Citation
The water footprint of urban energy systems

Material Information

Title:
The water footprint of urban energy systems concepts, methods and applications for assessing electricity supply risk factors
Creator:
Cohen, Elliot J. ( author )
Place of Publication:
Denver, Colo.
Publisher:
University of Colorado Denver
Publication Date:
Language:
English
Physical Description:
1 electronic file (165 pagers). : ;

Thesis/Dissertation Information

Degree:
Doctorate ( Doctor of Philosophy)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Civil Engineering, CU Denver
Degree Disciplines:
Civil Engineering
Committee Chair:
Rajagopalan, Balaji
Committee Members:
Ramaswami, Anu
Heikkila, Tanya
Karunanithi, Arun
Janson, Bruce

Subjects

Subjects / Keywords:
Water withdrawals ( lcsh )
Electric power distribution -- Reliability ( lcsh )
Water -- United States ( lcsh )
Electric power failures -- Risk assessment ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
his dissertation adds to the body of knowledge of the water-energy nexus in four measurable ways. First, a water withdrawal footprint of energy supply (WWFES) to cities was developed, and placed it in the context of other water footprints defined in the literature. The WWFES provides a novel way to quantify direct and indirect water requirements to satisfy urban energy demand. The magnitude of the WWFES for Denver, Colorado was found to be 381 liters/person/day and 66% as large as all direct water uses in the city combined (mean estimate). This finding is relevant to urban sustainability planning as it shows significant water conservation may be achieved through energy efficiency and energy conservation. Next, we demonstrate the robustness of the WWFES method for a rapidly developing city (Delhi) with unique energy requirements, energy infrastructure and data availability compared to the initial test case (Denver). Data collected for the Indian power sector enabled exploration of spatial- and temporal-variability of electricity supply to cities and the associated dynamic WWFES. Integrating over both space and time for one year, we estimate the water requirements of electricity production alone to be 36% as large as municipal water supply for Delhi, compared to 16% for Denver. In both cases, this highlights that electricity supply, like municipal supply, can be at risk during drought or other hydrological extremes, corroborated by interviews with industry experts. The third and fourth contributions of this dissertation are to place water-related constraints to power generation in the context of other system risks using both social science methods and data-driven statistical analysis. For the former, a survey was administered to electricity infrastructure operators serving Delhi with three objectives: (1) identify and rank system risks to power supply reliability based on industry perceptions of risk; (2) identify and rank current and future service provision priorities; and (3) collect social network data regarding interaction between infrastructure operators. For the latter, an empirical study of electricity supply reliability in Northern India was conducted in a hierarchical modeling framework to assess the contribution of structural, environmental and supply-chain constraints to grid reliability. Model results indicate the WWFES is a statistically significant predictor of power supply reliability in Northern India when we control for structural, climate and supply-chain covariates. These results highlight the importance of the WWFES when evaluating risks to, and reliability of, trans-boundary energy systems.
Thesis:
Thesis (P.h.D.)--University of Colorado Denver. Civil engineering
Bibliography:
Includes bibliographic references.
System Details:
System requirements; Adobe Reader.
General Note:
Department of Civil Engineering
Statement of Responsibility:
by Elliot J. Cohen.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
902739339 ( OCLC )
ocn902739339

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

THE WATER FOOTPRINT OF URBAN ENERGY SYSTEMS: CONCEPTS, METHODS AND APPLICATIONS FOR ASSESSING ELECTRICITY SUPPLY RISK FACTORS by ELLIOT J. COHEN B.S., University of Maryland at College Park 2009, Mechanical Engineering M.S. University of Colorado at Denver 2011, Environmental Engineering A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Civil Engineering 2014

PAGE 2

ii This thesis for the Doctor of Philosophy degree by Elliot J. Cohen has been approved for the Civil Engineering by Balaji Rajagopalan Chair Anu Ramaswami Advisor Tanya Heikkila Arun Karunanithi Bruce Janson Date : May 2 201 4

PAGE 3

iii Cohen Elliot J., ( Ph.D., Civil Engineering ) The Water Footprint of Urban Energy Systems: Concepts, Methods and A pplications for Assessing Electricity Supply Risk F actors Thesis directed by Professor Anu Ramaswami ABSTRACT This dissertation adds to the body of knowledge of the water energy nexus in four measurable ways. First, a water withdrawal footprint of energy supply (WWFES) to cities was developed, and placed it in the context of other water footprints defined in the l iterature. The WWFES provides a novel way to quantify direct and indirect water requirements to satisfy urban energy demand. The magnitude of the WWFES for Denver, Colorado was found to be 381 liters/person/ day and 66% as large as all direct water uses in the city combined (mean estimate). This finding is relevant to urban sustainability planning as it shows significant water conservation may be achieved through energy efficiency and energy cons ervation. Next, w e demonstrate the robustness of the WWFES method for a rapidly developing city (Delhi) with unique energy requirements, energy infrastructure and data availability compared to the initial test case (Denver). Data collected for the Indian power sector enabled exploration of spatial and temporal variability of electricity supply to cities and the associated dynamic WWFES Integrating over both space and time for one year, we estimate the water requirements of electricity production alone to be 36% as large as municipal water supply for Delhi, compared to 16% for Denver. In both cases, this highlights that electricity supply like municipal supply can be at risk during drought or other hydrological extremes, corroborated by interviews with industry experts The third contribution of this dissertation is to place water related constraints to power generation in the context of other system risks using both social science methods and data driven

PAGE 4

iv statistical analysis For the former, a survey was administered to electricity infrastructure operator s serving Delhi with three objectives : (1) identify and rank system risks to power supply reliability based on industry perceptions of risk; (2) identify and rank current and future service provision priorities; and (3) collect social network data regarding frequency of interaction between infrastructure operators. Finally an empirical study of electricity supply reliabili ty in Northern India was conducted in a hierarchical modeling framework to assess the contribution of structural, environmental and supply chain constrain t s to grid reliability Model results indicate the WWFES is a statistically significant predictor of power supply reliability in Northern India when we control for structural climate and supply chain covariates These results highlight the importance of the WWFES when evaluating risks to, and reliability of, trans boundary energy systems. The form and content of this abstract are approved. I recommend its publication. Approved: Anu Ramaswami

PAGE 5

v DEDICATION I dedicate this work to my beautiful wife Livia, whose strength, integrity and compassion are an inspiration to me everyday. Without her, this work would be for naught.

PAGE 6

vi ACKNOWLEDGMENTS I would like to thank the National Science Foundation Integrative Graduate Education and Research Traineeship (IGERT), and the University of Colorado Denver program on Sustai nable Urban Infrastructure that it funded, for providing a unique graduate opportunity for rigorous, interdisciplinary study of global sustainability challenges. I would like to recognize the Lead PI, Anu Ramaswami, whose vision and leadership transformed an ambitious idea into a full fledged program. I would also like to thank the faculty that co created and co taught the sustainability curriculum, including Debbie Main, Saeed Barhaghi, Jason Ren, Arun Karunanithi, Tanya Hiekkila, Chris Weible, John B rett and Felix Flechas. Our incredible program managers over 5 years deserve endless praise for keep ing everything running smoothly Luann Rudol ph, Meghan Bernard, Alison Kent, Zeljko Spiric and Koben Calhoun. Specific to this thesis, I would like to acknowledg e the co creation of many ideas, co ncepts and methods. First, my advisor, Prof. Anu Ramaswami, and I spent many long sessions thinking critically together on the content of this thesis. The WWFES concept in particular is an extension of her work on GHG foo tprints for cities, applied here to the analogous case of water. Prof. Balaji Rajagopalan taught me many of the statistical methods found in chapter VI in his Advance Data Analysis class at CU Boulder. I enjoyed the class so much that I took it twice! Prof Tanya Hiekkila was instrumental in helping me think through the research question s and methods for the social network analysis in chapter V and more generally teaching an engineer how to think like a social scientist in her excellent class, Theories of Change of Infrastructure Management. Finally, I would like to thank the U.S. India Fulbright Nehru Scholars program for supporting 9 months of fieldwork in Delhi, India, which became a cornerstone of my dissertation. To the many professional contacts and generous hosts who are too numerous to name, I sincerely appreciate al l of your help and hospitality throughout my journey.

PAGE 7

vii TABLE OF CONTENTS CHAPTER INTRODUCTION TO THIS DISSERTATION ................................ ................................ .............. 1 REVIEW OF COOLING WA TER CONSTRAINTS TO P OWER GENERATION ..................... 3 Synthesis ................................ ................................ ................................ ................................ ...... 3 Regional Assessments of Water Use for Power Generation and Competing Uses ..................... 3 Technology Specific Assessments of Water Use in Power Generation ................................ ...... 7 Dynamic Modeling of Water Withdrawal Constraints and Temperature Constraints to Power Generation ................................ ................................ ................................ ................................ ... 8 THE WATER WITHDRAWAL FOOTPRINT OF ENERGY SUPPLY TO CITIES: CONCEPTUAL DEVELOPME NT AND APPLICATION T O DENVER, CO ........................... 10 Abstract ................................ ................................ ................................ ................................ ...... 10 The Need for Water Footprints of Urban Energy Systems ................................ ........................ 10 Objectives ................................ ................................ ................................ ................................ .. 15 Conceptual Development of the WWFES and its Place in the Water Footprint Literature ...... 16 Territorial Accounting of Direct Water Use in Regions ................................ ....................... 16 Equivalent Land Area Footprints for Direct Water Use ................................ ........................ 17 Economic Consumption Based Water "Loss" Footprints for Regions ................................ 17 Economic Production Based Water Withdrawal Footprints. ................................ ................ 18 Infrastructure Supply Chain Water Footprint for Cities ................................ ........................ 18 Methodology ................................ ................................ ................................ .............................. 20 Electricity Import Analysis ................................ ................................ ................................ ... 20 Computing the WWFES ................................ ................................ ................................ ........ 21 Sensitivty of the WWFES to Regional Variability and Energy Technology Production Pathways ................................ ................................ ................................ ............................... 22

PAGE 8

viii Application of the WWFES to Denver, CO ................................ ................................ .............. 25 Community wide Demand for Energy ................................ ................................ .................. 25 Electricity ................................ ................................ ................................ .............................. 25 Biofuels ................................ ................................ ................................ ................................ 26 Natural Gas ................................ ................................ ................................ ............................ 26 Petroleum Fuels ................................ ................................ ................................ ..................... 28 Scenario Modeling ................................ ................................ ................................ ..................... 29 Electricity ................................ ................................ ................................ .............................. 30 Transportation Fuels ................................ ................................ ................................ .............. 31 Buildings Natural Gas ................................ ................................ ................................ ........... 31 Municipal Water Supply ................................ ................................ ................................ ....... 31 Results ................................ ................................ ................................ ................................ ....... 32 Electricity Import Analysis ................................ ................................ ................................ ... 32 The Aggregate WWFES for Denver, Colorado ................................ ................................ .... 33 Conclusion ................................ ................................ ................................ ................................ 35 SPATIALLY AND TEMPORALLY DELINEATED WATER FOO TPRINTS FOR TRANS BOUNDARY ELECTRICITY SUPPLY T O CITIES: EXPLORING VIRTUAL WATER TRANSFERS TO DELHI V IA THE NORTHERN INDI A POWER GRID ................................ 38 Introduction to Embodied Water and Water Footprints ................................ ............................ 38 Application of the Water Footprint Concept to Urban Sustainability ................................ ....... 39 Objectives ................................ ................................ ................................ ................................ .. 40 Study Design ................................ ................................ ................................ ............................. 40 Context ................................ ................................ ................................ ................................ .. 41 Data ................................ ................................ ................................ ................................ ....... 42 Results ................................ ................................ ................................ ................................ ....... 49

PAGE 9

ix Spatial and Temporal Delineation ................................ ................................ ....................... 49 Seasonality ................................ ................................ ................................ ............................ 53 Integrating Over Space and Time to Yield City Wide Annual Totals ................................ .. 56 Discussion ................................ ................................ ................................ ................................ .. 59 FOOTPRINT INFORMED SOCIAL NETW ORK ANALYSIS ................................ .................. 63 Introduction ................................ ................................ ................................ ............................... 63 An Interdisciplinary Framework for Urban Sustainability ................................ ........................ 64 A Novel Approach: Combing Footprints with Social Network Analysis ................................ 65 Evolution of Infrastructure Supply Chain Footprints ................................ ................................ 65 Social Networks, the Diffusion of Innovation and Network Interventions ............................... 67 Social Networks ................................ ................................ ................................ .................... 67 Diffusion of Innovations ................................ ................................ ................................ ....... 68 Network Interventions ................................ ................................ ................................ ........... 69 Social Network Analysis Applied to SEIS Actor Categories: A Brief Review of the Literature ................................ ................................ ................................ ................................ ................... 70 Resource Users (e.g. Households, Businesses and Government) ................................ ......... 70 Policy Actors ................................ ................................ ................................ ......................... 71 Infra structure Designer Operators ................................ ................................ ......................... 71 Application to the Water Energy Nexus in Delhi, India ................................ ........................... 74 The National Power Grid of India ................................ ................................ ............................. 77 Study Design ................................ ................................ ................................ ............................. 79 Objectives ................................ ................................ ................................ .............................. 79 Methodology and Data Collection ................................ ................................ ........................ 79 Network Analysis Results and Discussion ................................ ................................ ................ 82 Survey Results and Conclussion ................................ ................................ ................................ 88

PAGE 10

x THE EFFECT OF CLIMAT E AND SUPPLY CHAIN CONSTRAINTS ON GRID SCALE ELECTRICITY SUPPLY R ELIABILITY: A HIERAR CHICAL LINEAR MODEL .................. 91 Introduction ................................ ................................ ................................ ............................... 91 Power Supply Reliability ................................ ................................ ................................ ........... 92 Defining and Measuring Power Supply Reliability ................................ ................................ ... 92 Risk Factors to Power Supply Reliability ................................ ................................ .................. 99 Data ................................ ................................ ................................ ................................ .......... 100 Structural Constraints ................................ ................................ ................................ .............. 103 Climate Effects ................................ ................................ ................................ ........................ 106 Supply Chain Constraints ................................ ................................ ................................ ........ 109 Motivation for, and Applicability of, Hierarchical Models ................................ ..................... 112 Hierarchical Model A ................................ ................................ ................................ .............. 114 HLM A, Level 1 ................................ ................................ ................................ .................. 115 HLM A, Level 2 ................................ ................................ ................................ .................. 117 HLM A, Level 3 ................................ ................................ ................................ .................. 120 Cumulative HLM A ................................ ................................ ................................ ............ 121 Hierarchical Model B ................................ ................................ ................................ .............. 123 Mixed Effects Models ................................ ................................ ................................ ......... 123 HLM B Results ................................ ................................ ................................ ................... 124 Discussion ................................ ................................ ................................ ................................ 130 APPENDIX DATA SOURCES FOR THE INDIAN POWER SECTOR ................................ ....................... 133 SURVEY OF ELECTRICIT Y INFRASTRUCTURE OPE RATORS SERVING DELHI ......... 135 PART I: Service Provision ................................ ................................ ................................ .. 135 PART II: Operations and Management ................................ ................................ ............... 136

PAGE 11

xi PART III: Future Electricit y Planning ................................ ................................ ................ 137 PART IV: Linkages With Infrastructure Providers and Regulatory Bodies ....................... 137 REFERENCES ................................ ................................ ................................ ............................ 140

PAGE 12

xii LIST OF TABLES TABLE III 1 Types of Water Footprints of Regions ................................ ................................ ................... 20 III 2 Material/Energy Flow Analysis (MEFA) and Life Cycle Assessment (LCA) Water Withdrawal Intensity Factors (WWIF) for Denver. ................................ ................................ ....... 28 IV 1 Operational Water Withdrawal and Consumption Factors for Power Stations Serving Delhi ( gal/MWH) ................................ ................................ ................................ ................................ ..... 43 IV 2 Water Withdrawal F ootprint of Energy Supplied to Delhi: Summary ................................ .. 56 IV 3 Water Withdrawal Footprint Benchmarks: Comparing Delhi & Denver .............................. 56 V 1 Overview of Actors: Number Identified Per Category ................................ .......................... 81 VI 1 Electricity Supply Reliability Risk Factors ................................ ................................ ........... 99 VI 2 HLM A Level 1 Model Coefficients and ANO VA ................................ ............................. 115 VI 3 HLM A Level 1 ANOVA Summary ................................ ................................ ................... 116 VI 4 HLM A Level 2 Initia l Model ................................ ................................ ............................. 118 VI 5 HLM A Level 2 "Best Model" After BIC Subset Selection ................................ ............... 118 VI 6 HLM A Level 2 ANOVA Summary ................................ ................................ ................... 119 VI 7 HLM A Level 3 Model Coefficients and ANOVA ................................ ............................. 120 VI 8 HLM A Leve l 3 ANOVA Summary ................................ ................................ ................... 120 VI 9 HLM A ANOVA Summary Table: All Levels ................................ ................................ ... 121 VI 10 Fixed Effects for HLM with Structural Predictors Only (Null Model) ............................. 127 VI 11 Fixed Effects for HLM with Structural, Climate & Supply Chain Predictors (Alter nate Model) ................................ ................................ ................................ ................................ .......... 127 VI 12 Fixed Effects for HLM with Climate & Supply Chain Predictors (Best Model) .............. 127 VI 13 Fixed and Random Intercepts for "Best Model" ................................ ............................... 128 VI 14 Hypothesis Test Comparing Log likelihood Ratio to Chi Square Distributi on ................ 128 VI 15 Subset Selection Using Log likelihood Ratio Test ................................ ............................ 129

PAGE 13

xiii LIST OF FIGURES FIGURE III 1 Water Levels Drop at a Large Power S tation in East Texas During Summer D rought. ........ 14 III 2 Illustration of the WWFES C oncept. ................................ ................................ ..................... 15 III 3 Trans Boundary Electricity I mports for 43 U.S. Cities ................................ ......................... 33 III 4 Water Withdrawal Footprint of Energy Supply (WWFES) Compared to Direct Water S upply for Denver, Colorado, USA. ................................ ................................ ................................ .......... 35 IV 1 Relationship Between Ambient Temperture and Energy Demand for Delhi, Inida. ............. 42 IV 2 Monthly Power Supply Position for Northern Region (NR) States ................................ ..... 48 IV 3 Monthly Energy Requirement Met by Net Drawal from Grid for NR States ....................... 49 IV 4 Trans boundary Water Footprint of Energy Supplied from Grid to Delhi, b y Source .......... 52 IV 5 In boundary Water Footprint of Energy Generation in Delhi, b y Source. ............................ 52 IV 6 Comparing In b oundary vs. Trans boundary Water Requirements of Energy Supplied to Delhi ................................ ................................ ................................ ................................ ............... 53 IV 7 Monsoon vs. Non Monsoon Seasonal Total Water Withdrawals and Consumption Related to Trans Boundary Energy S upplied to Delhi. ................................ ................................ ................... 54 IV 8 Calendar Seasonal Total Water Withdrawals and Consumption Related to Trans B oundar y Energy S upplied to Delhi ................................ ................................ ................................ ............... 55 IV 9 Monthly Energy Supplied to Delhi from Trans Boundary S ources ................................ ...... 55 IV 10. Fuelwise Summary of Delhi WWFES ................................ ................................ ............... 57 IV 12 The River Yamuna is Dammed and Diverted for Municipal Supply to Delhi .................... 61 IV 1 3 Recharged by Wastewater Effluent Through Delhi, the River Yamuna is Dammed and Diverted Again, This Time to Supply Cooling Water to a Large Power Station ...72 V 1 Supply Chain Network Analysis (Lazzarini et al. 2001) ................................ ........................ 73 V 2 Footprint Informed Networ k Analysis (Adapted for this T hesis) ................................ .......... 73 V 3 Network Graph of Electricity Infrastructure Operators S erving Delhi. ................................ .. 86 V 4 Ego Centric Network Analysis with Alter Connections ................................ ......................... 96 V 5 Priorities of Electricity Infrastructure Operators ................................ ................................ .... 89

PAGE 14

xiv V 6 Risk Perceptions of Electricity Infrastructure Operators (Present and Future Risks) ............. 90 V 7 Ranking of Risk Factors by Electricity Infrastructure Operators ................................ ........... 90 VI 1 Gamma D istribution F it to ENS (GWh) ................................ ................................ ................ 95 VI 2 Log Normal T ransformation of ENS ................................ ................................ .................... 95 VI 3 Gamm a Distribution Fit to RNS (D imensionless) ................................ ................................ 96 VI 4 Log Normal Transformation of RNS (D imensionless) ................................ ......................... 96 VI 5 Time Series of ENS (GWh) ................................ ................................ ................................ ... 97 VI 6 Time Series of RNS (%) ................................ ................................ ................................ ........ 97 VI 7 Group wise Boxplots of ENS (GWh) ................................ ................................ .................... 98 VI 8 Group wise Boxplots of RNS (%) ................................ ................................ ......................... 98 VI 9 Relationshi p Between ENS and Power System Structural C ovariates ................................ 105 VI 10 Relationship Between Log(ENS) and Power System Structural C ovariates ..................... 105 VI 11 Effect of Temperature on E ne rgy D emand in Delhi, India. ................................ .............. 107 VI 12 Precipitation at State Capital Nearest Neighbor Weather S tations (NOAA ISD) ............ 108 VI 13 Monthly Rainfall Accumulation (IITM) ................................ ................................ ........... 108 VI 14 Log(ENS) vs. Temperature Pre cipitation Index, Showing Group wise Linear R egressions ................................ ................................ ................................ ................................ ..................... 109 VI 15 Fuel Supply to Coal, Gas and Hydro Power S tations, All India, 2008 2013 .................... 111 VI 16 Scatterplot Matrix of Log(ENS) Versus Supply Chain V ariables ................................ .... 112 VI 17 Hierarchical Modeling Framework ................................ ................................ ................... 115 VI 18 HLM A Level 1 Model Diagnostics ................................ ................................ .................. 117 VI 19 HLM A Level 2 Model Diagnostics ................................ ................................ .................. 119 VI 20 HLM A Level 3 Model Diagnostics ................................ ................................ .................. 121 VI 21 HLM A Cumulative Fit ................................ ................................ ................................ ..... 122 VI 22 Model Comparison: HLM A vs. Single Level GLM w. All Predictor s ............................ 122 VI 23 The Effect of Capacity Adequacy on Grid Reliability ................................ ...................... 125

PAGE 15

xv V I 24 Comparing Fitte d vs. Observed Response in Log Space (left column) and Back Transformed to the Original Response S pace (right column) ................................ ...................... 130

PAGE 16

xvi LIST OF EQUATIONS EQUATION III 1 Computing the WWFES. ................................ ................................ ................................ ....... 21 IV 1 Generalized Water Footprint Equation ................................ ................................ .................. 39 IV 2 Energy Requirement ................................ ................................ ................................ .............. 45 IV 3 Energy Available ................................ ................................ ................................ ................... 45 IV 4 In Boundary Energy Generation ................................ ................................ ........................... 46 IV 5 Trans Boundary Energy Supply ................................ ................................ ............................ 46 VI 1 Energy Requirement ................................ ................................ ................................ .............. 93 VI 2 Energy Available ................................ ................................ ................................ ................... 93 VI 3 In Boundary Energy Generation ................................ ................................ ........................... 93 VI 4 Trans Boundary Energy Supply ................................ ................................ ............................ 93 VI 5 Energy Not Supplied [GWh] ................................ ................................ ................................ 93 VI 6 Requirement Not Supplied [Dimensionless] ................................ ................................ ......... 93 VI 7 Energy Index of Reliability [Dimensionless] ................................ ................................ ........ 93 VI 8 Final Equation for HLM A, Level 1 ................................ ................................ .................... 116 VI 9 Final E quation for HLM A, Level 2 ................................ ................................ .................... 119 VI 10 Final E quaiton for HLM A, Level 3 ................................ ................................ .................. 120 VI 11 Mixed Effects Model, Level 1 ................................ ................................ ........................... 124 VI 12 Mixed Effects Model, Level 2 ................................ ................................ ........................... 124

PAGE 17

1 CHAPTER I INTRODUCTION T O THIS DISSERTATION Water footprints t raditionally estimate water lost due to evapo transpiration associated with producing a certain good, and the same embodied in trade across regions is used to estimate regional/national water footprints (Hoekstra and Hung 2002) These footprints, however, do not address risk posed to cit y energy supplies characterized by insufficient water availability at points of production. For example, cooling water intake (e.g. withdrawals) at thermoelectric power plants would not be included in a standard water footprint only the fraction lost to ev aporation and drift from the cooling tower. Water withdrawal intensity factors for producing goods and services are being developed at the national scale (Blackhurst et al. 2010) but lack sufficient spatial resolution to address these types of city scale water energy tradeoffs. Untangling complex issues at the water energy nexus is the motivation for this dissertation. It is organized into four main chapters with the following objectives. C hapter III develops concepts and methods for a water withdrawal foo tprint of energy supply (WWFES) to cities, and places it in the context of other water footprints defined in the literature. The magnitude of the WWFES is computed for Denver, Colorado and compared to the city's direct use of water to offer perspective. Sc enario analysis is used to explore how energy technology and energy policy choices shape the growth trajectory of the water footprint of cities. C hapter IV tests the robustness of the WWFES methodology for a rapidly developing city subject to unique data availability compared to U.S cities. Utilizing in depth government reporting, we compute a dynamic water footprint of the electricity supply to Delhi, India. The objective is to quantitatively assess virtual water transfers embodied in electricity supply to cities. Further, we explore spatial and temporal delineation of the footprint giving special attention to seasonality.

PAGE 18

2 The objective of c hapter V is to investi gate risk perceptions, priorities and information sharing between electricity infrastructure operators Energy infrastructure is designed and operated by thousands of individuals working in myriad institutions operating across multiple scales. As such, me thods are needed to quickly and reliably identify who are the key actors in an infrastructure supply chain. One such method is applied to a case study of electricity infrastructure operators in Delhi, India. The objective of Chapter VI is to evaluate empi rical data gathered for the Northern Regional power grid of India to assess the statistical significance of structural, environmental and supply chain constraints to power supply reliability. The goal is to compare signals seen in the empirical data with the stated risk perceptions of electricity infrastructure operat ors and to contextualize the relative impact of water supply constraints on grid scale electricity systems subject to multiple and multi scale risk s.

