Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Modeling and mitigating the climate and health impacts of emissions from public transportation bus fleets.. Gouge, Brian D. 2012

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
24-ubc_2012_fall_gouge_brian.pdf [ 19.53MB ]
Metadata
JSON: 24-1.0073025.json
JSON-LD: 24-1.0073025-ld.json
RDF/XML (Pretty): 24-1.0073025-rdf.xml
RDF/JSON: 24-1.0073025-rdf.json
Turtle: 24-1.0073025-turtle.txt
N-Triples: 24-1.0073025-rdf-ntriples.txt
Original Record: 24-1.0073025-source.json
Full Text
24-1.0073025-fulltext.txt
Citation
24-1.0073025.ris

Full Text

MODELING AND MITIGATING THE CLIMATE AND HEALTH IMPACTS OF EMISSIONS FROM PUBLIC TRANSPORTATION BUS FLEETS: AN INTEGRATED APPROACH TO SUSTAINABLE PUBLIC TRANSPORTATION  by Brian D Gouge  BEng., The University of Victoria, 2001  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Resource Management and Environmental Studies)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  August, 2012  © Brian D Gouge, 2012  Abstract Public transportation has been widely promoted as a means of increasing the sustainability of urban transportation systems; however these systems also have adverse impacts. Further, although transit agencies are making efforts to address these impacts, the assessment tools and mitigation options available to them are limited. An integrated assessment model was developed to explicitly address the adverse climate and health impacts of the primary exhaust emissions from heavy-duty transit bus fleets. Models of the climate and health impact pathways were developed at several different spatial scales (e.g., macro, meso, and micro). These models were used to quantify the potential of a novel operational control strategy based on vehicle scheduling optimisation to reduce the impacts and costs of operating transit bus fleets. In addition to demonstrating the benefits of the vehicle scheduling optimisation, the results showed that transit agencies that optimise for operating costs and/or climate impacts alone may inadvertently increase health impacts and highlight the need for an integrated assessment approach. In developing the health impact pathway model, particular attention was devoted to evaluating methods of modeling vehicle activity and emissions and the implications of these methods on estimating exposure and health impacts. In comparison to micro and meso scale assessments, traditional regional/macro scale assessments based on emissions inventories were found to underestimate exposure and health impacts because they do not account for the intra-regional spatial variability in, and relationship between emissions and populations. Further, traditional distance-based emission factor modeling approaches were found to poorly characterise the spatial distribution of emissions as well as underestimate total emissions in comparison to modal modeling approaches because they do not fully account for the effects of vehicle activity. However, while modal modeling approaches likely have several advantages over emission factor modeling approaches, an evaluation of a major new modal emissions model developed by the United States Environmental Protection Agency, MOVES, revealed significant biases in the model’s predictions of NO X , PM, and THC emissions from both diesel and CNG transit buses. This suggests that with respect to transit buses, MOVES would benefit from further calibration and its predictions should be interpreted with care.  ii  Preface Each of the research chapters of this dissertation were originally written as manuscripts for publication in a peer-reviewed academic journal. The following details my specific contributions to each research chapter.  Chapter 2 is based on a multi-year collaborative research project led by Dr. Hadi Dowlatabadi that involved me, Mr. Francis Ries and two industry partners, GIRO Inc. and the South Coast British Columbia Transportation Authority, commonly known as TransLink. The conceptual idea discussed in this chapter was developed by Dr. Hadi Dowlatabadi, Mr. Francis Ries and me. The specific implementation of the idea discussed in Chapter 2 was developed by me. I developed the model and performed all calculations described with the exception of those related to dispersion and intake fractions, which were performed by Mr. Francis Ries. I was responsible for all aspects of writing an initial manuscript version of the Chapter which is in the process of being submitted to an academic journal. Comments on the manuscript were provided by Dr. Hadi Dowlatabadi, Dr. Milind Kandlikar, and Mr. Francis Ries.  A version of Chapter 3 has been published. Gouge B., Ries F.J., and Dowlatabadi H. (2010). Spatial distribution of diesel transit bus emissions and urban populations: Implications of coincidence and scale on exposure. Environmental Science and Technology. 44 (18): 7163-7168. I designed the study, supervised the collection of data, performed all analyses and wrote the manuscript. Comments on the manuscript were provided by Dr. Hadi Dowlatabadi, Dr. Milind Kandlikar, Dr. Andrew Grieshop, Dr. Steven Rogak, Mr. Francis Ries, and Mr. David Gourley (TransLink).  A version of Chapter 4 was developed as manuscript and is in the process of being submitted to an academic journal. I designed the study, acquired the data, performed the analysis, and wrote the manuscript. Comments on the manuscript were provided by Dr. Hadi Dowlatabadi, Dr. Milind Kandlikar and Mr. David Gourley (TransLink).  iii  Table of Contents Abstract ......................................................................................................................................... ii Preface .......................................................................................................................................... iii Table of Contents ..........................................................................................................................iv List of Tables .................................................................................................................................ix List of Figures ...............................................................................................................................xi Acknowledgements .....................................................................................................................xix Dedication ..................................................................................................................................... xx Chapter 1 : Introduction ............................................................................................................... 