063.00E-063.50E-064.00E-064.50E-065.00E-06Scenario1Scenario2Scenario3Scenario4Scenario5NCRP CRP CRP (Node 11)CRP (Node 17)CRP (Node 24)NCRP (Node 11)NCRP (Node 17)NCRP (Node 24)b) 141  head. Figure 7-7 shows location of the current boosters (denoted as CB, information collected by the city), proposed boosters (denoted and PB) (after optimization) and four nodes to observe the results (here forth will address as observed nodes). The figure also provides a brief description of seven scenarios to observe results (Figure 7-7).  Scenario 1: The current booster practice was modelled in this scenario. Three boosters denoted as CB1, CB 2, and CB3 were located in the EPANET model. 0.7mg/L dosage were assumed for each booster station.  Scenario 2: The study area was assumed to have no booster station. However, chlorinated water was supplied from the reservoir.  Scenario 3: PB1 was effective with a dosage of 0.6mg/L.  Scenario 4: PB1 and PB2 were effective in this scenario with 0.6mg/L of dosages. Scenario 5: Three boosters denoted as PB1, PB2 and PB3 were effective in this scenario with 0.6mg/L of dosage for each booster station. Scenario 6: This scenario had four boosters (PB1, PB2, PB3 and PB4) with 0.6mg/L of dosage. Scenario 7: Scenario 7 assumed five boosters (PB1, PB2, PB3, PB4 and PB5) for the study area with 0.6mg/L of dosage. Results were observed on node J1838, J1882, J1386 and J-6375 (ID used by the City). Nodes J1838, and J1882 are situated close to the proposed booster stations. Node J1386 represents an intermediate node and Node J-6375 is a proposed booster station.  Necessary assumptions were made in EPANET model to successfully run the EPANET analysis. kcl, and F were assumed from Elshorbagy (2000). There are many nodes with missing demand information. Therefore, optimization was implemented without any normalization with base demand. However, 0.1L/Sec of base demand was assumed for LCC estimation for any missing demand value.  142   Scenario no Brief description Water reservoir Booster 1 Booster 2 Booster 3 Booster 4 Booster 5 Dosage (mg/L) Location Dosage (mg/L) Location Dosage (mg/L) Location Dosage (mg/L) Location Dosage (mg/L) Location Dosage (mg/L) 1 Present practice by the city 0.6 CB1 0.7 CB2 0.7 CB3 0.7 N/A N/A 2 No booster station 0.6 N/A N/A N/A N/A N/A 3 One booster station 0.6 PB1 0.6 N/A N/A N/A N/A 4 Two boosters were used 0.6 PB1 0.6 PB2 0.6 N/A N/A N/A 5 Three booster were used 0.6 PB1 0.6 PB2 0.6 PB3 0.6 N/A N/A 6 Four booster were used 0.6 PB1 0.6 PB2 0.6 PB3 0.6 PB4 0.6 N/A 7 Five boosters were used 0.6 PB1 0.6 PB2 0.6 PB3 0.6 PB4 0.6 PB5 0.6  N/A- Not available Figure 7-7: Study area on the city of Kelowna DN with scenario details for the case study 143  7.4.1 Optimization results  Figure 7-8 represents the water quality change on the selected nodes for Scenario 1.  Scenario 1 represents the current booster practice by the city, where FRC concentrations were low or zero in certain nodes. For example, Node J1882 and J-6375 had fair amount of FRC but Node J1838 and J1386 had 0mg/L of FRC (Figure 7-8a). TTHM was not observed in Node J1838 and J1386 but higher TTHM was detected for Node J1882 and J-6375. Especially, Node J-6375 had TTHM above the regulatory threshold value (80µg/L) (Figure 7-8b).  Figure 7-9 represents Scenario 2, where no additional chlorine booster was used except the chlorinated water supply from the reservoir. The results clearly show the need of installing additional chlorine boosters in the DN. There was no detectable FRC to the extreme nodes of the DN. For example, FRC concentration was 0mg/L for Node J1386 and J1838 almost throughout the analysis period (Figure 7-9a). TTHM was detectable for Node J1882 and J-6375, however was less than the threshold value. Figure 7-10 represents Scenario 3, where the first booster was placed as PB1. It can be observed that this node is also used as a current booster station by the city. The first booster addition showed sharp increase in FRC for Node J-6375 (Figure 7-10a). The other observed nodes maintained the same FRC as Scenario 2. TTHM also showed sharp increase in concentrations for Node J-6375 (Figure 7-10b).  Figure 7-11shows water quality change for Scenario 4. The second booster station (PB2) was added in this scenario. PB2 is a tank and one of the farthest nodes to supply chlorinated water. FRC was fairly increased in this scenario showing better water quality compared to Scenario 1, 2, and 3. However, Node J1838 showed no FRC from 45 to 50hr simulation (Figure 7-11a). The EPANET has certain limitations such as negligible or no demand (Flow rate ≈0L/Sec) was observed in certain simulation steps, which could be one of the reasons for this variation.  Besides, Node J1882 and J-6375 had FRC lower than 0.2mg/L in certain hrs (Figure 7-11a). In terms of TTHM concentrations, most of them were below the threshold value except Node J-6375 where there were TTHM concentrations above 80µg/L (Figure 7-11b).  Scenario 5 is presented in Figure 7-12 where there were three boosters denoted as PB1, PB2, and PB3. Improvement was observed in this scenario compared to Scenario 4. Most FRC 144  concentrations were above 0.2mg/L, except Node J-6375 where initially the concentration was below or just 0.2mg/L (Figure 7-12a). Node J1838 also showed a few points below 0.2mg/L concentrations. Node J1882 had a few points with higher FRC spikes. Variations in nodal demands could be one of the reasons for these variations. TTHM concentrations were also under the threshold value except for a few spikes for Node J1882 (Figure 7-12b). Node J-6375 was almost unchanged in TTHM concentrations compared to Scenario 4.   Figure 7-8: Water quality change for Scenario 1, a) FRC and b) TTHM concentrations  00.20.40.60.811.225 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71FRC(mg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'a) 02040608010012014016018025 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71TTHM (µg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'b) 145    Figure 7-9: Water quality change for Scenario 2, a) FRC and b) TTHM concentrations  00.050.10.150.20.250.30.350.40.4525 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71FRC(mg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'a) 0510152025303540455025 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71TTHM(µg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'b) 146    Figure 7-10: Water quality change for Scenario 3, a) FRC and b) TTHM concentrations There were no significant change in between Scenario 6 and Scenario 7, therefore only Scenario 6 is presented in Figure 7-13. Scenario 6 had four boosters denoted as PB1, PB2, PB3, and PB4 with concentration of 0.6mg/L each. Most of the observed nodes remained unchanged except Node J-6375 where FRC concentrations were increased (Figure 7-13a). TTHM was also under control except for a few points just crossing the threshold line (Figure 7-13b).  00.050.10.150.20.250.30.350.40.4525 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71FRC(mg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'a) 010203040506070809010025 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71TTHM(µg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'b) 147    Figure 7-11: Water quality change for Scenario 4, a) FRC and b) TTHM concentrations    00.10.20.30.40.50.60.725 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71FRC(mg/L) Time step (hr) 'J1838''J1882''J1838''J-6375'a) 010203040506070809010025 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71TTHM(µg/L) Time step (hr) 'J1838''J1882''J1838''J-6375'b) 148    Figure 7-12: Water quality change for Scenario 5, a) FRC and b) TTHM concentrations   00.20.40.60.811.225 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71FRC(mg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'a) 010203040506070809010025 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71TTHM(µg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'b) 149    Figure 7-13: Water quality change for Scenario 6, a) FRC and b) TTHM concentrations Overall water quality change can be observed in Figure 7-14 after each booster addition.  For example, Scenario 2 to 3 represents WQI change when the first booster station (PB1) is placed (Scenario 3) compared to Scenario 2 when there was no booster. There were 211 nodes improved, 76 nodes unchanged and 11 nodes degraded in water quality. Similarly, Scenario 3 to 4 was compared when 23 nodes showed improvement. There were 274 nodes unchanged and 1 node degraded in water quality. 16 nodes still improved after adding the third booster station 00.20.40.60.811.225 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71FRC(mg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'a) 010203040506070809010025 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71TTHM(µg/L) Time step (hr) 'J1838''J1882''J1386''J-6375'b) 150  (PB3) (Scenario 5 to 4 in Figure 7-14). Almost all the nodes showed no change in WQI when there was a comparison between Scenario 6 to 5. However, some of the nodes were still improving in terms of increasing FRC (Node J-6375 in Figure 7-13). After analyzing all these figures, it can be concluded that four booster stations with 0.6mg/L of dosage can be a better solution compared to using no booster  (Scenario 2) or the current boosters (Scenario 1) adapted by the city. Further trade-off analysis was performed to get a better or more confirmed conclusion.    Figure 7-14: WQI change after booster station addition  7.4.2 Trade-off results WQI and LCC are presented for each scenario in Figure 7-15. Scenario 1 (current booster practice) showed no significant difference from Scenario 2 (without any booster station) on these observed nodes.  There was increase in WQI for Node J-6375 after adapting the first booster station (Scenario 3). There was significant improvement in water quality in Scenario 4 (two boosters) for most of the observed nodes except Node J-6375, which remained same as Scenario 3. Water quality kept improving in Node J1882 after adapting the third booster station (Scenario 050100150200250300350Scenario 2 to 3 Scenario 3 to 4 Scenario 4 to 5 Scenario 5 to 6 Scenario 6 to 7Water quality change Imporved WQIUnchanged WQIDegradation in WQI151  5). There was slight degradation in WQI in Node J-6375 after placing the fourth booster station. It should be noted that Node J-6375 is the fourth booster station and therefore water quality was degraded slightly because of the superimposing effect of the nearby water reservoir. There was no change in water quality after adding the fifth booster station in Scenario 7.  LCC costs were estimated for each scenario. There was no significant difference between the LCC in Scenario 1 and Scenario 5 through 0.7mg/L and 0.6m/L were the chlorine dosages in Scenario 1 and Scenario 5 respectively. The negligible chemical cost compared to maintenance and capital cost was the main reason behind this result. In both of the scenarios the LCC was around 0.6x 105 $/10yrs. There was significant addition in the LCC after adding each booster station.  Figure 7-16 represents variations in CRP, and NCRP for Scenario 1, Scenario 2, Scenario 3, Scenario 4, Scenario 5, Scenario 6, and Scenario 7. Scenario 1 shows highest values for both CRP (5.7x10-5) and NCRP (0.22) in Node J-6375. In other scenarios CRP (3.4x10-5) and NCRP (0.13) were lower and almost the same for Node J-6375. Negligible or very low risk potentials were noticed in observed nodes for Scenario 2. Risk potentials increased for Node J1838 from Scenario 2 to 4 however remained unchanged after Scenario 5. Risk potentials were increase in Scenario 3 but remained unchanged at Scenario 4. However, there was again increase in risk potentials for Node J1882 after adapting Scenario 5, but remained unchanged for Scenario 6 to 7. There were negligible risk potentials for Node J1386 in Scenario 2, however increased in Scenario 3 and 4. The risk potentials for this node remained unchanged for Scenario 5, 6, and 7. There was no difference in risk potentials for all observed nodes between Scenario 6 and 7. In future, decision makers may provide valuable conclusions if there were regulatory thresholds available for risk potentials. Finally, after considering these results, three to four booster stations are recommended with 0.6mg/L of dosage. More concise methodology is suggested for both location and optimal dosage selection for booster chlorination.  kcl and F values are also  recommended to be estimated from direct samples. Regression analysis can be done with sampling data in this respect. 152   Figure 7-15: Trade-offs between water quality index and life cycle cost (LCC) for the case study  3040506070809010000.20.40.60.811.2WQI Life cycle cost (LCC) x 100000 LCC ($/10yr)WQI (Node J1838)WQI (Node J1882)WQI (Node J1386)WQI (Node J-6375)153    Figure 7-16: Case study results, a) CRP, and b) NCRP variations  0E+001E-052E-053E-054E-055E-056E-05Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6 Scenario 7Cancer risk potential (CRP) CRP (Node J1838) CRP (Node J1882)CRP (Node J1386) CRP (Node J-6375)a) 0E+005E-021E-012E-012E-013E-01Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6 Scenario 7Non-cancer risk potential (NCRP) NCRP (Node J1838) NCRP (Node J1882)b) 154  7.5 Summary Detectable FRC concentrations can ensure microbiological water quality in DNs and can be effectively maintained with chlorine booster stations. Booster stations can also provide optimal levels of FRC with reduced DBPs, however require careful placement. This study proposed an optimization scheme to locate booster stations in small to medium sized DNs. A greedy algorithm was used to locate booster stations with an assumption of fixed booster dosage. A modified version of the popular CCME WQI was adapted as an objective function by considering regulatory threshold values for FRC and TTHM. The index was further normalized with respect to demand to be used as the objective function for optimization. Later, quadratic optimization was adapted to predict THM species and CRP, NCRP were calculated. Additionally, LCC for each booster solution was estimated.  