PAGE 19

3 CHAPTER II REVIEW O F C OOLING W ATER C ONSTRAINTS TO P OWER G ENERATION Synthesis The technical literature on water supply constraints to thermoelectric power generation can be summarized in three categories: (1) Regional (geographically bounded) assessments of water use for power generation and competing uses, (2) technology specific assessments of water use in power generation, and (3) dynamic modeling of water withdrawal constraints and temperature constraints to power generation. Regional Assessments of Water Use in P ower Generation and Competing U ses This category includes the works of Tidwell et al. (2012), Stillwell et al. (2011a, 2011b), Sovacool and Sovacool (2009), Sheldon (2008), EPRI (2005), Kruse et al. (2004), Curlee and Sale (2004) and the Land and Water Fund of the Rockies (2003). The Land and Water Fund of the Rockies (2003) report helped raise awareness to regional competition for water resources between thermoelectric power generation and competing uses such as municipal supply and agriculture. Since then, attention to water ener gy competition has grown tremendously and most papers on the subject now cite similar examples in the introductory and background sections. In particular, Woldeyesus (2012) offer s examples of water energy competition in the U.S. power sector. Additional ex amples for the U.S. are provided in the DOE/NETL (2006) Report to Congress on the Water Energy Nexus Examples for Europe are provided in the Rationale' section of Forster and Lilliestam (2010). Sheldon (2008) and Sovacool & Sovacool (2009) try to identi fy areas in the U.S. where water withdrawals for power generation may be at risk to drought. Sheldon (2008) overlays average water withdrawal rates at select power plants in the U.S. onto a drought severity map (Palmer Drought Severity Index; Fig. D2) offe ring a snapshot of where power generation may

PAGE 20

4 be at risk to drought. Similarly, Sovacool & Sovacool (2009) identify 22 counties in the U.S. where power generation is most at risk due to the combination of population growth, summer water deficits and planne d thermoelectric capacity additions. Sovacool and Sovacool (2009) offer a full county by county assessment of the entire U.S. A significa nt limitation, however, is their assumption that "the electricity for a given county comes from within that county and stays there [and] ignores the possibility of electricity imports and exports between counties". This simplification results in underestimation of supply chain risks to surrounding counties given that 93% of U.S. counties have little to no endogenous power generation capacity and thus must import electricity from surrounding areas (Cohen and Ramaswami 2014 ). Kruse et al (2004) evaluate potential (e.g. modeled) impacts on power plants due to minimum water releases from Gavins Poi nt [reservoir] to the Missour i R iver. This study examines effluent temperature and insufficient flow as potential constraints to power generation at 15 power plants along the Missouri River. Curlee and Sale (2004) provide a general overview of the interdependencies between water and energy production in the U.S, including thermoelectric generation, oil/gas extraction, hydropower, inland water navigation and the carbon economy. EPRI (2006) presents a detailed framework to evaluate water demands and availability for power generation in U.S. watershed s The culmination of the EPRI Framework is a water balance model for select watersheds with the ability to examine time series trends of available streamflow at power plant locations compared to theoretical flow requirements. The available s treamflow for use in power generation is computed as total streamflow less competing uses and instream requirements. Stillwell et al (2011a) study water energy relationships in Texas. They explore water requirements of electricity generation and electricity requirements of potable water/ wastewater treatment using a combination of national average and state specific values. Of interest to the present literature review is the former the water requirements of electricity generation. In that

PAGE 21

5 regard, t he authors estimate evaporative water consumption in statewide electricity generation in Texas for the period 2008 to 2060 under four technology scenarios. It is unclear what assumptions are used to project statewide electricity demand, fuel mix and cooli ng technologies beyond 2018 key elements of the water use forecast. Looking ahead to a carbon capped future, Stillwell et al. (2011a) provide useful assumptions regarding carbon capture and sequestration (CCS). For a traditional pulverized coal power plan t retrofitted with CCS, to produce the same net power output as an equivalent plant without CCS, an additional 25 40% more power is required. This translates to 50 90% increase in cooling water, accounting for the parasitic use of steam in thermal cycling for the CO 2 scrubbers and electrical power for compressing the CO 2 for pipeline transport and subsurface injection (Stillwell et al. 2011a) Finally Stillwell et al (2011a) provide a qualitative assessment of current and proposed water/energy policy. Stillwell et al (2011b) builds on the premise that water shortages from drought or instream flow requirements may reduce the amount of water available for power plant cooling, introducing vulnerabilities to the power system. Offering a fresh perspective, t hey ask the question in re verse: can conserving water in the power sector through the use of dry cooling and hybrid wet dry cooling technologies improve volume reliability (% of total water demand met over a given period of time) to other water rights hold ers in Texas? An interesting question indeed. The analysis includes three scenarios: (1) No open loop cooling, (2) No open loop cooling plus implementation of hybrid wet dry cooling at all power plants, resulting in 50% water savings compared to scenario 1 and (3) Implementation of dry cooling at all power plants, resulting in 90% water savings compared to scenario 1. The three scenarios are evaluated using two existing water availability models: (a) the full execution of water rights, and (b) current cond itions reflecting actual diversions and return flows for 11 major river basins in Texas. Their findings suggest that use of dry cooling alleviates drought concerns for municipal supply and instream flow requirements. The work of Stillwell et al (2011b) wa s facilitated by existing Water

PAGE 22

6 Availability Models (WAMs) that contain spatially resolved water rights for 1 1 major river basins in Texas a unique asset. Tidwell et al. (2012) present a similar analysis as Sovacool and Sovacool (2009), wherein they esti mate the potential impact of water availability on future expansion of thermoelectric power generation. Tidwell et al. (2012) point to a move in the electric power indust ry towards closed loop cooling ( wherein withdrawals are greatly reduced but ev aporativ e consumption increases compared to once through cooling ) but are m otivated by consumptive water use as the limiting factor for future thermoelectric capacity expansion Tidwell et al. (2012) analyzes watershed level consumptive water use compared to physi cal streamflow. Their analysis employs a System Dynamics m odel with five interacting systems: 1. Demography ~ fn (pop, gross state product) 2. Electric power production ~ fn (demand, plant retirements, retrofits and new construction according to Annual Energy Ou tlook and Electricity Market Module) 3. Thermoelectric water demand ~ fn (fuel type and cooling type of power production according to EIA 767 and NETL Coal Plant Database) 4. Non Thermoelectric water demand ~ fn (municipal, industrial, agricultural and mining wa ter use according to USGS county level estimates) 5. Water supply ~ fn (USGS stream gauge statistics) Water availability is then computed as a function of consumptive demand in the watershed plus upstream consumption Effectively, it is a ratio of water demand to water supply, although it arguably includes double counting of consumptive uses depending on the location of the stream gauge. For example if water is consumed upstream of a gauge, then the stream gauge already reflects this and thus the analysis should only include consumptive uses downstream of the gauge to avoid double counting W ater availability calculation s also include several strong assumptions regarding seasonality of consumptive losses that are not visible in the annual level data. The a uthors

PAGE 23

7 exogenously impose a 12% increase in thermoelectric water demand to reflect higher demand and evaporation rates in the summer; a factor 1.5 increase in irrigation consumptive use to re flect an 8 month growing season; and a factor 1.5 increase in mun icipal consumptive use to reflect an 8 month outdoor water season. Consumption in all other use sectors was maintained at average levels. These may be useful rules of thumb when data at higher temporal resolution is not available. Modeling exercises are th en carried out to 2035 under three scenarios: (1) AEO's reference case, (2) Low Fossil Technology Cost Case and (3) Low Renewable Technology Cost Case. The principal difference between the three scenarios is the mix of fuels adopted in future power plant c onstruction. The results are presented in two shade intensity maps of the U.S., one for non thermoelectric water consumption and the second for thermoelectric water consumption, both at the county scale. The most striking feature of the maps is the highly dissimilar spatial pattern of thermoelectric versus non thermoelectric water demand. Non thermoelectric consumption is clustered around cities and agricultural areas, whereas thermoelectric consumption has a seemingly random pattern. Finally Tidwell et al. (2012) identify counties where future expansion of thermoelectric generation may be limited by water availability. Water scarce counties are identified as having an availability metric of 0.7 or higher, e.g. where demand accounts for 70% or more of the physical supply. Technology S pecific Assessments of Water Use in Power G eneration This category includes the works of Fthenakis & Kim (2010), Mielke et al (2010), Macknick et al (2011) and Mekonnen & Hoekstra (2012). Ft henakis & Kim (2010) apply life cycle analysis to evaluate water withdrawal and consumption factors for conven tional and renewable electric power plants. Similarl y, Mielke et al. (2010) provide upstream and operation

PAGE 24

8 phase water use for a wide range of energy carriers used in both tran sportation and electricity generation. Mekonnen and Hoekstra (2012) estimate the blue water footprint of hydro power by estimating reservoir evaporation rates. Their empirical analysis reveals a positive linear relationship between evaporative water consump hydroelectric capacity. Macknick et al (2011) present a meta analysis and literature review of water withdrawal and consumption factors for electricity generation, including both conventional and renewab le tec hnologies Dynamic Modeling of Water Withdrawal Constraints and Temperature Constraints to Power G eneration This c ategory includes the works of Koch and Vogele (2009) and Forester and Lilliestam (2010). Both Koch and Vogele (2009) and Forster and Lillies tam (2010) derive equations for modeling temperature and volume constraints to power generation and estimate the cost of associated power capacity reductions. Forster and Lilliestam show power generation reductions and associated costs for the specific cas e of a model nuclear power plant in central Europe with once through cooling for a range of temperature and streamflow scenarios. They represent the cost of capacity reductions with the price of spot market power purchases at 60 euro/MWh. Koch and Vogele (2009) develop equations for estimating thermoelectric cooling water demand and economic evaluation of lost generation capacity due to water constraints for an existing regional water management model. By coupling these water and energy models, it is poss ible to evaluate a range of scenarios including socio economic trends, climate change and technology adoption. Koch and Vogele (2009) estimate the cost of capacity reductions as a function of the market price of electricity less the variable costs of production. Of particular interest, the analysis shows seasonality in thermoelectric cooling water demand linked to temperature and humidity. They do not, however, explore empirical relationship s between climate

PAGE 25

9 and water withdrawal intensity factors as done i n Woldeyesus et al. (2013 ). Both sets of authors Koch and Voegel (2009) and Forester and Lilliestam (2010) provi de robust modeling frameworks that are similar in structure and may be adapted for use in other regional contexts or modeling scenarios. Future work can improve upon these models by considering on site water storage for recirculating cooling systems which are common throughout the industry and is shown to be significant in estimating water related risk to power generation (Woldyesyus et al. 2013 ).

PAGE 26

10 CHAPTER III THE WATER WITHDRAWAL FOOTPRINT OF ENERGY SUPPLY TO CITIES: CONCEPTUAL DEVELOPME NT AND APPLICATION TO DENVE R, CO Abstract This article presents a water withdrawal footprint for energy supply (WWFES) to cities, and places it in the context of other water footprints defined in the literature. Analysis of electricity use versus electricity gene ration in 43 U.S. cities highlights the need for developing WWFES to estimate risks to trans boundary city energy supplies due to water constraints. The magnitude of the WWFES is computed for Denver, Colorado and compared to the city's direct use of water to offer perspective. The baseline WWFES for Denver is found to be 66% as large as all direct water uses in the city combined (mean estimate). Minimum, mean and maximum estimates are computed to demonstrate sensitivity of the WWFES to selection of water wi thdrawal intensity factors. Finally, scenario analysis explores the impact of energy technology and energy policy choices in shaping the future water footprint of cities 1 The Need for Water Footprints of Urban Energy Systems The urban metabolism literatu re contributes significantly to our understanding of water and energy flows to cities. Yet, in the context of water, these studies typically account only for direct water flows across city boundaries (e.g. Kennedy et al. 2007, Jenerette and Larson 2006, Je nerette et al. 2006; Luck et al. 2001), and do not consider water embodied in energy carriers entering cities. While important for other purposes, such focus on direct in boundary water use does not bring to light the indirect dependence of city energy sys tems on the use of water outside the city boundary to meet city energy demand. By quantifying the direct link between energy use in cities and the associated use of water elsewhere to meet that demand, it becomes clear that 1 This work has been published in the Journal of Industrial Ecology (Cohen and Ramaswami 2014).

PAGE 27

11 urban energy supplies may be vul nerable to climate extremes such as drought and heat waves occurring well beyond the city boundary. Over the past decade, this phenomenon has been well documented in the electric power sector in particular. For example, high stream temperatures and low streamflows limited thermoelectric power generation in the southeastern USA in 2007 and 2010 (NETL 2009; Fleischauer 2010), France in 2003 and 2009, and Germany and Spain in 2006 (Forster and Lilliestam 2010). In these instances, quantifying transboundary water requirements of city energy use can elucidate the extent to which cities may be vulnerable to drought not only in terms of direct water supply (e.g. Padowski and Jawitz 2012), but in terms of energy supply. This requires a better understanding of ene rgy carrier flows serving cities, as well as the water embodied in the supply of these energy carriers. The gap in understating water embodied in energy supplies to cities is also seen in the energy literature. For example, Sovacool & Sovacool (2009) ident ified 22 counties in the U.S. where power generation is most at risk due to the combination of population growth, summer water deficits and planned thermoelectric capacity additions. However, the study presumes that " electricity for a given county comes from within that county and stays there [and] ignores the possibility of electricity imports and exports between counties" (Sovacool and Sovacool 2009, pg. 2766). This simplification results in underestimation of supply chain risks to surrounding counties given that 93% of U.S. counties have little to no endogenous power generation capacity and thus rely on electricity imports from surrounding areas (eGRID 2010). Thus water supply chain risks to power generation in one county can have impacts on surrounding counties as well. In both the urban metabolism literature and the water energy literature there is a need for an analytic tool to represent the dependence of urban energy supply on water resources outside the city boundary, with explicit consideration of transboundary energy infrastructure. The purpose of this paper is to develop suitable water footprints of urban energy supply to support policy/planning at the water energy nexus in cities, and place this footprint in the context of other

PAGE 28

12 water footprints discussed in the literature. Developing water footprints of city energy systems enables cities to consider the potential impact of factors such as drought and climate change on their energy supply, whether that energy is produced within or outside the cit y boundary. Such water footprints also serve a useful communication vehicle for the general public by demonstrating that reducing energy use, and/or the water intensity of energy production, conserves water. Different types of water footprints of city ener gy supplies can be computed. We posit that water withdrawal footprints address operational risk to the production system, i.e., operational risk to power plants when streamflow is not adequate (in terms of quantity or quality) to meet water requirements fo r cooling and other critical operations in power generation. The importance of water withdrawals in characterizing operational risk to thermoelectric power generation can be seen by the U.S. Department of Energy report that focuses on the location of power plant water intake pipes in streams (NETL 2009), indicating withdrawal during low stream flow condi tions is a concern. See Figure III 1 for a visual illustration of low streamflows adjacent to a power plant. Discussions in India have also indicated that e lectric utility operators are concerned about water withdrawal limitations during periods of low streamflow (NTPC, 2013). Thus water withdrawal footprints address operational risk to energy producers. Water consumption (e.g. evaporative loss) footprints ar e also important in the context of watershed scale planning among multiple, competing users (e.g. power generation, agriculture, etc.), with consideration of return flows. However, in the context of city energy supply, water withdrawals directly address op erational risk posed by water constraints. Thus, this paper develops a water withdrawal footprint for energy supply (WWFES) to cities. The WWFES can serve two policy purposes in the context of sustainable cities, as shown in Figure III 2. First, for cities that own and operate their own water and energy utilities, the WWFES supports energy resource planning to minimize risk to power generation due to water

PAGE 29

13 supply shortages. For example, Woldeyesus and colleagues (2013) compute water withdrawals for power ge neration at a municipally owned utility in Colorado Springs, Colorado (Supplementary materials Fig. S1 ), and associated risk to power generation due to low stream flow. To manage such risk, electric utilities consider power generation from a portfolio p erspective of grid connected power plants that are dispatched and managed within a power control area (PCA) or other load balancing authority (NERC 1992). When any one power plant's generation is constrained by insufficient water supply, generation can be shifted to other power plants in the portfolio based on marginal least cost and marginal water requirement criteria. Woledeyesus and colleagues (2013) have applied such portfolio modeling using the MARKAL TIMES model, and found it useful to support both lo ng term energy resource planning in the case of extreme water scarcity, as well as co management of water and electricity for a municipal utility (the latter is ill ustrated conceptually in Fig. III 2a ). For cities that are served by larger, investor owned utilities (such as Xcel Energy, which serves Denver and other parts of Colorado), the footprint connects energy users in cities to the production system, whose operational risk can be estimated using the proposed WWFES concept. Transparency about producti on system risk can stimulate local governments to influence the water intensity of their energy suppliers by leveraging the city's purchasing power vis a vis long term contracts. This is already happening in the electricity sector with respect to GHG emiss ions factors and climate action planning. For example, cities such as Boulder, Colorado are influencing energy suppliers to reduce their GHG emissions intensity to help the city achieve climate action targets (Gallucci 2013; Jaffe 2013). Thus, for cities t hat are served by large electric utilities, the WWFES connects the use system to the production system, and can serve a risk communication p urpose, as illustrated in Fig III 2b

PAGE 30

14 Figure III 1 Water L evels D rop at a Large Coal Fired P ower Station in East Texas During Summer 2011 D rought. [Water levels drop at a large coal fired power station in East Texas during summer 2011 drought. MW=Megawatt; MLD=Million liters per day. Photo by Kevin Green/Copyright Long view News Journal, used by permission]

PAGE 31

15 Figure III 2 Illustration of the WWFES concept [ Illustration of the WWFES concept for (A) For a city that supports local energy demand with in boundary water resources, and (B) a city that relies on trans boundary energy infrastructures and the use of water outside the city boundary to support local energy demand. WWFES = Water withdrawal footprint of energy system; MEFA = Material/energy flow analysis ] Object ives The purpose of this paper is to articulate the water withdrawal footprint concept for a city's energy supply. Specifically: 1. Articulate a water withdrawal footprint for energy supply (WWFES) to cities and place it in the context of the existing literature on urban environmental footprints; 2. Conduct an electricity import analysis to estimate the proportion of cities in the U.S. that have significant electricity imports whose supply chain risks are not addressed using currently available analytic to ols; 3. Conduct a case study of Denver, CO, to compare the magnitude of the water withdrawal

PAGE 32

16 footprint of the city's energy system to its direct use of water (e.g. municipal supply). Drought vulnerability assessments for cities typically focus on direct water use only and ignore indirect water uses such as those embodied in critical energy flows serving the city. Sensitivity of the water footprint of the energy system to technology and policy choices in the energy sector is also explored. Conceptual D e velopmen t of the WWFES and its P lace in the W ater F ootprint L iterature The WWFES incorporates the use of electricity, heating fuels and transportation fuels in a city's residential, commercial and industrial sectors, and traces the supply chain for producing these energy carriers within and across city boundaries (hence trans boundary). The WWFES is computed as the product of community wide use of energy carriers and the life cycle water withdrawal intensity factor for producing these energy carriers. Fig III 2 ill ustrates how the WWFES can be used to communicate risk to city energy systems. In the next section we place the WWFES in the context of existing water footprints defined in the literature. Territorial Accounting of Direct Water Use in Regions Regional wa ter balance models typically address direct water use within a certain boundary and do not consider water embodied in energy, food or materials entering/leaving the system boundary. Territorial accounting and regional water balance models are crucial to wa tershed level planning. In the U.S., the largest data source for territorial accounting of direct water use is the USGS, which reports water withdrawals and consumption in eight sectors public supply, domestic, commercial, industrial, thermoelectric, irrig ation, mining and livestock with sector classifications changing somewhat over time. The most recent survey of direct water use in all sectors at the county level, including both water withdrawals and consumption, was conducted in 1995 (USGS 2005). Notably domestic, commercial and industrial sectors increasingly rely on water deliveries from public supply rather than self

PAGE 33

17 supplied withdrawals (USGS 1995). Territorial accounts report water withdrawals for thermoelectric generation if it occurs within the re porting boundary (geographic or hydrologic); they do not, however, account for water embodied in electricity produced elsewhere. As referenced previously, a large proportion of U.S. cities and counties import electricity and other energy carriers to suppor t production activities (eGRID 2010), making water embodied in the energy system an important missing consideration in territorial accounting of water use at the spatial scale of cities. Equivalent Land Area Footprints for Direct Water Use Some authors have taken the territorial accounting of direct water use within a region and then computed the equivalent land catchment area that would be required to obtain the same amount of water given local rainfall and evapo transpiration rates (see Luck et al. 200 1; Jenerette et al. 2006). These catchment area footprints show large land areas needed to capture rainfall in arid areas to meet direct water demand. Such methods help illustrate where direct water use exceeds local renewable supply, but miss the contribu tion of water embodied in energy imports critical to economic activity in the region. Economic Consumption Based Water "Loss" Footprints for Regions The water footprint concept was first developed and popularized by Hoekstra and colleagues (see Hoekstra a nd Hung 2002, 2005; Hoekstra and Chapagain 2007a, 2007b) to demonstrate the amount of consumptive water use embodied in international agricultural trade. Their work builds on the insights of Allan (1998) who observed that water scarce regions import water intensive goods to effectively increase access to water resources. To quantify this phenomenon, Hoekstra and colleagues' water footprint assigns consumptive water use related to evapotranspiration, pollution dilution and direct incorporation into agricultu ral and livestock products, not to the region where they are produced, but to where they are finally consumed by

PAGE 34

18 households, governments and capital investments, referred to as economic final consumption. We refer to this method as consumption based water loss footprinting, to highlight the economics definition for final consumption and re naming consumptive water use as water loss to avoid confusion. Consumption based accounting requires detailed and accurate monetary trade data that is reliably available only at the national scale. Consumption based footprints make it possible to link economic final consumption in one country to water loss in another (see Hoekstra and Hung 2002, 2005) but do not address operational risk to energy production (see Fig III 1) Economic Production B ased Water Withdrawal Footprints Blackhurst and colleagues (2010) developed national average water withdrawal intensity factors (water use per dollar of economic output) for all 428 sectors in the 2002 U.S. economic input output table. This approach offers an elegant solution to computing supply chain water withdrawal footprints for U.S. goods and services. However, the authors advise against using this tool to represent the water footprint of production activities at smaller (sub national) geographic scales due to large regional variation in water intensity parameters. For example, grid average water withdrawals per unit of electricity supplied to Denver, Colorado in 2005 were estimated here to be ~2,200 liters of water per Megawa tt hour (mean estimate; Lw/MWh), compared to the national average of ~94,000 Lw/MWh (CMU 2013; EIA 2013). Thus other water footprinting approaches that utilize the same water withdrawal concept are needed to describe water supply chain impacts at the city scale. Infrastructure Supply Chain Water Footprint for Cities A new approach, suitable to the city scale, is the infrastructure supply chain method, first articulated for greenhouse gas footprinting by Ramaswami and colleagues (2008), and since replicated in many cities (e.g., Hillman and Ramaswami 2008; Kennedy et al. 2009; Baynes et al. 2011, Chavez et al. 2012). Applied to the analogous case of water footprinting, the infrastructure

PAGE 35

19 supply chain method estimates direct and indirect (embodied) water use in key infrastructure serving a community. In this article, we focus on water footprints of energy supply infrastructures, developing the WWFES for cities. As noted in the introduction, we posit the WWFES to be policy relevant because it brings to the fore ground water supply risks to transboundary energy infrastructure that cities depend on. The WWFES informs cities that their energy supply, like their municipal water supply, can be at risk during drought. The footprint explicitly connects users of energy i n the city to the production system as illustrated in Fig III 2a Table III 1 presents a side by side comparison of the various definitions of water footprints in the literature, providing clarity on the conceptual development of the WWFES for cities in th e context of other footprint definitions. The rest of this paper further reinforces the need for developing WWFES for cities by conducting an energy import analysis for a set of 43 US cities, and computes the WWFES of Denver's energy supply to do a magnitu de comparison with the public water supply.

PAGE 36

20 Table III 1 Types of Water Footprints of Regions Type of Water Footprint Applications & Policy Relevance Limitations Refs. Territorial accounting of direct water withdrawals in all sectors within a given boundary Regional water balance; Addresses water supply for direct water use. Omits water embodied in trans boundary infrastructures, e.g. electricity and fuel imports; Solution space limited to conservation of direct water use. Kenny et al. 2009; Sovacool & Sovacool 2009; USGS 1995. Same as above plus computes equivalent catchment land area needed to meet direct water use. Luck et al. 2001; Jenerette et al. 2006. Economic consumption based water loss footprints for regions Educates the consumer about water loss embodied in their agricultural and livestock imports. Excludes water embodied in non biomass energy; Does not address water withdrawal constraints to production. Hoekstra & Hung 2002, 2005; Hoeskstra & Chapagain 2007a,b. Economic production based water withdrawal footprints of U.S. goods and services Tracks water withdrawals for production across all U.S. industrial sectors. Cannot be accurately downscaled to city level due to regional variation in technology & infrastructure norms. Blackhurst et al. 2010 Infrastructure supply chain water withdrawal footprint for cities Tracks direct and embodied water flow s in key energy infrastructures; Adjusts for local technologies; Addresses water flow constraints on energy generation; Informs conserving water by conserving energy. Requires local knowledge of study area, including interviews w. infrastructure designers & operators. This article Methodology Electricity Import Analysis Electricity use data was gathered for 43 U.S. city counties (e.g. cities that are also counties) that have completed a community wide greenhouse gas (GHG) inventory according to ICLEI Local Governments for Sustainability draft protocol (ICLEI USA 2010, 2011). The comm unity wide GHG inventories contain annual electricity use in the residential, commercial

PAGE 37

21 and industrial sectors. These data were cross referenced with a comprehensive national database of electricity production at the county level (eGRID 2010) to determine the ratio of electricity use compared to endogenous (i.e. local) production in each county. The difference betwe en in boundary electricity use and in boundary production is net electricity imports. This analysis represents the full set of available data at the intersection of ICLEI USA's 275 member cities that have completed community wide energy use and greenhouse gas emissions inventories and the U.S. EPA's eGRID database of county level energy production data (ICLEI USA 2010; eGRID 2010). Using data from these 43 city counties, we compute the percent that import varying levels of electricity as a proportion of th eir total electricity use. As more city counties publicly report energy/GHG inventories, it will be possible to increase the sample size and power of this analysis. National data already show that only ~7% of U.S. counties contain power plants larger than 1 MW nameplate capacity (eGRID 2010), indicating that at least 93% import electricity from beyond their geographic boundary. Computing the WWFES The WWFES integrates concepts from urban metabolism, greenhouse gas accounting and life cycle assessment (LCA) to estimate the water footprint of a city's energy system in terms of aggregate freshwater withdrawals from surface water and groundwater. The WWFES is computed as the product of material energy flows that support community wide energy use and the lifecycl e water withdrawal intensity of those flows. Equation III 1 Computing the WWFES. The WWF methodology is general and can be applied to any city where community wide energy use data is available. The first step is a material/energy flow analysis, which reveals community wide use of

PAGE 38

22 energy carriers (Ramaswmai et al. 2008, Kennedy et al. 2009, Chavez and Ramaswami 2011, Baynes et al. 2011). Key energy flows include electricity, heating/process fuels (e.g. natural gas) and transportation fuels (e.g. gasoline, diesel, bioethanol, biodiesel and jet fuel) as well as any other regionally significant energy carriers such as biomass, LPG, CNG or kerosene. Methods for allocating transportation fuel use to a p articular city within a larger commuter shed (including jet fuel use at regional airports) are available from Hillman and colleagues (2011). Bioethanol use (which is rarely reported directly because it is blended with gasoline prior to sale) can be estimat ed as a percentage of gasoline use. Inclusion of other biomass based fuels such as cellulosic ethanol, biodiesel and synfuels may be important in future scenarios as production volumes increase. Collectively, this energy use data is represented by the MEFA use term in equation 1. The second step is a water life cycle assessment of each energy carrier identified in MEFA use For each energy carrier i life cycle water withdrawals are summed over the energy supply chain from point of production to point of use (city energy use system), and expressed in terms of an aggregate water withdrawal intensity factor (WWIF production ) in equation 1. Supply chain analysis often requires interviews or surveys of energy service providers (e.g. electric utility, natural gas u tility) and bulk commodity suppliers (e.g. corn growers and biofuel producers) to identify production locations and production pathways. Using this information, literature estimates of life cycle water use in the production of energy carriers can be tailor ed to reflect local conditions. Methods for computing lifecycle WWIF from the LCA literature are discussed next, taking care to incorporate local conditions. Sensitivity of the WWFES to Regional Variability and Energy Technology Production Pathways Adjusti ng literature WWIF estimates to reflect local conditions is essential for energy carriers that exhibit regional variability (as is the case for biofuels) or sensitivity to infrastructure

PAGE 39

23 design (as is the case for electric power generation). Chiu and colle agues (2009) demonstrated that national average water use estimates for bioethanol are not representative because of high regional variability related to feedstock irrigation practices and climatic differences. For example, the irrigation requirement of co rn ethanol produced in Ohio is estimated at five liters of water per liter of ethanol (Lw/Le), whereas the same produced in California has an irrigation requirement of 2138 Lw/Le. Similarly for electricity generation, the WWIF can vary by 2 orders of magni tude between thermoelectric power plants with different cooling systems (e.g. once through vs. re circulating), and up to 3 orders of magnitude between once through thermoelectric power plants and renewable energy systems that do not rely on a thermodynami c power cycle, such as wind and solar photovoltaic (Macknick et al. 2011; Mielke et al. 2010; Fthenakis and Kim 2010; NETL 2008). By contrast, the WWIF of conventional petroleum fuels exhibit relatively low spatial and process variability, and are indisti nct from national supplies, thus industry average WWIF for petroleum products are appropriate. Chiu and Wu (2011) report water use in the production of petroleum gasoline in the range of 3.4 to 6.6 liters of water per liter fuel (Lw/Lf) for U.S. onshore pr oduction, 2.5 to 5.8 Lw/Lf for Saudi Arabian onshore production and 2.6 to 6.2 Lw/Lf for in situ recovery of Canadian Oil Sands demonstrating the relatively narrow band of water use estimates for petroleum production pathways as compared to the order of ma gnitude ranges observed for bioethanol production and electricity generation. Thus we proceed to develop locally fine tuned WWIF for electricity, biofuels and natural gas, while retaining as reported WWIF for petroleum fuels. For electricity, a grid averag e WWIF Elec can be estimated as a function of primary fuel source, combustion technology and cooling system type of thermal power stations located in the power control area (e.g. load balancing authority) or larger regional grid interconnection. Fthenakis and Kim (20 10), NETL (2008), Macknick and colleagues (2011) and Mielke and colleagues (2010) all serve as good references on the water intensity of electricity generation.