1 1.1  Climate and Health Impacts of Transit Bus Emissions .................................................. 3  1.1.1  Heavy-Duty Vehicle Emissions.................................................................................. 4  1.1.1.1  Exhaust ............................................................................................................... 4  1.1.1.2  Non-Exhaust ....................................................................................................... 6  1.1.1.3  Secondary Pollutants .......................................................................................... 7  1.1.2  Health and Climate Impacts........................................................................................ 7  1.1.2.1  Health.................................................................................................................. 7  1.1.2.2  Climate................................................................................................................ 9  1.1.3  Health and Climate Impact Pathways ....................................................................... 10  1.1.4  Scale and Variability................................................................................................. 13  1.1.5  Traditional Management and Quantification Approaches and Limitations ............. 14  1.2  Mitigating Health and Climate Impacts: Capital and Operational Control Strategies . 16  1.3  Integrated Assessment and Decision Making ............................................................... 18  1.4  Research Objectives and Scope .................................................................................... 22  1.5  Overview and Structure (Road-Map) of Dissertation................................................... 26  1.6  Literature Review ......................................................................................................... 29  1.6.1  Vehicle Activity ........................................................................................................ 29  1.6.2  Emissions .................................................................................................................. 32  1.6.2.1  Emissions Measurement and Sampling Methods ............................................. 32  1.6.2.2  Factors Affecting Heavy-Duty Vehicle Emissions........................................... 35  1.6.2.3  Emissions Modeling ......................................................................................... 40  1.6.2.4  Emissions Models ............................................................................................. 44  1.6.2.4.1 MOBILE ..................................................................................................... 45 iv  1.6.2.4.2 Integrated Bus Information System (IBIS) ................................................. 46 1.6.2.4.3 Comprehensive Modal Emissions Model (CMEM) ................................... 46 1.6.2.4.4 Motor Vehicle Emissions Simulator (MOVES) ......................................... 48 1.6.3  Air Quality (Dispersion and Transformation) .......................................................... 48  Chapter 2 : Integrated Assessment: Minimising the Climate and Health Impacts of Emissions from Heavy-Duty Public Transportation Bus Fleets through Vehicle Scheduling .................................................................................................................................... 50 2.1  Introduction................................................................................................................... 50  2.2  Materials and Methods ................................................................................................. 53  2.2.1  Transit System .......................................................................................................... 53  2.2.2  Emissions Model and Operating Costs ..................................................................... 53  2.2.3  Climate Impact Indicator .......................................................................................... 54  2.2.4  Public Health Impact Indicator ................................................................................. 55  2.2.5  Vehicle Scheduling and Assignment Optimisation .................................................. 57  2.3  Results and Discussion ................................................................................................. 58  2.3.1  Indicators and Optimisation Objectives ................................................................... 61  2.3.2  Climate...................................................................................................................... 62  2.3.3  Health and Exposure ................................................................................................. 62  2.3.3.1 2.3.4  Concentration-Response Function .................................................................... 64  Co-benefits and Trade-offs of Multi-Objective Optimisation .................................. 65  2.3.4.1  Climate and Health Impact Trade-offs ............................................................. 65  2.3.4.2  Valuation and Optimal Assignments ................................................................ 66  2.3.5  Factors Affecting Vehicle Scheduling ...................................................................... 68  2.3.6  Assumptions and Limitations ................................................................................... 70  2.3.6.1  Emissions Model .............................................................................................. 71  2.3.6.2  Intake Fraction .................................................................................................. 72  2.3.6.2.1 Population ................................................................................................... 72 2.3.6.2.2 Dispersion and Transformation................................................................... 73 2.3.6.3  Indicators and Valuation ................................................................................... 75  2.3.6.4  Optimisation Solutions ..................................................................................... 76  2.3.7  Implications .............................................................................................................. 77  Chapter 3 : Micro Scale Modeling: Spatial Distribution of Diesel Transit Bus Emissions and Urban Populations and the Implications of Coincidence and Scale on Exposure ......... 