Finally, total number of required booster stations was selected using trade-off analysis between WQI, LCC, CRP and NCRP. A small DN with 28 nodes was used to explain the method elaborately. Later, a case study was performed on a small part of the city of Kelowna DN. The analysis recommended one and three to four booster stations for the small network and the part of the city of Kelowna DN respectively.  In both cases, water quality was improved after placing the proposed booster stations. A more modified algorithm is required to select dosage and location simultaneously.       155  Chapter 8 : MINIMIZING THE IMPACTS OF CONTAMINANT INTRUSION A version of this chapter has been combined with Chapter 5 and under preparation for a possible publication to Water Research Journal with the title “Optimizing booster chlorination to minimize the impacts of contaminant intrusion in a small water distribution network” by Islam, N., Rodriguez, M. J., Farahat, A., and Sadiq, R.  (Islam et al. 2015d). 8.1 Background We have previously discussed about the regulations suggesting minimum thresholds for residual disinfectants.  For instance, 0.2 mg/L of FRC is recommended by the SWTR (USEPA 2004). It can inactivate bacteria and ensures acceptable water quality; however, higher levels of chlorination may generate unwanted DBPs, e.g., TTHM and HAAs. Some of the DBPs in drinking water may be linked to potential detrimental health issues such as reproductive effects, bladder cancer and child development effects (Nieuwenhuijsen et al. 2000; Richardson 2003). The highest level of 4 mg/L as FRC is suggested in Stage 1 DBP Rule (USEPA 2004). Regulations including the Interim Enhanced Surface Water Treatment rule recommends trade-off between DBPs and microbiological contamination (USEPA 2004).  However, a higher level of FRC can also be linked to T&O issues. This is much more profound in First Nation Communities, where T&O issues related to chlorine are not acceptable by the major population (Health Canada 2011). We have also discussed about the benefits of  using booster chlorination such as it can provide adequate levels of FRC, reduce  total chlorine usage, and cost for chemicals (Boccelli et al. 2003; Carrico and Singer 2009). However, placement of a new booster station will add on to the capital cost for installation and maintenance. These conflicting objectives warrant an in-depth study for the optimization of booster chlorination in distribution networks.  Moreover, there have been numerous studies on optimization of booster chlorination (Behzaidan et al. 2012; Boccelli et al. 1998; Carrico and Singer 2009; Cozzolino et al. 2005; Gibbs et al. 2010b; Harmant et al. 2004; Hongxiang et al. 2010; Kang and Lansey 2010b; Kansal and Arora 2004; Lansey et al. 2007; Meng et al. 2011; Ostfeld and Salomons 2006a; Parks et al. 2009; Prasad et al. 2004; Propato and Uber 2004a; Rico-Ramirez et al. 2007), but  very few studies have been reported on contaminant intrusion related risks (Parks et al. 2009) and costs 156  due to chlorination (Farmani et al. 2006).  This chapter proposes a new methodology to address the above mentioned issues in a DN.  Salient steps of this methodology are: 1. identifying potential intrusion points,  2. predicting contamination propagation by assuming intrusions in certain nodes, and 3. optimizing location and dosages for booster chlorination using LCC, microbiological risk and chemical risk trade-offs.   Chapter 5 has already described the first step by proposing a risk based model to identify potential intrusion points. This chapter provides details on the remaining methodology including microbiological intrusion and DBPs modelling, and multi-objective optimization specific to intrusions. The methodology has been implemented with a case study performed on the City of Kelowna water main systems (Province of British Columbia, Canada).  8.2 Methodology The proposed methodology consists of three components: Component 1 -identifying nodes with highest intrusion risk potential (IRP), Component 2- modelling contaminant intrusions and DBPs in DN, and Component 3- optimization for booster chlorination (Figure 8-1).  Figure 8-1: Proposed methodology with three components for analysis 157  Since the Chapter 5 has  already described the model used in Component 1,  only Component 2 and 3  have been discussed in this this chapter. 8.2.1 Intrusion and DBPs modelling EPANET-MSX programmers’ toolkit was used to model intrusions and DBP occurrence simultaneously, which was further connected to Component 3 for optimization.  Bacteria, e.g., Salmonella, Shigella, Vibrio cholerae, E. coli O157:H7, and Legionella pneumophila along with protozoa such as Cryptosporidium and Giardia can be some common microbes to predict for contaminant intrusions. Cryptosporidium is resistant to chlorination, therefore, E. coli O157:H7 is selected in this study to assess the microbiological water quality after adding booster chlorination.  Microbiological intrusion and propagation in DN requires defining: 1) intrusion condition, 2) amount of microbial pollutant, and 3) reaction kinetics. Proposed point scoring method determines the intrusion condition.   A risk based method to select the nodes for intrusions has earlier been proposed in Chapter 5.  The amount of pollutant can be calculated using an equation  proposed in Besner et al. (2002). Assumption can be made for pollutant concentration or mass rate from literature such as 5000 CFU/L  as assumed in this study for E. coli O157:H7 (USEPA 2005a).  Chick –Watson first order kinetics has been  adapted  as the intrusion kinetics from Betanzo et al. (2008): pdPk PCdt            [8-1]  where P = E. coli O157:H7 concentration  (CFU/L); C = FRC concentration (mg/L); kP = pathogen kinetic decay constant (assumed from Betanzo et al. (2008)).  For chlorine concentrations, first-order chlorine decay model (Courtis et al. 2009; Powell et al. 2000b), was adapted (Equation [3-2]). Chlorine decay co-efficient can be either bulk, or wall decay or both. However, this study has considered only bulk decay.   158   Previous Equation [3-5] has been adapted to model TTHM.  Methodology stated using Equation [7-6] to [7-12] has been adapted once again to model THM species. THM species simulation is required to estimate the related chemical risks. 8.2.2 Multi-objective optimization for booster chlorination E. coli O157:H7 and THM species concentrations have been converted to microbiological and chemical risk potentials, respectively. Optimal dosage and locations have been selected after considering these risk potentials and expected LCCs for booster chlorination.  Microbial risk potential (MRP) was estimated using quantitative microbial risk assessment (QMRA) framework described in Chapter 2. E. coli O157:H7concentrations have been obtained for 72 hours of simulations and the maximum concentration from 25th to 48th hr simulation was considered for risk potential estimation. 2L/day has been assumed by Benser et al. (2012) as the consumed water by individual.  Finally, the MRP has been estimated using the beta-poison model described in QMRA framework (DuPont et al. 1971): 1502 11 [1 ]MRP dN             [8-2] where MRP=  probability of infection per day; d= dose (E. coli O157:H7 concentration multiplied by the volume of ingestion (2L/d)); N50= the experimental dose at which 50% of the population is expected to be affected (2.11x 106 (DuPont et al. 1971)); α= Beta-poison distribution factors (1.55x 10-1 (DuPont et al. 1971)).  THM species concentrations have been converted to chemical risk potential (ChRP) using reference doses (RfD) collected from the Integrated Risk Information System (IRIS) database (USEPA 2014) (Microbial risk assessment Microbial risk assessment refers to the steps to understand, reduce, and prevent risks caused by hazardous microbes, whether natural or anthropogenic, intentional or unintended. Though microbial risk assessment evolves from chemical risk assessment, there are fundamental differences between them; e. g. the exposure of chemical is mainly chronic while microbial exposure is acute. Moreover, quantification of microbial risk is harder compared to the chemical 159  one as it depends on many factors such as microbial growth, inactivation by disinfection and treatments. Quantitative microbial risk assessment (QMRA) is a popular way to estimate the risk of infection by specific microbes (Austin et al. 2012; Besner et al. 2011). This approach is performed by the office of water to support SWTR and the LTESWTR (USEPA 2010b). The standard QMRA consists of four basic steps (Ahmed et al. 2010):  (1) Hazard Identification: Hazard identification is to define the potential microbes (e.g., protozoan, virus, and bacteria) which are harmful to the system and also investigates the effect of the microbes. Most drinking water related QMRA address enteric viruses (Hepatitis A, Norwalk virus, rotavirus), parasites (Ascaris, Eimeria, Cyclospora, Toxoplasma), protozoa (Cryptosporidium and Giardia) and bacteria (E. coli, Shigella, Salmonella, Vibrio cholera) (Haas et al. 1999). Unlike microbial water quality assessment, E. coli O157:H7 can be considered for simplicity to investigate the effect of chlorination with QMRA in DN. Table 2-3). Only RfDs have been considered to account acute toxicity effects of DBPs.  Slope factor (SF) has not been considered for chronic effects as it might not be comparable with acute microbiological effects. The exposure route can be ingestion, dermal adsorption, and inhalation (Hamidin et al. 2008).  However, only oral exposure has been considered.   Individual THM species exposure has been estimated as follows (Hamidin et al. 2008): nnTHMTHM IR CFExpBW          [8-3] where Exp = exposure dose (mg/kg/day); THMn= TTHM species concentration (mg/L); IR = ingestion rate (L/day) ; CF = conversion factor ; and BW = Body weight (kg). IR=2 L/day,  and BW=70 kg (assumed) (Hamidin et al. 2008). Finally, the ChRP has been estimated as follows: 160  30nnnTHMTTHMn THMExpChRPRfD          [8-4] where  RfD information was collected from the USEPA (2014) (Microbial risk assessment Microbial risk assessment refers to the steps to understand, reduce, and prevent risks caused by hazardous microbes, whether natural or anthropogenic, intentional or unintended. Though microbial risk assessment evolves from chemical risk assessment, there are fundamental differences between them; e. g. the exposure of chemical is mainly chronic while microbial exposure is acute. Moreover, quantification of microbial risk is harder compared to the chemical one as it depends on many factors such as microbial growth, inactivation by disinfection and treatments. Quantitative microbial risk assessment (QMRA) is a popular way to estimate the risk of infection by specific microbes (Austin et al. 2012; Besner et al. 2011). This approach is performed by the office of water to support SWTR and the LTESWTR (USEPA 2010b). The standard QMRA consists of four basic steps (Ahmed et al. 2010):  (1) Hazard Identification: Hazard identification is to define the potential microbes (e.g., protozoan, virus, and bacteria) which are harmful to the system and also investigates the effect of the microbes. Most drinking water related QMRA address enteric viruses (Hepatitis A, Norwalk virus, rotavirus), parasites (Ascaris, Eimeria, Cyclospora, Toxoplasma), protozoa (Cryptosporidium and Giardia) and bacteria (E. coli, Shigella, Salmonella, Vibrio cholera) (Haas et al. 1999). Unlike microbial water quality assessment, E. coli O157:H7 can be considered for simplicity to investigate the effect of chlorination with QMRA in DN. Table 2-3). Life-cycle cost (LCC) has been estimated for booster chlorination. The detailed methodology to estimate LCC for hypo-chlorination has been described in Chapter 7. Equation [7-17] has been adapted to estimate the cost with similar assumptions stated in Chapter 7. There are many solvers for obtaining an optimal solution, however, soft computing techniques such as Genetic algorithm (GA) finds near-optimal solutions quickly. Multi objective genetic 161  algorithm (MOGA) has been incorporated in this application of booster optimization, with global optimization toolbox in Matlab. A Matlab coding has been created in this respect using the following objective functions: Minimize f1= 1( ( ))( )tNN NNNMax ChRP QMax Q         [8-5] Minimize f2= 1( ( ))( )tNN NNNMax MRP QMax Q         [8-6] and Minimize f3= 1NbNbNbLCC           [8-7] where N= Order of nodes in the DN; t= time step for hydraulic and water quality simulation; QN= Base water demand for Node N; tNChRP = Chemical risk potential for Node N and time step t; tNMRP = Microbiological risk potential for Node N and time step t; Nb= Order of node selected as booster station; LCCNb= Estimated life -cycle cost for specific booster dosage and node Nb selection. These three-objective models are subject to the following constrains: 0≤ Nb ≤N           [8-8] 0.2mg/L ≤ DO ≤1mg/L         [8-9] 0.05mg/L≤ tNC   ≤ 1mg/L         [8-10] where D= Booster dosage (mg/L); C= FRC concentration at node N and at time step t (hr). Equation [8-9] presents the highest and the lowest chlorine dosages defined for optimization. 0.2mg/L has been selected as the lowest chlorine dosage as the USEPA suggested minimum 0.2mg/L of chlorine level in DNs (USEPA 2004). USEPA also suggested 4mg/L as the highest chlorine level (USEPA 2004) in DNs, which may create chlorine T&O complaints.  Previously, Australia suggested 0.6mg/L as the highest chlorine level (Health Canada 2009). However, we 162  are proposing a highest dosage of 1mg/L to manage issues like chlorine T&O complaints and higher DBPs.     8.3 Case Study The optimization scheme is specified for small DNs, therefore a small sector from the city of Kelowna DN was selected for the case study (Figure 8-2). The figure represents the Thiessen polygons created to analyze IRP inside the DN.  Figure 8-2b shows the small part of the DN selected for optimization. The part contained one reservoir, three tanks, four pumps, 340 water mains, and 298 nodes (Figure 3-6). 