PAGE 40

24 For bioethanol, the WWIF EtOH is determined as a function of feedstock type (e.g. corn grain ve rsus cellulosic), feedstock origin (e.g. grown in Iowa versus California) and irrigation requirement (e.g. rain fed versus irrigated), and to a lesser extent, the process for converting feedstock to fuel (e.g. dry mill refining for corn grain versus thermo chemical conversion for cellulosic feedstock). For further detail on water use in the production of fuel ethanol, refer to Wu and Chiu (2011), Wu and colleagues (2009), Chiu and colleagues (2009), Dominguez Faus and colleagues (2009) and Williams and colle agues (2009). In determining the WWIF NG for natural gas (both for use in buildings and electricity generation), regional prevalence of hydraulic fracturing versus conventional drilling is key. Regional estimates of water use for hydraulic fracturing based on industry data are now emerging, including Goodwin and colleagues (2012) and Chesapeake Energy (2011). For all literature based WWIF estimates, care must be taken to reconstruct life cycle water withdrawals and not just the water loss (e.g. evaporative ) component, which is often the focus of reporting. In certain cases, withdrawals and consumption are equivalent, such as in petroleum refining and biofuels refining. In these special cases, nearly all of the input water is consumed and none is available f or downstream uses, and thus consumption equals withdrawals from a water balance perspective. Finally, it is important to consider any policies, market forces or technology trends that may significantly alter the energy supply chain of a city, including t rans national pipelines for unconventional oil products (e.g. Canadian oil sands), a shift towards cellulosic rather than corn grain ethanol, a shift towards closed loop rather than open loop cooling at thermoelectric power plants, and higher penetration o f renewable electricity. The next section applies the generalized WWFES methodology to the specific case of Denver, Colorado to demonstrate the magnitude of the WWFES compared to public supply of water. We chose Denver because of the richness of available data, including a 7 year record of

PAGE 41

25 community wide energy use disaggregated by energy carrier and sector (Ramaswami et al. 2007). We compute a locally fine tuned baseline WWFES for Denver followed by scenario modeling and results. Applica tion of the WWFES to Denver, CO Community wide Demand for Energy Baseline community wide energy use in the residential, commercial and industrial sectors was obtained from Denver's 2005 GHG inventory (Ramaswami et al. 2008). The key energy carriers were electricity (6,333.2 GWh annual use), bioethanol (estimated at 5% of gasoline use for a tot al of 49.1 ML), petroleum fuels (1675.6 ML), and natural gas (41,846.1 TJ). See Table III 2. Computation of the WWIF production for each energy carrier is described next. Electricity Grid average WWIF Elec for Denver's electricity supply was computed as a fu nction of fuel type and cooling type of active power plants in the PCA (eGRID 2010). In the baseline year (2005) the mean estimate of WWIF Elec was 2168 Lw/MWh. The minimum and maximum estimates for that year were 1464 and 3518 Lw/MWh, respectively. Life cy cle water withdrawal estimates for thermoelectric production pathways include upstream water withdrawals for fuel extraction, dust suppression, cleaning/purification and transport to thermoelectric power plants, as well as operation phase withdrawals, most ly for cooling and ash handling. WWIF Elec for non hydro renewable energy technologies (e.g. PV and wind) include upstream device manufacturing and power plant construction, and are generally considered to have near zero operation phase water requirements (Fthenakis and Kim 2010). Hydropower, whic h contributes less than 2% of Denver's electricity supply in the baseline year, is assumed to have zero operational water withdrawals as the water requirement is fully in stream and non extractive. Special treatment of hydropower may be necessary for citie s that rely on this

PAGE 42

26 resource more heavily, but that remains outside the scope of this study. Biofuels Interviews with Colorado based bioethanol refineries confirmed the volume and feedstock origin of corn ethanol supplies to the Denver region. According t o the interviews, the vast majority of Denver bound bioethanol comes from one of three dry mill refineries located in Northeast Colorado. The three refineries report a combined annual production of 150 million gallons (MMG), of which 13 MMG is used by Denv er (FRE 2011; EAI 2011). To minimize the cost of transporting large volumes of low energy density feedstock, ethanol refineries in Colorado typically source feedstock from within a 150 mile radius (FRE 2011). Post refinery distribution of fuel ethanol with in Colorado is similarly constrained by the cost of truck transportation to less than 200 miles (FRE 2011). Based on supply chain data from these three refineries, the feedstock origin of Denver bound bioethanol is estimated to be 64% Colorado, 26% Nebrask a and 10% Kansas. Coupled with regional irrigation estimates (Wu and Chiu 2011) we arrive at an upstream WWIF EtOH ranging from 843 to 1433 liters of water per liter of ethanol (Lw/Le). An additional 3 Lw/Le is allocated for dry mill conversion of feedstock to fuel (Wu and Chiu 2011) (see Table 2). Cellulosic ethanol, which is not yet produced at commercial scale in the U.S., is projected to play a major role in future bioethanol use, as mandated by a federal renewable fuel standard ( EISA 2007 ). The WWIF EtOH Cell for next generation cellulosic ethanol is adopted from Williams and colleagues (2009) and is found to be in the range of 2.2 to 8.6 Lw/Le with a mean value of 4.9 Lw/Le. Natural Gas Natural gas use in Denver is grouped into two categories. The first is direct use of natural gas in buildings, including all domestic, commercial and industrial direct use of natural gas (e.g. for space heating, water heating and industrial processes), as reported in Ramaswami and

PAGE 43

27 colleagues (2008). The second is indirect use of natural gas through use of electricity (e.g. natural gas use in combustion turbine and combined cycle power generation). The amount of natural gas embodied in electricity is back calculated from the portfolio mix of Denver's electricity supply in a given year and a thermal efficiency of 45.8% for natural gas combined cycle power plants (Koomey et al. 2010). For direct use of natural gas in buildings, the use phase (e.g. combustion) is assumed to have zero water requirement, thus WWIF NG is based sole ly on upstream water use. For natural gas embodied in electricity, the combustion phase water requirement is determined according to the type of power plant (e.g. natural gas combined cycle) and cooling type (e.g. forced draft cooling tower), as described previously in the subsection on WWIF Elec Upstream water requirements for producing natural gas are assumed to be the same for both direct and indirect use of natural gas. Upstream water use in the production distribution of natural gas depends heavily on the production pathway. For conventional natural gas wells, water is used for lubrication and cooling during the drilling process, as well as for ancillary purposes such as dust suppression on unpaved roads. However, normalized over the expected lifetime production of the well, WWIF NG conventional is near zero (Mielke et al. 2010). By contrast, for multi stage hydraulic fracturing which is quickly overtaking conventional drilling as the preferred natural gas extraction method the life cycle WWIF NG frac is substantially higher. In Colorado, hydraulic fracturing is already the predominant natural gas extraction method, accounting for over 95% of all active wells in the state (COGCC, 2011). Previous studies have put the water requirement for hydraulic fracturi ng in the range of 2.3 to 6.8 liters per MMBTU (Lw/MMBTU) (Mielke et al. 2010; Soeder and Kappel 2009; Chesapeake Energy 2011). A more recent study, specific to Colorado (Goodwin et al. 2012), provides water intensity estimates for hydraulic fracturing bas ed on high, medium and low lifetime well production scenarios. These values range from 11.0 to 36.7 Lw/MMBTU for

PAGE 44

28 horizontal wells and 20.4 to 53.0 Lw/MMBTU for vertical wells. For use in this article, we assume a 50/50 split between vertical and horizontal hydraulic fracturing and adopt the minimum, mean and maximum water intensity estimates for extraction plus an additional 3.8 Lw/MMBTU for pipeline transport, as reported by Mielke and colleagues (2010). Petroleum Fuels Wu and Chiu (2011) offer an excellent analysis on the water intensity of transportation fuels that builds on both authors' previous work in this area. From their analysis, we adopt minimum, mean and maximum lifecycle estimates of WWIF Petrol based on net water use for well injection ( e.g. recovery) and refining. Net well injection water use is calculated as the production weighted average of injection water use for primary, secondary and tertiary oil recovery technologies, less the amount of produced water re injected for oil recovery. Analysis is based on data from PADD (Petroleum Administration for Defense Districts) Regions II, III and V, which represent 89.5% of U.S. onshore crude oil production (Wu and Chiu 2011). The use phase (e.g. combustion in an internal combustion engine) and end of life phase (e.g. as tailpipe emissions) are regarded to have zero additional water requirements. Table III 2 Material/Energy Flow Analysis (MEFA) and Life Cycle Assessment (LCA) Water Withdrawal Intensity Factors (WWIF ) for Denver. [Material/Energy Flow Analysis (M/EFA) and Life Cycle Assessment (LCA) Water Withdrawal Intensity Factors (WWIF) for Denver's Baseline (2005) Water Withdrawal Footprint for Energy Supply (WWFES) Showing Mean Estimates Only.] M EFA Use x WWIF Production,Total = WWFES Energy End Use Sector Energy Carrier M/EFA Use WWIF Production Contribution to WWFES in ML [% of total] Life Cycle Stage Total Units Refs. Upstream & Transport^ Energy Conversion* Electricity Coal 3.610 TWh 169 2743 2912 Lw/MWh (a) 10,512 [12.6%] NG 2.502 TWh 186 1092 1278 Lw/MWh (a, b) 3,197 [3.8%] Wind 0.095 TWh 230 0.0 230 Lw/MWh (a) 22 [0.0%] PV 0.000 TWh 1996 0.0 1996 Lw/MWh (a) 0 [0.0%] Buildings NG 41.846 PJ 23.7 0.0 23.7 Lw/GJ (b) 992 [1.2%]

PAGE 45

29 Transport. Gasoline 982.2 ML 3.75 1.5 5.25 Lw/L (c) 5,157 [6.2%] Diesel 365.5 ML 3.75 1.5 5.25 Lw/L (c) 1,919 [2.3%] Jet Fuel 327.9 ML 3.75 1.5 5.25 Lw/L (c) 1,721 [2.1%] Corn EtOH 49.1 ML 1219 3.0 1222 Lw/L (c) 60,000 [71.8%] Cell. EtOH 0.0 ML 0.0 4.9 4.9 Lw/L (d) 0 [0.0%] TOTAL 83,545 [100%] *Energy Conversion (EC) refers to the transformation of primary energy to end use energy. ^Upstream (U) refers to all life cycle activities preceding the energy conversion step. (a) Fthenakis and Kim 2010; (b) Goodwin et al. 2012; (c) Wu and Chiu 2011; (d) Williams et al. 2009 NG = natural gas; PV = photovoltaic; EtOH = fuel ethanol; TWh = terawatt hour; PJ = petajoule; ML = million liters; Lw/MWh = liters water per megawatt hour; Lw/GJ = liters water per g igajoule; Lw/L = liters water per liter fuel. WWIF Production reflects U and EC life cycle activities in the energy supply chain serving Denver, as described below: Coal : (U) Surface mining, benefication, train transport and plant construction; (EC) Forced draft cooling tower. NG : (U) On shore extraction from shale formation via hydraulic fracturing plus pipeline transport; (EC) Combined cycle power generation with forced draft cooling tower (for electricity only). PV : (U) Manufacturing of multi crystalline silicon modules, mounting frame and balance of system. Wind : (U) Manufacturing of wind turbines and construction of wind farm. Gasoline, Diesel & Jet Fuel : (U) Production weighted avg. of U.S. onshore crude recovery technologies (e.g. primary, secondary a nd tertiary production); (EC) U.S. refinery avg. Corn EtOH : (U) Production weighted avg. of feedstock irrigation requirements; (EC) Dry mill refinery. Cell. EtOH : (U) Non irrigated feedstock; (EC) Avg. of thermochemical and biochemical conversion technolo gies Scenario Modeling The water withdrawal footprint of Denver's energy supply was computed for a baseline year (2005) and two supply side scenarios for 2030 business as usual (BAU) and increased penetration of renewable energy (RE). Both the BAU and RE scenarios (henceforth BAU 2030 and RE 2030) assume the same amount of total community wide energy demand in 2030. Supply characteristics, such as fuel mix, production pathways and energy conversion technologies, are what differ. Demand side strategies a re not considered in these scenarios given the focus of this section on understanding the effect of technology choices on water embodied in community wide energy use.

PAGE 46

30 In estimating the WWFES out to 2030, sectorwise demand growth is modeled separately from changes in the fuel mix and changes in the production pathway of that sector. Demand growth is extrapolated from observed trends in Denver's per capita disaggregated energy use from 2000 to 2007 (Ramaswami et al. 2007), coupled with official estimates of p opulation growth (SDO 2012). Model assumptions related to changes in the fuel mix and production pathway within each sector (e.g. supply side characteristics) are described next. Electricity Both the BAU and RE scenarios assume a transition away from coal for Denver metropolitan power plants as mandated by existing Colorado legislation (Clean Air Clean Jobs Act 2010). The current grid mix serving the area is approximately 57% coal, 39.5% natural gas, 2% hydro and 1.5% non hydro renewable energy (mostly wind ) (GEO 2010). The BAU scenario assumes a transition towards more natural gas and less coal, as is already under way with three small coal fired power plants in the Denver metropolitan area scheduled to close over the next few years and Denver's largest coa l plant to transition towards natural gas combined cycle while expanding capacity to make up for the decommissioned coal at neighboring plants (Xcel Energy 2011). Thus the grid mix in BAU 2030 will be 31.5% coal, 65% natural gas, 2% hydro and 1.5% non hydr o renewable energy. In contrast, the RE 2030 scenario assumes a near complete phase out of coal, but instead of relying fully on natural gas, 30% penetration of non hydro renewable electricity (mostly wind) is assumed. This is consistent with a successful Colorado ballot initiative (Renewable Energy Standard 2004) that calls for 30% renewable electricity generation by 2020 for investor owned utilities, with a 3% carve out for distributed generation (e.g. rooftop PV). Thus the grid mix in RE 2030 will be 5% coal, 65% natural gas, 27% wind, 3% PV and 0% hydro.

PAGE 47

31 Transportation Fuels The RE scenario assumes displacement of 16.8% of all petroleum based gasoline with biofuels (on an energy content basis) as mandated by the federal renewable fuel standard (RFS) ( EI SA 2007 ). Due to the highly commoditized nature of gasoline, it is assumed that the share of gasoline displaced by biofuels will be the same in Colorado as for the U.S. as a whole (e.g. 16.8%). The BAU scenario, by contrast, assumes that biofuel penetratio n in the gasoline market will remain constant at 2007 levels (just prior to the RFS taking effect) at 4.8% on an energy content basis (EIA 2010). Diesel and jet fuel use are assumed to remain unchanged between the BAU and RE scenarios. Buildings Natural Ga s Natural gas use in buildings is unchanged between the BAU and RE scenarios for the year 2030, both following observed annual per capita decreases in natural gas use in buildings as older buildings are retrofitted or replaced with newer ones (Ramaswami et al. 2012 a ). Additionally, the WWIF NG Frac for natural gas use in buildings is assumed to remain constant from 2005 to 2030 since nearly all Colorado natural gas production already employs the more water intensive hydraulic fracturing production pathway. Recall that both BAU 2030 and RE 2030 reflect extrapolation of current trends in per capita energy use based on existing demand side management in the energy sector, but do not incorporate more aggressive policies that may stimulate further reductions in e nergy use such as carbon taxes or ordinances that require all homes be retrofitted in 5 years (Ramaswami et al. 2012 a ); such policies are beyond the scope and objectives of this article. Municipal Water Supply Over the course of a decade, Denver's per capita municipal water use decreased from 202 gallons per day (gpd) in 2000 to 132 gpd in 2010 (the most recent year where data is available;

PAGE 48

32 Denver Water 2010). The sharp decline in per capita water use was catalyzed by a severe drought in 2002 that requi red emergency conservation measures throughout the city, and although those restrictions have since been lifted, conservation behaviors and low flow water fixtures have remained. Going forward to 2030, we assume per capita water use remains constant at cur rent levels of 132 gpd to reflect official conservation goals and projections by Denver's water utility (Denver Water 2011). In a detailed study on residential water use trends across North America, Rockaway and colleagues (2011) conclude "Although there h as been a clear trend of declining residential customer water use over the past 25 years, this trend may be flattening over the next 20 years". This supports our relatively simple assumption that per capita water use in Denver after falling significantly o ver the last decade will remain flat over the next 20 years. Results Electricity Import Analysis In Figure III 3 shows the percentage of cities importing electricity at different levels (based on a set of 43 cities where both electricity production and electricity use data are available). About 60% of the cities import more than half (>50%) of their electricity. Nearly 80% of cities rely on electricity imports to some extent. The remaining 20% of cities are net producers, meaning they export electricity to neighboring areas. These results emphasize that trans boundary flows of water embodied in electricity are significant and should not be ignored when computing a water footprint for cities or other small geographic regions.

PAGE 49

33 Figure III 3 Trans boundary Electricity Imports for 43 U.S. Cities The Aggregate WWFES for Denver, Colorado The aggregate WWFES for Denver, Colorado is shown in Figure III 4 with side by side comparison to Denver's municipal supply not to suggest direct competition between the two but to provide a useful magnitude comparison. In the baseline year (2005), the aggregate WWFES is estimated to be 44% to 84% as large as the city's direct use of water, with a mean estimate of 64%. This indica tes that the total amount of freshwater required to meet city energy demand is of the same scale as direct water use in the residential, commercial and industrial sectors, combined. This is a key take away message considering that many cities already have drought action plans in place for the municipal water sector (e.g. restrictions on lawn watering and car washing) but not for the energy sector (e.g. restrictions on household or business electricity use). Looking ahead, we see that for BAU 2030, the WWFES remains nearly unchanged when compared to direct water use retaining a lower bound around 45% and an upper bound around 87% of the size of direct water use. By contrast, for RE 2030, the range of estimates of the WWFES shifts upwards, with min, mean and m ax values of 54%, 78% and 101% the size of

PAGE 50

34 direct water use, respectively. Percent contribution to the WWFES from electricity is markedly lower in RE 2030 as compared to BAU 2030 (10% versus 20% when comparing mean estimates), whereas the reverse is true for corn ethanol. Corn ethanol is responsible for nearly 82% of the aggregate WWFES in RE 2030, compared to 71% in BAU 2030. Reductions in electricity related water withdrawals result from displacing water intensive coal steam power plants with less water intensive technologies such as natural gas combined cycle, and wind. However, these savings are more than offset (in an aggregate sense) by increases in corn ethanol related water withdrawals. Recall that in the WWFES, water withdrawals are aggregated acro ss both space and time and thus water savings are not commutative in a physical sense, but rather from a footprint perspective only. In terms of policy, it is important to note that while water withdrawals indeed increase with increased penetration of biof uels, the extent is greatly attenuated by focusing future production on next generation cellulosic ethanol rather than corn ethanol (as already mandated in the federal renewable fuel standard). Here, WWIF are illustrative: WWIF EtOH Cell is similar in magni tude to WWIF Petrol indicating that displacing petroleum fuels with non irrigated cellulosic ethanol will have relatively little net effect on aggregate water withdrawals. The same is obviously untrue for corn ethanol, with WWIF EtOH Corn being 200 300 times as large as WWIF Petrol (see Table III 2). This illustrates an important use of the WWFES to inform the development of GHG mitigation policies that avoid shifting the burden to water resources.

PAGE 51

35 Figure III 4 Water withdrawal footprint of energy supply (WWFES) compared to direct water su pply for Denver, Colorado, USA. Min/mean/ max estimates provided for baseline (2005) and two scenarios (BAU 2030; RE 2030). Dotted line denotes mean estimate of WWFES for each scenario Scenarios: Baseline : Grid (57% coal, 40% NG; 3% RE); Fuel (95% oil, 5% corn EtOH) BAU 2030 : Grid (31% coal, 65% NG, 4% RE); Fuel (95% oil, 5% corn EtOH) RE 2030 : Grid (5% coal, 65% NG, 30% RE); Fuel (83% oil, 7% corn EtOH, 10% cellulosic EtOH) Abbreviations : BAU = business as usual; RE = renewable energy; BLY = billion liters per year; EtOH = fuel ethanol; NG = natural gas; Grid = electricity grid mix; Fuel = vehicle fleet fuel mix Conclusion This article articulates a novel water withdrawal footprint of city energy supply and

PAGE 52

36 places it in the context of other water footprints. While water consumption footprints (which focus on evapo transpiration and account for return flows) are important to watershed level planning among mu ltiple water users, water withdrawal footprints address operational risk from a producer perspective. In particular, a water withdrawal footprint can be applied to understand how operational risk to thermoelectric power generation not only for a single pow er plant, but for a portfolio of grid connected plants can be linked back to the city. By focusing on embodied water withdrawal requirements to meet city energy demand, the WWFES connects energy users in cities to their energy suppliers. The WWFES takes a life cycle perspective on the water requirements of city energy demand, including upstream and operation phase water withdrawals occurring both within and outside the city boundary. Electricity import analysis showing ~60% of cities in our 43 city sample i mported more than half their electricity highlights the need for cities to develop transboundary water footprints of their energy systems. These results suggest that transboundary flows of water embodied in electricity are significant and should not be ign ored when computing a water footprint for cities or similarly small geographic regions. A case study of Denver, Colorado, USA reveals that aggregate water withdrawals required to meet city energy demand are 66% as large (mean estimate) as the city's direct use of water for the year 2005. Projections to 2030 show that the magnitude of the WWFES remains nearly constant as compared to direct water use under a business as usual scenario, but grows to 81% as large as direct water use (mean estimate) in a renewab le energy scenario. From a footprint perspective, reductions in life cycle water use in the electricity sector are offset by increased penetration of biofuels in the transportation sector. For cities that operate their own water and electric utilities, th e WWFES can be applied directly to help inform long term co management and infrastructure investment. For cities that are served by large investor owned energy utilities, the WWFES broadens the decision making

PAGE 53

37 space for water conservation by including cons ideration of water embodied in energy supply. Potential risk to city energy supply due to water shortages is also made visible by connecting energy users in cities to water use in the energy production system. More broadly, the WWFES contributes to our und erstanding of the full scale and scope of water requirements to meet urban energy demand, and the role of infrastructure design and policy options in influencing the magnitude of those water requirements. The WWFES provides a conceptual framework for trans lating multiple and multi scale water resource constraints to a city's ability to produce or import energy, and vice versa, linking changes in a city's energy use and energy infrastructure to demand for water withdrawals within and across watersheds and ju risdictional boundaries. Future work will build on this framework by incorporating seasonal variability of both energy and water use patterns (e.g. Woldeyesus 2012) and the role of adaptive responses in times of drought or exigency.

PAGE 54

38 CHAPTER IV S PATIALLY AND T EMPORALLY D ELINEATED W ATER F OOTPRINT S FOR T RANS B OUNDARY E LECTRICITY S UPPLY TO C ITIES: EXPLORING V IRTUAL WATER TRANSFERS T O DELHI V IA T HE NORTHERN INDIA PO WER GRID Introduction to Embodied Water and Water Footprints Allan (1998) observed that water scarce regions in the Middle East and North Africa import water intensive agricultural commodities to compensate for lack of local water resources. By importing water intensive goods, a region effectively gains access to the use of water resourc es located elsewhere. While the amount of water physically embodied in imported goods may be small (e.g. the moisture content of rice), the volume of water that went into producing the imported goods (e.g. flooding a rice patty) may be vast. It is the amou nt the volume of water used to produce a good that is now known as embodied water. Hoekstra and colleagues (e.g. Hoekstra and Hung 2002, 2005; Hoekstra and Chapagain 2007a,b) built on the insights of Allan (1998) by computing the water requirements of imp ortant global food crops and then applying import export data to assess water embodied in international crop trade. The amount of water embodied in net imports, plus the volume of water used domestically to produce goods for domestic consumption, yields th e Water Footprint (WF) of a nation. The water footprint concept can also been applied to the analogous case of water embodied in transboundary energy infrastructure (Cohen and Ramaswami 201 4 ). Cohen and Ramaswami (2014 ) show that 93% of U.S. counties have little to no in boundary electricity generation (i.e. power plants) and thus rely on grid electricity produced outside the city boundary. Since energy (like food) is highly water intensive (e.g. NETL 2009), cities and other small geographical areas therefo re rely on external water resources to meet local energy demand.

PAGE 55

39 Application of the Water Footprint Concept to Urban Sustainability Life cycle analysis (LCA) is used to describe the environmental impact of a specific technology, process or product ( NREL 2 010 ). Material Energy Flow Analysis (MEFA) rooted in the study of urban metabolism quantifies the amount of materials and energy used by a particular population or community ( Hillman and Ramaswami 2010 ). Cohen and Ramaswami (2014 ) perform an urban MEFA to identify and quantify key energy flows serving cities (e.g. electricity, transportation fuels, heating and process fuels), coupled with LCA based estimates of the water intensity of those energy production pathways, to compute a water footprint of city energy use. Equation IV 1 Generalized Water Footprint Equation Urban environmental footprints often focus on a single metric or set of closely related metrics (e.g. water withdrawals and consumptio n; energy use and GHG emissions) both for simplicity and saliency. F ootprints have served as a n important communication tool informing communities of their environmental impact across scales from local to global ( Hillman and Ramaswami 2010 ). Chapter III of this thesis articulated how footprints can also be used for risk communication informing communities how they may be vulnerable to resource competition or production constraints outside their physical/political boundary, and thus outside their direct operation control/jurisdiction (Cohen and Ramaswami 201 4 ). For example, Scope 2 electricity purchases may constitute a significant chunk of a city's GHG footprint, but with major coal fired power plants increasingly relocated away from major population cen ters to reduce public health impacts ( House Bill 10 1365 2010), the footprint becomes increasingly important to connect energy users to trans boundary producers. Application of this concept to Denver, Colorado, USA, revealed that the water withdrawal footprint of the city's energy supply is on the same order of magnitude as all direct water use in the city combined (see chapter III of this thesis ). Thus the

PAGE 56

40 water f ootprint makes it possible to consider how city energy use impacts regional water resources, as well as the converse, how changes in regional water resources may effect city energy supply. To quantify such relationships, however, the water footprint must b e spatially disaggregated to link energy demand centers (e.g. cities) to points of energy production where water is a key input (e.g. cooling water at thermal power stations). To date, there are no such spatially delineated water footprints that link energ y demand centers to the use of water elsewhere. This article intends to fill that gap. Objectives 1. Delineate the share of Delhi's electricity supply that comes from within the city boundary (in boundary) versus outside the city boundary (trans boundary). 2. Es timate operational water withdrawals and consumption associated with in boundary and trans boundary electricity generation supplied to Delhi. 3. Benchmark the magnitude of the water footprint of Delhi's electricity supply to the magnitude of the city's munici pal water supply. Also evaluate on a per capita basis for perspective. 4. Spatially delineate the water footprint of Delhi's trans boundary electricity supply in terms of water withdrawals and consumption at specific power plant locations. 5. Explore seasonalit y of the spatially delineated water footprint in terms of both magnitude and geography. Study Design This article puts forth reproducible methods for estimating spatially and temporally delineated water footprints of city electricity use. Analysis focuses on identifying and interpreting salient features of the water footprint such as seasonality. Data and context are based on

PAGE 57

41 fieldwork in Delhi, India. For methodological details on how to compute the water footprint of urban energy demand, including quantifying material/energy flows and estimating water intensity of those flows for a variety of energy technologies, plea se see Cohen and Ramaswami (2014 ). The present article selects just one of the energy flows studied previously electricity and delves more deeply into two key areas: spatial and temporal delineation. Context As the capital of India and the second largest urban agglomeration in the world with a population of over 22 million (UN 2010), Delhi is a microcosm for global sustainability challenges including rapid population growth, rural to urban migration, and risin g standards of living, all in the context of natural resource scarcity and degradation ( Upadhyay 2011; Paliwal et al. 2006; CPCB 2006 ). An emerging piece to the puzzle is the complex relationship and inherent tradeoffs between water and energy resources. Delhi like much of India is undergoing a dramatic transformation in the way it uses energy, and how much. Between 2002 and 2012, power generation across India increased by 70%, growing 5.5% annually. Going forward to 2030, energy generation and consumptio n are expected to have a doubling time of 12 years (CEA 2012). Urbanization, commercialization and economic development have transformed former luxuries such as air conditioning into everyday necessities (Personnel Communication, 2013). Delhi in particular has one of the most extreme climates of any major city on earth, with daily summer temperatures in excess of 40 deg C. (104 F.), and heatwaves up to 50 deg C. (122 F.). Delhi's demand for energy could one day surpass per capita levels found in major weste rn cities that have mild summer temperatures by comparison. Fig. IV 1 highlights the relationship between energy demand and temperature for Delhi. The correlation between climate and energy demand will only strengthen as more of the population adopts air c onditioning and gains access to 24 hour electricity supply (Personal Communication with SLDC, 2013

PAGE 58

42 Figure IV 1 Temperature Energy Demand Relationship for Delhi, India Data Two types of data are required to compute the water footprint of urban energy supply energy use data disaggregated by fuel type and production pathway (MEFA), and the water intensity of each production pathway (LCA). Let's begin with the latter. Water Intensity of Power Generation : I n recent years, a wealth of research has emerged on the water requirements of energy production, especially for biofuels (see Dominguez Faus 2009; Wu et al. 2009; Chiu et al. 2009; Williams et al 2009; Wu and Chiu 2011), unconventional fossil fuels (see Mie lke et al. 2010; Goodwin et al. 2012) and electricity generation (see Fthenakis and Kim 2010; NETL 2008; Macknick et al 2011; Meldrum et al. 2013). The literature includes bottom up/process based LCA estimates (e.g. Williams et al. 2009), as well as top do wn estimates based on industrial sector data (e.g. Blackhurst et al. 2009; Wu and Chiu 2011). For electricity, there is a wide range of published estimates across many technology sub types, fuel types, cooling types and production pathways. To bring clarity and consistency, the

PAGE 59

43 U.S. National Renewable Energy Laboratory (NREL), performed a thorough review of the literature (Macknick et al. 2011), followed by a statistical harmonization of quality screened estimates (Meldrum et al. 2013). Values reporte d by NREL are adopted for use in this article (see Table IV 1 ). In all of the studies described above, the water intensity of electricity generation is time invariant. Woldeyesus et al. (2013) demonstrate seasonal variability in the water intensity of ele ctricity generation for a municipal utility in Colorado and derive empirical relationships between water intensity and climate variables such as ambient temperature and precipitation. Seasonal variability was previously identified by Rubbelke and Vogele (2 010), but no derivation was provided. In both cases, seasonal adjustments to the water intensity of electricity generation are not applicable to climates or conditions outside the specific study areas. A generalized equation rooted in thermodynamics, yet m indful of the limitations of data availability, is an important area for future research. Table IV 1 Operational W ater W it hdrawal and Consumption F actors for Power Stations S erving Delhi (gal/MWH) Fuel Sub Typ e Cooling Metric Min Median Max n** Coal PC Cooling tower Withdrawals 460 660 1200 21 Coal PC Open loop Withdrawals 15000 3500 57000 16 NG CC Cooling tower Withdrawals 150 250 760 16 NG CT NA Withdrawals 430 430 430 1 Hydro Generic NA Withdrawals^ 0 0 0 0 Nuclear All types Cooling tower Withdrawals 800 1100 2600 7 Coal PC Cooling tower Consumption 200 530 1300 20 Coal PC Open loop Consumption 71 140 350 11 NG CC Cooling tower Consumption 47 210 300 19 NG CT NA Consumption 50 50 340 3 Hydro Generic NA Consumption# 1425 4491 18000 3 Nuclear All types Cooling tower Consumption 580 720 890 9 Relevant values selected from Meldrum et al. (2013) and Macknick et al. (2011). ** n = number of literature estimates used to compute statistics. ^ Withdrawals for hydropower are considered zero because the use of water is fully in stream and non extractive.