78 v  3.1  Introduction................................................................................................................... 78  3.2  Materials and Methods ................................................................................................. 79  3.2.1  Route and Vehicle..................................................................................................... 80  3.2.2  Data Collection ......................................................................................................... 80  3.2.3  Vehicle Activity Data Point ...................................................................................... 81  3.2.3.1  Sampling Interval (t k ) ....................................................................................... 81  3.2.3.2  Linear Distance (d k ) ......................................................................................... 81  3.2.3.3  Road Grade (g k ) ................................................................................................ 81  3.2.3.4  Vehicle Dynamics (v k , a k ) ................................................................................ 83  3.2.3.5  Vehicle Specific Power (VSP k ) ........................................................................ 83  3.2.3.6  Emission Rate Model (ER c,k ) ........................................................................... 84  3.2.4  Evaluation of GPS Data ............................................................................................ 84  3.2.5  Total Emissions (TE) ................................................................................................ 85  3.2.6  Total Interval Emissions (TIE) and Instantaneous Emission Factor ........................ 85  3.2.7  Impacted Population (P) ........................................................................................... 86  3.2.7.1  Zone 1 ............................................................................................................... 86  3.2.7.2  Zones 2-7 .......................................................................................................... 87  3.2.8  Spatial Coincidence Factor (SCF) ............................................................................ 87  3.2.9  Uncertainty and Importance Analysis ...................................................................... 87  3.3  Results .......................................................................................................................... 89  3.3.1  Spatial Distributions of Emissions and Impacted Populations ................................. 89  3.3.2  Spatial Coincidence and Exposure ........................................................................... 95  3.3.3  Importance ................................................................................................................ 95  3.3.4  Evaluation of GPS Data ............................................................................................ 96  3.4  Discussion ..................................................................................................................... 98  3.4.1  Emissions Modeling and Explanatory Variables ...................................................... 98  3.4.2  Importance of Explanatory Variables ....................................................................... 99  3.4.3  Receptor and Population Modeling .......................................................................... 99  3.4.4  Exposure and Spatial Heterogeneity....................................................................... 100  3.4.5  Assumptions and Limitations ................................................................................. 101  3.4.6  Implications and Recommendations ....................................................................... 103  Chapter 4 : Evaluating the MOVES Emissions Model using Heavy-Duty Transit Bus Data ............................................................................................................................................. 104 vi  4.1  Introduction................................................................................................................. 104  4.2  Materials and Methods ............................................................................................... 105  4.2.1  Emissions Measurement Dataset ............................................................................ 106  4.2.1.1  West Virginia University Data ....................................................................... 108  4.2.1.2  TransLink Data ............................................................................................... 108  4.2.2  Emissions Modeling and MOVES ......................................................................... 109  4.2.3  Evaluating the MOVES Model ............................................................................... 112  4.3  Results and Discussion ............................................................................................... 113  4.3.1  Prediction Errors ..................................................................................................... 113  4.3.2  OpMode and Second-by-Second Emission Rates .................................................. 116  4.3.3  Sources of Prediction Error Bias and Uncertainty.................................................. 119  4.3.3.1  Explanatory Variables .................................................................................... 119  4.3.3.2  Datasets and Base Emission Rates ................................................................. 120  4.3.4  Correcting Prediction Error Bias and Uncertainty.................................................. 122  4.3.5  Implications and Recommendations ....................................................................... 123  Chapter 5 : Conclusion.............................................................................................................. 125 5.1  Contributions .............................................................................................................. 125  5.2  Findings and Recommendations ................................................................................. 127  5.2.1  Scale and Level of Detail ........................................................................................ 127  5.2.2  Vehicle Activity ...................................................................................................... 129  5.2.3  Emissions ................................................................................................................ 