0.6 to 1 mg/L chlorine concentration is applied as the booster dosage (information collected from the City). Therefore, 0.05 to 1 mg/L booster dosages were assumed for analysis. 0.033 (1/hr) and 0.651 were assumed as kcl and F from Elshorbagy (2000). 0.7 mg/L chlorine concentration and 342 m head were assumed for the reservoir (information collected by the city). 660 L/hr.mg was assumed as kp from Betanzo et al. (2008). Previously, a case study  as described in Chapter 5,  represented intrusion risk results for 2093 nodes,  out of which only 298 nodes  were considered as a continuation for optimization (Figure 8-3).  A closer view of the intrusion risk potential rankings is presented for this sector in Figure 8-3.  Estimated IRP showed higher value for some of the polygons. Especially, out of 298 nodes, some nodes were ranked between Class 1 and Class 2 ranges.  Finally, Figure 8-3 presents top ten selected intrusion points to model contaminant intrusion. 163     Figure 8-2: City of Kelowna – a) water main system with Thiessen polygons, and b) a small part for performing optimizationa) b) 164    Figure 8-3: Intrusion risk potential (IRP) ranks on a part of the City of Kelowna DN A sustained event was assumed for optimization.  Ideally, intrusions should take place when the pressure is negative or low. An initial hydraulic analysis was performed; however previously selected ten DN nodes did not have a time step with negative pressure. Therefore, continuous E. coli O157:H7 intrusions were assumed to enter through the top ten selected nodes (Figure 8-3). E. coli O157:H7 concentration of 5000CFU/L was considered from USEPA (2005b). Finally, E. coli O157:H7, FRC, TTHM, and THM species were predicted using Equation [8-1], [3-5] and the quadratic optimization using Equation [7-6] to [7-12].  EPANET multi species extension was used in this respect using kinetic simulation and quadratic optimization code created in Matlab. E. coli O157:H7 and THM species were converted to MRP and ChRP respectively. Finally, MOGA performed the optimization using Equation [8-5] to [8-10]. 165  8.3.1 Selecting the “best” solution The optimization generated 70 solutions for minimizing ChRP, MRP and LCCs. Figure 8-4, Figure 8-5 and Figure 8-6 present Pareto fronts for ChRP vs MRP, ChRP vs LCC, and MRP vs LCC respectively.  Figure 8-4: Pareto fronts for average ChRP vs average MRP Here, CRP and MRP were normalized with base demand using Equation [8-5], and [8-6], respectively. Following this, average MRP and ChRP for 298 nodes were estimated and represented in these Pareto fronts (Figure 8-4). After considering these Pareto fronts (ChRP vs MRP, ChRP vs LCC, and MRP vs LCC), three solutions were shortlisted from the 70 solutions. Two other solutions were selected for minimum ChRP and MRP. 0 0.005 0.01 0.015 0.02 0.025 0.03 0.0350.0530.0540.0550.0560.0570.0580.059Average ChRPAverage MRPOptimal solution for min average ChRP vs MRP 166   Figure 8-5: Pareto fronts for average ChRP vs life cycle cost (LCC) There was no separate solution for minimum LCC as one of the Pareto solution was representing minimum LCC. Finally, Figure 8-7 compares these five solutions, where dosage, node ID, graphical locations are also presented for each solution. Solution 1 is focused on minimizing MRP and ChRP selected from Figure 8-4. Figure 8-4 presents an optimal area from which Solution 1 is selected based on the highest lateral distance between boosters. The booster stations are uniformly distributed with dosage ranges from 0.06-0.96mg/L. Solution 2 is representing a solution for minimizing ChRP and LCC, where booster stations were less uniformly distributed than Solution 1. The solution may create more DBPs in long run because of the superposition effects among the closely placed booster stations. Solution 4 has selected water reservoir as one of the four booster stations, however chlorinated water will be always supplied from water reservoir. Solution 5 was focused on minimum MRP; therefore booster stations were located close to the intrusion points. Figure 8-7 also represents the current booster practice adapted by the city, which shows some similarities with Solution 3. Typically, detectable FRC and customer complaints are prime concerns for the current practice adapted by the city. Therefore, current 0 0.005 0.01 0.015 0.02 0.025 0.03 0.0350.07780.07790.0780.07810.07820.07830.07840.07850.0786Average ChRP 10 yr LCCOptimal solution for min average ChRP vs LCC, min LCC 167  practice is similarity with Solution 3 where minimization of MRP and LCC is the prime concern. Figure 8-6: Pareto fronts for average MRP vs life cycle cost (LCC) Comparisons among these five solutions can be observed in Figure 8-8. The figure also compares them with the initial condition when there was no booster station except the water reservoir. Solution 1 shows improvement from the initial condition by decreasing both the average MRP at a 10years LCC of 0.0779x1000000$. The average ChRP increases with addition of booster stations in Solution 1, however, MRP decreased from 0.041 to 0.038.  0.037 0.0375 0.038 0.0385 0.039 0.0395 0.04 0.0405 0.0410.07780.07790.0780.07810.07820.07830.07840.07850.0786Average MRP 10 yr LCCOptimal solution for min MRP vs LCC 168   Figure 8-7: Proposed and present booster stations in the City of Kelowna 169     Figure 8-8: Average ChRP, average MRP and LCC comparisons for five selected optimal solutions0.006 0.007 0.006 0.016 0.004 0.013 0.041 0.038 0.039 0.037 0.040 0.037 0.0000 0.07813 0.07794 0.07796 0.07812 0.07825 0.0000.0100.0200.0300.0400.0500.0600.0700.0800.090Initial Condition Solution 1 Solution 2 Solution 3 Solution 4 Solution 5Average CRPAverage MRPLCCx 1000000$170  Solution 2 shows degradation in microbiological water quality but slight improvement in chemical water quality. Solutions 3 and 5 significantly decrease the MRP at the cost of higher ChRP. Solution 4 shows the lowest ChRP, but higher MRP. However, Solution 1 was selected as it suggests both lesser ChRP and MRP with uniformly located booster stations. Utility planers also emphasize on improving more microbiological water quality and Solution 1 decreases more MRP compared to Solution 2.  8.3.2 Water quality improvement Solution 1, called proposed booster stations, was applied on the selected sector of the DN and the results were compared with simulation results assuming: 1. Initial condition- when there was no booster chlorination except chlorinated water supply from reservoir, and 2. Present booster solution- when current booster stations were placed on the EPANET model to observe water quality changes. Three booster stations (Figure 8-7), with a dosage of 0.7 mg/L each, were assumed resembling City of Kelowna’s current booster practice.  Figure 8-9 represents the result comparisons for average ChRP, average MRP and LCC. Average ChRP was increased slight after adapting Solution 1. On the other hand, average MRP was significantly decreased after adapting Solution 1.Uniformly distributed booster stations with optimal dosage (Solution 1) improved the microbiological water quality. However, the LCC was higher for Solution 1 (Four booster stations) compared to the present booster stations (three booster stations). The additional booster station was responsible to increase this cost. To have a closer view, water quality change is presented for E. coli O157:H7, FRC, and TTHM concentrations in two selected nodes, Node ET67 and J-6355, where:  Node ET 67 has 2 L/min demand.  It is closely located to one of the contamination intrusion points. It is also closely located to one of the proposed booster stations.  J-6355 has a demand of 12 L/min and is located close to the reservoir and one of the booster stations.  Figure 8-10 represents water quality changes in Node 67. Initially E. coli O157:H7 was detected even at 39th hr of analysis (Figure 8-10a). The present booster stations were unable to change the 171  E. coli O157:H7 concentrations at this node.  E. coli O157:H7 was inactivated at 24th hr of simulation after adapting proposed booster stations (Figure 8-10a).  Detectable FRC was visible from 40th hr for the initial conditions or with present booster stations (Figure 8-10b). The proposed booster stations ensured safety at the DN even from the 24th hr of simulation with detectable FRC concentration. The FRC concentration was increased up to 0.065mg/L by the end of 48th hr of simulation. However, TTHM concentration was increased up to 31µg/L after adapting proposed booster stations (Figure 8-10c), however, still under control as this is lesser than the TTHM regulatory threshold value of 80µg/L. It can be concluded that the proposed booster stations were quite effective in controlling both microbiological and chemical water quality at node ET67. For node J-6355, present booster stations were less effective in inactivating E. coli O157:H7 compared to the proposed booster stations (Figure 8-11a). Initially, E. coli O157:H7 was active even at the end of the 48th hr simulation. But, for present booster solution, E. coli O157:H7 was mainly inactivate but with a small peak at 26th hr time step. The proposed booster stations inactivated E. coli O157:H7 from the very beginning and continued up to 48hr simulation. There was no detectable FRC until simulation time 37hr for initial conditions (Figure 8-11b). There was detectable FRC after adapting both proposed (max. 0.30 mg/L) and present booster solution (max. 0.6 mg/L). FRC concentration was 0.6 mg/L for present booster solution, which is the Australian highest threshold (0.6mg/L) value for FRC (Health Canada 2009) and hence may be unacceptable to many consumers in terms of taste & odors. Maximum TTHM was approximate 40µg/L for proposed booster stations, but was 50µg/L for present booster solution. Therefore, the proposed scheme was more effective in controlling chemical and microbiological water quality for node J-6355, compared to the present solution. Decision-makers can take their decision based on selected nodes or by analyzing a combined effect. 172    Figure 8-9: Average ChRP, average MRP and LCC comparisons among initial conditions, present booster stations and proposed solution after optimization 0.0000.0100.0200.0300.0400.0500.0600.0700.080Average CRPAverage MRPLCCx1000000$0.006 0.041 0.000 0.006 0.040 0.059 0.007 0.038 0.078 Initial ConditionPresent booster solutionsProposed solutions after opt.Average ChRP 173       Figure 8-10: Proposed booster effect on Node ET67 for, a) E.coli O157:H7, b) Free residual chlorine, and c) TTHM concentrations  0500100015002000250030003500400024 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48E. coli O157:H7 (CFU/L) Time step (hr) Initial E. coli O157:H7E. coli O157:H7 with presentbooster solutionE. coli O157:H7 afterproposed booster stationsa) 00.010.020.030.040.050.060.0724 26 28 30 32 34 36 38 40 42 44 46 48FRC (mg/L) Time step (hr) Initial FRCFRC with present booster solutionFRC after proposed booster stationsb) 0510152025303524 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48TTHM (µg/L) Time step (hr) Initial TTHMTTHM with present booster solutionTTHM after proposed booster stationsc) 174       Figure 8-11: Proposed booster effect on Node J-6355 for, a) E. coli O157:H7, b) Free residual chlorine, and c) TTHM concentrations010002000300040005000600024 26 28 30 32 34 36 38 40 42 44 46 48E. coli O157:H7 (CFU/L) Time step (hr) Initial E. coli O157:H7E. coli O157:H7 with presentbooster solutionE. coli O157:H7 after proposedbooster stationsa) -0.100.10.20.30.40.50.60.724 26 28 30 32 34 36 38 40 42 44 46 48FRC (mg/L) Time step (hr) Initial FRCFRC with present booster solutionFRC after proposed booster stationsb) 010203040506024252627282930313233343536373839404142434445464748TTHM (µg/L) Time step (hr) Initial TTHMTTHM with present booster solutionTTHM after proposed boosterstationsc) 175  8.4 Summary This part of the research proposed a model for optimizing booster dosage and locations by minimizing chemical, microbiological risks and LCC for booster chlorination. A Matlab coding was created to model E. coli O156: H7, TTHM, and THM species using EPANET-MSX functions. Booster location, and dosages were selected using optimization with Multi-Objective Genetic Algorithm (MOGA). Minimizing microbiological, chemical risk potentials and LCCs were the objective functions. Traditional QMRA and chemical risk assessment methods were adapted for microbiological risk (MRP) and chemical risk potential (ChRP) estimation. A methodology was also presented to estimate 10years LCCs for booster chlorination. A tool was developed in Matlab integrating these steps. Previously, a methodology was implemented on the City of Kelowna (British Columbia, Canada) DN for IRP estimation (Chapter 5); however a small part was selected for optimization in this chapter. The optimization selected 70 solutions for booster chlorination out of which five solutions were short listed after observing Pareto fronts. Finally, after detailed observations a solution with four booster stations with dosage between 0.06-0.96mg/L was selected. The proposed booster stations were compared with the initial condition (no booster), and present booster solution (three booster with 0.7mg/L of dosage each) adapted by the city. The present booster practice is more on managing both microbiological and chemical water quality. The proposed scheme is expected to be beneficial for the small and first Nations communities, where better management is required for the booster stations to control any taste & odor complaints with chemical and microbiological risk minimization.    176  Chapter 9 : CONCLUSIONS AND RECOMMENDATIONS 9.1 Summary and Conclusions The main objective of this research was to guide decision-making for ensuring water quality safety in the DN using innovative optimization schemes/ algorithms for booster chlorination. Major strength and contributions of this research are:  Model applicability under both normal operating situation as well as under emergency situation, which will ensure not only water quality safety but also resilience at DNs.  