PAGE 60

44 # Consumption for hydropower is based on reservoir evaporation rates. The range of estimates reflects different methods of allocating evaporatio n to electricity generation versus other uses such as water storage and flood control. Delhi Energy Use and Overview of the Indian Power Sector : Energy data for Delhi was compiled and cross referenced from a variety of official Government of India sources including the Central Electricity Authority (CEA), the Northern Region Power Committee (NRPC) and the Delhi State Load Dispatch Center (SLDC). To interpret Delhi's energy data, it helps to first understand the broader structure of energy supply and demand in India. As a basis for comparison, let's first look to the United States. In the U.S., 71% of ultimate electricity customers are served by a single, vertically integrated, investor owned power utility that owns/operates the full supply chain of electr icity infrastructure from power generation to transmission, distribution, load dispatch and end use metering. By contrast, the power sector in India is modular, with clear distinctions between generation companies, transmission utilities, load dispatch cen ters and distribution companies operating across scales. Collectively, we refer to these as electricity infrastructure operators ( EIO's ) applying terminology from the Social Ecological Infrastructural Systems (SEIS) Framework for the study of trans boundar y infrastructure (Ramaswami et al. 2012 b ). India is divided into five regional power grids N, NE, E, W and S with transmission between regions controlled by the National Load Dispatch Center (NLDC). The N, NE, E and W regional grids are synchronously inter connected, meaning that power flows freely between them as a function of load balance and subject to the transfer capacity of inter regional ties. The Southern regional grid, by contrast, is asynchronously connected to the four other regions via HVDC lines meaning that the volume and direction of power flow is controlled manually ( Barpanda 2013 ). Each of the five regions has its own Regional Load Dispatch Centers (RLDC), which assumes operational control from the NLDC at the regional periphery (boundary). RLDCs manage inter state transmission and load balance among States and Union Territories located

PAGE 61

45 within the region. In turn, the RLDC cedes operational control to constituent State Load Dispatch Centers (SLDC) at each State periphery. The SLDC then overs ees intra state energy exchanges among distribution companies (e.g. beneficiaries) and generating assets (e.g. power plants) located within the State. Finally, distribution companies manage local distribution, demand forecasting, scheduling and load sheddi ng within their service territory, subject to system constraints at the State, Regional and National levels. With a four tier system Central, Regional, State and Local coordination among hierarchical levels is key. For this express purpose, each region has a Regional Power Committee (RPC) that convenes various technical, operational and commercial coordination committees with major electricity infrastructure operators in that region represented by a permanent or rotating seat ( Pahwa 2013). In terms of gen eration, there are three major categories of power producers Central Generating Stations (CGS), State Generating Stations (SGS) and Private Generating Stations (PGS). In general, CGS enter into long term contracts (~25 years) to sell bulk power to benefici aries (e.g. states and distribution companies) located across the country. Likewise, beneficiaries maintain a portfolio of contracts with various CGS to hedge against price spikes and load shedding from any one supplier. SGS and most PGS, by comparison, se ll bulk power to the state in which they are located, and thus are considered that state's "own generation". In this article, we will use the term "own generation" synonymously with "in boundary" or "local" production. Analogously, we will refer to a beneficiary State's allocations from CGS as "trans boundary supply", as it is by definition coming across the state's periphery (boundary). Equations ( IV 2 IV 3, IV 4 and IV 5) show the mathematical relationships between various technical terms used throu ghout this article. Equation IV 2 Energy Requirement Requirement = Available + Load Shedding Equation IV 3 Energy Available Available = Own Generation + N et Drawal from Grid

PAGE 62

46 Equation IV 4 In Boundary Energy Generation Own Generation = SGS + PGS + dedicated CGS located in the State Equation IV 5 Trans Bounda ry Energy Supply Net Drawal from Grid = Scheduled Entitlement from CGS located outside boundary + Scheduled Bilateral Purchases (Imports) Scheduled Bilateral Sales (Exports) + Unscheduled Interchanges (UI) *Note that UI is the difference between scheduled and actual energy exchanges, where the schedule is set day ahead as per declared capacity of CGS. Actual energy exchanges reflect the portion of the entitlement from CGS and of bilateral purchases that a beneficiary chooses to accept on that part icular day based on day of changes in their own generation and/or requirement. Publically available energy accounting data at the state, regional and central levels of the Indian power sector is a tremendous boon to researchers offering a unique opportunity to track origin destination of bulk power transmission across scales. [Of course it is not possible (nor relevant) to track the flow of electrons, but rather the relationship between users and producers is key.] Figure IV 2 shows the power supp ly position (e.g. energy portfolio) of Northern Region states, highlighting seasonality of supply/demand characteristics and the relative contribution of each state's own generation versus reliance on trans boundary grid infrastructure. Figure IV 3 further distills salient information from Figure IV 2, showing the percentage of each state's total energy requirement met by trans boundary production. Across India, 75% of transmission capacity is reserved for long term contracts, such as between states and CG S. The remaining 25% is reserved for medium term (3 month to 3 year) and short term (1 day to 3 month) contracts. Particular to Delhi, long term contracts account for ~90% of total energy supply, with short and medium term contracts accounting for ~8%, an d unscheduled interchanges (UI) the remaining 2%. Of the medium term and short term contracts, at least half are day ahead purchas es from power exchange markets ( Sharma 2013). Medium term contracts fill a seasonal niche, such as moving excess summertime h ydroelectric capacity via the Northern Regional Grid from temperate Himalayan states to hotter, more population dense states below. Beneficiary states possess characteristic summer demand

PAGE 63

47 curves, for example, driven by air conditioning in urban areas such as Delhi, and by groundwater pumping in agricultural areas such as Uttar Pradesh, Haryana and Punjab. Short term contracts (e.g. day ahead purchases from power exchanges) can be used to satisfy any remaining unmet demand after both long term and medium te rm contracts have been fulfilled. On any given day, however, it is the prerogative of the beneficiary to accept or decline power from contractual suppliers based on cost. For example, a state may choose to decline readily available yet expensive power duri ng peak demand or fuel shortages if the additional cost is unlikely to be recovered from customers ( Sharma 2013). The difference between a state's energy requirement (e.g. demand) and the amount available to customers is defined as load shedding (e.g. pow er cuts). Figure IV 2 shows the requirement available profile of Northern Region states from April 2011 to March 2012, along with the amount of energy supplied from in boundary production (e.g. own generation) and trans boundary production (e.g. net drawal from grid), respectively. Figure IV 3 further explores the significance of trans boundary production in meeting state energy requirements. A wealth of information can be gleaned from Figures IV 2 and IV 3. First, we see that Delhi and Chandigarh (both U nion Territories; e.g. city states like Washington, DC) receive a majority of their power from trans boundary sources. In 2011 12, Delhi relied on net drawal from the grid to meet over half of its energy requirement throughout the winter and up to 74% in p eak summer months. In Chandigarh, reliance on transboundary energy production is virtually 100% all year around. This situation is emblematic of cities worldwide, which tend to rely heavily on transboundary infrastructures due to their small spatial scale relative to high demand for resources. On the other end of the spectrum are comparatively large and resource rich Himalayan states such as Himachal Pradesh (HP), Jammu and Kashmir (J&K), and Uttarakhand, which are net energy exporters in the summer when sn owmelt and monsoon rains ramp up hydropower generation. A third trend, which requires additional information to arrive at, but provides important context, is that peak energy demand in Punjab, Uttar Pradesh (UP) and Haryana

PAGE 64

48 correspond to peak summer irriga tion requirements. These three states have the highest irrigated acreage in India (State of Indian Agriculture, 2012), and the power requirement of groundwater pumping in particular has a noticeable effect on statewide energy requirement. The seasonal effe ct is attenuated in UP, perhaps because of other major industries that require power consistently throughout the year. Figure IV 2 Monthly Power Supply Position of Northern Region (NR) States

PAGE 65

49 Figure IV 3 Monthly Energy Requirement Met by Net Drawal from Grid Results Combining detailed energy use data for Delhi (described in section 4.2.2) with literature estimates of the water intensity of power generatio n (section 4.2.1) and applying equation IV 1, we computed a spatially and temporally delineated energy and water footprint for Delhi. The result is a geo referenced look at the water inputs to Delhi's energy system over one year. Adopting the language of Allen (1998), this represents virtual water transfers embodied in ele ctricity trade between regions. Spatial and Temporal Delineation Power station m etadata (e.g. fuel type, technology type cooling type ) were used to assign water intensity factors to each power station serving Delhi (gal/MWh e; see Table IV 1 ) The water footprint was then estimated ( bottom up ) using these water intensity values and monthly energy allocations to Delhi. Total water withdrawals and consumption at any power sta tion may be calculated as the summation of energy allocations to all beneficiaries times that station's water intensity factor Analysis and figures presented here represent allocations to Delhi only and thus a fraction of total water use. Figures IV 4 an d IV 5 show the in boundary and

PAGE 66

50 trans boundary water footprint of Delhi's electricity supply, disaggregated by source (individual power stations). Notably, in Figure IV 5 we see that one of the in boundary power stations (BTPS) has monthly water withdrawal requirements two orders of magnitude higher than the rest. BTPS the largest single power provider to Delhi is a 715MW coal fired station with an open loop cooling system that draws approximately 10^5 (1 lakh) cubic meters per hour from the Yamun a River (Agra Canal), according to plant operators ( NTPC 2013). Even though most of that water is returned to the river 10 deg. C. warmer but still available for downstream uses the immense water withdrawal requirement puts power generation at risk to low streamflow caused by dry spells, drought or upstream over appropriation ( NTPC 2013). In fact, insufficient water supply leads to partial de rating (decreased available capacity) and/or costly adaptive measures approximately 20 days per year, in recent yea rs ( NTPC 2013). Water scarcity is not the only water related concern, however. During monsoon season (June Sept), water turbidity and salinity can increase so high that the open loop cooling system must be shut off to prevent scouring in the condensers an d other equipment. This leads to an additional 10 days per year of reduced output or expensive adaptive measures such as switching to a closed loop system, which brings the withdrawal requirement down from 100,000 cubic meters per hour to 4000 cubic meters per hour. Not only is there a time and cost penalty for such switching (which is a unique adaptive capacity in of itself), but there is also a water consumption penalty. Closed loop cooling systems consume nearly all of the raw water input ( consumption~= w ithdrawals). Thus there is an inherent tradeoff between water supply risk to power plant s and risk to downstream users higher withdrawals put the power plant at risk, higher consumption means less available downstream. Throughout the electric power indust ry (both in the U.S. and India),

PAGE 67

51 there is a shift away from open loop cooling towards closed loop cooling to mitigate water supply risks to production ( NTPC 2013; NETL 2009 ). Returning to Figure IV 4, hydropower stations (HPS) have distinct signatures fro m thermal power stations (TPS). Water consumption at HPS peaks in the summer when reservoirs are full, and drops in the fall / winter when monsoon rains are gone and the catchment area returns to snow and ice. Recall that water withdrawals for HPS are consid ered zero because the use of water is assumed to be fully in stream and non extractive. For TPS in Figure IV 4 it is hard to discern the underlying pattern of withdrawals and consumption because the shape is an artifact of energy allocations to Delhi, not the full energy generation (and thus full water use ). However, by summing over all trans boundary virtual water transfers to Delhi we can reconstruct the true water footprint signature (Figure IV 6). Figure IV 6 shows the total in boundary and trans boundary water footprint of Delhi's electricity supply. Analyzing the data in this way, it becomes evident that the in boundary and trans boundary components of the water footprint of Delhi's electricity supply are fundamentally distinct. The in boundary w ater footprint is characterized by very high withdrawals driven by the use of open loop cooling at a single large power station (BTPS, described previously) while consumption is orders of magnitude lower. The trans boundary water footprint, by comparison, is characterized by high consumption driven by evaporation from large hydropower reservoirs as well as evaporation and drift from thermal power station cooling towers. Interpretation of these results suggest that Delhi's in boundary power generation may be vulnerable to insufficient or poor quality (e.g. high turbidity) water supply whereas trans boundary generation may be more concerned with total evaporative losses in accordance with watershed level agreements to ensure adequate supply among competing us es.

PAGE 68

52 Figure IV 4 Trans boundary Water Footprin t of Energy Supplied from Grid to Delhi, Disaggregated by Source (2011 12) Figure IV 5 In boundary Water Footprint of Energy Generation in Delhi, Disaggregated by Source (2011 12)

PAGE 69

53 Figure IV 6 Comparing In boundary vs. Trans boundary WWFES for Delhi (2011/12) S easonality The objective of s patial and temporal delineation was to understand the water footprint as dynamic in both space and time. Shifts in location, timing and magnitude of water withdrawals and consumption are driven by a multitude of factors at across scales. For example, on hourly and daily timescales, shifts occur according to load dispatch wherein certain power stations are ramped up and others ramped down to create the least cost generation mix. At slightly longer time scales (e.g. weeks, months) variations are driven by s easonal fluctuations in resource availability (e.g. hydropower), comparative fuel prices (e.g. use of natural gas peaking plants increases when prices are low) and scheduled maintenance outages. In the medium to long term (e.g. years to decades), variatio n is a function of new power plant construction and retrofitting of older plants, including changes in fuel type or cooling type. At the multi decadal scale are large technological shifts that minimize water withdrawal requirements, including adoption of c ooling towers, hybrid wet dry cooling and air condensed cooling systems at thermal power stations, and increased use of renewable energy technologies with near zero operational water withdrawal requirements such as wind and solar PV (Meldrum et al. 2013).

PAGE 70

54 Figures IV 7 and IV 8 spatially and temporally delineate the trans boundary component of the water footprint of Delhi's electricity supply, highlighting seasonal shifts in the location and magnitude of water withdrawals and consumption. Figure IV 7 illust rates variation between monsoon season (June Sept) and non monsoon seasons (e.g. dry seasons, Oct May). Figure 8 shows variation at three month time steps, by calendar season. A clear trend emerges, particularly in Figure IV 8, of increasing summertime wat er consumption associated with hydropower generation. Withdrawals and consumption associated with thermal power generation (both coal and gas) remain relatively constant. This pattern is a reflection of the changing composition of Delhi's power supply thro ughout the year (Figure IV 9). Figure IV 9 shows the monthly fuel wise contribution of trans boundary power generation to Delhi. Spatial delineation of the in boundary component of the water footprint has been omitted due to its comparatively small geograp hic scale and single water source, the Yamuna River. As such, the city's in boundary water withdrawals and consumption may be considered a point process. Figure IV 7 Monsoon vs. Non M onsoon S eason al Total W ater W ithdrawals and C onsumption R elated to T rans boundary Power Generation S upplied to Delhi

PAGE 71

55 Figure IV 8 Calendar seasonal total water withdrawals and consumption related to trans boundary power generation supplied to Delhi Figure IV 9 M onthly energy s upplied to Delhi from t rans boundary s ources

PAGE 72

56 Integrating Over Space and Time to Yield City Wide Annual T otals To better understand the water footprint of Delhi's electricity demand in the context of competing uses, we integrate over both space and time for o ne year to arrive at an annual total. Table IV 2 shows a summary of the water withdrawal requirement of Delhi's electricity system compared to life cycle water withdrawals embodied in all other major energ y carriers, including fuel use in the commercial, industrial and transportation sectors. A magnitude comparison with De lhi's municipal supply is provided to offer a sense of scale. The comparison does not imply direct comp eti ti on between water for energy and water for potable supply but rather provides a useful benchmark. Table IV 3 provides additional benchmarks through comparison with Denver. Finally, Table IV 4 provides a more detailed breakdown of the WWFES for Delhi Table IV 2 Water Withdrawal Footprint of En ergy Supplied to Delhi: Summary Table IV 3 Water Withdrawal Footprint Benchmarks: Comparing Delhi & Denver DELHI DENVER Energy Carrier WWF (ML/yr) WWF (L/p/d) WWF (%) Compare w. Muni. Supply WW F (ML/yr) WWF (L/p/d) WWF (%) Compare w. Muni. Supply Power 407,388 63.4 68.2% 35.6% 13,731 62.7 16.4% 12.6% Fuels 189,850 29.6 31.8% 16.6% 69,789 318.7 83.6% 63.9% Total 597,237 93.0 100 % 52.2% 83,520 381.4 100 % 76.4%

PAGE 73

57 Figure IV 10 Fuelwise Summary of Delhi WWFES (in l/p/d)

PAGE 74

58 Figure IV 11 Breakdown of Water Withdrawal Footprint of Energy Supplied To Delhi

PAGE 75

59 Discussion Urban metabolism is the study of material and energy flows used, transformed and produced by cities. Environmental footprints describe upstream and downstream impacts from the use of a particular material/energy flow by combining material energ y data from urban metabolism studies with LCA In this analysis, a water footprint was used to describe demand on local and trans boundary water resources from city wide electricity use occurring in Delhi, India for the year April 1 st 2011 to March 31 st 2012. Results from the footprint analysis can be used for a range of scholarly and practical applications. First and foremost, the footprint communicates to urban energy users their collective impact on local and regional water resources in terms of bill ions of liters of freshwater withdrawals and consumption per month. On a per capita basis, in boundary (e.g. local) water withdrawals from the Yamuna River range from 55 to 80 liters per person per day (lpd; mean estimates only), of which ~1 lpd is consume d. By comparison, official targets of direct (e.g. municipal) water supply to residents of Delhi is 150 lpd (Delhi Statistical Handbook 2012 Tables 1.27, 20.1, 20.2). Delhi Government considers the amount of piped water supplied to residents as "water cons umption" but in fact most of the water is returned to the Yamuna River in the form of wastewater roughly 60% receiving primary and secondary treatment via one of twenty sewage treatment plants located in the city, and the remaining 40% raw via open channel s. This means that the volume of water appropriated from the Yamuna River in Delhi each day for power generation is roughly half the volume appropriated for all residential and commercial direct uses combined. In terms of water consumption, power generatio n [given current use of open loop cooling at the largest power station] has only minimal impact as compared to municipal use, primarily due to differences in the extent of resource degradation. Water used in open loop cooling systems is degraded only insof ar as it is heated above ambient stream temperatures by ~10 deg. C (CEA Minimization of Water Use Report), whereas water used for domestic and

PAGE 76

60 commercial purposes clearly has significant biological and chemical loading. The use of water in trans boundary electricity production supplied to Delhi adds an additional 4 5 lpd of withdrawals from TPS, and 5 15 lpd in consumption, accounting for both evaporation and drift from TPS cooling towers and evaporation from hydropower reservoirs. Taken together, the in b oundary and trans boundary components of the water footprint of electricity supply to Delhi represent ~60 to 85 lpd of water withdrawals and 6 16 lpd of water consumption. In the context of a city where many residents routinely face seasonal shortages and daily rationing of both water and electricity supply, the tradeoff between water and energy is all the more apparent. Figure IV 12 shows where the Yamuna River the largest tributary of the Ganges ceases to be a river and becomes a wastewater drainage at th e Wazirabad barrage (weir dam) in North Delhi. For nine months each year (all but Monsoon season ) the remaining freshwater flow of the Yamuna River at Wazirabad is appropriated for municipal supply to Delhi (Delhi Jal Board 2013). Slowly the river is recharged by wastewater effluent as it flows through the city only to be dammed and diverted again for cooling at Badarpur Thermal Power Station (Fig. IV 13 ).

PAGE 77

61 Figure IV 12 The River Yamuna is Dammed and Diverted for Municipal Supply Figure IV 13 Re charged by Wastewater Effluent T hrough Delhi, the Yamuna River is Dammed and Diverted Again, This Time to S upply Cooling Water to a Large Power Station

PAGE 78

62 In addition to its utility as a communication tool, the footprint can be used by infrastructure designers, operators, and policy makers to identify water related risk factors to critical infrastructure. Uninterrupted electr icity supply is critical not only to economic activity and everyday conveniences, but also underpins the functioning of many other critical urban infrastructures including public transit (e.g. subway, railway), hospitals, water treatment and distribution, wastewater sanitation, building ventilation and cooling, information and communication technology, street lights, traffic lights and much more. By spatially and temporally delineating the water footprint of electricity supply, a new area of risk assessmen t becomes possible namely how changes in the availability, quality and timing of in boundary and trans boundary water resources may effect critical city services. Potential applications of a spatially and temporally delineated water footprint include: S eason ahead forecasting of regional water demand vis ˆ vis energy forecasting Load dispatch that incorporates water minimization and virtual water transfers between regions Identifying hotspots of water scarcity or water competition when applied in conjunction with drought severity maps or regional water balance models Long term water resource planning that considers allocations to multiple competing uses such as power generation, agricultural irrigation, municipal supply and industrial production Integrated scenario modeling with input from regional climate models and economic growth models that inform the future availability and competition for water resources.

PAGE 79

63 CHAPTER V FOOTPRINT INFORMED SOCIAL NETW ORK ANALYSIS Introduction Over three billion people living in cities worldwide depend on the steady supply of material and energy flows coming from outside the urban system boundary (UN Habitat 2013 State of the World's Cities). These resource flows including water, electricity, food, fuel and const ruction materials are supplied via trans boundary, engineered infrastructures such as pipelines, power lines, road networks and railways (Ramaswami et al. 2012 b ). The conceptualization of trans boundary infrastructures as supply lines to cities comes largely from the field of Urban Metabolism, which views cities as living organisms with metabolic demands for energy and materials, and anabolic processes of resource t ransformation, accumulation and value added production (Kennedy et al 2007). A new thread in the urban metabolism literature has set out to understand these material/energy flows not just in terms of bulk quantity, but as supply chains that are designed a nd operated by humans, that is, individuals, organizations and institutions endowed with agency (Ramaswami et al. 2012 b ). This distinction is important because it opens up the decision making space with respect to sustainability interventions from the phys ical/technical only, to include social dimensions of human behavior, collective action, institutional design and public policy (Ramaswami et al. 2012 b ). As a step in that direction, this article develops a novel approach for linking physical infrastructur e to the social network of supply chain actors that design and operate those infrastructures. An exploratory method is proposed, followed by an illustrative case study of infrastructure designer operators in the electricity supply chain serving Delhi, Indi a. Delhi is the world's second largest city (UN 2009) with a population of 23 million and an observed growth rate of nearly 40% over the last decade (Delhi Statistical Handbook 2012). The

PAGE 80

64 megacity's shear size and rapid growth are compounded by increasing per capita demand for energy, water and materials (Statistical Handbook 2012); placing ever increasing stress on finite local resources (e.g. State of the Yamuna) and increasing the city's reliance on trans boundary engineered infrastructures. Taken toget her, these characteristics make Delhi a compelling cast study in urban sustainability. An Interdisciplinary Framework for U rban S ustainability Ramaswami et al. (2012 b ) introduce the So cial Ecological Infrastructural Systems (SEIS) Framework to study urban sustainability from an interdisciplinary perspective. The SEIS Framework brings together social science theories and models that describe the social system with engineering principles and physical based laws that describe the biophysical system. Inte raction between the social and biophysical systems is what shapes urban sustainability outcomes. An excerpt from the article describes the premise succinctly: "Cities are embedded within larger scale engineered infrastructures (e.g., electric power, water supply, and transportation networks) that convey natural resources over large distances for use by people in cities. The sustainability of city systems therefore depends upon complex, cross scale interactions between the natural system, the transboundary e ngineered infrastructures, and the multiple social actors and institutions that govern these infrastructures" (Ramaswami et al. 2012 b p1). The present article provides a demonstration of the SEIS Framework bringing together environmental footprint and supply chain analysis (physical system) with social network analysis (social system) as a potential way of informing sustainability interventions. Broadly, urban supply chain footprints quantify direct and indirect material/energy flows (or the environment al impact thereof) that support urban activities, and social network analysis is used to explore the relationship between organizations in the supply chain in terms of influence, access to information and speed of communication.

PAGE 81

65 A Novel Approach: Combing Footprints with Social Network Analysis The goal of this research is to explore how information regarding social network structure of infrastructure designer operators can be integrated with information regarding their respective environmental impact to improve sustainability outcomes. As motivation, Ramaswami et al. (2012 a ) show that adoption rates (e.g. the rate of diffusion of innovation) is a key limiting factor in the effectiveness of many sustainability interventions. For example, they find that vol untary home insulation and weatherization programs that reduce household energy use for heating and cooling had adoption rates of only 2 4%. Similarly, adoption of voluntary green building design by infrastructure designers/operators in the commercial buil ding sector of Denver, CO represented just 5.4% of new construction (Ramaswami et al. 2012 a ). Adoption rates observed in Denver are in line with national estimates (Simmons et al. 2009). This means that every year in cities across the U.S., 19 of every 20 new commercial buildings are still designed and built without the use of readily available and highly effective sustainability innovations locking in unnecessarily high embodied and operational energy demand for the next 40 60 years of expected lifespan o f the building. The goal of a footprint informed network analysis is thus to identify actors (nodes) with high potential for reducing resource use/environmental impact in their own operations and for influencing others in the supply chain to do the same (e .g. diffusion of innovation). Given the examples provided above, this could take the form of encouraging (or incentivizing) a particularly influential construction company to adopt a sustainability innovation of interest with the goal of creating a new, in dustry wide, de facto standard. Evolution of Infrastructure Supply Chain Footprints Environmental footprints traditionally serve as a communication tool informing communities of their environmental impact across scales from local to global. Acknowledge ment of cities as drivers of change in solving global sustainability challenges (UN Habitat 2013 State of the World's Cities) has led to the development of urban supply chain footprints that connect

PAGE 82

66 cities to the trans boundary engineered infrastructures that bring energy, water, food and materials from outside the system boundary (Chavez and Ramaswami 2012). For example, city scale greenhouse gas footprints inform communities of their contribution to global climate change as a result of direct and indire ct fossil fuel combustion, industrial production (e.g. cement), agricultural activity (e.g. nitrogen fixing and soil carbon sequestration) and wastewater treatment (e.g. anaerobic digestion of organic waste) (Ramaswami et al. 2008). In addition to commun icating environmental impact, Cohen and Ramaswami (201 4 ) articulate how footprints can be used for risk communication informing communities how they may be vulnerable to trans boundary resource competition and production constraints. For example, heat wa ve and drought have led to electricity supply shortfalls in U.S. and E.U. cities due to cooling water constraints at regional thermal power stations (Forster and Lilliestam 2010; Flieschauer 2010; NETL 2009). Therefore, cities not only influence sustainability outcomes in locations where their resources are produced, but may also be at risk to production constraints in those places. However, footprints based on annual aggregate data which have been the mainstay of the literature are insufficient for uncovering these types of dynamic, two way relationships between the use system (e.g. city) and the production system (e.g. infrastructure supply chains). To address this shortcoming, Chapter IV of this thesis propose s a spatially delineated supply cha in footprint. The idea is to move from a single "black box" representing upstream activities to a network of supply chain producers operating across scales. Spatial delineation enables researchers to evaluate risk at individual production nodes and then tr anslate those risks back to the use system (e.g. city) based on network characteristics such as diversity and redundancy of supply. In addition to applications for risk assessment, spatial delineation is a novel way to identify relevant actors and identif y system boundaries the first step in any social network analysis. Leinert et al. (2013) assert that the first crucial steps to social network analysis

PAGE 83

67 identifying network boundaries and actors can be accomplished via stakeholder analysis and thus the two methodologies are complimentary. Stakeholder analysis, by definition, is an approach for identifying key actors in a system and assessing their priorities (Grimble and Wellard 1997). Here we propose spatially delineated supply chain footprints as an altern ative approach to stakeholder analysis, applicable to issues of urban sustainability. We posit that the combination of footprint information regarding environmental impact of producer nodes, with social network data regarding the influence of those nodes, creates opportunities for network interventions, described next. Social Networks, the Diffusion of Innovation and Network Interventions Social Networks Social networks have been studied for over 100 years, beginning with the work of sociologist Georg Sim mel on social geometry (Scott and Carington 2011). Since then, social network analysis (SNA) has become a science unto itself with applications in many disciplines ranging from health and behavioral science, to ecology, economics, public administration, la w enforcement and counter terrorism. A foundational concept in social network theory is that not all actors are equally important within a network (Tichy et al 1979). Importance, and more specifically, influence in social networks has been studied extensively (e.g. Freeman 1979; Bonacich 1972, 1987; Scott 2000; Wasserman and Faust 1994; Costenbader and Valente 2003) and several measures of influence are in common use, most notably related to the concept of centra lity (Freeman 1979). We provide a very brief definition of three centrality measures used here, and refer the reader to the literature for further details. Degree Centrality is the simplest measure of centrality: the number of ties incident upon a node. D egree centrality represents the number of direct links to other nodes and does not consider how those nodes are subsequently connected to other nodes, and is thus a

PAGE 84

68 measure of local influence within a network. Degree Centrality is also an indicator of how fast a node can give or receive information from others. A contemporary way to think about it is having more followers on twitter you can share information directly with others very quickly. Betweeness centrality represents the number of times a node serve s as a bridge between other nodes. It is an indicator of how a node controls the flow of information within a network, akin to the concept of a gatekeeper. Nodes with high betweeness centrality can also be viewed as linchpins without them the network may f all apart. Eigenvector centrality is a more sophisticated measure of influence in a network as compared to Degree Centrality. It assigns relative scores to all nodes in the network based on the concept that connections to high scoring nodes contribute mor e to the score of the node in question than equal connections to low scoring nodes (Xuguang et al. 2012). It is akin to the lay concept of being "well connected", in which knowing the "right" people is more important than how many people you know. These measures of centrality will be used in the Results section as part of a descriptive analysis of infrastructure operators (e.g. actors) in the electricity supply chain serving Delhi, India. Diffusion of Innovations Diffusion of innovations is the st udy of how new ideas and new practices spread within a population (Valente 2012). Formalized in a seminal work by Rogers (1963), the theory of diffusion of innovations has been used widely to analyze, model and predict the rate of adoption of myriad new te chnologies, policies, practices and social movements. The rate of adoption is determined by the perceived costs, benefits and legitimacy of innovations by potential adopters (Abrahamson and Roskenkopf 1997). Abrahamson and Roskenkopf (1997) go onto argue t hat network structure affects not only the rate of diffusion of innovation, but also the extent, e.g. why some innovations become the new de facto standard and others falter, never fully taking hold.