131  5.2.4  Air Quality, Population, and Exposure ................................................................... 134  5.2.5  Impact Indicators .................................................................................................... 135  5.2.6  Control Strategies and Decision Making ................................................................ 135  5.3  Strengths and Limitations ........................................................................................... 137  5.4  Future Research .......................................................................................................... 139  References ................................................................................................................................... 141 Appendices ................................................................................................................................. 164 Appendix A : Integrated Assessment ..................................................................................... 164 A.1  Vehicle Fleet and Categories .................................................................................. 164  A.2  Emissions Model .................................................................................................... 166  A.3  Operating Costs ...................................................................................................... 167  A.4  Climate Metric (GWP) ........................................................................................... 168 vii  A.5  PM 2.5 Intake Calculation Workflow ....................................................................... 169  A.6  Concentration to Emissions Ratio (C·Ê-1) and Dispersion Modeling ..................... 170  Appendix B : Micro Scale Modeling ...................................................................................... 176 B.1  Route Description ................................................................................................... 176  B.2  Vehicle Activity Data Point .................................................................................... 178  B.3  Total Interval Emissions (TIE) ............................................................................... 186  B.4  Impacted Population ............................................................................................... 188  B.5  Emissions Model .................................................................................................... 190  B.6  MOBILE6.2 Estimates ........................................................................................... 190  B.7  Idle Emissions......................................................................................................... 192  B.8  Spatial Distribution of Vehicle Activity and Emissions ......................................... 193  B.9  Total Emission (TE) Distributions ......................................................................... 194  B.10  Total Interval Emissions (TIE) Distributions ......................................................... 194  B.11  Spatial Coincidence Factor (SCF) Derivation ........................................................ 197  Appendix C – Evaluating the MOVES Model ....................................................................... 199 C.1  MOVES Pollutants and Processes .......................................................................... 199  C.2  MOVES Operating Modes (OpMode).................................................................... 200  C.3  West Virginia University Emissions Measurement Data ....................................... 201  C.4  TransLink Emissions Measurement Data ............................................................... 207  C.5  CNG Prediction Errors ........................................................................................... 218  C.6  Advanced Technology Group Prediction Errors .................................................... 218  C.7  OpMode and Second-by-Second Emission Rates .................................................. 221  C.8  OpModes 12-16 Emission Rate Error Analysis...................................................... 224  C.9  OpMode Distributions ............................................................................................ 227  viii  List of Tables Table 1.1 - Particulate Matter (PM) classifications by particle diameter (Kittelson,1998). ........... 6 Table 1.2 - Definitions of scales. .................................................................................................. 14 Table 1.3 – Average capital costs of bus technologies in 2010 (APTA,2010a) ........................... 17 Table 1.4 - Hypotheses. ................................................................................................................ 25 Table 1.5 – Vehicle activity measurement methods. .................................................................... 30 Table 1.6 - Emission sampling methods. ...................................................................................... 33 Table 1.7 – Factors affecting heavy-duty vehicle emissions. ....................................................... 36 Table 1.8 – Emissions models and categories. ............................................................................. 43 Table 1.9 - Heavy-duty vehicle emissions models. ...................................................................... 44 Table 2.1 - Bus categories, operating costs, emission factors and Global Warming Commitment. ...................................................................................................................................................... 54 Table 2.2 - Vehicle assignment optimisation results. ................................................................... 59 Table 2.3 - Percent contribution of climate forcing compounds to the total GWC·km-1 . ............ 62 Table 2.4 - Optimisation results based on intake fractions calculated using different spatial extents ........................................................................................................................................... 75 Table 3.1 - Uncertainty Analysis (UA) and Importance Analysis (IA) input variable distributions. ...................................................................................................................................................... 