Model flexibility to utilize minimum data, expertise, and therefore suitable for small to medium municipalities. To achieve this various strategies/ algorithms have been developed for evaluating microbiological, chemical (less DBPs) and aesthetic water quality (taste & odor complaints) in DNs. This research has developed indices specific for DBPs related regulatory violations and potential intrusions in DNs. Though these indices can integrate complex data series, but are simple and can be helpful in informed decision-making. There have been limited applications of water quality and risk based indices in DNs. The research is focused on small to medium sized DNs, where data availability and lack of technical and financial resources are major concerns. Moreover, the USEPA suggested 4 mg/L of maximum chlorine may cause taste & odor complaints, which are not acceptable in rural and First Nation communities, especially. Therefore, intelligent decision making is required to optimize booster chlorine dosage as well as their locations. The developed framework consists of six ‘novel’ models to ensure water quality safety in DNs using booster chlorination. Among these models, three of them are based on indices to assess the water quality, and remaining are schemes/ algorithms to optimize water quality in DNs.   The first model, non-compliance potential index (NCP index) evaluates DBP related regulatory violations. The model has a flexibility to be implemented for both sampled and modelled data. Sampled data will be used when required data are available and modelled values can be used under data limited conditions. The proposed index effectively predicts 177  how many parameter are violating or at the brink of violation. This can be very helpful for decision-makers in suggesting a suitable solution for managing water quality in DNs.   The second model is based on modified CCME WQI, specific for DBPs related regulatory violations. Chlorination in DNs ensures high microbiological water quality but may degrade chemical water quality with formation of DBPs. CCME WQI is generic to any regulatory violations but requires specific modification to evaluate chemical water quality. The methodology suggests guidelines to select specific parameters, estimates relative weights for parameters, and modifies the aggregation formulation to estimate the proposed index. Various conditions of Stage 1 and Stage 2 DBP regulations have been considered to develop this index.  The third model is related to intrusion risk potential (IRP) in a DN. The model utilized the popular definition of risk by accounting both, likelihood and consequences of intrusion. The methodology provides guidance to select variables and preparing them from a common municipal inventory data base. Variables representing water main characteristics, pollutant sources, pressure gradient, soil corrosivity, and consequences have been determined to estimate the IRP index.   The fourth model is an optimization scheme to predict booster dosage in DNs. The methodology used an index specific to chemical, microbiological and aesthetic water quality in DNs. In this index, only FRC has been utilized after considering various thresholds. The CCME WQI has been modified to account time series data of FRC. Optimization based on response surface optimization has been performed to ensure high water quality in a DN. The methodology is especially useful for small, rural and First nation’s communities where data availability is the main issue, and FRC is frequently the only parameter to manage water quality.   The fifth model is developed for locating booster stations using an index based optimization scheme. The index has been developed using FRC and TTHM time series data. A modification has been suggested in the CCME WQI. An algorithm has been proposed for maximizing water quality after normalization with nodal demands. The algorithm is a modified version of maximum covering location problem (MCLP) to locate booster stations. Later, the model has been used to conduct a trade-off analysis using estimated chemical risks and LCC for booster chlorination. The trade-off analysis 178  also suggests the total number of booster stations required. The methodology predicts the kinetics for FRC and TTHMs and individual species using quadratic optimization. The model needs experimental data of kinetic coefficients otherwise limited information is required to select the booster locations.   The sixth model is an optimization scheme to select both location and dosages for booster chlorination.  Previously developed risk based index has been implemented to locate potential intrusion points. Later, E. coli O156: H7 was assumed to enter in those intrusion points. Minimization of chemical and microbiological risk potentials and LCC for booster chlorination were defined as objective functions. These risk potentials were normalized with respect to nodal demands and incorporated as objective functions using MOGA. The scheme accounts health risks and can be applicable for decision making in emergency situation.  9.2 Limitations and Recommendations DN safety is essential as this is the closest point before consumption. Microbiological water quality can be compromised in DNs through treatment breakthrough, contaminant intrusion and biofilm growth. Maintaining detectable FRC in the DN is one of the best practices. Higher dosages are generally applied to ensure detectable FRC at extreme DN nodes. It may result in complaints from residents with regards to taste & odor and the formation of detrimental DBPs. Optimal levels of booster dosages in appropriate locations can minimize these effects. The research guides decision making based on consideration of prescribed regulatory thresholds, chemical and microbiological risk potentials, and LCC for booster chlorination. The decision making has been carried out by either evaluating the DN with indices or by proposing optimization schemes/ algorithms for booster chlorination. Due to the scale of the problem, the research has been conducted under certain limitations and assumptions.  9.2.1 Indices  The methodology for NCP index includes commonly available DBPs, and operational parameters. Basic indicators, e.g., water age and upstream chlorine dosage can be incorporated into the index for regression analysis. These will account for water residence time and initial chlorine concentrations. Larger sampling size (n>100) is also 179  suggested for an effective regression analysis. Authentic laboratory studies to predict site-specific kinetics is recommended. Municipalities are recommended to track this information for effective model prediction. The index was specific for DBP regulations; however a model predicting trade-off analysis between microbiological and chemical NCP indices can be a possibility.   The modified CCME WQI developed can evaluate chemical water quality. The index has used a weighting methodology to assign parameter weights. It has considered various perspectives concerning DBP rules. The weights can be assigned after soliciting numbers of experts. The kriging methodology can be more conservative for identifying problematic areas in a DN. It does not account for the impacts of reservoir, tanks, and pump operations which could possibly affect the water quality.  The intrusion risk potential model is a generic model to integrate common municipal inventory data. Various assumptions were made to make it suitable for the case study. However, direct variables such as ground water table, water main distance from the water table, pollutant source indicators such as distance from land fill and farm lands could be accounted based on available information. Similar approach can be used for them to assign index value in each DN node. The model predicts potential intrusion points in the DN. The node will be identified on the built-in EPANET model. The methodology cannot detect intrusion points within water mains rather identifies DN nodes.  None of the indices is a substitute for decision making based on individual parameter or variables. This has been noted to be a common argument among peers defending the need of index.  9.2.2 Optimization schemes and algorithms  The proposed optimization scheme using response surface optimization is capable of suggesting only dosage not locations. The boosters were located hypothetically on extreme areas where low FRC was suspected. The scheme was not dynamic in terms of integrating EPANET model simulations with other analysis in real time. The scheme was based on the temporal data for FRC only. This makes it suitable for small to medium sized municipalities where FRC could be only sampled. However, incorporating other 180  parameters such as TTHMs can be a future approach which was adopted in other two optimization schemes.  Optimization scheme to locate booster stations assumed a reasonable dosage of 0.6mg/L for each booster station. The optimization was defined as an unconstrained optimization using MCLP. A more powerful algorithm is required to assign constrains inside the optimization formulation. The simulation time was very long for solving a network of over 2000 nodes. Consequently, a smaller network with 300 nodes was selected for the case study. The model was, therefore, limited to a small sized DN and required longer simulation time for large networks.   Optimization scheme for selecting dosage and locations for booster chlorination used MOGA. The model is also linked to the third model where we have selected potential intrusion points in a DN. Microbiological contamination was assumed at certain nodes and then optimization was performed considering microbiological risk, chemical risk, and LCC. Equal importance has been given to microbiological risk, chemical risk and LCC. However, utilities can be more concerned about microbiological risk than chemical risk and LCC. Multi-objective functions can be proposed after assigning highest importance to the microbiological risk than to other objective functions. Application of this model is also limited to small to medium sized DNs.   Bulk decay coefficients were assumed in the last two models. These values should be site specific.   Simplistic kinetics was assumed for TTHM concentrations, which were based on chlorine consumption. More sophisticated model, depending on TOC, FRC, pH, temperature can be used in future based on site specific data.   Last two optimization models depend on nodal demands, therefore contain uncertainty because of the nodal demand variation.  Using booster chlorination may not be feasible for extremely small DN with simple hydraulics. These DNs can maintain water quality after adding an optimal dosage at the point of entry. Adding additional chlorination point may unnecessarily increase the costs for booster chlorination.   Same optimization models can be applied in larger DNs, however required stronger algorithms to cope up with the size of the DN. 181   The third and the sixth models were specific to microbiological intrusion. Optimization specific to other contaminants such as petroleum (hydrocarbons) can be a topic of future research.   The proposed models are expected to be beneficial in policy-making by suggesting required amount of booster dosage and numbers for a specific sized DN.   These optimization schemes can be incorporated in Model Predictive Control system and can help in selecting different dosages based on seasonal source water variations. 9.2.3 Decision support tool In this research, various aspects of water quality management in small to medium sized networks have been discussed.  A software tool which can provide single platform to perform these analyses has not been developed. However a conceptual framework that can integrate key components is given in Figure 9-1, which can provide a blue print for proposed decision support tool. Regulatory requirements, water quality monitoring, water quality assessment, water quality management using disinfection and other methodologies are interconnected. SWTR, Stage 1 & 2 D/DBPR, IESWTR, TCR, and SDWA are the principle regulations. This research proposed various index and risk based approaches to select dosage and locations for booster chlorination. Pressure maintenance, back flow detection, flushing, and effective valve and fire hydrants operations can be also incorporated with booster chlorination. This will collectively ensure water quality safety in DNs. This integrated software tool has a potential to make intelligent decision making such as determining flushing frequency, and valve operations and SCADA based decision making etc.  182     Figure 9-1: Conceptual framework for decision support tool     183  References  AGRA. (1998). Ground Water Assessment for Drinking Water for the City of Kelowna, BC. Ahmed, W., Vieritz, A., Goonetilleke, A., and Gardner, T. (2010). “Risk Assessment Determined Using Quantitative Microbial Australia, as Potable or Nonpotable Water, Rainwater in Southeast Queensland, Health Risk from the Use of Roof-Harvested.” Applied and Environmental Microbiology, 76(22), 7382–7391. Al-Jasser, A. O. (2007). “Chlorine decay in drinking-water transmission and distribution systems: pipe service age effect.” Water research, 41(2), 387–396. Al-zahrani, M. A., and Moied, K. (2003). “Optimizing water quality monitoring stations using genetic algorithms.” The Arabian Journal for Science and Engineering, 28(1B), 57–75. Aral, M. M., and Maslia, M. L. (1997). “Exposure assessment using simulation and GIS.” CSCE/ASCE Environmental Engineering Conf., Canadian Society of Civil Engineers, Edmonton, Alberta, Canada, 885– 892. Austin, R. G., Choi, C. Y., Balagopal, T. C., and Gurian, P. L. (2012). “Event-based microbial risk assessment and response analysis of cryptosporidium in municipal water distribution networks.” 