PAGE 85

69 Their analysis suggests that diffusion of innovation is highly sensitive to initial conditions such as the embedded location and centrality of early adopters in the social network (Abrahamson and Rosenkopf 1997). In the context of sustainability, this means that success depends not only on having the right poli cy, technology or behavior change, but having the right early adopters. Network Interventions Network interventions are "the process of using social network data to accelerate behavior change or improve organizational performance" (Valente 2012). Network interventions build on two well supported premises. First, that social networks influence individual and organizational decisions/behaviors (Valente 2012), and second, that social networks are the medium for diffusion of innovation; e.g. the flow of infor mation regarding costs, benefits and legitimacy of adoption is channeled via network ties (Rogers 2003; Valente 1995). Thus, network interventions operationalize the theory of diffusion of innovation by applying social network information to actively inter vene in the system to propagate a desired outcome (Valente 2012). For example, network interventions have been applied in over 20 randomized control trials to increase the adoption of evidence based medical practices and behaviors (Valente 2012). In one su ch example, Kelly et al. (2006) conduct a peer leader nomination social network analysis to identify influential candidates within a population at high risk of contracting HIV to serve as peer educators to encourage the use of prophylactics with success. S imilar to public health, environmental sustainability faces many intractable problems such as high personal vehicle use in industrialized countries (particularly the U.S.) and rapid adoption of personal vehicles in developing countries (particularly India and China). Single occupancy vehicles, in addition to being highly energy intensive per km travelled, lead to increased congestion, air pollution and other deleterious effects over efficient public transit and non motorized transportation. Given the pervas iveness and difficulty in changing (or even nudging) these trends, network interventions may be a highly effective yet largely untapped

PAGE 86

70 strategy for affecting bottom up behavior change. While the present study provides a descriptive analysis only and stops short of any network interventions in the electricity supply chain serving Delhi, the potential for future network interventions based on the descriptive analysis provided here is one of the key motivators for this research. This research is part of a lar ge, multi year research project in Delhi and elsewhere that seeks to inform sustainability interventions by local governments ( NSF Grant No. 1243535 ). Social Network Analysis Applied to SEIS Actor Categories: A Brief Review of the Literature Among the myriad applications of social network analysis, sustainability has become an increasingly important one. Here we provide a brief overview of network studies in the sustainability literature that relate to one or more of the three social actor cat egories defined in the SEIS framework: (1) resource users, (2) infrastructure designer operators and (3) policy actors. Resource Users (e.g. Households, Businesses and Government) The role of social networks in the adoption of household energy efficiency innovations is of enduring interest, with recent works by McMichael and Shipworth (2013), McEachern and Hanson (2008) and Egmond et al. (2005), as well as historical works without formal use of network analysis but in the context of communication channels in the diffusion of innovation by Ball et al. (1999), Coltrane et al. (1996), Dennis et al. (1990), Castanzo et al. (1986), Darley and Beniger (1981), Darley (1977) and others. In terms of businesses as resource users, Howard Grenville and Paquin (2006), Ashton (2008), Zheng et al. (2009), DomÂŽnech and Davies (2011) and Zhang et al. (2012) apply network analysis to study directional resource flows in eco industrial parks (e.g. for industrial symbiosis). Abrahamson and Rosenkopf (1997) explore network effe cts on Diffusion of Innovation in producer supply chains. Lazzarini et al. (2001) develop a conceptual model that distinguishes between horizontal and vertical connectivity in supply chains to better understand value added

PAGE 87

71 processes. Further, on the topic of robustness and resilience in supply chain design, Strozzi and Colicchia (2012) find 345 references using a citation network analysis of Google Scholar. Given the wide variety and large number of network studies pertaining to the "resource user" actor ca tegory, not every reference has been, nor could reasonably be, catalogued here. Policy Actors Lee and van de Meene (2012) study policy learning and policy adoption within the C40 network the world's 40 largest cities working to address climate change. Particular to the U.S., Ragland et al. (2011) investigate perceptions and communication strategies regarding carbon capture and sequestration (CCS) among state level energy policy actors as a proxy for network connectedness among potential change agents. R utland and Aylett (2008) offer a detailed study of how policy actor networks brought about climate policy in Portland, Oregon. Infrastructure Designer Operators Social Network Analysis of infrastructure designer operators is an emerging topic with only a handful of studies to date. In delineating this infrastructure designer operator actor category, we make a distinction between critical infrastructures (e.g. electricity, fuel, food, potable water, wastewater sanitation and transportation networks), and ot her supply chains that bring us consumer goods and commercial services. Here we distinguish supply chains based on their overall impact to city functionality; e.g. supply chains that if disrupted for a few hours or a few days would not cause the city to co me to a halt, can be considered non critical. For example, the supply of clothing, consumer electronics and durable goods are non critical supply chains and are more aptly defined as resource users rather than infrastructure designers because they depend o n (rather than provide) critical infrastructures as would any business or household. For this reason, social network analyses of business supply chains are categorized under the first

PAGE 88

72 subsection (above) on resource users, unless that supply chain is explic itly a critical infrastructure. Given these criteria, only one peer reviewed study was found to be relevant to our analysis of electricity infrastructure operators. Lienert et al. (2013) investigate fragmentation in Swiss water infrastructure planning vi s ˆ vis cooperation between organizational actors at different decisional levels (local, regional, national) and across sectors (potable supply vs. wastewater treatment). They find high levels of fragmentation in both the vertical and horizontal directions (e.g. between decisional levels and across sectors, respectively), indicating a need for a more integrated planning processes. In terms of study design, Lienert et al. (2013) offer credence to certain methodological decisions chosen here, such as intervi ewing a single representative for each identified organization (with one exception in the Lienert study in which two were interviewed) and merging similar actors into a single group. For example, in our study, one of the nine identified actor sub categorie s is electricity distribution companies, of which there are five operating in Delhi yet network data was collected for the group as a whole rather than for each company individually due to practical limitations in survey length and respondent fatigue. The remainder of the actor sub categories represented single organizations and thus no grouping was necessary. More broadly, both Lienert et al. (2013) and our study chose to interview actors across decisional scales (from local to national) and to differentia te actor sub categories based on function. Both studies provide descriptive analysis of a particular infrastructure supply chain, but differ in terms of objective, geographical context, the type of network data collected and analyzed, and intended applicat ions.

PAGE 89

73 Figure V 1 Supply Chain Network Analysis (Lazzarini et al. 2001) Figure V 2 Footprint Informed Network Analysis (adapted to this thesis)

PAGE 90

74 Application to the Water Energy Nexus in Delhi, India As one of the largest and fastest growing me gacities in the world (UN 2010) is Delhi India a microcosm for global sustainability challenges including rapid population growth, rural to urban migration, and rising standards of living, all in the context of natural resource scarcity and ecosystem degradation (UN 2013). An emerging piece to the puzzle is the complex relationship and inherent tradeoffs between water and energy supply. Over 50 million people in the Northern Indian States of Uttarakhand, Himachal Pradesh, Uttar Pradesh, Haryana and Delhi depend on the River Yamuna for water. The river be gins as a pristine glacial stream at 6320 meters above sea level in the Mussoorie range of the Lower Himalayas. From there, the Yamuna drops elevation and flows 120 kilometers towards the opening of the Indo Gangetic Plain home to approximately one third o f all grain production in India. Some 50 kilometers later in Dak Patthar, the Yamuna encounters the first of many weir dams and barrages, thus beginning its monumental task of providing water to tens of millions of people for everything from drinking and b athing to irrigation, electricity generation and industrial production. By the time the Yamuna reaches Delhi 224 kilometers later, whatever is left of the river is dammed and siphoned off for municipal supply to Delhi (CPCB 2006; Upadhyay et al. 2011). The river slowly regains volume as it meanders through the city the municipal supply diverted above is returned via open sewers and wastewater canals, only to be dammed and diverted again, this time for cooling at a major coal fired power station located in t he south of the city (Upadhyay et al. 2011). In a coupled water e nergy footprint study of Delhi (Chapter IV of this dissertation) find that local water withdrawals from the Yamuna River to support power generation range from 55 to 80 liters per person per day (lpd), of which ~1 lpd is consumed (evaporated) and the rest is returned ~ 10 degrees C warmer. By comparison, water withdrawals for direct use in the residential and commercial sectors of the city (e.g. piped municipal supply) are 150 lpd (Delhi

PAGE 91

75 Govern ment 2012), of which roughly 20% is consumed and 80% is returned to the river (60% treated and 40% raw; Delhi Jal Board 2013). From an in stream flow perspective, the volume of water extracted from the Yamuna River each day for power generation (55 80 lpd) is roughly half the amount required for all direct residential and commercial uses in the city, combined (150 lpd). Several studies have cited lack of in stream flow as a major contributor to serious and endemic water quality problems in the River Yamuna ( Upadhyay 2011; Paliwal et al. 2006; CPCB 2006). With no streamflow, the river has no pollution assimilation capacity nor any ecosystem services to offer. So then, what can a city like Delhi do to address such water scarcity? Most cities consider four opti ons: (1) increasing local surface water withdrawals, (2) importing surface water (e.g. trans boundary water diversions), (3) increasing local groundwater withdrawals, or (4) reducing demand via direct water use conservation efforts (e.g. restrictions on ou tdoor water use, low flow water fixtures, shorter showers, etc. ). Options (1) and (2) are not options for Delhi, according to officials at the Delhi municipal water board, because the city already utilizes its full allocation from both the Yamuna and Ganges rivers in accordance with inter state river compacts (Delhi Jal Board, 2013). Option (3) is also not viable groundwater is already being pumped at far above sustainable levels, as indicated by depression of the local water table at an average rate o f three feet per year, every year, for the past 30 (Sharma 2013). Option (4) is not tenable either in a city where 20% of the population has yet to gain access to piped water supply (Delhi Government 2012) and promises of more, not less, water are yet to b e realized. While some direct water savings are possible by fixing leaky pipes and curtailing excess usage in high income residential and commercial areas, these savings are surely to be overwhelmed by population growth and increased access to piped water supply for millions of more residents. So alas, what can a city like Delhi do? The value of a footprint is that it reveals cross sector and cross scale linkages and thus opens up the solution space to include more options. Thus instead of focusing only on minimizing

PAGE 92

76 direct water usage by households and businesses in the city, an integrated footprint informed strategy will include minimizing indirect water usage vis ˆ vis conserving electricity. For example, a typical 750 MW coal fired power station in Ind ia uses anywhere from 4,000 to 100,000 cubic meters of water per hour, depending on the type of cooling system (NTPC 2013). This represents an additional, indirect demand for water by every electricity user in the city that remains unaccounted for by water utilities, and thus remains invisible to most city wide water conservation efforts. Simply put, a city can conserve water by saving electricity (Cohen and Ramaswami 2014 ). Herein lies the value of a footprint: the ability to identify and quantify cross se ctor strategies for improving resource efficiency and mitigating environmental impact. After conducting a footprint to identify cross sector strategies (e.g. innovations), diffusion of innovation theory and social network analysis can be applied to identi fy potential drivers and barriers to adoption. Diffusion of innovation theory tells us that the decision to adopt is essentially an information seeking and information processing activity in which an individual is motivated to reduce uncertainty about th e advantages and disadvantages of an innovation'' (Rogers, 2003, p.172). Therefore, if within a network of infrastructure operators there is only a low level of information sharing and few opportunities for "trial by others" (e.g. vicarious experience with an innovation via a close network tie; Rogers 1983) then diffusion and adoption may be slow. Conversely, if there are high levels of information sharing within a network, or if information sharing can be induced via network interventions (Valente 2012), t hen diffusion is accelerated. As another example, social network analysis of major electricity users in a city (e.g. heavy industry, government offices, high rise residential buildings, universities, hospitals, hotels, shopping malls, etc. ) can reveal how information is shared between nodes and thus how to design a network intervention or social norming mechanism to promote an environmentally preferred behavior. Because large energy users are indirectly large water users given the significant water requirem ents of power generation, this becomes a win win strategy reducing both energy use

PAGE 93

77 and associated impacts as well as reducing water use and its associated impacts. Similarly, a footprint informed network analysis could lead to the development of a program that reduces water use vis ˆ vis advanced cooling technologies at thermal power stations and strategically targets potential early adopters with significant network influence to increase the rate and extent of diffusion of innovation. With methodological d etails of infrastructure supply chain footprints provided elsewhere (see Cohen and Ramaswami 2014 ; Chavez et al. 201 3) this article focuses on describing the social system of infrastructure operators and integrating the physical and social components of the SEIS Framework. Social network analysis is proposed as one way of understanding how potential sustainability interventions may diffuse throughout the electricity infrastructure supply chain serving Delhi. Before delving into further methodological details of the study, it is helpful to provide an overview of the Indian power sector with a focus on identifying and differentiating the roles of various electricity infrastructure operators. The National Power Grid of India The power sector in India is highly modular, with clear distinctions between generation companies, transmission utilities, load dispatch centers and distribution companies operating across scales. This is in contrast to the U.S. where 71% of ultimate electricity customers are served b y a single, vertically integrated, investor owned power utility that owns/operates the full supply chain of electricity infrastructure from power generation to transmission, distribution, load dispatch and end use metering. In both vertically integrated an d modular systems, we refer to each organization that operates a major components of the electricity supply chain as an electricity infrastructure operator ( EIO ) applying terminology from the Social Ecological Infrastructural Systems (SEIS) Framework (Rama swami et al. 2012 b ).

PAGE 94

78 The national power grid of India (technically termed the Inter State Transmission System or ISTS) is divided into five regional power grids, N, NE, E, W and S. Transmission between regions is managed by the National Load Dispatch Center (NLDC). The N, NE, E and W regional grids are synchronously interconnected, meaning that power flows freely between them as a function of load balance and the transfer capacity of inter regional ties. The S regional grid, by contrast, is asynchronou sly connected to the four other regions via HVDC lines, meaning that the quantum and direction of power flow is controlled manually (CEA 2013). Each region has its own Regional Load Dispatch Centers (RLDC), which assumes operational control from the NLDC at its regional periphery (boundary). The RLDC manages inter state transmission and load balance among States and Union Territories located within the region. Within each state, grid operation is managed by a State Load Dispatch Center (SLDC). The SLDC ove rsees intra state energy exchanges among distribution companies (e.g. beneficiaries) and generating assets (e.g. power plants) located within the State. Finally, at the sub state level (e.g. city level) distribution companies manage distribution, demand fo recasting, scheduling and load shedding within their service territory, subject to system constraints at the State, Regional and National levels. With a four tier system Central, Regional, State and Local coordination among hierarchical levels is key. For this express purpose, each region has a Regional Power Committee (RPC) that convenes various technical, operational and commercial coordination committees. Every electricity infrastructure operator in the region is represented on a regional power committee on either a permanent or rotating basis (NRPC, 2013). In terms of power generation, there are three major categories of power producers Central Generating Stations (CGS), State Generating Stations (SGS) and Private Generating Stations (PGS), depending on ownership. In general, CGS have long term contracts to sell bulk power to multiple beneficiaries (e.g. states and distribution companies) located throughout the country, whereas state and private generating stations sell power directly to the state in whi ch they are located.

PAGE 95

79 Study Design Social network analysis can be a powerful tool for understanding how infrastructure designer operators are interconnected with respect to communication, information sharing and influence. To date, the only network repres entation available for most infrastructure designer operators are formal organizational charts (e.g. positional analysis). Organizational charts depict official hierarchy but provide little insight into how such a structure is operationalized, and if and h ow it is carried out in practice. By comparison, an interactional network analysis reveals how actors interact with one another and thus provides a more realistic representation of network structure. Objectives The objective of this study is to demonstrate what a footprint informed actor analysis may l ook l ike through a pilot study of electricity infrastructure operators serving Delhi, India. There are four main components to this objective: 1. Identify key actors in the infrastructure supply chain of interest (e.g. determine the population of actors, N); 2. Develop a survey instrument to collect network data and relevant node attributes including each organization's primary duty, spatial scale of operations, decisional level, top priorities and perceiv ed risks to service provision; 3. Map the social structure of the actors according to interactional network data; 4. Develop hypotheses of how the information collected in components 1 3 can be used to inform sustainability interventions in the supply chain of i nterest. Methodology and Data Collection An ego centric network analysis of electricity infrastructure operators (IOs) serving the National Capital Territory of Delhi (NCT Delhi) was preformed over the course of three months

PAGE 96

80 from May to July 2013. 12 IOs were identified as "egos" of the study. Semi structured interviews and a written survey were administered to each ego to obtain attribution and network data. Several s ub categories of IOs were ide ntified, including power generating companies (GENCOs ) transmission utilities (TX) distribution companies (DISCOMs) and load dispatch centers ( Grid O Ps ) operating across four geographical/hierarchal scales (local, state, regional and national/central ). In most cases, the subcategory corresponds to a single entity (e.g. the National Load Dispatch Center, which manages inter regional electricity exchanges and injection load balance for the national grid). In the remaining few cases, the subcategory is comprised of several entities sharing the same primary func tion (e.g. DISCOMs, which forecast load and distribute electricity to end users at the local level ) Combing highly similar actors into a single actor subcategory was a tactical decision to keep survey length and respondent fatigue to a minimum. This practice was also employed by Lienert et al (2012). Network data collection focused on how egos are connected to one another and to identified policy actors (PAs) and dependent cr itical infrastructures (DCIs). Dependent critical infrastructures, as the name implies, are critical infrastructures such as water supply, wastewater sanitation and public transit, which depend on a steady supply of electricity. For example, Delhi Metrorail (the city's underground subway system) relies on multiple 66kv and h igher electrical connections to keep trains running and prevent passengers fr om being stranded underground. Similarly, a one hour power cut to water treatment plants results in a three hour disruption of water supply (Delhi Jal Board). We use the term pol icy actors, in this context, to describe any organization identified by electricity infrastructure operators to materially effect their day to day operation and/or longer term planning. That can include formal policy actors such as State Electricity Regula tory Commissions which set tariffs, or indirect policy actors such as the State Public Works Departments, which must approve the digging up of streets to build/modify/repair electricity infrastructure on public easements Policy actors linked to the elect ricity infrastructure operators

PAGE 97

81 were identified by an iterative process of literature review, website hyperlinks, expert consultation and snowball sampling (e.g. respondent driven sampling). Table 1 provides an overview of all the actors: IOs, Pas and DCIs Policy actors and dependent critical infrastructures are considered alters to the egos. Egos a re also alters of one another. The result is bi directional network data for ego ego ties, and unidirectional ne twork data for ego alter ties. One main differe nce between an ego centric network analysis and a full network analysis is that only ties emanating from an ego are included. In a full network analysis, by contrast, all node ties would be included, subject to boundary conditions. I n an ego centric networ k analysis, only the egos are surveyed/interviewed regarding their network connections alters are asked about but not asked directly. Table V 1 Overview of Actors: Number Identified Per Category Egos Alters Alters Total Actors Hierarchical Level EIO PA DCI EIO +PA+DCI Local 6 (4) 1 1 8 State 3 (2) 8 5 17 Regional/Central 6 (6) 7 (1) 5 14 TOTAL 15 (12) 16 11 39 ( ) Parenthesis indicate number of survey completed Table V 2 Description of Actors Spatial Scale Actor Type Sector Primary Duty Actor Name (Organization) Abbrev. Local EIO Power Distribution New Delhi Muni. Corp. Elec. Dept. NDMC Local EIO Power Distribution North Delhi Power Ltd. NDPL Local EIO Power Distribution BSES Rajdani Power Ltd. BRPL Local EIO Power Distribution BSES Yamuna Power Ltd. BYPL State EIO Power Transmission State Transmission Utility STU State EIO Power Grid operation State Load Dispatch Center SLDC Regional EIO Power Grid operation N. Regional Load Dispatch Center NRLDC Regional EIO Power Analysis of grid N. Regional Power Committee NRPC National EIO Power Transmission Central Transmission Utility Powergrid National EIO Power Grid operation National Load Dispatch Center NLDC National EIO Power Generation National Thermal Power Corp. NTPC

PAGE 98

82 National EIO Power Generation National Hydro Power Corp. NHPC State PA Power Tariff setting State Electricity Reg. Comm. SERC State PA Power EE/RE program admin. Energy Efficiency and Renewable Energy Management EEREM National PA Power E nergy policy integration & rulemaking Ministry of Power MOP National PA Power Power planning, policy & technical supervision Central Electricity Authority CEA National PA Power Tariff setting Central Electricity Reg. Comm. CERC National PA Power Regulate atomic energy ; civilian & defense activities Dept. of Atomic Energy Atomic Local DCI Health Health provider Private Sector Hospitals PSH State DCI Health Health provider Delhi Government Hospitals DGH State DCI Water Water supply & sanitation Delhi Jal Board Water State DCI Transport Transit operator Delhi Metrorail Metro State PA General Public housing, road & flyover construction Delhi Public Works Department PWD State PA Law Law enforcement Delhi Police Police State PA Public Admin. Municipal governance Delhi Municipal Corporation MCD State PA Housing Coordination of infrastructure Delhi Dept. of Urban Development DUD State PA Housing Urban p lanning & construction Delhi Development Authority DDA State PA Env't Law enforcement Delhi Pollution Control Committee DPCC National PA General Funding for urban infrastructure development Jawaharlal Nehru National Urban Renewal Mission JNNURM National DCI Transport Transit operator Indian Railways Rail National DCI Health Health provider Central Sector Hospitals CSH National PA Env't Law e nforcement Central Pollution Control Board CPCB *Note: two other organizations were included in the network data collection survey but have are not included here, nor in further analysis. One was trifurcated into three organizations (captured above) and is now defunct, and the second never existed. Network Analysis Results and Discussion Interactional network data collected in this study was analyzed using the statistical network analysis package, UCINET ( Borgatti et al. 2002 ). Overall, physical knowledge of the

PAGE 99

83 energy system tracks well with network analysis results, lending credence to its veracity and supporting the concept of network analysis for analyzing built infrastructure. Figure V 3 shows a bi directional network graph for electricity infrastructure operators serving Delhi, India Each e lectricity infrastructure operator is represented by a node, with the interactions bet ween nodes repr esented by directional arrows. If a stated interaction between two nodes exists then a tie (line) is drawn Arrows indicate the direction of the flow of information, with arrowhead size indicating tie strength. As an example, in Figure V 3 BRPL (a local distribution company, or DISCOM) indicated a strong tie (e.g. frequent interaction ) with POWERGRID (the central transmission utility of India), and thus a large arrowhead points from BRPL to POWERGRID. In the other direction POWERGRID stat ed a weaker tie ( fewer interactions ) with BRPL, and thu s a smaller arrowhead is used. There are many potential explanations for non uniform directional information flow, not the least of which is the relative importance of one node relative to the other. T his may be particularly true in the context of physical infrastructure. For example we expect a DISCOM to share more information up the supply chain to the transmission utility than the transmission ut ility may share down ward to the DISCOM because the DIS COM is at the terminus of the infrastructure supply chain and fully dependent on upstream nodes. The transmission utility, by contrast is not dependent on any single downstream node (e.g. DISCOM ) to fulfill its primary duty of transferring bulk power betw een regio ns. Upstream nodes may have less an incentive to share information with downstream nodes. Similarly, downstream nodes may perceive a stronger tie with upstream nodes than is reciprocated. Verifying this observation would require a larger study in cluding more DISCOMs and transmission utilities. In addition to analyzing the strength of ties between nodes (indicated by arrowhead size, as described above) the centrality of individual nodes within a network is of primary interest. There are several measures of node centrality (described in section 5.5). H ere we examine

PAGE 100

84 betweeness centrality an indicator of how a node controls the flow of information within a network, akin to the concept of a gatekeeper Returning to Figure V 3 node size is computed and shown as a function of betweeness centrality the larger the node, the more frequently it serves as a bridge between other nodes. Here we see that the national transmission utility (POWERGRID) state transmission utilities (STU), state load dispatch centers (SLDC) and a large DISCOM (NDPL/TATA) are the most central to the network. Interestingly, the nation's largest energy producer (NTPC) is less central to the network ; perhaps ( again ) because it is at one end of the infrastructure supply chain (the top, as oppo sed to DISCOMs at the bottom). This finding may have implications for catalyzing change in the energy sector. Energy producers are often the focus of new regulations and policies aimed a t improving environmental performance largely because they are the physical point sources. However, grid operators (e.g. load dispatch centers) may be more effective lever age points given their network centrality dominance. Rather than a top down, command and control style solution to environmental performance vis ˆ vis strict regulation of power generation and notwithstanding a bottom up market incentive such as a price on carbon, there may be overlooked opportunities for a technocratic grid optimization approach. For example, introducing multi objective optimization criteria into existing electricity disp atch models to simultaneously solve for least cost solutions (as they do now ), and minimize emissions (which they currently do not). A second opportunity may be found with transmission /distribution utilities which again are not point sources, but are gate keepers connecting energy producers with energy consumers. One can imagine a policy in which transmission utilities must meet environmental performance portfolio targets, and thus are more likely to enter into long term contracts with energy producers that help them achieve those targets. This in turn incentivizes power generation companies to improve the environmental performance of their fleet in a least cost way. These are in effect market solutions that allow for innovation, optimization and private sec t or investment without having to tackle the Sisyphean task of putting a price on carbon

PAGE 101

85 Finally we look at the full ego centric network including alter connections Figure V 5 shows the network of electricity infrastructure operators (EIO's) serving Delhi, India plus all identified policy actors (PA's) and dependent critical infrastructures (DCI's), excluding isolates. Compared to Figure V 1 (which considers EIO's only), there is a shift in the betweeness centrality of EIO's when we consider a b r o ader societal network including actors outside the energy sector DISCOM s, which in Figure V 1 appeared to be at the terminus of the energy supply chain, are now the bridge towards a di verse group of actors and resource users including public works department s police department s, and urban development authorities as well as other critical infrastructures including hospitals, railways and water supply. Interestingly, there is only one p endant, coal/gas suppliers, which are connected to power generation companies only. On the other end of the supply chain, however, there are multiple, multi faceted connections between EIO's PA's and DCI's at many levels. One would expect the Central Pollution Control Board (CPCB) to regulate emissions from thermal power generation and thus have network ties to the National Thermal Power Corporation (NTPC), but the fact that they are also connected to a large DISCOM would b e less evident otherwise. Similarly, Delhi Metro Rail, Indian Railways, hospitals of all kinds, and the public water utility (Delhi Jal Board) have multi level connections to EIO's, driving home the interconnectedness of critical unban infrastructure. Thes e findings and the ideas they illuminate support our premise that social network analysis can be a useful tool for mapping the underlying connectivity of phy sical infrastructure networks, and ultimately, affecting change by finding optimal starting point s for sustainability and resilience interventions We believe p otential abounds for footprint informed social network analysis in studying critical infrastructures and that more work will be done to carry this concept further.