88 Table 3.2 - Total, idle, near bus stop and near intersection emissions as well as emissions factors of CO, NO X , and HC for east- and west- bound traversals of the 99 B-Line bus route estimated using macro scale and micro scale modeling approaches. ........................................................... 90 Table 4.1 - Emissions measurement data summary.................................................................... 107 Table 4.2 - MOVES explanatory variables for running exhaust emissions. ............................. 110 Table A.1 – TransLink vehicle fleet and categories. .................................................................. 164 Table A.2 - USEPA regulatory emission standards for urban transit buses (Barnitt,2008, DieselNet,2007a). ....................................................................................................................... 165 Table A.3 - Emission factors. ..................................................................................................... 166 Table A.4 - Bus operating costs. ................................................................................................. 168 Table A.5 – Global Warming Potentials (GWP) ........................................................................ 168 Table A.6 - CALINE4 Configuration Parameters ...................................................................... 172 Table B.1 - Summary of the 99 B-Line GPS data collection campaign. .................................... 176 Table B.2 - Distribution of t k before and after correction for invalid time-stamps. ................... 178 ix  Table B.3- Two examples of the GPS time-stamp corruption. .................................................. 179 Table B.4 – Definitions of VSP Modes (Zhai, et al.,2008). ....................................................... 190 Table B.5 - MOBILE6.2 CO, NO X , and HC emission factors for a fleet of 1996-1995 diesel transit buses. ............................................................................................................................... 190 Table B.6 - MOBILE6.2 emission rate estimates of CO, NO X , and HC for diesel transit buses between 2000 and 1995. ............................................................................................................. 191 Table B.7 - Total, idle, near bus stop and near intersection emissions as well as emissions factors of CO, NO X , and HC for east-bound and west-bound traversals of the 99 B-Line bus route estimated using micro scale modeling approaches. Total emissions estimated using macro scale modeling approaches (MOBILE6.2). ......................................................................................... 196 Table C.1- Pollutant names. abbreviations and MOVES PollutantIDs. ..................................... 199 Table C.2 – Emission process names and MOVES ProcessIDs. ................................................ 199 Table C.3 - MOVES operating modes (OpMode) (USEPA,2009a)........................................... 200 Table C.4 – WVU Vehicle and Test Filter ................................................................................. 202 Table C.5 - Data Quality Tests ................................................................................................... 202 Table C.6 – WVU Emission Measurement Adjustments ........................................................... 202 Table C.7 - Drive cycles that speed-time traces were obtained for. ........................................... 203 Table C.8 - TransLink vehicles emission tested. ........................................................................ 208 Table C.9 - TransLink emissions measurements and calculations summary. ............................ 214 Table C.10 - Molecular mass. ..................................................................................................... 215 Table C.11 – TransLink data quality tests. ................................................................................. 216 Table C.12 – TransLink emission measurement adjustments. ................................................... 217  x  List of Figures Figure 1.1 - Public transportation ridership in the United States after (APTA,2006, 2011, United States Census Bureau,2000, 2012). ................................................................................................ 2 Figure 1.2 - Transit bus distribution by energy source (APTA,2011). ........................................... 3 Figure 1.3 – Climate and health impact pathways and integrated assessment framework (Fuglestvedt, et al.,2010, Marshall, et al.,2006, Smith,1993). Grey indicates non-focus areas. . 12 Figure 1.4 - Integrated assessment process after (Rotmans, et al.,2003)...................................... 19 Figure 1.5 - Decision making context. Highlighted ring indicates the scope of this dissertation. 24 Figure 1.6 - Dissertation overview and structure (road-map)....................................................... 26 Figure 1.7 – Comprehensive Modal Emissions Model (CMEM) structure for heavy-duty vehicles after (Barth, et al.,2004). ............................................................................................................... 47 Figure 2.1 – Scatter plots of the total distance traveled on a block and the intake fraction of the block. Each point represents a specific block and is color coded by bus category. The scatter plots show the bus-block assignments for Fleet A and the scenarios that minimised operating costs (a), climate impacts (b), health impacts (c), and total emissions of PM 2.5 (d). ................... 60 Figure 2.2 - Scatter plots of the total distance traveled on a block and the intake fraction of the block. Each point represents a specific block and is color coded by bus category. The scatter plots show the bus-block assignments for Fleet B and the scenarios that minimised operating costs (a), climate impacts (b), health impacts (c), and total emissions of PM 2.5 (d). ................... 61 Figure 2.3 - Histograms of the primary PM 2.5 intake fractions (exposure potential) (a) and distance-weighted intake fractions (b) for bus routes in Vancouver, Canada. ............................. 63 Figure 2.4 – Map of population density and bus routes in Vancouver, Canada. Routes are color coded and grouped based on the distance-weighted primary PM 2.5 intake fraction. ................... 64 Figure 2.5 - Relationship between climate and health impacts for Fleet A (a) and Fleet B (b). The change in the minimum health impact is expressed as a function of the change in the climate impact. Bottom and left axes show the change in absolute value and top and right axes show the corresponding percent change. ..................................................................................................... 