12th Annual International Conference on Water Distribution Systems Analysis 2010, WDSA 2010, Tucson, AZ, United states, 459–470. AWWARF. (2003). Case study for a distribution system emergency response tool. Denver, Colorado. Baalousha, H. (2010). “Assessment of a groundwater quality monitoring network using vulnerability mapping and geostatistics: A case study from Heretaunga Plains, New Zealand.” Agricultural Water Management, 97(2), 240–246. Baranowski, T. M., and LeBoeuf, E. J. (2008). “Consequence management utilizing optimization.” J. Water Resour. Plann. Manage. Div., 134(4), 386–394. 184  Bayraktar, H., and Turalioglu, F. S. (2005). “A Kriging-based approach for locating a sampling site—in the assessment of air quality.” Stochastic Environmental Research and Risk Assessment, 19(4), 301–305. Behzaidan, K., Alimohammadnejad, M., Ardeshir, A., and Vasheghani, H. (2012). “A novel approach for water quality management in water distribution systems by multi-objective booster chlorination.” International Journal of Civil engineering, 10(1), 51–60. Berardi, L., Giustolisi, O., Kapelan, Z., and Savic, D. a. (2008). “Development of pipe deterioration models for water distribution systems using EPR.” Journal of Hydroinformatics, 10(2), 113. Berry, J., and Hart, W. E. (2008). User’s manual TEVA-SPOT toolkit 2.2. Cincinnati, OH, United states. Berry, J. W., Fleiscer, L., Hart, W. E., Phillips, C. A., and Watson, J. P. (2005). “Sensor Placement in Municipal Water Networks.” Journal of Water Resources Planning and Management, 131(3), 237– 243. Berry, J. W., Hart, W. E., Phillips, C. A., Uber, J. G., and Watson, J. P. (2006). “Sensor placement in municipal water networks with temporal integer programming models.” J. Water Resour. Plann. Manage. Div., 132(4), 218–224. Besner, M. C., Gauthier, V., Servais, P., and Camper, A. (2002). “Explaining the Occurrence of Coliforms in Distribution Systems.” AWWA, 94(8), 95–109. Besner, M. C., Messner, M., and Regli, S. (2012). “Pathogen intrusion in distribution systems : model to assess the potential health risks.” 12th Annual International Conference on Water Distribution Systems Analysis 2010, WDSA 2010, Tucson, AZ, United states, 484–493. Besner, M.-C., Prévost, M., and Regli, S. (2011). “Assessing the public health risk of microbial intrusion events in distribution systems: conceptual model, available data, and challenges.” Water research, 45(3), 961–979. 185  Betanzo, E. W., Hofmann, R., Hu, Z., Baribeau, H., and Alam, Z. (2008). “Modeling the Impact of Microbial Intrusion on Secondary Disinfection in a Drinking Water Distribution System.” Journal of Environmental Engineering., 134(4), 231–237. Boccelli, D. L., Tryby, M. E., Uber, J. G., Rossman, L. A., Zierolf, M. L., and Polycarpou, M. M. (1998). “Optimal scheduling of booster disinfection in water distribution systems.” Journal of water resources planning and management, 124(2), 99–111. Boccelli, D. L., Tryby, M. E., Uber, J. G., and Summers, R. S. (2003). “A reactive species model for chlorine decay and THM formation under rechlorination conditions.” Water research, 37(11), 2654–66. Bogdan, L., and Studzinski, J. (2007). “Modeling of water pressure distribution in water nets using the kriging algorithm.” 5th International Industrial Simulation Conference, Delft, Netherlands, 52–56. Boyacioglu, H. (2007). “Development of a water quality index based on a European classification scheme.” Water SA, 33(1), 101–106. Brown, D., Bridgeman, J., and West, J. R. (2011). “Predicting chlorine decay and THM formation in water supply systems.” Reviews in Environmental Science and Bio/Technology, 10(1), 79–99. Brown, R. M., McClelland, N. I., Deininger, R. A., and Tozer, R. G. (1970). “A water quality index: do we dare?” Water Sewage Works, 117, 339–343. Butera, I., Boano, F., Revelli, R., and Ridolfi, L. (2013). “Recovering the Release History of a Pollutant Intrusion into a Water Supply System through a Geostatistical Approach.” Journal of Water Resources Planning and Management, 139(4), 418–425. Bylaws. (2012). “City of Kelowna Bylaws.” <http://www.kelowna.ca/cm/page1329.aspx>. 186  Card, H., Goulding, R., Blanchard, C., Penney, B., Rumbolt, G., and Bradley, A. (2015). “Chlorination equipment selection guidelines.” Department of Environmental and Conservation, <http://www.env.gov.nl.ca/env/waterres/waste/chlorination.html>. Carr, R. D., Greenberg, H. J., Hart, W. E., Konjevod, G., Lauer, E., Lin, H., Morrison, T., Phillips, C., and A. (2006). “Robust optimization of contaminant sensor placement for community water systems.” Mathematical Programming, 107(1), 337– 356. Carrico, B., and Singer, P. C. (2009). “Impact of Booster Chlorination on Chlorine Decay and THM Production: Simulated Analysis.” Journal of Environmental Engineering, 135(10), 928. Carrico, B., and Singer, P. C. (2010). “Impact of Booster Chlorination on Chlorine Decay and THM Production : Simulated Analysis.” October, (October 2009), 2005–2010. CCME. (2001). Canadian water quality guidelines for the protection of aquatic life: Canadian water quality index 1.0 technical report. Winnipeg, Manitoba. Chang, Y. C., Randall, A. A., Choi, O. Y., Choi, D., Cho, H., and Kikuchi, S. (2010). “Influence of disinfection on bacterial regrowth in pilot distribution system.” Korean J. Chem. Eng., 27(6), 1860–1863. Chung, E., and Lee, K. (2009). “Prioritization of water management for sustainability using hydrologic simulation model and multicriteria decision making techniques.” Journal of Environmental Management, 90(3), 1502–1511. Clark, R. (1998). “Chlorine Demand and TTHM Formation Kinetics: A Second-Order Model.” Journal of Environmental Engineering, 124(1), 16–24. Clark, R. M., Chandrasekaran, L., and Buchberger, S. (2007). “Modeling the Propagation of Waterborne Disease in Water Distribution Systems: Results from a Case Study.” Water Distribution Systems Analysis Symposium 2006, American Society of Civil Engineers, Cincinnati, OH, United states, 71. 187  Clark, R. M., and Sivaganesan, M. (1998). “Predicting Chlorine Residuals and Formation of TTHMs in Drinking Water.” Journal of Environmental Engineering, 124(12), 1203–1210. Coelho, S. T., James, S., Sunna, N., Jaish, A. A., and Chatiia, J. (2003). “Controlling water quality in intermittent supply systems.” Water Science and Technology: Water Supply, 3(1-2), 119–125. Comboul, M., and Ghanem, R. (2013). “Value of Information in the Design of Resilient Water Distribution Sensor Networks.” Journal of Water Resources Planning and Management, 139(4), 449–455. Courtis, B. J., West, J. R., and Bridgeman, J. (2009). “Temporal and Spatial Variations in Bulk Chlorine Decay within a Water Supply System.” Journal of Environmental Engineering, 135(3), 147–152. Cozzolino, L., Pianese, D., and Pirozzi, F. (2005). “Control of DBPs in water distribution systems through optimal chlorine dosage and disinfection station allocation.” Desalination, 176(1-3), 113–125. Cristo, C. D., and Leopardi, A. (2007). “Uncertainty effects on pollution source location in water networks.” 8th Annual Water Distribution Systems Analysis Symposium 2006, Cincinnati, OH, United states, 125. Cristo, C. D., and Leopardi, A. (2008). “Pollution Source Identification of Accidental Contamination in Water Distribution Networks.” Journal of Water Resources Planning and Management, 134(2), 197–202. Dempster, A. (1967). “Upper and lower probabilities induced by a multi-valued mapping.” Ann. Stat., 28, 325–339. Deng, Y., Jiang, W., and Sadiq, R. (2011). “Modeling contaminant intrusion in water distribution networks: A new similarity-based DST method.” Expert Systems with Applications, Elsevier Ltd, 38(1), 571–578. 188  Dunnette, D. A. (1979). “A geographically viable water quality index used in Oregon.” J. Water Pollut. Control Fed., 51, 53–61. DuPont, H. L., Formal, S. B., Hornick, R. B., Snyder, M. J., Libonati, J. P., Sheahan, D. G., LaBrec, E. H., and Kalas, J. P. (1971). “Pathogenesis of Escherichia coli diarrhea.” The New England Journal of Medicine, 285(1), 1–9. DuPont, H. L., Hornick, R. B., Snyder, M. J., Libonati, J. P., Samuel, B. F., and Gangarosa, E. J. (1972). “Immunity in shigellosis. II. Protection induced by oral live vaccine or primary infection.” The Journal of Infectious Diseases, 125(1), 12–16. Ebacher, G., Besner, M., Prévost, M., and Allard, D. (2012). “Negative pressure events in water distribution systems: Public health risk assessment based on transient analysis outputs.” 12th Annual International Conference on Water Distribution Systems Analysis 2010, WDSA 2010, Tucson, AZ, United states, 471–483. Eliades, D., and Polycarpou, M. (2006). “Iterative deepening of Pareto solutions in water sensor networks.” Proc., 8th Annual Water Distribution System Analysis Symp., Cincinnati., 114. Elshorbagy, W. A. (2000). “Kinetics of THM Species in Finished Drinking Water.” Journal of Water Resources Planning and Management-asce, 126(1), 21–28. EnvironmentCanada. (2015). “Drinking Water Advisories in Canada.” <https://www.ec.gc.ca/indicateurs-indicators/default.asp?lang=en&n=2C75C17A-1> (Jan. 20, 2015). Espejo, L., Kretschmer, N., Oyarzún, J., Meza, F., Núñez, J., Maturana, H., Soto, G., Oyarzo, P., Garrido, M., Suckel, F., Amezaga, J., and Oyarzún, R. (2012). “Application of water quality indices and analysis of the surface water quality monitoring network in semiarid North-Central Chile.” Environmental monitoring and assessment, 184(9), 5571–5588. Farmani, R., Walters, G., and Savic, D. (2006). “Evolutionary multi-objective optimization of the design and operation of water distribution network: total cost vs. reliability vs. water quality.” Journal of Hydroinformatics, 9(1), 79. 189  Fisher, I., Kastl, G., and Sathasivan, A. (2011). “Evaluation of suitable chlorine bulk-decay models for water distribution systems.” Water research, Elsevier Ltd, 45(16), 4896–4908. Fisher, I., Kastl, G., Sathasivan, A., and Jegatheesan, V. (2011). “Suitability of Chlorine Bulk Decay Models for Planning and Management of Water Distribution Systems.” Critical Reviews in Environmental Science and Technology, 41(20), 1843–1882. Ford, T. E., and MacKenzie, W. R. (2000). How safe is our drinking water? Despite technologic advances, waterborne disease is still a threat, Postgrad. Med. 108(4). Francisque, A., Rodriguez, M. J., Sadiq, R., Miranda, L. F., and Proulx, F. (2009). “Prioritizing monitoring locations in a water distribution network: a fuzzy risk approach.” Journal of Water Supply: Research and Technology- AQUA, IWA, 58(7), 488–509. Francisque, A., Shahriar, A., Islam, N., Betrie, G., Siddiqui, R. B., Tesfamariam, S., and Sadiq, R. (2014). “A decision support tool for water mains renewal for small to medium sized utilities: a risk index approach.” Journal of Water Supply: Research and Technology—AQUA, 63(4), 281–302. Garreta, V., Monestiez, P., and Hoef ver, J. M. (2010). “Spatial modelling and prediction on river networks: Up model, down model or hybrid?” Environmetrics, 21(5), 439–456. Gibbs, M., Maier, H. R., and Dandy, G. C. (2010). “Comparison of genetic algorithm parameter setting methods for chlorine injection optimization.” Journal of Water Resources Planning and Management, 136(2), 288. Gibbs, M. S., Dandy, G. C., and Maier, H. R. (2010). “Calibration and Optimization of the Pumping and Disinfection of a Real Water Supply System.” Journal of Water Resources Planning and Management, 136(4), 493. Goldberg, D. E. (1989). Genetic algorithms in search, optimization, and machine learning. Addison- Wesley, New York. ISBN: 0201157675. 190  Golfinopoulos, S. K., and Arhonditsis, G. B. (2002). “Multiple regression models: a methodology for evaluating trihalomethane concentrations in drinking water from raw water characteristics.” Chemosphere, 47(9), 1007–1018. Gould, J. P., Fitchhorn, L. E., and Urheim, E. (1981). Formation of brominated trihalomethanes: Extent and kinetics. (R. L. Jolley, ed.), Ann Arbor Science, Ann Arbor, Mich., 297–310. Grayman, W. M., Speight, V. L., and Uber, J. G. (2007). “Use of distribution system modeling in designing microbial monitoring programs.” Restoring Our Natural Habitat - Proceedings of the 2007 World Environmental and Water Resources Congress, Tampa, FL, United states. Guan, J., Aral, M. M., Maslia, M. L., and Grayman, W. M. (2006). “Identification of contaminant sources in water distribution systems using simulation—Optimization method: Case study.” J. Water Resour. Plann. Manage., 132(4), 252–262. Haas, C. (2012). “Quantitative Microbial Risk Assessment (QMRA) Wiki.” <http://wiki.camra.msu.edu/index.php?title=Quantitative_Microbial_Risk_Assessment_(QMRA)_Wiki>. Haas, C. N., Rose, J. B., and Gerba, C. P. (Eds.). (1999). Quantitative Microbial Risk Assessment. John Wiley and Sons., New York, N.Y., USA. Hallam, N. B., Hua, F., West, J. R., Forster, C. F., and Simms, J. (2003). “Bulk Decay of Chlorine in Water Distribution Systems.” Journal of Water Resources Planning and Management, 129(1), 78–81. Hallam, N. B., West, J. R., Forster, C. F., Powell, J. C., and Spencer, I. (2002). “The decay of chlorine associated with the pipe wall in water distribution systems.” Water research, 36(14), 3479–3488. Hamidin, N., Yu, Q. J., and Connell, D. W. (2008). “Human health risk assessment of chlorinated disinfection by-products in drinking water using a probabilistic approach.” Water research, 42(13), 3263–3274. 191  Harmant, P., Nace, A., and Kiene, L. (2004). “An algorithm to optimize booster chlorination in water distribution network.” Joint Conference on Water Resource Engineering and Water Resources Planning and Management 2000: Building Partnerships, Minneapolis, MN, United states, 1–10. Hart, D. B., and McKenna, S. A. (2009). CANARY user’s manual. Version 4.1. Office of Research and Development, National Homeland Security Research Center, U.S. EPA, Cincinnati., EPA–600–R–08–040A. Hart, W., and Murray, R. (2010). “Review of Sensor Placement Strategies for Contamination Warning Systems in Drinking Water Distribution Systems.” J. Water Resour. Plann. Manage., 136(6), 611–619. Health Canada. (2007). “First Nations and Inuit Health: Drinking Water Quality.” <http://www.hc-sc.gc.ca/fnihspni/ promotion/water-eau>. Health Canada. (2009). “Guidelines for Canadian Drinking Water Quality.” Ottawa, Ontario. Health Canada. (2011). Perceptions of Drinking Water Quality in First Nations Communities and General Population. Ontario, 8. Health Canada. (2012). Guidelines for Canadian Drinking Water Quality Summary Table. Ottawa, Ontario Canada. Hebert, A., Forestier, D., Lenes, D., Benanou, D., Jacob, S., Arfi, C., Lambolez, L., and Levi, Y. (2010). “Innovative method for prioritizing emerging disinfection by-products (DBPs) in drinking water on the basis of their potential impact on public health.” Water Research, 44(10), 3147–3165. Helbling, D. E., and VanBriesen, J. M. (2009). “Propagation of Chlorine Demand Signals Induced by Microbial Contaminants in a Drinking Water Distribution System.” World Environmental and Water Resources Congress 2009: Great Rivers, American Society of Civil Engineers, Kansas City, MO, United states, 515–524. 