PAGE 102

86 Figure V 3 Network Graph of Electricity Infrastructure Operators S erving Delhi [ Node size represents betweeness centrality. Arrowhead size represents strength of tie. Color coded as follows: green = DISCOM; blue = Transmission Utili ty; black = Grid Operator; yellow = Central Administration] BRPL BYPL CEA NDMC NDPL/TATA NLDC NRLDC NTPC POWERGRID SLDC STU

PAGE 103

87 Figure V 4 Ego Centric Network Analysis with Alters. [Blue squares represent egos, red circles are alters]

PAGE 104

88 Survey Results and Conclu sion In addition to mapping the social network of electricity infrastructure operators, we wish to know more about their respective priorities and perceptions of risk This information can help motivate sustainability interventions in the power sector, which is after all the goal of this thesis. For example, knowing that a power generation company has identified reducing water use at thermal power stations as a priority and is risk averse to coal supply shortages, may indicate a willing ness to pursue alternatives to coal that combine the benefits of lower water requirements with fewer supply chain constraints, such as increased utilization of wind power. This will of co urse have co benefits for GHG mitigation but the catalyst/ driver may be water or supply chain constraints, and not Climate Change. Figure V 5 shows how a set of seven priorities specific to electricity infrastructure operators were ranked as important at the present time and in future planning. Interestingly, future concerns over water are ranked high, but concerns remain low at the present time. Similar results were observed for GHG mitigation, perhaps indicating a pattern of deferment when dealing with environmental externalities. The stated priorities of electricity infrastructure operators are summarized in Figure V 6 (comparing current and future priorities) and V 7 ( overall ranki ng of current priorities ). These results offer insight into answering two key questions : (1) Which factors may be contributing to electricity supply shortfalls (a.k.a power cuts, a.k.a. load shedding, a.k.a. reduced reliability) in Northern India? (2) What is the relative importance of various, interconnected system risk s to power supply reliability? These two questions may begin to be answered based on the opinions of expert s surveyed and presented here. However, if we wish to corroborate these findings but do not wish to conduct a second, larger survey perhaps we can test them empirically instead?

PAGE 105

89 In the final chapter of this dissertation (next), we draw on survey results presented in Figure V 7 to design an observational study to test th e significance of identified risk factors to electricity supply reliability in Northern India. Figure V 5 Priorities of Electricity Infrastructure Operators

PAGE 106

90 Figure V 6 Risk P erceptions of Electricity Infrastructure Operators (Present and Future) Figure V 7 Ranking of Risk Factors by Electricity Infrastructure Operators

PAGE 107

91 CHAPTER VI THE EFFECT OF CLIMAT E AND SUPPLY CHAIN CONSTRAINTS ON GRID SCALE ELECTRICITY SUPPLY RELIABILITY: A HIERA RCHICAL LINEAR MODEL Introduction For the first time, this research brings together detailed energy use, energy production, energy exchange and load dispatch data with geophysical climate information to explore the effects of climate and infrastructure supply chains on electricity supply r eliability for a large regional power grid located in India. The Northern Region (NR) power grid serves nine States and Union Territories (UT's; provincial city states) in Northern India with a combined population of nearly half a billion. Each State/UT se rves as a power control area (PCA), responsible for load generation balance within its territory. Defined broadly, load generation balance is the instantaneous matching of electricity supply (e.g. generation) and demand (e.g. load) in an interconnected gri d network. In preparation for this study we surveyed 14 organizations responsible for electricity generation, transmission, distribution and load dispatch in the Northern Region grid. Respondents were asked to rank 10 risk factors contributing to electric ity supply shortfalls (e.g. scheduled and unscheduled power cuts) based on their experience and expertise. Now we wish to test if these risk factors are observable and their effect measurable from available data. This article presents an empirical analysis of the effect of climate and supply chain constraints on electricity supply reliability in Northern India, with state power control areas as the units of analysis. An important distinction between this study and many others studying power supply reliabili ty, is that we do not attempt to model the power grid itself, but rather, evaluate the explanatory power of particular sets of exogenous variables, namely climate and supply chain constraints. Structural characteristics endogenous to a power system, such a s capacity adequacy

PAGE 108

92 and capacity availability, are included as control variables in order to facilitate comparison between PCAs with varying baseline reliability. Other system risks such as technical faults (e.g. incorrect protection operations), operator error and cyber attack, are not considered. Power Supply Reliability There is a vast technical literature on the reliability of interconnected power systems that spans multiple decades and multiple disciplines. We will not attempt to summarize the literat ure here, but rather, provide a simple heuristic describing where this study lands. In the broadest sense, power supply reliability is modeled in one of two ways. The first are physical based models, fundamentally governed by Ohm's Law and Kirchhoff's Law, which describe power flow in electrical networks (e.g. Serway and Jewett 2004). The second modeling paradigm is topological governed by network theory, which describes how nodes in a network are interconnected (Pines et al. 2010). This study is neither. I nstead of modeling the power grid directly, we present a multi level regression to test the effect of climate and supply chain variables on electricity supply reliability while controlling for structural differences between power control areas. This inters ection of climate, energy and infrastructure may be of particular interest to policy actors, infrastructure designer/operators, and end users of electricity (Ramaswami et al. 2012b). Defining and Measuring Power Supply Reliability Electricity supply reliab ility can be measured in many ways, and there is an abundance of literature dedicated to evaluating and comparing various metrics (see Tollefson et al. 1991; Debnath and Goal 1995; EPRI 2005; Johannson et al. 2013; to name just a few). In this study, we em ploy a simple and intuitive reliability metric adopted from Roy Billinton's foundational work, Power System Reliability Evaluation (1970). The chosen metric, energy not supplied (ENS; see equation VI 5) represents total disconnected load experienced by end users, also known as load

PAGE 109

93 shedding It is equivalent to the integral of demand not met over a given period of time. Using data reported by the Indian power sector, it is given by the total energy requirement for a State/UT less the total energy available in a given month. ENS can be standardized by dividing by total energy requirement, yielding RNS ( energy not supplied as a fraction of total requirement ; see equation VI 6). RNS is a scale invariant reliability metric useful for comparing power systems of different size. An RNS of 0 represents perfect reliability. An RNS of 1 would indicate that no demand was met. Equations VI 1 to VI 4 show the mathematical relationship between energy terminology in use by the Ministry of Power (MoP) and Central Electricity Authority (CEA) of India. Equation VI 1 Energy Requirement # Equation VI 2 Energy Available # ! Equation VI 3 In Boundary Energy Generation # # ! ! Equation VI 4 Trans Boundary Energy Supply ! ! ! ! ! ! # ! ! $ % & ! ! $ % # ! $ % *UI is the difference between scheduled and actual energy exchanges. The schedule is set day ahead as per declared capacity of all generating stations and load forecast of beneficiary States. Actual energy exchanges reflect the entitlements from CGS and bilateral purchases that a beneficiary State accepts based on current day changes in own generation and/or requirement. Equation VI 5 Energy Not Supplied [GWh] $ & % ! ! Equation VI 6 Requirement Not Supplied [Dimensionless] $ % ( ! ( Equation VI 7 Energy Index of Reliability [Dimensionless] $ ( % $ %

PAGE 110

94 ENS is reported directly by CEA for each State/UT in the country at monthly time steps (see CEA 2013). A histogram of ENS for the two most recent years where data is available (2011 13) reveals a positively skewed distribution with most of the data tending towards zero, similar to geophysical phenomena such as precipitation (Figure VI 1, upper left quadrant). Data of this form is well reproduced by a gamma distribution or a log normal transformation. Four distributional properties of ENS are shown in Figure VI 1. Figure VI 2 illustrates the distributional properties of ENS after a lognormal transformation. Similarly, Figure VI 3 shows the distributional properties of RNS, a normalized alternative to ENS, and Figure VI 4, a lognormal transformation of RNS. We see that the distributional properties of ENS and RNS are nearly identical, and the pair remain nearly identical when we compare log transformations Given the goodness of fit for ENS and RNS, it is reasonable to proceed with either response variable, and with or without a log transformation. Exploratory analysis was performed in parallel for all combinations and then compared based on covariate response scatterplots and distributional fits. Ultimately, ENS was chosen over RNS to preserve information regarding the magnitude of response (RNS is normalized). A lognormal transformation was then applied to give more space to observations clustered in the lower tail and to lin earize X Y relationships (see Figure V I 9 and VI 10 in next section). Finally, we sho w timeseries of ENS (Figure VI 5) and RNS (Figure VI 6 ) as well as boxplots (Figure VI 7 and VI 8) for the two most recen t years where data is available.

PAGE 111

95 Figure VI 1 Gamma distribution fit to ENS (GWh) Figure VI 2 Log Normal transformation of ENS

PAGE 112

96 Figure VI 3 Gamma distribution fit to RNS (dimensionless) Figure VI 4 Log Normal transformation of RNS (dimensionless)

PAGE 113

97 Figure VI 5 Time Series of Energy Not Supplied to NR States (GWh) Figure VI 6 Time Series of Requirement Not Supplied to 5 NR States (%)

PAGE 114

98 Figure VI 7 Group wise Boxplots of Energy Not Supplied to 5 NR States Figure VI 8 Group wise Boxplots of Requirement Not Supplied to 5 NR States

PAGE 115

99 Risk Factors to Power Supply Reliability In spring 2013 we surveyed 14 organizations responsible for electricity generation, transmission, distribution and load dispatch in the NR (see chapter 4 of this dissertation). Respondents were asked to rank 10 risk factors contributing to electricity supp ly shortfalls (e.g. scheduled and unscheduled power cuts), based on their expertise and experience. The set of 10 risk factors was generated a priori based on investigator knowledge of the system, review of peer reviewed literature and government reports, and pilot testing. Pilot tests included a seed list of risk factors followed by a prompt for respondents to expand or modify the list. Table VI 1 highlights results from the survey: a ranking of 10 risk factors. The survey provides valuable information on perceptions of risk by electricity infrastructure operators, but due to the small sample size (n=14), survey results are intended as heuristic rather than conclusive. As such, rankings and scores provided in table VI 1 have not been subject to statistical significance testing. Table VI 1 Electricity Supply Reliability Risk Factors Risk Factor Rank Score Insufficient generating capacity 1*** 92% Heatwave/drought 2*** 92% Fuel supply disruptions to thermal power stations (TPS) 3** 85% Under utilization of generating capacity 4** 85% Lack of demand side management 5* 69% Monsoon rains disrupt T&D infrastructure 6* 54% Insufficient/poor quality cooling water for TPS 7 46% Insufficient water supply for hydropower 8 38% Insufficient T&D infrastructure 9 31% Difficult securing necessary permits 10 23% Ranking (1 10) based on survey of electricity infrastructure operators (n=14) ** Starred pairs indicate tied rank Score represents (%) of respondents that identify a risk factor as important at the present time. Building on the results of our survey, we wish to test if these relationships are observable and measurable from available data. Moreover, can we corroborate or inform the risk perceptions

PAGE 116

100 of electricity infrastruc ture operators through empirical evidence? The first step of any statistical test is to define the dependent and independent variables. The previous section detailed the dependent (e.g. response) variable of interest the amount of energy not supplied to S tates/UTs over a given time period ENS. Next, we define the set of independent (predictor) variables. Our starting point (initial condition) is the set of 10 risk factors identified by electricity infrastructure operators (Table VI 1). For every identified risk factor, we propose an associated metric commensurate with publically available data (Table VI 4). The following section summarizes the extensive data requirements needed to translate risk factors in Table VI 1 into reproducible metrics in Table VI 4. Data The crux of this research, like many interdisciplinary studies, is data availability. Extensive data collection, cleaning, compilation, triangulation, cross reference and quality control measures were undertaken. Standard quality control (QC) measur es including NA queries, complete record checks, summary statistics and visual inspection were applied to every dataset. Additional QC, including multi scale aggregation/disaggregation/summation checks, benchmarking, cross referencing and interpolation wer e applied as necessary. Table VI 2 and Table VI 3 summarize the 15+ datasets synthesized for this study. Table VI 4 shows the final metrics used in our model.

PAGE 117

101 Table VI 2 Power Sector Data Availability and Sources Category Data Units Years Source System wide Power Supply Position Requirement, Available, Surplus, Deficit GWh '05 13 CEA Monthly Reports > Power Supply Position > Year > Energy (PDF) Plant Load Factors (PLF) MW 05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > All India (PDF) Inter Regional Exchange Inter Regional Energy Exchanges GWh 05 13 CEA Monthly Reports > Inter Regional Energy Exchanges (PDF) Price Rate of Sale of Power Paise/ KWh 10 11 CEA Annual Reports (PDF) Grid Operation Station wise Injection Schedule/Actual 10 5 KWh 08 13 NRPC > Commercial Activities > UI Account > Supporting Files (xls) State wise Drawal Schedule/Actual " Unscheduled Interchange " Frequency Hz " Cost of UI ~fn(Hz) Rupee " State wise Power Supply Position Power Requirement (Peak Demand) MW 05 13 CEA Monthly Reports > Power Supply Position > Peak (PDF) Electricity Requirement GWh 05 13 CEA Monthly Reports > Power Supply Position > Energy (PDF) Net Drawal from Grid GWh 05 13 CEA Monthly Reports > Inter Regional Energy Exchanges (PDF) Gen eration CGS Allocated Shares MW 05 13 CEA Monthly Reports > All India Installed Capacity: Statewise/Utilitywise Electricity Generation GWh 05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > Regionwise, Statewise, Stationwise Installed Capacity located in State/UT MW 05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > Regionwise, Statewise, Stationwise Station Wise Gen eration Gen. Stns. Name 05 13 Monthly Generation MWh 05 13 Capacity MW 05 13 Plant Load Factor (PLF) % 05 13 Plant Availability Factor (PAF) % NRPC > Archive > Regional Energy Account > Month (PDF) Supply Chain Fuel type type 05 13 Cooling type type NA NOT AVAILABLE Water withdrawals / consumption (actual) M 3 /hr NA NOT AVAILABLE Environ ment Water withdrawal/ consumption (design) M 3 /hr Once CEA Report on minimization of water requirement in coal TPS (appendix 11)

PAGE 118

102 Table VI 3 Climate Data Availability and Sources Table VI 4 Reliability Risk Factors and Corresponding Metrics Risk Factor Metric Metric Description Data Avail able Data Incl uded Insufficient Gen. Capacity Capacity Adequacy State wise Peak demand / (Total Installed + Allocated Capacity) Yes Yes Heatwave/ Drought Max Temperature State wise max. temp. from NN stn. Yes Yes Precipitation Accumulation State wise precip from area weighted stn data Yes Yes Precipitation Anomaly State wise precip anomaly Yes Yes Max Temp. / Precip. Heatwave/drought combo [derived] Yes Yes Fuel Supply Disruptions Coal Stock Days All India coal stock position Yes Yes Gas Fleet Availability Factor (FAF) All India gas cap acity de rating due to fuel shortage Yes Yes Hydro Storage All India reservoir storage Yes Yes Underutilization of Gen. Capacity Fleet Load Factor (FLF) Producion weighted sum of PLF Yes Yes Fleet Availability Factor ( FAF) Production weighted sum of PAF Yes Yes Unpredictable load/Lack of DSM Cost of Unscheduled Interchanges (UI) Deviations from day ahead load schedule priced ~ fn of grid frequency at 15 min dt. Yes Yes Monsoon disrupts T&D Precip Accumulation State wise monthly accumulation Yes Yes Precip Anomaly State wise monthly anomaly Yes Yes Insufficient/poor quality cooling water at TPS Water supply at TPS Actual station wise water supply No No Water availability at TPS Precipitation compared to climatology for that location Partial No Water Withdrawal Footprint of Energy Supply (WWFES) Statewise WWFES computed ~ fn of water intensity of production Yes Yes Insufficient water for hydropower Station wise storage Station wise reservoir level Partial No Aggregate Storage All India reservoir storage Yes Yes Insufficient T&D Outside of scope Outside of scope NA NA Securing Permits Outside of scope Outside of scope NA NA [Crossed off items indicate omission from further analysis due to data limitations] Category Data Units Available Source System wide Station data Hourly temperature Celsius 1973 2012 http://www.metoffice.g ov.uk/hadobs/hadisd/in dex.html Hourly precipitation mm 1973 2009 Area weighted station data Monthly precipitati on accumulation/ anomaly mm 1871 2013 ftp://www.tropmet.res.i n/pub/data/rain/iitm subdivrf.txt

PAGE 119

103 Structural Constraints The motivation for this study is to estimate the effect of climate and supply chain constraints on power supply reliability. To obtain a richer distribution of observe d responses, and to increase the sample size and generalizability of our study, we analyze data from multiple Power Control Areas with varying baseline reliabilities. For example, the national capital of India (Delhi) and the provincial capital of the stat es of Punjab and Haryana (Chandigarh) have far higher power supply reliability than neighboring States (see Figures VI 5 and VI 6). Such differences in baseline reliability may be explained by structural differences in electricity infrastructure. Recall th at each State/Province is a Power Control Area with unique load generation balance characteristics. To account for these differences, we consider a set of key structural characteristics: capacity adequacy, capacity utilization, capacity availability, reli ance on energy imports, and demand side management. Intuitively we expect basic power system adequacy concepts such as capacity adequacy, capacity utilization and capacity availability to be well correlated to reliability. Reliance on energy imports and th e availability of demand side management have been added as additional possible explanatory variables. Estimates of capacity adequacy, capacity utilization, capacity availability and import reliance are based on production weighted summations of the same form. For example, we measure capacity utilization in terms of a Fleet Load Factor (FLF) akin to the fam iliar Plant Load Factor (PLF). FLF is the summation of all PLFs, scaled by their relative contribution to the power supply position of each beneficiary state in each month. Likewise, we compute Fleet Availability Factors (FAF) by substituting availability for load in the above expression to yield an estimate of total installed and allocated capacity available to a beneficiary state in a given month. The final structural variable considered is demand side management. Here we use a proxy metric: a frequency d ependent dynamic pricing mechanism that is levied on deviations from

PAGE 120

104 injection/load schedule for generators/beneficiaries. This load balancing mechanism is structured such that the cost of under generating or over drawing is inversely proportional to grid frequency, within a permissible operating bandwidth. Dynamic pricing helps alleviate stress to the grid by penalizing exacerbations and incentivizing corrective action. Figure VI 10 is a scatterplot matrix of the log response versus each of the structural variables described above. We see strong correlations between reliability, capacity utilization, capacity adequacy and the cost of unscheduled interchanges (our metric for DSM), with weaker relationships observed for trans boundary supply and fleet availa bility. Note that capacity adequacy and utilization are highly co linear, and thus only one (adequacy) is used in subsequent modeling. Interestingly, we see an inverse relationship between reliability and trans boundary supply: as the fraction of demand me t by imports increases relative to in boundary production, reliability improves. However, this may be due to confounding factors, such as the political importance of Delhi and Chandigarh, the two States/UTs with the highest reliability and the highest prop ortion of imports. It should be noted that other structural variables can be included in addition to or instead of the ones proposed here. Ultimately, the inclusion or omission of any variable is subject to data availability.

PAGE 121

1 05 Figure VI 9 Relationship between ENS and power system structural covariates Figure VI 10 Relationship between l og(ENS) and power system structural covariates L og transformation linearizes the relationship between response and structural covariates. Non linear relationship between response and structural covariates

PAGE 122

106 Climate Effects To test the effect of climate on energy supply reliability, we start with a set of metrics with known relationships to energy demand. It is well known that temperature is a primary driver of electricity demand in urbanized areas (see Figure VI 11 ), yet the extent to which temperature may correlate to electricity reliability has yet to be explored at grid scale for India. Similarly, it is well known (in India) that below average rainfall during the summer monsoon leads to increased energy demand for groundwa ter pumping and irrigation, yet any potential effect on grid reliability remains unknown. Moreover, the combination of extreme temperatures with low rainfall may have a multiplicative effect of increasing energy demand in both rural (irrigation) and urban (air conditioning) areas while simultaneously reducing power generating capacity due to decreased efficiency of cooling towers and, potentially, outright cooling water shortages. In the other extreme, high rainfall can lead to flood damage to transmission and distribution infrastructure (e.g. substation flooding) and disrupt fuel supply chains, such as rail transport of coal from pithead to thermal power station. Daily temperature and precipitation data were collected from the National Oceanic and Atmosph eric Administration (NOAA) Integrated Surface Database (ISD) using a nearest neighbor algorithm to find the closest weather station by greater circle distance to the state capital of each State/UT included in this study. This approach worked well for tempe rature, but revealed a lack of available data for precipitation. None of the weather stations located in India have a complete precipitation record in recent years. Data is purportedly logged every three hours, but most of the stations located in India con tain fewer than 400 observations over the 672 day period of interest (March 1 2011 Jan 1 2013), indicating substantial missing data. The temporal distribution is also uneven; observations are more frequent during Monsoon season and sparse during times of less rainfall (see Figure VI 12 a dot plot of precipitation versus time for 11 candidate nearest neighbor weather stations). Given the paucity of precipitation data for the

PAGE 123

107 Indian subcontinent contained in NOAA ISD, additional global climate datasets with daily or sub daily observation frequencies were queried, but to no avail. At the time of analysis, the NOAA GHCN database did not contain precipitation data for India past 2010, and Hadley ISD is based on the same raw data as NOAA ISD, thus subject to the same missing data. With daily station data unavailable, we opted for the next best thing: monthly accumulation over spatially aggregated climate subdivisions from the Indian Institute of Tropical Meteorology (IITM). IITM subdivision data are based on area weighted aggregations of station data. However, the mountainous states of Uttarakhand, Himachal Pradesh and Jammu & Kashmir (three of nine states/UT's in northern India) are omitted by IITM due to sparse data and complex terrain. As such, these three stat es are omitted from subsequent modeling due to lack of precipitation data. Figure VI 13 shows monthly rainfall accumulation for five NR states where data is available. Figure VI 14 shows the X Y relationship between a temperature precipitation index and lo g reliability. High temperatures combined with large negative precipitation anomalies correlate to higher levels of energy not supplied (e.g. reduced reliability). In the Results section we will see if this relationship is statistically significant. Fi gure VI 11 Effect of temperature on energy demand in Delhi, India.

PAGE 124

108 Figure VI 12 Precipitation at State capital nearest neighbor weather stations (NOAA ISD) Figure VI 13 Monthly rainfall accumulation for 5 NR states (IITM)

PAGE 125

109 Figure VI 14 Log Reliability vs. Temperature Precipitation Index, Showing Group wise Line ar R egressions. Supply Chain Constraints The third and final set of variables we wish to consider are supply chain constraints. In this category we consider the primary raw materials for power generation: fuel and water. From a material balance perspective, the largest two inputs required for the rmal power generation are fuel for combustion and water for cooling. A standard 2x500 MW coal fired power station with wet cooling towers requires approximately 72 million liters of freshwater withdrawals per day, 90% of which is lost to evaporation and dr ift from the cooling towers, as well as smaller amounts to ash handling, dust suppression and other plant processes (CEA 2012b). Likewise, the same plant requires ~19,000 metric tones of coal per day (CEA 2010). Short supply of either raw material can lead to capacity de rating and/or unscheduled deviations from injection schedule. To represent the possibility of water supply constraints to thermal power generation in our model, we introduce a derived metric based on previous research. The proposed metric is a production weighted, supply chain water footprint of energy supply to beneficiary states. It is based on the capacity, fuel type and cooling type of contributing power stations. The water

PAGE 126

110 footprint is estimated in terms of both water withdrawals and w ater consumption for in boundary production (power stations located within the State) and trans boundary supply (energy allocated from power stations located outside the state). Water withdrawal intensity factors (WWIF) and water consumption intensity fact ors (WCIF), reported in units of liters of water per MWh e, are adopted from Cohen and Ramaswami (2014), with an India specific update for thermal power stations based on CEA (2013). For further explanation and methodological details on the water footprint of energy supply, please refer to Cohen and Ramaswami (2014). To represent the possibility of coal supply constraints to thermal power generation in our model, we compute monthly median coal stock position (in terms of days of coal on hand), based on sta tion wise daily coal reports from CEA. Due to irregular reporting frequencies by some stations, monthly median values are computed as a robust alternative to monthly mean. Figure VI 15 a shows the aggregate coal stock position of all India thermal power sta tions (TPS) for the last 5 years. The blue line represents the overall trend in fuel supply availability with 95% confidence intervals denoted by gray shading. Similarly, for gas power plants (GPP), we compute fleet capacity de ratings due to fuel supply s hortages, estimated as the daily gas supply on hand divid ed by daily requirement (Fig. 15c ). Capacity de ratings due to fuel supply shortages are represented in our model by a fleet availability factor (FAF), computed as the production weighted contributio n of each GPP to each beneficiary state in each month. For hydro, we compute the potential energy storage in all India HEP reservoirs at monthly intervals (Fig. VI 15b ). In the next section we introduce a hierarchical modeling framework to test the effect of these variables on electricity supply reliability. We will use the terms "PCA", "State/UT" and "group" interchangeably to describe the unit of analysis, depending on the context.

PAGE 127

111 Figure VI 15 Fuel supply to coal, gas and hydro po wer stations, All India, 2008 13

PAGE 128

112 Figure VI 16 Scatterplot matrix of log(ENS) versus supply chain variables Motivation for, and Applicability of, Hierarchical Models The underlying data used in this study consists of repeated observations for a set of (j) states/provinces over (i) occasions. Accordingly, the response vector Y and predictor matrix X each contain (i) x (j) = (n) rows. Response vector Y (log reliability) is observed at (j=5) states on (i=21) occasions, yielding (n=105) records. Predictor matrix X contains (n=105) rows of (p=12) predictor columns. From a statistical perspective, the rows of Y are not strictly independent because for each state (j) there are (i) repeated measurements subject to some set of shared characteristics inherent to group (j). This is analogous to studies in education research wherein standardized test scores of

PAGE 129

113 students within schools cannot be considered independent because each stu dent (i) attends some school (j) and is therefore subject to some school level characteristics shared by fellow students attending the same school. This non independence of observations violates one of the fundamental assumptions of single level regression techniques including ordinary least squares, weighted least squares and GLM (Valente and Oliveira 2011). As a result, hierarchical methods suited to nested data are required. According to Fox (2002, appendix p.1) hierarchical models (also known as mixed effect models ) are "appropriate for representing clustered, and therefore dependent, data arising, for example, when data are collected hierarchically, when observations are taken on related individuals (such as siblings), or when data are gathered over time on the same individuals." The analysis presented here belongs to the latter category repeated measurements over time for a set of power control areas. In addition to appropriate handling of nested data, hierarchical models provide explicit decomposit ion of the total variance of a dependent variable into within group and between group variance. Again providing an example from education research, hierarchical models are used extensively to test the significance of individual versus school level predicto rs of academic achievement (see Lubienski C. and S. 2006; Braun et al. 2006). In summary, hierarchical models offer two distinct advantages over single level regression: (1) appropriate handling of non independence due to repeated observations over time fo r a set of individuals/groups; and (2) straightforward application of control variables to adjust for differences in expectation between individuals/groups. Next, we present two different hierarchical models. The first tests the effect of structural PCA characteristics on log reliability and then passes the residuals onto second and third level models that apply climate and supply chain characteristics, resp ectively, to explain variance remaining in the error term. This we will call hierarchical linear model A (HLM A); the order of

PAGE 130

114 the models is hierarchical (structural, climate and supply chain). The second hierarchical approach is a mixed effects model designed to explicitly control for structural differences between groups while testing a set of explanatory variables. Rather than try to explain differences in the expectation between groups by regressing the response onto a set of control variables and t hen passing the residuals onto subsequent models, we instead allow random intercepts for each group to adjust for structural differences and test fixed effects of explanatory variables. Thus the random intercept captures variation between groups not explai ned by covariates in the model. We call this hierarchical linear model B (HLM B). In the next two sections we present results from HLM A, followed by HLM B. Hierarchical Model A The following vignette depicts the hierarchical framework in HLM A. Variance is conserved between multiple hierarchical levels through the error term each successive level tries to explain residual variance from the prior model. To avoid over fitting, we include a fit complexity tradeoff objective criteria that rewards for explai ning additional variance but penalizes for additional model terms. Here we use Akaike Information Criteria (AIC) and Bayesian Information Criteria (BIC) for model subset selection at each hierarchical level.

PAGE 131

115 Figure VI 17 Hierarchical Modeling Framework HLM A, Level 1 Level 1 of HLM A is simply a linear regression of the form $ % where the response vector Y is log(RNS) and the columns of predictor matrix X are the structural covariates: capacity adequacy (CapAdeq), fleet availability (FAF), the cost of unscheduled interchanges (UI_cost) and the percent of energy supplied from the grid (as opposed to in boundary generation; Pct_Grid). Solving the linear regression yields model coefficients in Table VI 5. Table VI 6 shows the ANOVA summary. Equation VI 8 shows the final model for HLM A level 1 (statistically significant variables on ly). Table VI 5 HLM A Level 1 Model Coefficients and ANOVA Estimate Std. Error t value Pr(>|t|) Signif. Code (Intercept) 3.62 2.98 1.214 2.27E 01 Cap_Adeq 3.81 0.586 6.488 3.37E 09 *** FAF 0.0112 0.0312 0.359 0.719993 UI_Cost 4.80E 05 1.25E 05 3.825 2.27E 04 *** Pct_Grid 5.22E 02 9.59E 03 5.441 3.78E 07 ***

PAGE 132

116 Table VI 6 HLM A Level 1 ANOVA Summary SST SSR SSE R 2 262 132 130 0.503 Equation VI 8 Final Equation for HLM A, Level 1 $ % $ ) +, % # $ / 0 % # $ 0 ,, 1, % # After fitting a linear model, it is imperative to check the assumptions of the residuals: normality, heteroscadasticity and independence. We also check for und ue influence of outliers (e.g. Cook's Distance) an d autocorrelation. Figure VI 18 shows a suite of model diagnostics designed to allow visual inspection of these assumptions. We repeat these checks at each hierarchical level. The residuals from level 1 p ass inspection: Normally distributed residuals (Q Q plot), No significant heteroscadasticity (residuals vs. covariate scatterplots are largely random), mild autocorrelation (up to lag 3 and some cyclical behavior at higher lags, although below a standard c ritical threshold), and no undue influence of outliers (all Cook's distance << 10).