66 Figure 2.6 - Relationship between health and operating and climate costs for Fleet A (a) and Fleet B (b). The change in the minimum health impact is expressed as a function of the sum of the operating and climate costs assuming a value of $25 per tCO 2 e. Bottom and left axes show the change in absolute value and top and right axes show the corresponding percent change. ... 67  xi  Figure 2.7 - Influence diagram of factors affecting vehicle scheduling optimisation that incorporates climate, health, and operating costs and impacts. .................................................... 68 Figure 3.1 - Map of the 99 B-Line Bus Rapid Transit (BRT) route from the University of British Columbia at the western terminus to Commercial Drive at the eastern Terminus in Vancouver, Canada. ......................................................................................................................................... 80 Figure 3.2 - Grade distribution for the west-bound route after filtering (a) and before filtering (b) with a 51-point central averaging filter. ....................................................................................... 82 Figure 3.3 - Grade distribution for the west-bound route after filtering (a) and before filtering (b) with a 51-point central averaging filter. ....................................................................................... 82 Figure 3.4 – Final elevation and grade profiles of the west-bound (a) and east-bound (b) route after the elevation profile was filtered with a 51-point central averaging filter. .......................... 83 Figure 3.5 - Final elevation and grade profiles of the east-bound route after the elevation profile was filtered with a 51-point central averaging filter. .................................................................... 83 Figure 3.6 - Diagram of the model used to calculate emissions over an interval of length x. The function f(x,v,a) solves –x + v⋅t + ½⋅a⋅t2 for t............................................................................... 86 Figure 3.7 - Histograms of the spatial distributions of estimated emissions of CO, NO X , and HC for east-bound traversals of the 99 B-Line bus route. The probability of the estimated instantaneous emission factor along the route is indicated by the color bar. The direction of travel is indicated by the red arrows. Bus stops are indicated by vertical red lines. Major intersections are indicated by vertical black lines. Minor intersections are indicated by vertical dashed lines. The elevation is indicated by the grey profile. Populations in zone 1 (pedestrian) and zone 5 (200-500 m) are shown as indicated ............................................................................................. 91 Figure 3.8 - Histograms of the spatial distributions of estimated emissions of CO, NO X , and HC for west-bound traversals of the 99 B-Line bus route. The probability of the estimated instantaneous emission factor along the route is indicated by the color bar. The direction of travel is indicated by the red arrows. Bus stops are indicated by vertical red lines. Major intersections are indicated by vertical black lines. Minor intersections are indicated by vertical dashed lines. The elevation is indicated by the gray profile. Populations in zone 1 (pedestrian) and zone 5 (200-500 m) are shown as indicated........................................................................... 92 Figure 3.9 - Histograms of the spatial distributions of vehicle activity for west-bound traversals showing the probability of vehicle activity in 50 m intervals. The direction of travel is indicated by the red arrows. Bus stops are indicated by vertical red lines. Major intersections are indicated  xii  by vertical black lines. Minor intersections are indicated by vertical dashed lines. The elevation is indicated by the grey profile. .................................................................................................... 93 Figure 3.10 – Population within the 7 zones (Pedestrian, 0-50 m, 50-100 m, 100-200 m, 200-500 m, 500-1000 m, 1000-5000m) along the west-bound 99 B-Line bus route. Major intersections are indicated by solid vertical grey lines. Minor intersections are indicated by dashed vertical grey lines. Bus stops are indicated by vertical red lines. ............................................................. 94 Figure 3.11 - Population within the 7 zones (Pedestrian, 0-50 m, 50-100 m, 100-200 m, 200-500 m, 500-1000 m, 1000-5000m) along the east-bound 99 B-Line bus route. Minor intersections are indicated by dashed vertical grey lines. Bus stops are indicated by vertical red lines. ......... 94 Figure 3.12 - Boxplots of the spatial coincidence factor (SCF) showing the change in exposure due to the spatial coincidence of the impacted populations in seven zones around the route and estimated emissions of CO, NO X , and HC for west-bound and east-bound traversals of the 99 BLine bus route. Median values are indicated by bars and mean values by dots. Changes in exposure due to MOBILE6.2 underestimates are not shown. ...................................................... 95 Figure 3.13 - Importance analysis of the input variables used to estimate the spatial distribution of emissions of CO (a), NO X (b), and HC (c) along the 99 B-Line bus route. Black bars indicate the partial rank correlation coefficients (PRCC) values for primary input variables: velocity (v), acceleration (a), and grade (g) and secondary input variables: interval traversal time (t) and emission rate (ER). Gray bars indicate negative PRCC values. Numerical rankings of importance were based on the absolute values of the PRCC. Lower ranks indicated greater importance. ................................................................................................................................... 96 Figure 3.14 - Map-matching errors for west-bound (a) and east-bound (b) traversals. The road centre lines are indicated by the dotted yellow line and the outer edges of the roads are indicated by the vertical black lines. ............................................................................................................ 