192  Hernandez-Castro, S. (2007). “Two-stage stochastic approach to the optimal location of booster disinfection stations.” Industrial & engineering chemistry research, 46(19), 6284–6292. Hongxiang, W., Wenxian, G., Jianxin, X., and Hongmei, G. (2010). “A Hybrid PSO for Optimizing Locations of Booster Chlorination Stations in Water Distribution Systems.” 2010 International Conference on Intelligent Computation Technology and Automation, ICICTA 2010, Ieee, Changsha, China, 126–129. Hornick, R. B., Music, S. I., Wenzel, R., Cash, R. A., Libonati, J. P., Snyder, M. J., and Woodward, T. E. (1971). “The Broad Street Pump Revisited: Response of Volunteers to Ingested Cholera Vibrios.” Bulletin New York Academy of Medicine., 47(10), 1181–1191. HSPD-9. (2006). “Homeland security presidential directive/HSPD-9.” <www.whitehouse.gov/news/releases/2004/02/20040203-2.html>. Hu, Y., and Hubble, D. (2007). “Factors contributing to the failure of asbestos cement water mains.” Canadian Journal of Civil Engineering, 34(5), 608–621. Huang, J. J., and McBean, E. A. (2007). “Using Bayesian statistics to estimate the coefficients of a two- component second-order chlorine bulk decay model for a water distribution system.” Water research, 41(2), 287–294. Huang, J. J., Mcbean, E. A., and James, W. (2007). “Multi-Objective Optimization for Monitoring Sensor Placement in Water Distribution Systems.” 8th Annual Water Distribution Systems Analysis Symposium 2006, Cincinnati, OH, United states, 113. Huang, J. J., McBean, E. a., and Shen, H. (2009). “Data Mining as a Tool to Identify Contaminant Sources in Water Distribution Systems.” 10th Annual Water Distribution Systems Analysis Conference, WDSA 2008, American Society of Civil Engineers, Kruger National Park, South africa, 1091–1101. Hurley, T., Sadiq, R., and Mazumder, A. (2012). “Adaptation and evaluation of the Canadian Council of Ministers of the Environment Water Quality Index (CCME WQI) for use as an 193  effective tool to characterize drinking source water quality.” Water research, Elsevier Ltd, 46(11), 3544–3552. Islam, M. S., Sadiq, R., Rodriguez, M. J., Francisque, A., Najjaran, H., and Hoorfar, M. (2011). “Leakage Detection and Location in Water Distribution System using a Fuzzy-Based Methodology.” Urban Water Journal, 8(6), 351–365. Islam, N., Farahat, A., Al-zahrani, M. A. M., Rodriguez, M. J., and Sadiq, R. (2015a). “Contaminant intrusion in water distribution networks : review and proposal of an integrated model for decision making.” Environmental Reviews (Accepted). Islam, N., Sadiq, R., Rodriguez, M. J., and Legay, C. (2015b). “Assessing regulatory violations of disinfection by-products in water distribution networks using a non-compliance potential index.” Environmental Monitoring and Assessment (Submitted). Islam, N., Sadiq, R., Rodriguez, M. J., and Legay, C. (2015c). “Assessment of water quality in distribution networks through the lens of DBP Rules.” Water SA (Submitted). Islam, N., Rodriguez, M. J., Farahat, A., and Sadiq, R. (2015d). “Optimizing booster chlorination to minimize the impacts of contaminant intrusion in a small water distribution network.” Water Research (Submitted). Islam, N., Sadiq, R., and Rodriguez, M. J. (2015e). “Locating chlorine booster stations in small water distribution networks : a methodology using optimization and trade-off analysis.” Water Research (Submitted). Islam, N., Sadiq, R., and Rodriguez, M., J. (2012a). “Locating optimal locations for booster chlorination using a modified CCME water quality index.” 15th Canadian national Conference & 6th Policy Forum on Drinking Water, October 21-24, Kelowna, BC. Islam, N., Sadiq, R., and Rodriguez, M., J. (2012b). “Managing water quality in small distribution networks using booster chlorination: a water quality index approach.” 1st International Conference on Advances in Civil Engineering, Chittagong, Bangladesh. 194  Islam, N., Sadiq, R., and Rodriguez, M., J. (2013). “Optimizing booster chlorination in water distribution networks: a water quality index approach.” Environmental Monitoring and Assessment, 185(10), 8035–8050. Isovitsch, S., and Vanbriesen, J. (2008). “Sensor Placement and Optimization Criteria Dependencies in a Water Distribution System.” Journal of Water Resources Planning and Management, 134(2), 186–196. Jabari Kohpaei, A., and Sathasivan, A. (2011). “Chlorine decay prediction in bulk water using the parallel second order model: An analytical solution development.” Chemical Engineering Journal, Elsevier B.V., 171(1), 232–241. Jaeger, C., Hightower, M., and Torres, T. (2010). “Evolution of Sandia’s Risk Assessment Methodology for Water and Wastewater Utilities (RAM-W).” World Environmental and Water Resources Congress 2010: Challenges of Change - Proceedings of the World Environmental and Water Resources Congress 2010, Providence, RI, United states, 3804–3810. Jeffrey Yang, Y., Goodrich, J. a, Clark, R. M., and Li, S. Y. (2008). “Modeling and testing of reactive contaminant transport in drinking water pipes: chlorine response and implications for online contaminant detection.” Water research, 42(6-7), 1397–412. Joseph, S. A., Adams, B. J., and McCabe, B. (2010). “Methodology for Bayesian belief network development to facilitate compliance with water quality regulations.” Journal of Infrastructure Systems, 16(1), 58–65. Kang, D., and Lansey, K. (2010). “Real-Time Optimal Valve Operation and Booster Disinfection for Water Quality in Water Distribution Systems.” Journal of Water Resources Planning and Management, 136(4), 463. Kansal, M. L., and Arora, G. (2004). “Reliability Analysis of Booster Chlorination in Urban Water Distribution Networks.” Bridging the Gap: Meeting the World’s Water and 195  Environmental Resources Challenges - Proceedings of the World Water and Environmental Resources Congress 2001, Orlando, FL, United states, 1–10. Karim, M., Abbaszadegan, M., and LeChevallier, M. W. (2003). “Potential for pathogen intrusion.” J. Am.WaterWorks Assoc., 95(5), 134–146. Kessler, A., Ostfeld, A., and Sinai, G. (1998). “Detecting accidental contaminations in municipal water networks.” J. Water Resour. Plann. Manage. Div., 124(4), 192–198. Khan, F., Husain, T., and Lumb, A. (2003). “Water quality evaluation and trend analysis in selected watersheds of the Atlantic region of Canada.” Environmental Monitoring and assessment, Springer, 88(1), 221–248. Khanal, N., Buchberger, S. G., Mckenna, S. A., Clark, R. M., and Grayman, W. M. (2005). “Vulnerability assessment of water distribution system to chemical intrusions.” 2005 World Water and Environmental Resources Congress, Anchorage, AK, United states, 56. Kim, J. H., Tran, T. V. T., and Chung, G. (2012). “Optimization of water quality sensor locations in water distribution systems considering imperfect mixing.” 12th Annual International Conference on Water Distribution Systems Analysis 2010, WDSA 2010, Tucson, AZ, United states, 317–326. Kim, M., Choi, C. Y., and Gerba, C. P. (2008). “Source tracking of microbial intrusion in water systems using artificial neural networks.” Water research, 42(4-5), 1308–14. Kirmeyer, G. J., Friedman, M., Martel, K., Howie, D., LeChevallier, M., Abbaszadegan, M., Karim, M., Funk, J., and Harbour, J. (2001). Pathogen Intrusion into the Distribution System. American Water Works Association Research Foundation. Denver, CO. Kleiner, Y., and Rajani, B. (2002). “Forecasting Variations and Trends in Water-Main Breaks.” Journal of Infrastructure Systems, 8(4), 122. Kumar, A., Ostfeld, A., and Sinai, G. (1997). “Identification of monitoring stations in water distribution system.” Journal of Environmental Engineering, 123(8), 746–752. 196  Laird, C. L., Biegler, L. T., van Bloemen Waanders, B. G., and Bartlett A. R. (2005). “Contamination source characterization for water networks.” J. Water Resour. Plann. Manage., 131(2), 125–134. Lansey, K., Pasha, F., Pool, S., Elshorbagy, W., and Uber, J. (2007). “Locating Satellite Booster Disinfectant Stations.” Journal of Water Resources Planning and Management, 133(4), 372. Lawrence, W. W. (1976). Of acceptable risk. William Kaufmann Inc., Los Altos, CA. Lee, B. H., and Deininger, R. A. (1992). “Optimal Locations of Monitoring Stations in Water Distribution System.” Journal of Environmental Engineering, 118(1), 4–17. Legay, C., Rodriguez, M. J., Miranda-Moreno, L., Sérodes, J.-B., and Levallois, P. (2011). “Multi-level modelling of chlorination by-product presence in drinking water distribution systems for human exposure assessment purposes.” Environmental monitoring and assessment, 178(1-4), 507–24. Legay, C., Rodriguez, M. J., Sérodes, J. B., and Levallois, P. (2010). “Estimation of chlorination by-products presence in drinking water in epidemiological studies on adverse reproductive outcomes: a review.” The Science of the total environment, Elsevier B.V., 408(3), 456–472. Lin, Y.-C., and Yeh, H.-D. (2005). “Trihalomethane Species Forecast Using Optimization Methods: Genetic Algorithms and Simulated Annealing.” Journal of Computing in Civil Engineering, 19(3), 248–257. Lindley, T. R. (2001). “A framework to protect water distribution systems against potential intrusions.” University of Cincinnati, Cincinnati. Liou, S., Lo, S., and Wang, S. (2004). “A generalized water quality index for Taiwan.” Environmental Monitoring and Assessment, 96(1-3), 35–52. 197  Liu, W. Y., Yue, K., Su, J. Y., and Yao, Y. (2009). “Probabilistic representation and approximate inference of type-2 fuzzy events in Bayesian networks with interval probability parameters.” Expert Systems with Applications, 36(4), 8076–8083. Loureiro, D. (2002). “The influence of transients in characteristic hydraulic and water quality parameters.” MSc. Thesis. IST, Lisbon (in Portuguese). Lumb, A., Halliwell, D., and Sharma, T. (2006). “Application of CCME Water Quality Index to monitor water quality: A case study of the Mackenzie River basin, Canada.” Environmental Monitoring and Assessment, Springer, 113(1), 411–429. McCullough, N. B., and Elsele, C. W. (1951). “Experimental Human Salmonellosis: I. Pathogenicity of Strains of Salmonella meleagridis and Salmonella anatum Obtained from Spray-Dried Whole Egg.” The Journal of Infectious Diseases, 88(3), 278–289. Meng, F., Liu, S., Auckenthaler, P., Bai, L., Wang, H., and Wu, X. (2011). “Optimization of booster disinfection with backtracking algorithm.” 2011 International Conference on Consumer Electronics, Communications and Networks, CECNet 2011 - Proceedings, XianNing, China, 1458–1461. Messner, M. J., Chappell, C. L., and Okhuysen, P. C. (2001). “Risk assessment for Cryptosporidium: a hierarchical Bayesian analysis of human dose response data.” Water Research, 35(16), 3934–3940. MMCD. (2015). “Master Municipal Construction Documents Associations.” <https://www.mmcd.net/>. Muller, D., Edwards, M. l L., and Smith, D. W. (1983). “Changes in iron and transferrin levels and body temperature in experimental airborne legionellosis.” Journal of Infectious Diseases, 147, 302–307. Murphy, R. R., Curriero, F. C., and Ball, W. P. (2010). “Comparison of Spatial Interpolation Methods for Water Quality Evaluation in the Chesapeake Bay.” Journal of Environmental Engineering, 136(2), 160–171. 198  Nadkarni, S., and Shenoy, P. P. (2001). “A Bayesian network approach to making inferences in causal maps.” European Journal of Operational Research, 128(3), 479–498. NHMRC. (2004). Australia drinking water guidelines: Fact sheets—inorganic chemicals. National Health and Medical Research Council. Nieuwenhuijsen, M. J., Toledano, M. B., and Elliott, P. (2000). “Uptake of chlorination disinfection by-products; a review and a discussion of its implications for exposure assessment in epidemiological studies.” Journal of exposure analysis and environmental epidemiology, 10(6 Pt 1), 586–599. Nilsson, K. A., Buchberger, S. G., and Clark, R. M. (2004). “Simulating accidental exposures to deliberate intrusions.” Proceedings of the 2004 World Water and Environmetal Resources Congress: Critical Transitions in Water and Environmetal Resources Management, Salt Lake City, UT, United states, 4749–4758. Nilsson, K. A., Buchberger, S. G., and Clark, R. M. (2005). “Simulating Exposures to Deliberate Intrusions into Water Distribution Systems.” Journal of Water Resources Planning and Management, 131(3), 228–236. NRC. (1983). Risk assessment in the federal government: managing the process. NAS Press, Washington. NRC. (2003). Water quality in distribution systems. 1. Ostfeld, A., and Salomons, E. (2004). “Optimal layout of early warning detection stations for water distribution systems security.” J. Water Resour. Plann. Manage. Div., 130(5), 377–385. Ostfeld, A., and Salomons, E. (2005). “Optimal early warning monitoring system layout for water networks security: Inclusion of sensors sensitivities and response delays.” Civ. Eng. Environ. Syst., 22(3), 151–169. 