PAGE 133

117 Figure VI 18 HLM A Level 1 Model Diagnostics HLM A, Level 2 Level 2 takes the residuals from level 1 and applies a secon d linear regression using environmental and climate explanatory variables. Again we employ the regression solution $ % where our response vector Y is now the residuals from level 1 and the columns of X are environment/climate variables: In boundary monthly maximum temperature (IB_MaxTemp) In boundary water withdrawal footprint (IB_WWF) Trans boundary water withdrawal footprint (TB_WWF) Monthly precipitation anomaly (P_Anomaly) Monthly precipitation accumulation (P_Act) Linear combinations (e.g. multiplicative effects) of these variables.

PAGE 134

118 Solving the linear regression yields model coefficients in Table VI 7. Table VI 8 shows the "best" model coefficients based on BIC subset selection. Table VI 9 shows the ANOVA summary. Equation VI 9 shows the final model for HLM A level 2 (statistically significant variables only) Table VI 7 HLM A Level 2 Initial Model Model Term Estimate Std. Error t value Pr(>|t|) Signif. Code (Intercept) 3.89E 02 1.15E 01 0.339 0.73565 IB_MAXTEMP 7.43E 04 3.23E 03 0.23 0.81869 IB_WWF 1.32E 05 2.26E 05 0.583 0.56126 TB_WWF 8.79E 05 1.39E 04 0.634 0.52817 P_Anomaly 7.81E 05 5.15E 04 0.152 0.87979 P_Act 1.25E 04 9.39E 05 1.336 0.18526 IB_MAXTEMP:IB_WWF 3.23E 07 6.29E 07 0.513 0.60917 IB_MAXTEMP:TB_WWF 1.79E 06 3.86E 06 0.463 0.64464 IB_WWF:TB_WWF 3.94E 08 3.81E 08 1.035 0.30385 IB_MAXTEMP:P_Anomaly 5.21E 06 1.35E 05 0.388 0.69933 IB_WWF:P_Anomaly 1.04E 07 1.09E 07 0.952 0.34405 TB_WWF:P_Anomaly 5.85E 07 6.78E 07 0.862 0.39099 IB_MAXTEMP:P_Act 5.30E 06 2.47E 06 2.148 0.03469 IB_WWFmean:P_Act 1.12E 08 3.55E 09 3.162 0.0022 ** TB_WWFmean:P_Act 8.35E 08 3.21E 08 2.602 0.01099 P_Anomaly_mm:P_Act 4.63E 07 2.45E 07 1.891 0.06217 IB_MAXTEMP:IB_WWF:TB_WWF 8.82E 10 1.07E 09 0.828 0.41011 IB_MAXTEMP:IB_WWF:P_Anomaly 3.43E 09 2.83E 09 1.214 0.22834 IB_MAXTEMP:TB_WWF:P_Anomaly 1.91E 08 1.73E 08 1.109 0.27055 IB_WWF:TB_WWF:P_Anomaly 2.57E 10 1.91E 10 1.346 0.18216 IB_WWF:TB_WWF:P_Act 2.02E 11 6.29E 12 3.204 0.00193 ** IB_MAXTEMP:P_Anomaly:P_Act 1.29E 08 6.40E 09 2.009 0.0478 IB_MAXTEMP:IB_WWF:TB_WWF:P_Anom 7.72E 12 4.87E 12 1.584 0.11694 Significant variables highlighted and bolded. Signif. codes: 0 *** 0.001 ** 0.01 ' 0.05 '.' 0.1 Table VI 8 HLM A Level 2 "Best Model" After BIC Subset Selection Estimate Std. Error t value Pr(>|t|) Signif. Code (Intercept) 1.42E+00 1.01E+00 1.407 0.163 IB_MAXTEMP 5.44E 02 2.80E 02 1.943 0.055 IB_WWF 9.35E 06 9.30E 05 0.101 0.920 P_Anomaly 1.84E 02 4.82E 03 3.824 0.000 *** IB_MAXTEMP:IB_WWF 1.37E 06 2.58E 06 0.531 0.596 IB_MAXTEMP:P_Anomaly 4.82E 04 1.22E 04 3.964 0.000 *** IB_WWF:P_Anomaly 1.43E 06 4.73E 07 3.028 0.003 ** IB_MAXTEMP:IB_WWF:P_Anom. 3.63E 08 1.16E 08 3.115 0.002 ** Significant variables highlighted and bolded. Signif. codes: 0 *** 0.001 ** 0.01 ' 0.05 '.' 0.1

PAGE 135

119 Table VI 9 HLM A Level 2 ANOVA Summary SST SSR SSE R 2 0.270 0.144 0.127 0.532 Equation VI 9 Final equation for HLM A, Level 2 2 $ 0 )1 1+ % # 2 $ 3 3, 14 % # 2 $ 4 )0 14 % # 2 2 $ 1, 33 % # 2 2 $ 3 ,/ 14 % Figure VI 19 HLM A Level 2 Model Diagnostics Again, we check modeling assumptions. The residuals appear normally distributed except at the very upper tail (see Q Q p lot, top left). Clustering arises in the TB_WWF vs. Residuals scatterplot, but this is mostly due to clustering in the X data and not heterscadasticity with the residuals. Other covariate vs. residual scatterplots pass inspection, but again we see clusteri ng due

PAGE 136

120 to the underlying nature of the covariate data -precipitation anomaly and accumulation tend towards zero and have positively skewed distributions. Autocorrelation attenuated from level 1 diagnostics and are minimal. There is no undue influence of outliers (all Cook's distance << 10). HLM A, Level 3 Level 3 takes the residuals from level 2 and applies a third linear regression, this time with supply chain variables. We start with the general form $ % where response vector Y is now the residuals from level 2 and predictor matrix X contains fuel supply chain variables: coal stock at thermal power stations (Coal_Stock), gas fleet availability reductions due to gas supply shortages (Gas_FAF) and hydroelectric reservoir storage (Hydr o_Storage). Solving the linear regression yields model coef ficients presented in Table VI 10. Table VI 11 shows the ANOVA summary. Equation VI 10 shows the final equation for HLM A level 3 (statistically significant variables only). Table VI 10 HLM A Level 3 Model Coefficients and ANOVA Estimate Std. Error T value Pr(>|t|) Signif. Code (Intercept) 0.070 0.046 1.525 0.130 Gas_FAF 0.129 0.069 1.869 0.065 Hydro_Storage 0.029 0.018 1.647 0.103 Table VI 11 HLM A Level 3 ANOVA Summary SST SSR SSE R 2 262 132 130 0.503 Equation VI 10 Final equaiton for HLM A, Level 3 $ 1 3,/ % # The residuals from level 3 pass inspection: Normally distributed residuals (Q Q plot), No significant heteroscadasticity (residuals vs. covariate scatterplots are largely random), mild autocorrelation (cyclical behavior but mostly below the critical thresh old), and no undue influence of outliers (all Cook's distance << 10).

PAGE 137

121 Figure VI 20 HLM A Level 3 Model Diagnostics Cumulative HLM A Each level of the hierarchical model provides new information and improves goodness of fit Figure VI 21 shows the model fit progression at each level. Table VI 12 provides a summary table of ANOVA at each hierarchical level and the cumulative model. Note that SST, SST and SSE are identical for models evaluated in the same space (e.g. original or log transfo rm). Table VI 12 HLM A ANOVA Summary Table: All Levels Model SST SSR SSE Pearson R 2 L1 Gamma GLM 0.29 0.15 0.25 0.22 L1 Lognormal GLM (log space) 262.36 132.13 130.23 0.50 L1 Lognormal "Best" GLM (log space) 262.36 131.96 130.40 0.50 L2 Best model (log space) 130.40 45.57 84.82 0.35 L3 Best model (log space) 84.82 13.69 71.13 0.16 L1 + L2 cumulative (log space) 262.36 220.51 84.82 0.68 L1 + L2 + L3 cumulative (log space) 262.36 235.83 71.13 0.74 L1 + L2 + L3 cumulative (back transformed) 0.295 0.648 0.336 0.483 L1 Lognormal Only (back transformed) 0.295 0.235 0.291 0.225

PAGE 138

122 Figure VI 21 HLM A Cumulative Fit Finally, we compare HLM A against a single level GLM with all of the predictors. Figure VI 22 Model Comparison: HLM A vs. Single Level GLM w. All Predictors

PAGE 139

123 Hierarchical Model B Mixed Effects Model s Consider we wish to estimate the energy not supplied to a given state in a given month. Let this be our response variable with subscripts i and j denoting the state and month, respectively. In the simplest case, we could estimate as the expec ted value of all the observations (the grand mean ), plus the expected deviation of an observation for $ % from the grand mean, plus a normally distributed error term $ % The error term represents deviations from the expectation not captured by the model. Mathematically, this is: (eqn. VI 12 ), where is normally distributed with mean zero and variance sigma squared. Now suppose we have additional knowledge of the system (and requisite data) to help explain more of the varian ce in That is, reduce the variance contained in the error term. Our hypothesis was that climate and supply chain constraints can help explain variance in electricity supply reliability after adjusting for structural differences between power contro l areas. To do so, and keeping generalizability in mind, we propose a random intercepts, fixed slope hierarchical model in which the intercepts are allowed to vary by group to adjust for differences in response expectation between groups, and the slope coe fficients of the predictor variables are fixed. Fixed slopes (as opposed to random slopes) facilitate generalizability by reducing the number of estimated parameters and thus the amount of data required to fit the model. Additionally, fixed slopes facilit ate interpretability by offering a more physical based explanation of the model parameters. For example, for a one standard deviation increase in predictor we expect a

PAGE 140

124 change in the log response equivalent to the slope for all groups. If we had random slopes, this effect would be different for each group, confounding generalizability. To turn the simplest case hierarchical model (equation VI 11 ) into a fixed slope, random intercepts model, we decompose the error term with a second level r egression (equation VI 12 ). We interpret as the deviation from the expected value of $ # ) for occasion i,j. Equation VI 11 Mixed Effects Model, Level 1 Equation VI 12 Mixed Effects Model, Level 2 5 5 5 # 5 Finally, we fit the hierarchical model using the lme4 package in the statistical computing environment R, and examine the results, described next. HLM B Results Model results support our hypothes is that climate and supply chain constraints help explain variability in electricity supply reliability in Northern India. In fact, climate and supply chain variables explained more of the variance in log reliability than structural characteristics after a djusting for differences in the expectation between groups.. Combined with results from HLM A where structural variables were highly significant (previous section), this suggests that climate and supply chain variables are particularly suited to explaining month to month variability (e.g. deviations from the mean), while structural variables are more useful in explaining the mean itself. In other words, climate and supply chain variables perform well in predicting deviations from the expected reliability fo r a given state, while structural variables explain why some states have higher mean reliability than others. As a quick illustration, Figure

PAGE 141

125 VI 23 shows state average ENS versus state average Capacity Adequacy over a two year period. Indeed we see that Capacity Adequacy is positively correlated with log reliability. Figure VI 23 The Effect of Capacity Adequacy on Grid Reliability The statistical significance of climate and supply chain variables in explaining deviations from expected grid reliability are based on three lines of corroborating evidence analysis of variance (Tables VI 13 VI 15 ), log likelihood ratio tests comparing nested models (Tables VI 17 VI 18 ), and visual inspection of observed versus fitted response in log (model) space and back transformed (response) space (Figures VI 24 ). Starting with Table VI 13 (ANOVA for the null model: structural predictors only) cap acity adequacy is significant at the 90% conf idence interval (CI). In Table VI 14 (ANOVA for

PAGE 142

126 the alternate model) capacity adequacy remains significant, plus all of the introduced climate predictors (temperature, precipitation anomaly, precipitation accumu lation and a joint temperature precipitation anomaly index), and three of the four supply chain predictors (water requirement of the energy system, coal stock on hand at thermal power stations, and hydroelectric potential energy storage). To test the over all significance of adding climate and supply chain predictors, we perform a log likelihood ratio test between the null (simple) model and alternate (full) model. The test is highly significant (p value<0.001), indicating the inclusion of climate and suppl y chain predictors is justified (Table VI 17 ). Now knowing that that the alternate (full) model is a significant improvement over the null (simple) model, we search for the most parsimonious nested subset of the full model. To do so, we remove predictor s ets one at a time, compute the log likelihood and compare the test statistic with that of the full model. We do this for sets of predictors (e.g. structural, climate, supply chain) rather than for every combination of individual predictors to avoid a so ca lled "fishing expedition". Exhaustive subset selection (over all combinations of predictors), particularly for large numbers of predictors, has been criticized for lacking hypothesis. By searching a large k dimensional space for the best combination of pre dictors, we increase the probability of finding a good fit by chance rather than by hypothesis. To avoid this potential pitfall, we test only subset combinations motivated by our research question: Do climate and supply chain information help explain varia bility in power supply reliability observed in Northern India if we control for differences in the expectation between states? To answer that question, we test a priori combinations of predictor sets given by (2^n) 1, where n=3 sets and excluding the empty set with no predictors. If, by contrast, we searched all combinations of

PAGE 143

127 predictors and did not constrain the model to a fixed number of parameters, then we would have (2^n) 1 = 4,095 combinations to compute and interpret. Table VI 13 Fixed Effects for HL M with Structural Predictors Only (Null Model) Coef. Std. Error DF t value p value Signif. Code (Intercept) 4.54363 0.64739 96 7.018 0.000 *** Cap. Adequacy 0.33682 0.16978 96 1.984 0.050 ** Cost of UI 0.08951 0.12210 96 0.733 0.465 Cap. Availability 0.02428 0.10736 96 0.226 0.822 Pct. Imports 0.08913 0.18837 96 0.473 0.637 Table VI 14 Fixed Effects for HLM with Structural, Climate & Supply Chain Predictors (Alternate Model) Coef. Std. Error DF t value p value Signif. Code (Intercept) 4.54363 0.45318 88 10.026 0.000 *** Temperature 0.41973 0.15300 88 2.743 0.007 *** Precip. Anomaly 1.77549 0.85703 88 2.072 0.041 ** Precip. Accum. 0.26554 0.11866 88 2.238 0.028 ** Temp:Precip 2.07318 0.85696 88 2.419 0.018 ** Water Requirement 1.06114 0.37391 88 2.838 0.006 *** Coal Stock 0.60416 0.17397 88 3.473 0.001 *** Gas Stock 0.02283 0.11132 88 0.205 0.838 Hydro Storage 0.24384 0.12384 88 1.969 0.052 Cap. Adequacy 0.36399 0.14847 88 2.452 0.016 ** Cost of UI 0.01747 0.10399 88 0.168 0.867 Cap. Availability 0.07168 0.09054 88 0.792 0.431 Pct. Imports 0.05384 0.20068 88 0.268 0.789 Table VI 15 Fixed Effects for HLM with Climate & Supply Chain Pred ictors (Best Model ) Coef. Std. Error DF t value p value Signif. code (Intercept) 4.54363 0.53071 92 8.561 0.000 *** Temperature 0.42468 0.13710 92 3.097 0.003 *** Precip. Anomaly 1.87030 0.84927 92 2.202 0.030 ** Precip. Accum. 0.28073 0.10596 92 2.649 0.010 *** Temp:Precip 2.24776 0.85020 92 2.644 0.010 *** Water Requirement 1.36248 0.39147 92 3.480 0.001 *** Coal Stock 0.50099 0.16883 92 2.967 0.004 *** Gas Stock 0.01707 0.10925 92 0.156 0.876

PAGE 144

128 Hydro Storage 0.28905 0.11940 92 2.421 0.018 ** Table VI 16 Fixed and Random Intercepts for "Best Model" Log Reliability (Log of Energy Not Supplied in GWh) Energy Not Supplied (GWh shortage/month) (Intercept) Grand Mean Expected Deviation from Grand Mean Combined Expectation Grand Mean Expected Deviation from Grand Mean Combined Expectation Delhi 4.54 1.82 2.73 94.03 0.16 15.30 Haryana 4.54 0.97 5.51 94.03 2.63 247.55 Punjab 4.54 1.34 5.88 94.03 3.82 359.30 Rajasthan 4.54 0.14 4.41 94.03 0.87 82.11 UP 4.54 0.36 4.19 94.03 0.70 65.78 Subset selection results are shown in table VI 18 The log likelihood ratio tests indicate that, in all cases, the full model is justified based on fit complexity tradeoff at the 90% confidence interval. At the 95% and 99% CI this holds true for all but one model (model 7, wherein structural predictors are omitted altogether). For model 7, at the 95% CI we cannot reject the null hypothesis that it is different from the full model. That is, the model fit is not significantly diminished when we omit structural predictors and use only climate and supply chain predictors. This agrees with inference from analysis of variance presented in tables VI 13 VI 15 which indicated that climate and supply chain predictors explained most of the variance in log reliability, and structural predictors explained little more. Table VI 17 Hypothesis Test Comparing Log likelihood Ratio to Chi Square Distribution Mod Description df AIC BIC logLik Test L.Ratio p value Null ST 7 321 339 153 Alt ST + CL + SC 15 277 317 123 1vs2 59.999 <.0001 Legend: ST = structural; CL= Climate; SC=supply chain

PAGE 145

129 Table VI 18 Subset Selection Using Log likelihood Ratio Test Model Description df AIC BIC logLik Test L.Ratio p value 1 ST + CL + SC 15 277 317 123 2 ST + CL 11 306 335 142 1vs2 36.96 <.0001 3 ST + SC 11 296 326 137 1vs3 27.67 <.0001 4 ST 7 321 339 153 1vs4 60.00 <.0001 5 CL 7 302 320 144 1vs5 41.19 <.0001 6 SC 7 310 329 148 1vs6 49.70 <.0001 7 CL + SC 11 278 307 128 1vs7 8.79 0.0664 In addition to statistical inference vis ˆ vis analysis of variance and likelihood ratio tests, we present visualizations of the model fit. Figure VI 10 compares fitted versus observed response in model space (lognormal) for the null model, alternate model and "best" subset model based on the Bayesian Information Criteria (BIC) and the log likelihood ratio test. Figure VI 12 shows fitted versus observed response for the same three models after back transforming from log space. In both spaces, we see signifi cant improvement from the null model to the alternate model when climate and supply chain predictors are introduced. Removing structural predictors has little to no impact on model fit.

PAGE 146

130 Figure VI 24 Comparing Fitted vs. Observed Response in log space (left column) and back transformed to the original response space (right column) Discussion This study highlights the need for expanding the system boundaries to include climate and infrastructure sup ply chains when evaluating risks to, and reliability of, energy systems. R 2 =.727 R 2 =.787 R 2 =.786 R 2 =.764 R 2 = .865 R 2 =. 855

PAGE 147

131 Reliability analysis and/or planning focused only on endogenous attributes such as capacity adequacy, availability and utilization miss the contribution of exogenous factors such as c limate and infrastructure supply chains, and the potential relationships between the two. For example, low capacity utilization in a power control area with unmet demand likely indicates uneconomic dispatch of costly generating units. Uneconomic dispatch, in turn, may be a manifestation of higher fuel costs due to fuel supply chain constraints. In fact, many GPP in northern India are equipped to handle fuel switching in case of short supply of the primary fuel (CEA 2012c). In addition to the extra capital c ost for fuel switching capability, replacement fuels such as naphtha are often more expensive than the primary fuel (CEA 2012c). This added cost of adaptation is also true for the second input to thermal power generation: water. Some very large thermal pow er stations in India are currently undergoing expensive retrofits to accommodate cooling water switching to hedge against decreased supply and increased competition for water. Switching from open loop cooling with an instantaneous water demand of ~10^5 m^3 /hr for a 750MW TPS to closed loop cooling towers with an instantaneous water demand ~4000 m^3/hr alleviates risk to production due to water withdrawal constraints while actually increasing total watershed consumption (NTPC 2013). Based on empirical data, this study found the water intensity of power generation to be significantly and inversely correlated to reliability: for a 1 standard deviation (sd) increase in water withdrawals, we expect a 1.36 sd increase in energy not supplied. Thus, reliabilit y goes down as water requirements go up. Similarly, the concurrence of high temperatures and below average precipitation has a negative effect on reliability. In terms of fuel, we see that coal stock is inversely correlated to energy not supplied as the co al supply on hand at TPS dwindles, reliability suffers.

PAGE 148

132 These are important considerations, particularly in the context of energy planning and evaluation of competing technology scenarios to meet future demand. In the India context, electricity use has inc reased an astounding 70% over the past decade and is projected to grow faster still, doubling from the current total in the next decade (Ministry of Power 2012). To meet such rapid growth, it is important to consider the upper limits of biophysical and inf rastructural support systems that may have significant impacts on reliability. For example, is there enough water available to meet future cooling loads if most of the growth in installed capacity will come from large thermal power stations? The Indian Min istry of Power calls for an additional 118 GW of thermal power generation by 2022 (Ministry of Power 2012). Assuming a capacity factor of 85%, this proposed capacity expansion would require an additional 3.5 billion cubic meters of water per year, nearly a ll of which would be lost to evaporation and drift from cooling towers. Similarly, can the aging rail network, much of it built by the British over 100 years ago, consistently deliver coal on time and in adequate quantity to meet the demands of the dozens of new TPS across the country? In light of these "exogenous" considerations, alternative energy technology scenarios may look increasingly attractive. How do cooling water constraints and fuel supply constraints change the expected reliability of thermal p ower generation, and thus the reliability comparison with wind and solar, which have near zero operational requirements for fuel and water? These questions are worth special consideration in light of ambitious targets to meet continued, exponential growth in energy demand while simultaneously improving reliability and staying within the limits of biophysical and infrastructural support systems that make energy delivery possible.

PAGE 149

133 APPENDIX A DATA SOURCES FOR THE INDIAN POWER SEC TOR Sub category Data Unit Avail. Source System wide Power Supply Position Monthly generation GWh '05 13 CEA Monthly Reports > Power Supply Position > Year > Energy (PDF) PLF (e.g. capacity factor) MW '05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > All India (PDF) Inter Regional Exchanges Monthly Inter Regional Energy Exchanges GWh '05 13 CEA Monthly Reports > Inter Regional Energy Exchanges (PDF) Price Rate of Sale of Power Paise/ KWh '10 11 CEA Annual Reports Frequency dependent UI rate Paise/ KWh 08 13 NRPC > Commercial Activities > UI account State wise Power Supply Position Peak Requirement MW '05 13 CEA Monthly Reports > Power Supply Position > Peak Energy Requirement MU '05 13 CEA Monthly Reports > Power Supply Position > Energy Generation Net Drawal from Grid MU '05 13 CEA Monthly Reports > Inter Regional Energy Exchanges (PDF) Monthly Generation MU '05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > Regionwise, Statewise, Stationwise Details Installed Capacity located in State/UT MW '05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > Regionwise, Statewise, Stationwise Details including allocated shares in Central Sector (aggregated, not station wise) MW '05 13 CEA Monthly Reports > All India Installed Capacity State Wise/Utility Wise Station Wise List of stations located in State/UT Stn name '05 13 CEA Monthly Reports > Generation Overview Report > Year > Month > Regionwise, State wise, Station wise Detail Monthly Generation MWh '05 13 Capacity MW '05 13 PLF (Plant Load Factor) % 2005 '05 13

PAGE 150

134 PAFM (Plant Availability Factor Monthly) % '05 13 NRPC > Archive > Regional Energy Account > Month Fuel type (e.g. coal, gas, hydro) type 05 13 Cooling type (e.g. once through, cooling tower, cooling pond, etc..) type NOT AVAILABLE W ater withdrawals & consumption (actual) Lw/M Wh NOT AVAILABLE Environment Water withdrawals & consumption (d esign) M^3/ hr One time CEA Report on minimization of water requirement in coal based thermal power stations, appendix 11 City wide Consumption Electricity use GWh 05 12 Delhi Statistical Handbook Consumption Electricity use MWh 2009 Chavez et al. (2012)

PAGE 151

135 APPENDIX B SURVEY OF ELECTRICIT Y INFRASTRUCTURE OPE RATORS SERVING DELHI Given the strategic role of developing world cities in addressing global sustainability challenges from supporting human development aspirations and improving public health to minimizi ng environment impact (locally and globally) and protecting critical ecosystems we seek to understand how infrastructure designers, operators and related policymakers balance these often competing priorities. As part of this broader research initiative, I am conducting a survey of electricity infrastructure providers serving NCT Delhi. From this process, we hope to: Identify key priorities in current operations and future planning of electricity infrastructure in India Map the key players (infrastructure designers/operators and regulatory/policy organizations) and understand how they interact with one another as well as with users of electricity One potential outcome of this research is to facilitate communication and c oordination between infrastructure providers and policymakers working independently towards similar goals. One potential outcome of this research is to identify sustainability initiatives aligned with the priorities of key actors in the electricity supply chain serving Delhi. The following questionnaire has been developed to specifically address these questions. PART I: Service Provision 1. Please indicate if your organization is involved primarily with electricity generation, transmission, distribution or gri d operation (circle all that apply). 2. What types of customers or clients do you serve? a. Electricity end use customers in the residential, commercial, industrial and government sectors (e.g. households, businesses and government offices) b. Critical infrastructures that rely on electricity (e.g. railway, metro, traffic lights, water/sewage utilities, hospitals, information/communication systems) c. Intermediary electricity infrastructure providers such as transmission and distribution companies. 3. Are any groups of customers prioritized in terms of rate structure ? If so, please explain the motivation for such a decision. 4. Are any groups of customers prioritized during periods of limited supply; e.g. are any customers protected from load shedding ?

PAGE 152

136 PART II: Operations and Management 1. Please indicate which of the following are priorities for your organization at the present time and in future planning. Please indicate Yes/No and rank the top 3. Current Future Improving reliability of electricity supply to existing customers Expanding service coverage to new or under served areas Lowering cost to customers Reducing local/regional pollution Reducing global greenhouse gas emissions Reducing water use at thermal power stations Other Please specify For every Yes indicated above, please provide at least one specific example of how that priority influences your day to day operational decisions. 2. Which factors have contributed (or you expect will contribute in the future) to power cuts (both notified and unscheduled)? Please indicate Yes/No and rank the top 3. Current Future Insufficient generation capacity Under utilization of existing genera tion capacity (e.g. too expensive for dispatch) Insufficient transmission/distribution infrastructure Fuel supply disruptions to thermal power stations (e.g. coal/gas shortages) Insufficient or poor quality cooling water at thermal power stations Insufficient water supply for hydroelectric power generation Difficulty securing necessary permits (e.g. land permit or water use permit for new thermal power stations) Very high or unpredictable consumer demand (e.g. lack of demand side manageme nt) Extreme weather such as heavy monsoon rains that disrupt transmission/distribution infrastructure Extreme weather such as heatwave that increases demand 5. In general, how does water quality and availability relate to any of the priorities or risk factors described above? Who do you interact with regarding water issues? For example, is water critical to securing environmental permits? Do thermal power stations engage with State irrigation departments during t imes of drought/flooding regarding water quantity/quality?