97 Figure 3.15 – Scatter plots comparing GPS and VSS velocity measurements (a) and acceleration estimates (b). Points located on the identity line indicate agreement between the GPS and VSS data................................................................................................................................................ 97 Figure 4.1 – Mean prediction errors for CO 2 (a), NO X (b), and PM (c) emissions from diesel buses in the conventional technology group by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval. .................................................................................................................................................... 114  xiii  Figure 4.2 –Mean prediction errors for CO 2 (a), NO X (b), and THC (c) emissions from CNG buses by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval........................................................................... 115 Figure 4.3 - (a,d) Measured (blue box plots) and predicted (green points) CO 2 and NO X emissions rates by OpMode for 2007 diesel buses in the conventional technology group from the TransLink dataset. Upper, middle, and lower horizontal lines of the boxes indicate the 75th, 50th (median), and 25th percentiles. Whiskers extend to the furthest data point within 1.5 times the interquartile range. Diamonds indicate the mean. (b, e) Contributions to the total prediction error from each OpMode. (c, f) Sample of the first 200 s of second-by-second predicted and measured CO 2 and NO X emission rates as well as speed. All other panels show all data. ........................ 118 Figure A.1 – Route Intake (I) and Intake Fraction (iF) calculation workflow and associated data sources. ....................................................................................................................................... 169 Figure A.2 – Mean unidirectional dispersion curve resulting from on-route emissions of 112 g⋅hr1  developed using CALINE4. The data points correspond to the concentration at the midpoint of  each of the six zones. Error bars represent the range of variation in the 8 directional dispersion curves (0, 45, 90, 135, 180, 225, 270, and 315°). ....................................................................... 173 Figure A.3 – Estimated concentration resulting from 112 g⋅hr-1 on-route emissions and the population in the six zones around the 99 B-Line bus route in Vancouver, Canada. ................. 173 Figure A.4 - The distributions of the total distance travelled (a,f), operating costs (b,g), climate impacts (c,h), total PM 2.5 emissions (d,h), and the health impacts (e,i) by bus category for optimisation scenarios A,B,E, and F and both of Fleet A and B. ............................................... 175 Figure B.1 – 5000 m section of the west-bound route showing the elevation profile filtered using a 101-point, 51-point, and 21-point central averaging filter. ...................................................... 180 Figure B.2 – 5000 m section of the west-bound route showing the resulting grade profile after the elevation was filtered using a 101-point, 51-point, and 21-point central averaging filter. ... 180 Figure B.3 – Histograms of velocity (v k ) for all west-bound (a) and east-bound (b) traversals. 181 Figure B.4 - Histograms of acceleration (a k ) for all west-bound (a) and east-bound (b) traversals. .................................................................................................................................................... 181 Figure B.5 - Histogram of the joint velocity - acceleration probability distribution for west-bound 99 B-Line traversals. Only velocities greater than 0.0 kph are shown. The distribution is comparable to the distribution found by (Yoon, et al.,2005b). .................................................. 181  xiv  Figure B.6 - Analysis of the GPS and VSS velocity data collected on May 9, 2009. Upper plots show the error between the VSS and GPS velocity data. The lower plot shows the raw GPS and VSS data. MAE = mean absolute error. MSE = mean squared error. ....................................... 182 Figure B.7 - Analysis of the GPS and VSS acceleration estimates derived from the data collected on May 9, 2009. Upper plots show the error between the VSS and GPS acceleration estimates. The lower plot shows the raw GPS and VSS data. MAE = mean absolute error. MSE = mean squared error. .............................................................................................................................. 182 Figure B.8 - Histograms of vehicle specific power (VSP k ) estimates for all west-bound (a) and east-bound (b) traversals. ............................................................................................................ 184 Figure B.9 - Histograms of engine power-demand (P k ) estimates for all west-bound (a) and eastbound (b) traversals. ................................................................................................................... 184 Figure B.10 - Histograms of CO emission rate estimates for all west-bound (a) and east-bound (b) traversals where each traversal was sampled 100 times. ...................................................... 185 Figure B.11 - Histograms of NO X emission rate estimates for all west-bound (a) and east-bound (b) traversals where each traversal was sampled 100 times. ...................................................... 185 Figure B.12 - Histograms of HC emission rate estimates for all west-bound (a) and east-bound (b) traversals where each traversal was sampled 100 times. ...................................................... 185 Figure B.13 – Graphical depiction of one iteration (b=4) of the algorithm used to estimate the total interval emissions (TIE). There are two vehicle activity data points Φ k and Φ k-1 and four (1...4) 50 m intervals over which the emissions are distributed. The values x 1 ...x 4 are the linear distances from the start of the route to the interval boundaries. ................................................. 187 Figure B.14 – Flowchart describing the pedestrian model. Walking speeds were dervied from (Willis, et al.,2004). .................................................................................................................... 189 Figure B.15 - MOBILE6.2 input file. ......................................................................................... 191 Figure B.16 - MOBILE6.2 registration distribution input file. .................................................. 192 Figure B.17 – One-way ANOVA test performed using R. ANOVA test showed that 58% of the variance in velocity was explained by the location along the route (i.e., the interval index or “bin”). ......................................................................................................................................... 