199  Ostfeld, A., and Salomons, E. (2006). “Conjunctive optimal scheduling of pumping and booster chlorine injections in water distribution systems.” Engineering Optimization, Taylor & Francis, 38(3), 337–352. Ozdemir, O. N., and Ucak, A. (2002). “Simulation of Chlorine Decay in Drinking-Water Distribution Systems.” Journal of Environmental Engineering, 128(1), 31. Parks, I., Shannon, L., and VanBriesen, J. M. (2009). “Booster Disinfection for Response to Contamination in a Drinking Water Distribution System.” Journal of Water Resources Planning and Management, 135(6), 502–511. Pearl, J. (1988). Probabilistic reasoning in intelligent systems: Networks of plausible inference. Morgan Kaufmann, San Francisco. Perelman, L., and Ostfeld, A. (2011). “Topological clustering for water distribution systems analysis.” Environmental Modelling & Software, Elsevier Ltd, 26(7), 969–972. Perelman, L., Ostfeld, A., and Asce, F. (2013). “Bayesian Networks for Source Intrusion Detection.” Journal of Water Resources Planning and Management, 139(4), 426–432. Pesce, S. F., and Wunderlin, D. A. (2000). “Use of water quality indices to verify the impact of Cordoba city (Argentina) on Suquia River.” Water Research, 34(11), 2915–2926. Powell, J. C., Hallam, N. B., West, J. R., Forster, C. F., and Simms, J. (2000a). “Factors which control bulk chlorine decay rates.” Water Research, 34(1), 117–126. Powell, J. C., West, J. R., Hallam, N. B., Forster, C. F., and Simms, J. (2000b). “Performance of various kinetic models for chlorine decay.” Journal of water resources planning and management, 126(1), 13–20. Prasad, T. D., Walters, G., and Savic, D. (2004). “Booster disinfection of water supply networks: Multiobjective approach.” Journal of Water Resources Planning and Management, 130(5), 367–376. 200  Preis, A., and Ostfeld, A. (2006a). “Contamination source characterization in water systems: A hybrid model trees-linear programming scheme.” J. Water Resour. Plann. Manage, 132(4), 263–273. Preis, A., and Ostfeld, A. (2006b). “Optimal sensors layout for contamination source identification in water distribution systems.” Proc., 8th Annual Water Distribution Systems Analysis Symp., ASCE, Reston, Va. Preis, A., and Ostfeld, A. (2008). “Multiobjective Contaminant Sensor Network Design for Water Distribution Systems.” Journal of Water Resources Planning and Management, 134(4), 366–377. Preis, A., Perelman, L., and Ostfeld, A. (2008). “Uncertainty Quantification of Contamination Source Identification.” World Environmental and Water Resources Congress 2008, American Society of Civil Engineers, Honolulu, HI, United states, 1–6. Propato, M. (2006). “Contamination warning in water networks: general mixed-integer linear models for sensor location design.” Journal of water resources planning and management, 132(4), 225. Propato, M., and Uber, J. G. (2004a). “Linear Least-Squares Formulation for Operation of Booster Disinfection Systems.” Journal of Water Resources Planning and Management, 130(1), 53–62. Propato, M., and Uber, J. G. (2004b). “Vulnerability of water distribution systems to pathogen intrusion: how effective is a disinfectant residual?” Environmental science & technology, 38(13), 3713–3722. Ren, J., Jenkinson, I., Wang, J., Xu, D. L., and Yang, J. B. (2009). “An Offshore Risk Analysis Method Using Fuzzy Bayesian Network.” Journal of Offshore Mechanics and Arctic Engineering, 131(4), 1–12. 201  Rendtorff, R. C. (1954). “The experimental transmission of human intestinal protozoan parasites. II. Giardia lamblia cysts given in capsules.” American Journal of Epidemiology, 59(2), 209–220. Richardson, S. (2003). “Disinfection by-products and other emerging contaminants in drinking water.” TrAC Trends in Analytical Chemistry, 22(10), 666–684. Richardson, S. D., Plewa, M. J., Wagner, E. D., Schoeny, R., and Demarini, D. M. (2007). “Occurrence, genotoxicity, and carcinogenicity of regulated and emerging disinfection by-products in drinking water: a review and roadmap for research.” Mutation research, 636(1-3), 178–242. Richardson, S. D., and Ternes, T. A. (2011). “Water Analysis: Emerging Contaminants and Current Issues.” Anal. Chem, 83, 4614–4648. Richardson, S. D., Thruston, A. D., Collette, T. W., and Ireland, J. C. (1996). “Identification of TiO 2 / UV Disinfection Byproducts in Drinking Water.” Environ. Sci. Technol, 30(11), 3327–3334. Rico-Ramirez, V., Iglesias-Silva, G. A., De, L. C., Francisco, G., and Hernandez-Castro, S. (2007). “Two-Stage Stochastic Approach to the Optimal Location of Booster Disinfection Stations.” Industrial and Engineering Chemistry Research, 46(19), 6284–6292. Rodriguez, M. J., and Sérodes, J. (2005). “Laboratory-scale chlorination to estimate the levels of halogenated DBPs in full-scale distribution systems.” Environmental monitoring and assessment, 110(1-3), 323–340. Rossman, L. A. (2000). EPANET 2 Users Manual. U.S. Environmental Protection Agency, Cincinnati, OH, EPA/600/R–00/057. Saaty, T. (1977). “A scaling method for priorities in hierarchical structures.” Journal of mathematical psychology, 15, 57–68. 202  Saaty, T. L. (1988). “Multicriteria Decision-Making: The Analytic Hierarchy Process.” University of Pittsburgh, Pittsburgh, PA. Sadiq, R., Kleiner, Y., and Rajani, B. (2006). “Estimating risk of contaminant intrusion in water distribution networks using Dempster–Shafer theory of evidence.” Civil Engineering and Environmental Systems, 23(3), 129–141. Sadiq, R., Kleiner, Y., and Rajani, B. (2007). “Water quality failures in distribution networks-risk analysis using fuzzy logic and evidential reasoning.” Risk analysis : an official publication of the Society for Risk Analysis, 27(5), 1381–1394. Sadiq, R., Kleiner, Y., and Rajani, B. (2010). Proof-of-Concept Model to Predict Water Quality Changes in Distribution Pipe Networks (Q-WARP). Ottawa, ON, 14. Sadiq, R., and Rodriguez, M. J. (2004a). “Disinfection by-products (DBPs) in drinking water and predictive models for their occurrence: a review.” Science of the Total Environment, Citeseer, 321(1-3), 21–46. Sadiq, R., and Rodriguez, M. J. (2004b). “Fuzzy synthetic evaluation of disinfection by-products--a risk-based indexing system.” Journal of environmental management, 73(1), 1–13. Sadiq, R., Tesfamariam, S., Shahriar, A., Francisque, A., Islam, N., and Betrie, G. (2012). Prioritization of Water Main Replacement Program: A Risk-based Decision Support Tool. Kelowna, BC, 67–85. Sadiq, R., Veitch, B., Husain, T., and Bose, N. (2005). Prioritizing environmental effects monitoring (EEM) programs: A risk-based strategy. , Offshore oil and gas environmental effects monitoring: Approaches and technologies (pp. 95–110). OH: Springer. (S. L. Armsworthy, P. J. Cranford, and . Lee, eds.), 95–110. Said, A., Stevens, D. K., and Sehlke, G. (2004). “An innovative index for evaluating water quality in streams.” Environmental Management, 34(4), 406–414. 203  Sanctis, A. E. De, Shang, F., Uber, J. G., and Asce, M. (2010). “Real-Time Identification of Possible Contamination Sources Using Network Backtracking Methods.” Journal of Water Resources Planning and Management, 136(4), 444–4453. Sarkar, C., and Abbasi, S. A. (2006). “Qualidex - A new software for generating water quality indice.” Environmental Monitoring and Assessment, 119(1-3), 201–231. Shafer, G. (1976). A Mathematical Theory of Evidence. Shang, F., Uber, J. G., and Polycarpou, M. M. (2002). “Particle back tracking algorithm for water distribution system analysis.” ASCE Journal of Environment Engineering, 128(5), 441–450. Shang, F., Uber, J. G., and Rossman, L. a. (2008). “Modeling reaction and transport of multiple species in water distribution systems.” Environmental science & technology, 42(3), 808–14. Shastri, Y., and Diwekar, U. (2006). “Sensor placement in water networks: A stochastic programming approach.” J. Water Resour. Plann. Manage., 132(3), 192–203. Shen, H., and Mcbean, E. (2012). “An arcGIS tool for rapid contaminant source identification.” 12th Annual International Conference on Water Distribution Systems Analysis 2010, WDSA 2010, Tucson, AZ, United states, 436–446. SierraLegalDefenceFund. (2006). Canada’s Drinking Water Report Card. Waterproof 2. Spruill, T. B., and Candela, L. (1990). “Two approaches to design of monitoring networks.” Ground Water, 28(3), 430–442. Subramaniam. (1996). “Optimal Locations of Booster Stations in Water Distribution Systems.” University of Cincinnati, 2001. Cincinnati, Ohio. Swamee, P. K., and Tyagi, A. (2007). “Improved method for aggregation of water quality subindices.” Journal of Environmental Engineering, 133(2), 220–225. 204  Tamminen, S., and Ramos, H. (2008). “Water supply system performance for different pipe materials Part I: water quality analysis.” Water resources management, Springer, 22(11), 1579–1607. Taylor, P., Ostfeld, A., and Salomons, E. (2006). “Conjunctive optimal scheduling of pumping and booster chlorine injections in water distribution systems.” Engineering Optimization, 38(3), 337–352. Taylor, P., Ozdemir, O. N., and Demir, E. (2010). “Experimental study of chlorine bulk decay in water supply pipes Experimental study of chlorine bulk decay in water supply pipes Etude expérimentale de la décroissance volumique du chlore dans les canalisations d ’ alimentation en eau.” Engineering, (November 2011), 37–41. Tesfamariam, S., and Sadiq, R. (2006). “Risk-based environmental decision-making using fuzzy analytic hierarchy process (F-AHP).” Stochastic Environmental Research and Risk Assessment, 21(1), 35–50. Teunis, P. F. M., Xu, M., Fleming, K. K., Yang, J., Moe, C. L., and Lechevallier, M. W. (2010). “Enteric virus infection risk from intrusion of sewage into a drinking water distribution network.” Environmental science & technology, 44(22), 8561–8566. Trussel, R. R., and Umphres, M. D. (1978). “The formation of trihalomethanes.” J. AWWA, 70(11), 604–612. Tryby, M. E., Boccelli, D. L., Uber, J. G., and Rossman, L. a. (2002). “Facility Location Model for Booster Disinfection of Water Supply Networks.” Journal of Water Resources Planning and Management, 128(5), 322-333. USEPA. (1995a). Determination of chlorination disinfection byproducts, chlorinated solvents, and halogenated pesticides/herbicides in drinking water by liquid-liquid extraction and gas chromatography with electron-capture detection. Rev. 1. 205  USEPA. (1995b). Determination of haloacetic acids and dalapon in drinking water by liquid-liquid extraction, derivatization and gas chromatography with electron capture detection. Cincinnati, Ohio 45268, Rev. 1. USEPA. (2004). The Effectiveness of Disinfectant Residuals in the Distribution System. Environmental Protection, 1200 Pennsylvania Ave., NW Washington DC. USEPA. (2005a). Economic Analysis for the Final Stage 2 Disinfectants and Disinfection Byproducts Rule. 815–R–05–010. USEPA. (2005b). Impact of Distribution System Water Quality on Disinfection Efficacy. USEPA. (2006). Comprehensive Disinfectants and Disinfection Byproducts Rules ( Stage 1 and Stage 2): Quick Reference Guide Overview of the Rules. 2–5. USEPA. (2009). “Drinking Water Contaminants.” <http://water.epa.gov/drink/contaminants/index.cfm>. USEPA. (2010a). VSAT – Risk Assessment Tool for Water Sector Utilities. EPA 817–F–10–015. USEPA. (2010b). U.S. Environmental Protection Agency Science Advisory Board Drinking Water Committee for the Review of EPA’s Microbial Risk Assessment Protocol. 17. USEPA. (2014). “Integrated Risk Information System (IRIS).” <http://www.epa.gov/IRIS/>. Uyak, V., Ozdemir, K., and Toroz, I. (2007). “Multiple linear regression modeling of disinfection by-products formation in Istanbul drinking water reservoirs.” The Science of the total environment, 378(3), 269–280. Wang, W., Ye, B., Yang, L., Li, Y., and Wang, Y. (2007). “Risk assessment on disinfection by-products of drinking water of different water sources and disinfection processes.” Environment International, 33(2), 219–225. 206  Watson, J. P., Greenberg, H. J., and Hart, W. E. (2004). “A multipleobjective analysis of sensor placement optimization in water networks.” Proc., ASCE EWRI Annual Conf. on Critical Transitions in Water and Environmental Resources Management, Salt Lake City, Utah. Weickgenannt, M., Kapelan, Z., Blokker, M., and Savic, D. A. (2010). “Risk-Based Sensor Placement for Contaminant Detection in Water Distribution Systems.” Journal of Water Resources Planning and Management, 136(6), 629–636. Weickgenannt, M., Kapelan, Z., Blokker, M., Savic, D. A., and Sawodny, O. (2010). “Optimization of sensor locations for contaminant detection in water distribution networks.” 2010 IEEE International Conference on Control Applications, Ieee, 2160–2165. WHO. (2004). Guidelines for drinking water quality. World Health Organization, Geneva, Switzerland, 3rd edition. WHO. (2008). Guidelines for Drinking-water Quality. Woo, H. M., Yoon, J. H., and Choi, D. Y. (2001). “Optimal monitoring sites based on water quality and quantity in water distribution systems.” Proc., ASCE EWRI Annual Conf. on Bridging the Gap: Meeting the World’s Water and Environmental Resources Challenges, Orlando, Fl. Xu, J. H., Fischbeck, P. S., Small, M. J., VanBriesen, J. M., and Casman, E. (2008). “Identifying sets of key nodes for placing sensors in dynamic water distribution networks.” J. Water Resour. Plann. Manage., 134(4), 378–385. Yan, J., Vairavamoorthy, K., and Gorantiwar, S. D. (2007). “Modelling the risk of contaminant intrusion in water mains.” Proceedings of the ICE - Water Management, 123–132. Zadeh, L. A. (1984). “Review of books: a mathematical theory of evidence.” AI Mag., 5(3), 81–83. 207  Zhang, W., Miller, C. T., and DiGiano, F. A. (2004). “Bacterial regrowth model for water distribution systems incorporating alternating split-operator solution technique.” Jour. Environmental Engineering, 130(9), 932–941. Zhang, Y., Martinez, D., Collins, C., Graham, N., Templeton, M. R., Huang, J., and Nieuwenhuijsen, M. (2011). “Modelling of haloacetic acid concentrations in a United Kingdom drinking water system.” Journal of Water Supply: Research and Technology—AQUA, 60(5), 275–285. 208  Appendices Appendix A: Water quality parameter, significance and monitoring locations Parameter Significance of monitoring Suggested monitoring location by regulations Point of entry Distribution system sites Reservoir Dead end places Low flow areas High flow areas Coliform sampling area E. coli Human health concern due to fecal contamination.  √      Chlorine residual Decreased value indicates contamination and nitrification. √  √ √ √  √ Heterotrophic plate counts (HPC) Indicates   Nitrification   Chlorine residual decay, and  Biofilm growth √  √ √ √  √ Ammonia  Used in systems where chloramine is used as a disinfectant and with high ammonia concentrations  Represents possible nitrification √  √ √ √  √ Nitrate and nitrite Same as ammonia √  √ √ √  √ Turbidity Indirect measure of chlorine demand. √ √      Flow Important operational parameter such as:  High flow zones represents area with high population density, and  Low flow represents poor hydraulics     √ √  Pressure Extreme low pressure (<5psi) represents possible water main break. Monitor after a breakage and after incidents which may result in low pressure (e.g., after flushing and firefighting) Temperature Higher temperature may increase reaction rate and create:  Taste and odor  Nitrification, and   Lower residual disinfection  √  √ √    209  pH  Extreme low and high pH represents different contamination occurrence  Rapid pH change represents chemical overfeed at the treatment √ √      Alkalinity Same and pH √ √      Conductivity Represents:   Chemical overfeed at the treatment   Corrosion and scale formation, and   Sudden change indirectly represents contamination √ √      Color Aesthetics √ √      Lead Lead-related health concern √ √      Iron  Corrosion   Chlorine demand, and  Aesthetics  √      Manganese Aesthetics  √      Taste and odor  Aesthetics and   Secondary impact of chlorination  √      Hydrocarbon Accidental spill of chemicals such as petroleum.  √        210  Appendix B: Selected water quality parameter measurement details Parameter Measurement method Measurement type Manual Sensor based E. coli  Membrane filtration (USEPA 2009) √  Free chlorine  Membrane covered polarographic sensor  Potentiometric  Colorimetric- diethyl-p-phenylene diamine method  √ Heterotrophic plate counts (HPC) HPC test or standard plate count method √  Ammonia Colorimetric - requires calibration standard and reagent solutions  √ Nitrate and nitrite  Ultraviolet absorbance  Colorimetric   Ion selective methods  √ Turbidity  Nephelometer  Incandescent light  Scattered light method  √ Flow Different operation principles are available for flow measurement.  √ Pressure  Manometer  Mechanical  Electronic  √ Temperature Four-electrode conductivity sensor-temperature measure using temperature sensing element  √ pH  Differential pH sensor  Amperometric method  Differential pH electrode  Different emerging sensors are used such as potentiometric pH sensor   √ Alkalinity  Titration  Occasionally, online titration units are used √ √ 211  Conductivity  Four-electrode conductivity method  Conductance method  Two electrode cells  √ Color Standard methods using wavelength, e.g., spectrophotometry are available for colour measurement  √ Taste and odor Comparison with distilled water and rating are two available method (HealthCanada 1979) √  Hydrocarbon  Freon and hexane based USEPA method (Reeves 2000)  Online measurement: - Gravimetric (weight)  - Colorimetric  - Infrared  - UV Absorption  - Nephelometry  - Fluorescence √ √ 212  Appendix C: Risk analysis approaches used for contaminant intrusion Reference Methodology Contaminant type General objective Brief description Propato and Uber (2004b) EPANET -MSX simulation NS Vulnerability assessment with free residual chlorine and microbial interaction. EPANET-MSX simulation with first order Chick-Watson law. Sadiq et al. 2006 Dempster-Shafer theory (DST) of evidence NS Contaminant intrusion likelihood measurement. Estimated using DST with parameters like breakage rate, transient pressure and distance between the source and the water main. Sadiq et al. 2007 Fuzzy logic, AHP, and Dempster-Shafer rule of combination NS Water quality failure analysis Aggregative risk estimation using various soft-computing and multi-criteria decision making method. Teunis et al. 2010 Hydraulic and Monte Carlo simulations Norovirus, rotavirus Enteric virus infection risk from sewage Considering the dilution factors using Monte Carlo simulation, estimated the risk of enteric virus infection risk. Austin et al. (2012) Monte Carlo simulation with EPANET and QMRA Cryptosporidium Exposure assessment and risk characterization with QMRA Simulating event based microbial intrusion using Monte Carlo simulation with EPANET. QMRA was also assessed for DN. Ebacher et al. (2012) Analysis using commercially available transient software and negative exponential model Cryptosporidium Public health risk assessment using transient analysis output Leakage orifices and submerged air vacuum valves (AVVs) were considered as the intrusion paths while negative exponential model was used to calculate the risk after analysis using transient analysis software. Perelman et al. (2013) Bayesian belief network (BBN) NS Source detection Cluster analysis, BBN and evaluation using real sensor data. NS- not specified; QMRA- Quantitative microbial risk assessment; BBN- Bayesian belief network213  Appendix D: Geostatistical approaches used in contaminant intrusion studies Reference Method General objective Brief description Helbling and VanBriesen (2009) Nonlinear regression Chlorine demand signal after analyzing microbial contaminations. Minimizing the observed and the predicted values are performed in this regard. Shen and McBean (2010) Data integration using ArcGIS tool Contaminant source identification Data integration such as land usage, and aging of the water mains are performed with ArcGIS tool. Perelman and Ostfeld (2011) Topography and connectivity analysis System analysis such as sensor placements and contaminant isolation. Clusters were made using graph theory based on flow direction in pipes. Butera et al. (2013) Geostatistical approach with numerical hydraulic modelling. Assess the uncertainty in the concentration measurements. The uncertainty of the concentration and the release function using statistical moments are used. The results are in confidential intervals. Both steady and unsteady state flow conditions.    214  Appendix E: Software tools used for predicting contaminant intrusion Name Reference Developer Objective and methods used Risk Assessment Methodology for Water Utilities (RAM-W) Jaeger et al. (2010) Awwa Research Foundation (AwwaRF) and Sandia National Laboratories, the Environmental Protection Agency (EPA) undertook a program in 2000 Risk assessment for water utilities Vulnerability Self-Assessment Tool (VSAT)  USEPA (2010b) USEPA Contains performing security threats, natural  hazards risk assessments, and utility Emergency Response Plans (ERPs) Security Emergency Management System (SEMS) Khanal et al. (2005) National Rural Water Association (NRWA)  Vulnerability assessment tool for small water utilities Threat Ensemble Vulnerability Assessment (TEVA) program Khanal et al. (2005) USEPA Probabilistic framework for the vulnerability assessment Integrated risk assessment of water distribution systems (IRA-WDS) Yan et al. (2007) UK Department for International Development (DFID) Estimates the risk of contaminant intrusion considering contaminant ingress, and the pipe conditions.   CANARY event detection software Hart and Mckenna (2009) USEPA Provides response for the accidental contamination. Used statistical methods to look parameter patterns for chlorine, total organic carbon.  The Threat Ensemble Vulnerability Assessment and Sensor Placement Optimization Tool (TEVA-SPOT) Berry & Hart, (2008)  USEPA Provides contamination warning in distribution network. PipelineNet software AWWARF, (2003)  USEPA Assess the effects of terrorist attack in a DN. 215   Appendix F: Proposed integrated model for contaminant detection and mitigation Model  Introduction A conceptual model is presented based on various aspects of intrusion in a DN. DN management involves in proper detection and mitigation against potential contaminant intrusions. The proposed conceptual model requires following five elements: 1) source identification, 2) monitoring station / sensor placement, 3) system analysis, 4) effect analysis, and 5) mitigation and control. Source identification and monitoring station selection and sensor design will be performed by adapting the desired optimization scheme. System analysis is required to evaluate a DN any time and will be performed using geographical and statistical analysis. Effect analysis will be performed using risk analysis to identify potential intrusion points or zones. Moreover, some minimum mitigation and control measures must protect the system from any epidemical situation.   The model provides the basic steps required to perform the above mentioned analysis and decision making. It should be noted that many of these analytical steps are interrelated and require multiple interrelated analysis.  216  Optimization Software-based hydraulic and water quality simulations are necessary for optimization. A DN model for hydraulic and water quality analysis is required in this respect. Initial hydraulic simulation will estimate nodal pressures, where negative or lower pressure (lower than a threshold, e.g., <20psi) will identify the possible pollutant entry points. Decision makers will assign pollutant concentration and flow rate at these points. For example, the pollutant flow rate can be estimated using the following equation  (Teunis et al. 2010) assuming the intrusion point an orifice opening. Finally, the volume of the contaminated water will be estimated by multiplying the flow rate with the duration of the incident. One can also assume pollutant concentrations such as microbiological parameter concentrations from the below table. The table lists typical concentrations for microbial pollutants in water/soil samples around water mains.  Typical microbial concentration in water/ soil samples around water mains Organism Reference Water (CFU/ PFU) per 100 ml Soil (CFU/ PFU) per 100 ml Bacillus Karim et al. (2003) < 1 to 4.6 x 106 < 1 to 1.3 x 108 Clostridium Karim et al. (2003) < 1 to 2.5 x 103 < 1 to 1 x 105 Coli phage Karim et al. (2003) < 1 to 1 x 104 < 1 Cryptosporidium Ebacher et al. (2012) 10 oocysts/L - Fecal coliforms Karim et al. (2003) < 2 to ³ 1.6 x 103 < 2 to ³ 1.6 x104 S. typhimurium Clark et al. (2007) 2.5 x  105 - Total coliforms Karim et al. (2003) < 2 to ³ 1.6 x 103 < 2 to ³ 1.6 x 104 CFU: Colony-forming units; PFU: Plaque-forming units  Hydraulic and water quality analysis will be performed in EPANET-MSX mode using the Chick-Watson first order kinetics and typical first order kinetics for microbiological growth (Betanzo et al. 2008) and free residual chlorine respectively:   pdPk PCdt      bdCk Cdt   where P= microbial population (Microbiological unit/L); C= free residual chlorine concentration (mg/L); kP= pathogen kinetic decay constant (L/hr. mg); kb = bulk-decay co-efficient for residual chlorine (1/hr). kP value is dependent on the type of microbes, pH, and temperature and the type of the pollutant source. Moreover, Betanzo et al. (2008) have also stated typical kP values for common microbes such as for Salmonella, Shigella, Vibrio cholerae, E. coli O157:H7, and Legionella pneumophila.  Optimization will be performed using an appropriate algorithm such as GA and objective functions stated in Table 2-6.  Water main condition will be assessed using the breakage rate, soil conditions and repair history to decide whether there might be a possible contaminant path. Soft-computing based methods such as fuzzy logic and DST can be used to estimate the contamination likelihood and will be integrated into the optimization. Finally, the analysis results will conclude an optimal solution for identifying the pollutant source or designing sensor and monitoring stations.  217  Risk analysis & other analysis Apart from likelihood estimation, the QMRA/ exponential model will be used for risk assessment assuming a beta-poison or exponential model (Equation[2.5] and [2.6]): Table 2-4 lists distribution parameter values for commonly available microbes.  Additionally, a factor representing risk consequences called significant factor will be estimated from land use and population density information. A simplistic weighted average or arithmetic mean will be adapted to estimate the significant factor. Finally, the risk assessment will be completed after combining the likelihood and the significant factor. Additionally, municipal inventory data, e.g., breakage rate, soil conditions and repair history will be integrated using advanced GIS operations or regression analysis. These geographical and statistical analyses will provide a simplistic system analysis and can help decision makers with protection and mitigation decisions.    Protection & mitigation After effective detection using the above mentioned methods mitigation should take place to protect the system. First, it requires setting boundary for contamination achieved by system analysis and source identification with optimization (Figure 3). The boundary will protect the system from further contamination. The next step is the affected area shutdown with appropriate valve operations. Warnings will be sent out to consumers with explanations of the situation. Further contamination will be prevented by inspecting the area of contamination. A water main with frequent breakage will require repairs or replacement, while a contaminated source will require immediate pollutant elimination. Additionally, chlorine dosage will be increased in case of microbiological pollutants. Adequate flushing will eliminate pollutants already in the DN. Finally, the system analysis will be repeated to ensure the DN safety.     218  Appendix G: Life cycle cost estimation for gas chlorination (1 )69.53(1 ) 1ygc gc gc gcyir irLCC PC Q D MCir        Symbol Unit Abbreviation LCCgc $/yr 10 years annual life-cycle cost  for gas chlorination PCgc $ Amount of capital installment payments for gas chlorination over y years αgc $/Pound Cost of chlorine MCgc $/yr Maintenance cost for gas chlorination ir % Annual interest rate Q L/Sec Flow rate for the node selected for booster station D mg/L Dosage for the booster station    

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0166702/manifest

Comment

Related Items