PAGE 153

137 6. In the past year, how many days have water related issues disrupted your organization's power generation, transmission or distribution? PART III: Future Electricity Planning 1. What factors will contribute to expected growth in energy demand over the next decade? 2. Is water scarcity or poor water quality a risk factor to your organization's activities? 3. How is your organization addressing water in future resource planning? a. What is the time horizon for these plans (e.g. 5 years, 10 years, 30 years)? b. What organizations are involved? c. Is your organization making changes to the way it secures water rights? d. Is your organization planning to modify the fuel type, cooling typ e, location or operation of any new or existing thermal power stations because of water availability or water quality? e. Is your organization planning any new hydroelectric projects or planning to modify any existing ones? If so, how and why? PART IV: Linkages With Infrastructure Providers and Regulatory Bodies In the first half of Part IV, we are trying to understand how your organization is connected to infrastructure designers and operators. We are interested in the frequency and type of interaction. For example how often does a power generation company coordinate with a water utility? Please indicate how often you interact with each organization listed below in their capacity as an INFRASTRUCTURE DESIGNER or OPERATOR, using a 0 to 4 scale: 0 -No interaction 1 -Rare or Infrequent 2 -Annual 3 -Monthly

PAGE 154

138 4 -Daily Organizations within individual cities such as the municipal corporation : a. MCD Please specify department (0 1 2 3 4) b. NDMC -Please specify department (0 1 2 3 4) c. City government hospitals (0 1 2 3 4) d. Distribution companies (DISCOMS) operating in Delhi: BRPL (0 1 2 3 4) BYPL (0 1 2 3 4) NDPL (0 1 2 3 4) NDMC (0 1 2 3 4) MES (0 1 2 3 4) e. Other -Please specify (0 1 2 3 4) State Level Organizations a. State generating stations (GENco's, e.g. IPGCL) (0 1 2 3 4) b. State Transmission Utility (Delhi Transco Limited) (0 1 2 3 4) c. State Load Dispatch Center (SLDC) (0 1 2 3 4) d. Area Load Dispatch Centers (A LDC) (0 1 2 3 4) e. Water infrastructure (e.g. Delhi Jal Board) (0 1 2 3 4) f. Transportation infrastructure (e.g. Delhi Metro) (0 1 2 3 4) g. Delhi Government hospitals (e.g. LNJP Hospital) (0 1 2 3 4) h. Delhi Police (0 1 2 3 4) i. Delhi Public Works Department (PWD) (0 1 2 3 4) j. Other Please Specify Central or Regional Level Organizations a. National Load Dispatch Center (NLDC) (0 1 2 3 4) b. Regional Load Dispatch Center (RLDC) (0 1 2 3 4) c. POWEGRID (CTU) (0 1 2 3 4) d. Central Sector Generating Stations (CSGS) (0 1 2 3 4) e. National Thermal Power Corporation (NTPC) (0 1 2 3 4) f. National Hydro Power Corporation (NHPC) (0 1 2 3 4) g. India Railways (0 1 2 3 4) h. Central Sector hospitals (e.g. AIIMS) (0 1 2 3 4) i. Other Please Specify Private a. Coal/gas suppliers (0 1 2 3 4) b. Equipment manufacturers (0 1 2 3 4) c. Other Please specify (0 1 2 3 4) In the second half of part IV, we are trying to understand how your organization is connected to regulatory and policy making organizations at different levels of government. Please indicate how often you interact with each REGULATORY or POLICY MAKING organization listed below, using a 0 to 4 scale: 0 -No interaction 1 -Rare or Infrequent

PAGE 155

139 2 -Annual 3 -Monthly 4 -Dai ly City Level Governmental Organizations a. NDMC -Please specify department (0 1 2 3 4) b. MCD Please specify department (0 1 2 3 4) c. District Public Grievance Redressal Committee (e.g. subcommittee to monitor the availability of proper power supply) (0 1 2 3 4) d. Designated development zones (e.g. industrial clusters) (0 1 2 3 4) e. Other -Please specify State Government (Delhi NCT Government) a. State Electricity Regulatory Commission (SERC) (0 1 2 3 4) b. Energy Efficiency & Renewable Energy Management C entre (0 1 2 3 4) c. Delhi Development Authority (DDA) (0 1 2 3 4) d. Delhi Pollution Control Committee (DPCC) (0 1 2 3 4) e. Delhi Environment Department (0 1 2 3 4) f. Delhi Climate Planning Unit (0 1 2 3 4) g. Delhi Urban Development Department (0 1 2 3 4) Central Government (Government of India) a. Ministry of Power (MoP) (0 1 2 3 4) b. Central Electricity Authority (CEA) (0 1 2 3 4) c. Central Electricity Regulatory Commission (CERC) (0 1 2 3 4) d. Department of Atomic Energy (0 1 2 3 4) e. Central Pollution Control Board (CPCB) (0 1 2 3 4) f. Jawaharlal Nehru Urban Renewal Mission (JNNURM) (0 1 2 3 4) Please indicate which of the organizations identified above does your organization interact with regarding water issues (e.g. quantity, quality, permits, payments, etc. ). Is there anything else you would like to tell us about your organization's interaction with energy users infrastructure providers or regulatory / policy organizations that we may have missed? We sincerely thank you for your time and contribution to this study. We will keep you informed of any scholarly publications arising from this work.

PAGE 156

140 REFERENCES Allan, J.A. Virtual Water; A Strategic Resource Global Solutions to Regional Deficits. Groundwater 36(4): 545 546. Ball, Richard; Ross Cullen and Christopher Gan (1999) The diffusion of energy efficiency innovations among residential energy consumers, Ne w Zealand Economic Papers, 33:1, 115 135, DOI: 10.1080/00779959909544300 Barpanda, S.S., 2013. Personal communication with Area General Manager, Operations Department, National Load Dispatch Center, Delhi, India March 2013. Barton, Brooke. 2010. Murky Wat ers? Corporate Reporting on Water Risk: A benchmarking study of 100 companies Ceres. Boston, MA. Available online: http://www.waterfootprint.org/Reports/Barton_2010.pdf Baynes, T., M. Lenzen, J. K. Steinberger and X. Bai. 2011. Comparison of household consumption and regional production approaches to assess urban energy use and implication for policy. Energy Policy 39: 7298 7309. Billinton, R. 1970. Reliability evaluation of power systems: First Edition Amsterdam: Gordon and Breach Science Publishers. Blackhurst, M.; C. Hendrickson, and J. Sels I Vidal. 2010. Direct and Indirect Water Withdrawals for the U.S. Industrial Sectors. Environmental Science & Technology 44 : 2126 2130 Braun, H. I.; A. Wang; F. Jenkins; and E. Weinbaum. 2006. The Black White Achievement Gap: Do State Policies Matter? Education Policy Analysis Archives 2006, Vol. 14 Issue 8, p1 107. 107p. CEA [Central Electricity Authority]. 2013. Load Generation Balance Report for the year 2012 13 Annual Reports of the Central Electricity Authority of India: New Delhi, May 2013. CEA. 2012a. Operation performance of generating stations in the country during the year 2011 12 Grid Operation & Distribution Wing, Op eration Performance Monitoring Division: New Delhi, April 2012. CEA. 2012b. Report on minimization of water requirement in coal thermal power stations Water Systems, Thermal Engineering & Technology Division : New Delhi, January 2012. CEA. 2012c. Monthly f uel supply and consumption report for gas based power station Monthly Reports of the Central Electricity Authority of India: New Delhi, January 2012. Central Electricity Authority. 2012. Operation performance of generating stations in the country during t he year 2011 12. Grid Operation & Distribution Wing, Operation Performance Monitoring Division: New Delhi, April, 2012.

PAGE 157

141 Chavez, A. and A. Ramaswami. 2011. Progress Towards Low Carbon: Approaches for Trans Boundary Greenhouse Gas Emissions Footprinting for Cities. Carbon Management. 2(4): 471 482. Chavez, A., A. Ramaswami, N. Dwarakanath, R. Ranjan and E. Kumar. 2012. Implementing Trans Boundary Infrastructure Based Greenhouse Gas Accounting for Delhi, India. Journal of Industrial Ecology 16(6): 814 828. Che sapeake Energy. 2011. Water Use in Barnett Deep Shale Gas Exploration: Fact Sheet. www.chk.com/media/educational library/fact sheets/barnett/bar nett_water_use_fact_sheet.pdf Accessed April 2013. Chiu, Yi Wen, B. Walseth and Sangwon Suh. 2009. Water Embodied in Bioethanol in the United States. Environmental Science & Technology 43: 2688 2692. Clean Air Clean Jobs Act 2010. House Bill 10 1365, Gen eral Assembly of the State of Colorado. www.leg.state.co.us/CLICS/CLICS2010A/csl.nsf/fsbillcont3/0CA296732C8CEF4D8725 76E40 0641B74?Open&file=1365_rer.pdf Accessed March 2013. CMU (Carnegie Mellon University) Green Design Institute. 2013. Economic Input Output Life Cycle Assessment (EIO LCA) US 2002 (428) model [Internet]. www.eiolca.net/. Accessed 24 March 2013. COGCC (Color ado Oil and Gas Conservation Commission). 2011. Personal communication with on duty engineering manager, Colorado Oil and Gas Conservation Commission, Denver, CO, April 2011. Cohen, E and A. Ramaswami. 2014. The Water Withdrawal Footprint of Energy Suppl y to Cities. Journal of Industrial Ecology 18: 26 39. doi: 10.1111/jiec.12086 Coltrane, S., Archer, D., Aronson, E., 1986. The social psychological foundations of successful energy conservation programs. Energy Policy 14 (2), 133 148. Compston, Hugh (2009) Networks, resources, political strategy and climate policy, Environmental Politics, 18:5, 727 746, DOI: 10.1080/09644010903157032 Costanzo, M; Archer, D; Aronson, E; Pettigrew, T (1986): Energy Conservation Behaviour: the difficult path from inform ation to action. American Psychologist, 41(5), 521 528. CPCB (Central Pollution Control Board), 2006 2007. Water Quality Status of Yamuna River, Assessment and Development of River Basin Series: ADSORBS/41/2006 2007. Published by Central Pollution Control Board (CPCB), Delhi, India. Darley, J M (1977): Energy conservation techniques as innovations, and their diffusion. Energy and Buildings, 1, 339 343. Darley, J M; Beniger, J R (1981): Diffusion of energy conserving innovations. Journal of Social Issues, 37(2), 151 171.

PAGE 158

142 Debnath, K. and L. Goal. 1995. Power system planning a reliability perspective. Electric Power Systems Research 34 (1995): 179 185. Dennis, M L; Soderstrom, E J; Koncinski, W S; Cavanaugh, B (1990): Effective dissemination of energy r elated information: Applying social psychology and evaluation research. American Psychologis, 45(10), 1109 1117. Denver Water. 2010. Comprehensive Annual Financial Report for the year ending Dec. 31, 2009. www.denverwater.org/docs/assets/69255A63 E313 AE72 AAD0D59021152AE4/2009_annual_report.pdf Accessed April 2012. Denver Water. 2011. Denver Water's Conservation Plan, 2011 edition. www.denverwater.org/Conservation/ConservationPlan/ Accessed April 2012. Dominguez Faus, R., S. Powers, J. Burken and P. Alvarez. 2009. The Water Footprint of Biofuels: A Drink or Drive Issue? Envi ronmental Science & Technology 43(9), 3005 3010. EAI (Energy Analysts International). 2011. Denver/North Front Range Fuel Supply Chain Costs and Impacts. EAI: Westminster, CO. www.ozoneaware.org/postfiles/2011/news/DenverNorthFrontRangeFuelSupplyCostImpa cts_EAIInc_2011_REV 2.pdf Accessed April 2012. eGRID (Emissions & Generation Resource Integrated Database). 2010. U.S. Environment al Protection Agency, eGRID2010 (year 2007), Version 1.1. http://www.epa.gov/egrid Accessed April 2011. EIA (Energy Information Administration). 2010. Annual Energy Outlook 2010 Washington, DC: EIA, U.S. Department of Energy. http://38.96.246.204/forecasts/aeo/data.cfm Accessed May 2012. EIA (Energy Information Administration). 2013. Electricity data: Average retail price of electricity to ultimate customers, 2013 http://www.eia.gov/electricity/sales_revenue_price/pdf/table4.pdf Accessed March 2013. Energy Independence and Security Act of 2007 (EISA 2007). 2007. Public Law 110 14 0, www.govtrack.us/congress/bills/110/hr6. Accessed March 2013. EPRI (Electric Power Research Institute). 2005. Framework to evaluate water demands and availability for electrical power production within watershed across the United States: development and applications. Available online: http://my.epri.com/portal/server.pt?Abstract_id=000000000001010116 EPRI (Electric Power Research Institute). 2005. Security, Quality, Reliabil ity and Availability: Metrics Definition: Progress Report EPRI, Palo Alto, CA: 2005. 1008569. Friedman, S R; Neaigus, A; Jose, B; Curtis, R; Goldstein, M; Ildefonso, G; Rothenberg, R B; Des

PAGE 159

143 Jarlais, D C. (1997). Sociometric risk networks and risk for HIV infection. American Journal of Public Health, ISSN 0090 0036, 08/1997, Volume 87, Issue 8, pp. 1289 1296 Fleischauer, E. 2010. Browns Ferry Too Hot to Handle. Decatur Daily [Texas], August 18 2010. http://www.decaturdaily.com/stories/Browns Ferry too hot to handle,66754 Accessed November 2011. Forster, H. and J. Lilliestam. 2010. Modeling thermoelectric power generation in view of climate change. Regional Environmental Change 10: 327 338. Fox, J. 2002. An R and S PLUS Companion to Applied Regression. Available online: http://cran.r project.org/doc/contrib/Fox Companion/appendix .html Accessed November 2013. Front Range Energy(FRE). 2011. Personal communication with on duty plant manager, Front Range Energy. Windsor, CO, January 2011. Fthenakis, V. and H. C. Kim. 2010. Life cycle uses of water in U.S. electricity generation. Rene wable and Sustainable Energy Reviews Elsevier 14(7): 2039 2046. Gallucci, M. 2013. Boulder likely to adopt its own green utility and risks of going solo. InsideClimate News Jan. 23 2013. http://insideclimatenews.org/news/20130120/clean energy renewable energy climate change global warming boulder municipalization xcel energy coal germany Accessed March 2013. GEO (Governor's Energy Office of Colorado). 2010. 2010 Colorado Utilities Report http://www.colorado.gov/cs/Satellite/GovEnergyOffice/CBON/125159777 4824 Accessed May 2012. Goodwin, S., K. Carlson, C. Douglas and K. Knox. 2012. Life cycle analysis of water use and intensity of oil and gas recovery in Wattenberg Field, Colorado. Oil and Gas Journal 7 May 2012. Hillman, T., and A. Ramaswami. 2010. Gree nhouse Gas Emission Footprints and Energy Use Benchmarks for Eight U.S. Cities. Environmental Science & Technology 44(6), 1902 1910. Hillman, T., Janson, B., Ramaswami, A. 2010. Spatial Allocation of Transportation Greenhouse Gas Emissions at the City Scal e. Journal of Transportation Eng ineering 137(6), 416 425. Hoekstra, A.Y., and A.K. Chapagain. 2007a. Water footprints of nations: water use by people as a function of their consumption pattern. Water Resources Management 21(1): 35 48. Hoekstra, A.Y., and A .K. Chapagain. 2007b. The water footprints of Morocco and the Netherlands: global water use as a result of domestic consumption of agricultural commodities. Ecological Economics 64(1): 143 151.

PAGE 160

144 Hoekstra, A.Y., and P.Q. Hung. 2002. Virtual water trade: a q uantification of virtual water flows between nations in relation to international crop trade, Value of Water Research Report Series No.11, UNESCO IHE, Delft. Hoekstra, A.Y., and P.Q. Hung. 2005. Globalisation of water resources: international virtual water flows in relation to crop trade. Global Environmental Change 15(1): 45 56. House Bill 10 1365: Clean Air Clean Jobs Act. 2010. General Assembly of the State of Colorado. Available online: http://www.leg.state.co.us/CLICS/CLICS2010A/csl.nsf/fsbillcont3/0CA296732C8CEF4 D872576E400641B74?Open&file=1365_rer.pdf ICLE I USA Local Governments for Sustainability. 2010. Annual Report 2010: Empowering Sustainable Communities www.icleiusa.org/library/documents/ICLEI_USA_Annual_Report_20 10.pdf Accessed May 2012. ICLEI USA Local Governments for Sustainability. 2011. Community Greenhouse Gas Emissions Inventory Protocol www.iclei usa.org/tools/ghg protocol/community greenhouse gas emissions inventory protocol Accessed May 2012. Jaffe, M. 2013. Boulder mulls Xcel Energy's future, The Denver Post January 06, 2013. http://www.denverpost.com/business/ci_22314716/boulder city mulls xcels future Accessed May 2013. Jenerette, D. G., W. Wu, S. Goldsmith, W. Marussich and J. Roach. 2006. Contrasting Water Footprints of Cities in China and the Unit ed States. Ecological Economics 57: 346 358. Jenerette, D., and L. Larson. 2006. A global perspective on changing sustainable urban water supplies. Global and Planetary Change 50: 202 211. Johansson, J; H. Hassel and E. Zio. 2013. Reliability and vulnerability analyses of critical infrastructures: Comparing two approaches in the context of power systems. Reliability Engineering and System Safety http://dx.doi.org/10.1016/j.ress.2013.02.027i Kelly JA, Amirkhanian YA, Kabakchieva E, Vassileva S, McAuliffe TL, DiFranceisco WJ, et al. Prevention of HIV and sexually transmitted diseases in high risk social networks of young Roma (Gypsy) men in Bulgaria: randomised controlled trial. BMJ 2006 doi: 10.1136/bmj.38992.478299.55. Kennedy, C., J. Cuddihy, and J. Engel Yan. 2007. The Changing Metabolism of Cities. Journal of Industrial Ecology 11(2): 43 59. Kennedy, C., J. Steinberger, B. Gasson, Y. Hansen, T. Hillman, M. Havranek, D. Pataki, A. Phdunglsip, A. Ramaswami, and G. V. Mendez. 2009. Greenhouse Gas Emissions from Global Cities. Environmental Science & Technology 42: 7297 7302.

PAGE 161

145 Kenny, J.F., N.L Barber, S.S. Hutson, K.S. Linsey, J.K. Lovelace and M.A. Maupin. 2009. Estimated use of water in the United States in 2005 U.S. Geological Survey Circular 1344, 52 p. Kohli, Rajesh. 2013. Personal communication with head of Water Systems, Thermal Engin eering & Technology Division, Central Electricity Authority, Delhi, India. March 2013. Koch H. and S. Vogele. 2009. Dynamic modeling of water demand, water availability and adaptation strategies for power plants to global change. Ecological Economics 68 ( 2009), 2031 2039. J. Koomey, H. Akbari, C. Blumstein, M. Brown, R. Brown, C. Calwell, S. Carter, R. Cavanagh, A. Chang, D. Claridge, P. Craig, R. Diamond, J. H. Eto, W. Fulkerson, A. Gadgil, H. Geller, J. Goldemberg, C. Goldman, D. B. Goldstein, S. Greenb erg, D. Hafemeister, J. Harris, H. Harvey, E. Heitz, E. Hirst, H. Hummel, D. Kammen, H. Kelly, S. Laitner, M. Levine, A. Lovins, G. Masters, J. E. McMahon, A. Meier, M. Messenger, J. Millhone, E. Mills, S. Nadel, B. Nordman, L. Price, J. Romm, M. Ross, M. Rufo, J. Sathaye, L. Schipper, S. H. Schneider, J. L. Sweeney, M. Verdict, D. Vorsatz, D. Wang, C. Weinberg, R. Wilk, J. Wilson and E. Worrell. 2010. Defining a Standard Metric for Electricity Savings. Environmental Research Letters 5(201) 014017, 10pp. K ruse, J., Womack, A., and S. Meyer. 2004. "Implications of Minimum Summer Water Releases from Gavins Point For Power Power Plants". Food and Agricultural Policy Research Institute at the University of Missouri. FAPRI UMC Report #11 04. Lubienski, C and S (2006). Review of "On the Public Private School Achievement Debate." Boulder and Tempe: Education and the Public Interest Center & Education Policy Research Unit. Retrieved [November 2013] from http://epicpolicy.org/thinktank/review on public private school achievement debate Luck, M., D. G. Jenerette, J. Wu, and N. B. Grimm. 2001. The Urban Funnel M odel and the Spatially Heterogeneous Ecological Footprint. Ecosystems 4: 782 796. Macknick, J., R. Newmark, G. Heath, and K. C. Hallett. 2011. A review of operational water consumption and withdrawal factors for electricity generating technologies NREL Te chnical Report NREL/TP 6A20 50900. Contract No. DE AC36 08GO28308. Golden, CO: National Renewable Energy Laboratory. Mielke, Erik; L. Diaz Anadon and V. Narayanamurti. 2010. Water Consumption of Energy Resource Extraction, Processing, and Conversion: A review of the literature for estimates of water intensity of energy resource extraction, processing to fuels, and conversion to electricity Energy Technology Innovation Policy Discussion Paper No. 2010 15, Belfer Center for Science and International Affai rs, Harvard Kennedy School, Harvard University, October 2010. Ministry of Power (MoP). 2012. Working group on power for the 12 th national electricity plan (2012 2017) Planning Commission, Ministry of Power: New Delhi, January 2012.

PAGE 162

146 NERC (North American E lectric Reliability Council). 1992. Control Area Concepts and Obligations Washington, DC: North American Electric Reliability Council, July 1992. NETL (National Energy Technology Lab), Impact of Drought on U.S. Steam Electric Power Plant Cooling Water Int akes and Related Water Resource Management Issues NETL, U.S. Department of Energy, 2009. NETL (National Energy Technology Laboratory). 2008. Estimating Freshwater Needs to Meet Future Thermoelectric Generation Requirements, 2008 Update. U.S. Department o f Energy. www.netl.doe.gov/technologies/coalpower/ewr/pubs/2008_Water_Needs_Analysis Final_10 2 2008.pdf Accessed May 2012. NREL (National Renewable Energy Laboratory). 2010. Life Cycle Assessments. http://www.nrel.gov/lci/assessments.html. Accessed October 2011. NSF ( Na tional Science Foundation) Grant no. 1243535. NSF PIRE: Developing Low Carbon Cities in the US, China and India Through Integration Across Engineering, Environmental Sciences, Social Sciences, and Public Health. NTPC (National Thermal Power Corporation of India). 2013. Personal com munication with thermal power plant operators and engineers, NTPC Badarpur Thermal Power Station, Delhi, India. March 2013. Padowski, J. C., and J. W. Jawitz. 2012. Water availability and vulnerability of 225 large cities in the United States, Water Resou rces Res earch, 48, W12529, doi:10.1029/2012WR012335. Pahwa, P.K. 2013. Personal communication with Member Secretary of the Northern Region Power Committee, Ministry of Power, Delhi, India, March 2013. Paliwal, R., Sharma, P. & Kansal, A. Water quality mod elling of the river Yamuna (India) using QUAL2E UNCAS. Journal of environmental management 83, 131 44 (2007). Perveen, Shama; Romit Sen and Mayuri Ghosh. 2012. India's Deepening Water Crisis: Water Risks for Indian Industries A preliminary study of 27 industrial sectors. Federation of Indian Chambers of Commerce and Industry (FICCI) and Columbia University Water Center (CWC) report. February 2012. Pines H., E. Cotilla Sanchez and S. Blumsack. 2010. Do topological models provide good information about electricity infrastructure vulnerability? CHAOS 20(033122): 1 5. Ragland, C.J.; A,Feldpausch Parkera, T. Rai Petersona, J. Stephensb and E. Wilson. Socio political dimensions of CCS deployment through the lens of social network analysis. Energy Procedia 4 (2011) 6210 6217 Ramaswami, A., Bernard, M., Chavez, A., Hillman, T., Whitaker, M., Thomas, G., & Marshall, M. (2012 a ). Quantifying Carbon Mitigation Wedges in US Cities: Near Term Strategy

PAGE 163

147 Analysis and Critical Review. Environmental Science & Technolology 46 (7), pp 3629 3642 Ramaswami, A., Weible, C., Main, D., Heikkila, T., Siddiki, S., Duvall, A., Pattison, A. and Bernard, M. (2012 b ), A Social Ecological Infrastructural Systems Framework for Interdisciplinary Study of Sustainable City Sys tems. Journal of Industrial Ecology, 16: 801 813. doi: 10.1111/j.1530 9290.2012.00566.x Ramaswami, A., Janson, B., Hillman, T., Wendrowski, J., Reiner, M., Posner, M., Thomas, G., Schlomberg, K. 2007. Greenhouse Gas Inventory for the City and County of Den ver. Prepared for Mayor John Hickenlooper's Greenprint Denver Initiative. www.greenprintdenver.org/docs/Denver_GHG_Inventory_Report.pdf Accessed May 2011. Ramaswami, A., T. Hillman, B. Janson, M. Reiner and G. Thomas. 2008. A Demand Centered, Hybrid Life Cycle Methodology for City Scale Greenhouse Gas Inventories. Environ. Sci. Technol ., 42: 6455 6461. Renewable Energy Standard 2004. C.R.S. (Colorado Revised Statute) 40 2 124. http://www.sos.state.co.us/pubs/elections/statutes.html Accessed May 2012. Rockaway, T. D., P. A. Coomes, J. Rivard and B. Kornstein. 2011. Residential water use trends in North Am erica. Journal of the American Water Works Association 103(2): 76 89. Rogers, E. M. (2003) Diffusion of Innovations (Free Press, New York, ed. 5, 2003) Rogers, Everett M (1983): Diffusion of Innovations. 3rd ed. Free Press, New York. Serway, R. and J. Jewett. 2004. Electricity and Magnetism. In Physics for Scientists and Engineers Vol. 2, 6 th Edition, edited by Michelle Julet. Belmont, CA: Brooks/Cole Thomas Learning. SDO (State Demography Office). 2012. Population Forecasts 2000 2040 Colorado Department of Local Affairs, The Official State Web Portal: http://www.colorado.gov/cs/Satellite?c=Page&childpag ename=DOLAMain%2FCBONL ayout&cid=1251593346867&pagename=CBONWrapper Accessed March 2013. Sharma, Deepak. 2013. Personal communication with Area Manager, State Load Dispatch Center, New Delhi, India March 2013. Sharma, Dinesh. 2013. Delhi fears water short age as groundwater level hits dangerous low. India Today [Delhi], March 6, 2013. http://indiatoday.intoday.i n/story/unregulated withdrawal of groundwater delhi brings levels to danger zone india today/1/256627.html Accessed August 8, 2013. Sheldon, Seth D. "Water Energy and Drought: Water Use for Energy Production in the U.S." University of Massachusetts Boston, EEOS Department, 100 Morrissey Blvd., Boston MA, 02125 3393, USA.

PAGE 164

148 Soeder, D.J., and W. M. Kappel. 2009. Water resources and natural gas production from the Marcellus Shale: U.S. Geological Survey Fact Sheet 2009 3032, 6pp. http://pubs.usgs.gov/fs/2009/3032/ Accessed March 2013. Sovacool, B. K. and K. E. Sovacool. 2009. Identifying future electricity water tradeoffs in the United States. Energy Policy 37: 2763 2773. Stillwell, A. S., C.W. King, M.E. W ebber, I. J. Duncan, and A. Hardberger. 2011a. The energy water nexus in Texas. Ecology and Society 16 (1): 2. Online at http://www.ecologyandsociety.org/col16/iss1/art2/ Stillwell, A., Clayto n, M., Webber, M. 2011b. Technical analysis of a river basin based model of advanced power plant cooling technologies for mitigating water management challenges Environ. Res. Lett 6 (2011): 034015 Tidwell, V. C., P. H. Kobos, L. A. Malczynski, G. Kilse a nd C. R. Castillo. 2012. Exploring the Water Thermoelectric Power Nexus Journal of Water Resources Planning and Management (2012) 138: 491 501. Tollefson, G; R. Billinton and G. Wacker. 1991. Comprehensive bibliography on reliability worth and electrical service consumer interruption costs: 1980 1990, IEEE Trans. Power Syst ., 6 (4): 1508 1514. United Nations, Department of Economic and Social Aff airs, Population Division (2010). World Urbanization Prospects: The 2009 Revision. CD ROM Edition Data in digital form (POP/ DB/WUP/Rev.2009). Upadhyay, R., Dasgupta, N., Hasan, A. & Upadhyay, S. K. Managing water quality of River Yamuna in NCR Delhi. Ph ysics and Chemistry of the Earth, Parts A/B/C 36, (2011). USGS (United State Geologic Survey). 1995. Estimated Use of Water in the United States in 1995, County Level Data http://water.usgs.gov/watu se/spread95.html Accessed April 2011. USGS (United States Geological Survey). 2005. Estimated Use of Water in the United States: County Level Data for 2005 http://water.usgs.gov/watuse/data/2005/ Acc essed April 2011. Valente, V. and T.A. Oliveira. 2011. Hierarchical Linear Models: Review and Applications. AIP Conference Proceedings 1389 1549 (2011); doi: 10.1063/1.3637924 Valente, T. W. 1995. Network Models of the Diffusion of Innovations (Hampton P ress, Cresskill, NJ, 1995). Williams, P., D. Inman, A. Aden and G. Heath. 2009. Environmental and Sustainability Factors Associated With Next Generation Biofuels in the U.S.: What Do We Really Know? Environmental Science and Technology 43 (13): 4763 4775.

PAGE 165

149 Woldeyesus, T. 2012. Electric Portfolio Modeling with Stochastic Water Climate Interactions: Implications for Co Management of Water and Electric Utilities. PhD Dissertation, University of Colorado at Denver, Denver, Colorado, USA. Woldeyesus, T., A. Rama swami, B. Rajagopalan, and B. Gracely. 2013. Seasonality of Water Intensity Factors and Risk to Cities at the Water Electricity Climate Nexus. Environmental Research Letters Submitted for Publication. Wu, M. and Y. Chiu. 2011. Consumptive Water Use in the Production of Ethanol and Petroleum Gasoline 2011 Update Center for Transportation Research, Energy Systems Division, Argonne National Laboratory. Oak Ridge, TN: U.S. Department of Energy. Wu, M., M. Mintz, M. Wang and S. Arora. 2009. Consumptive Water Use in the Production of Ethanol and Petroleum Gasoline Center for Transportation Research, Energy Systems Division, Argonne National Laboratory. Oak Ridge, TN: U.S. Department of Energy. htt p://www.ipd.anl.gov/anlpubs/2009/01/63493.pdf Accessed March 2013. Xcel Energy. 2011. Personal communication with on duty plant manager at the Cherokee coal fired power station, Xcel Energy, Commerce City, Colorado. May 17th, 2011.