193 Figure B.18 – Histograms of total emissions per traversal of CO, NO X , and HC for east-bound and west-bound traversals where each traversal was sampled 100 times. MOBILE6.2 estimates are indicated by the vertical magenta line. Total emissions of partial runs were not estimated. .................................................................................................................................................... 194  xv  Figure B.19 - Histograms of the total interval emissions (TIE) of CO for all west-bound (a) and east-bound (b) traversals where each traversal was sampled 100 times. MOBILE6.2 estimates are indicated by the vertical magenta line. ................................................................................. 194 Figure B.20 - Histograms of the total interval emissions (TIE) of NO X for all west-bound (a) and east-bound (b) traversals where each traversal was sampled 100 times. MOBILE6.2 estimates are indicated by the vertical magenta line. ................................................................................. 195 Figure B.21 - Histograms of the total interval emissions (TIE) of HC for all west-bound (a) and east-bound (b) traversals where each traversal was sampled 100 times. MOBILE6.2 estimates are indicated by the vertical magenta line. ................................................................................. 195 Figure B.22 – Spatial coincidence factor derivation: zones (z) and route intervals (i). ............. 197 Figure C.1- Drive cycle speed-time traces and MOVES OpMode distributions, part 1. ........... 204 Figure C.2- Drive cycle speed-time traces and MOVES OpMode distributions, part 2. ........... 205 Figure C.3- Drive cycle speed-time traces and MOVES OpMode distributions, part 3. ........... 206 Figure C.4- Phase 1 Drive Cycle sample. ................................................................................... 209 Figure C.5 - Phase 2 test track configuration (M.J. Bradley & Associates,2008) ...................... 210 Figure C.6 - Phase 2 drive cycle sample. One test consisted of five loops of the test track. .... 210 Figure C.7 – Phase 3 Flat test route. The route was 5.9 km long with an elevation range of 3 – 6 m. A total of 13 bus stops were located along the route and the average run time was 16.3 minutes (M.J. Bradley & Associates,2009). ............................................................................... 211 Figure C.8 - Phase 3 Hilly test route. The route was 9.9 km long with an elevation range of 50 – 360 m. A total of 19 bus stops were located along the route and the average run time was 31.1 minutes (M.J. Bradley & Associates,2009). ............................................................................... 211 Figure C.9 - Histograms of the grade on the Phase 3 Hilly Route. All other routes were assumed to have a grade of 0. .................................................................................................................... 212 Figure C.10 - Histograms of vehicle speed (a), acceleration (b), VSP (c), and power (d) after applying filter to the speed data. Data points where values were equal to zero were not shown. The plots suggest the vehicle activity data is valid and within the performance envelope of the buses tested. ................................................................................................................................ 213 Figure C.11 - Mean prediction errors for CH 4 (a) and PM (b) emissions from CNG buses by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval. .......................................................................................... 218  xvi  Figure C.12 - Mean prediction errors for CO 2 (a), NO X (b), and PM (c) emissions from hybrid diesel vehicles with DPFs (advanced technology group) by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval. .................................................................................................................................................... 219 Figure C.13 - Mean prediction errors for CO 2 (a), NO X (b), and PM (c) emissions from 2006 and prior model year diesel vehicles with DPFs (advanced technology group) by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval. .......................................................................................... 219 Figure C.14 - (a,d) Measured (blue box plots) and predicted (green points) CO 2 and NO X emissions rates by OpMode for 1998 diesel buses in the conventional technology group from the TransLink dataset. Upper, middle, and lower horizontal lines of the boxes indicate the 75th, 50th (median), and 25th percentiles. Whiskers extend to the furthest data point within 1.5 times the interquartile range. Diamonds indicate the mean. (b, e) Contributions to the total prediction error from each OpMode. (c, f) Sample of the first 200 s of second-by-second predicted and measured CO 2 and NO X emission rates as well as speed. All other panels show all data. ....................... 221 Figure C.15 – (a,d) Measured (blue box plots) and predicted (green points) CO 2 and NO X emissions rates by OpMode for 2001 diesel buses in the conventional technology group from the TransLink dataset. Upper, middle, and lower horizontal lines of the boxes indicate the 75th, 50th (median), and 25th percentiles. Whiskers extend to the furthest data point within 1.5 times the interquartile range. Diamonds indicate the mean. (b, e) Contributions to the total prediction error from each OpMode. (c, f) Sample of the first 200 s of second-by-second predicted and measured CO 2 and NO X emission rates as well as speed. All other panels show all data. ........................ 222 Figure C.16 – (a,d) Measured (blue box plots) and predicted (green points) CO 2 and NO X emissions rates by OpMode for 2005-2006 CNG buses from the TransLink dataset. Upper, middle, and lower horizontal lines of the boxes indicate the 75th, 50th (median), and 25th percentiles. Whiskers extend to the furthest data point within 1.5 times the interquartile range. Diamonds indicate the mean. (b, e) Contributions to the total prediction error from each OpMode. (c, f) Sample of the first 200 s of second-by-second predicted and measured CO 2 and NO X emission rates as well as speed. All other panels show all data. ....................................... 223 Figure C.17 – Scatter plots of CO 2 emission rates vs. OpMode (a), engine speed vs. vehicle speed (b), latitude vs. longitude (c), and second-by-second vehicle speed (d) for Bus 9755 (Test xvii  T2 and T3 from the Flat Route) from Phase 3. The green points in panel a) indicate predicted emission rates from MOVES. The points marked in red indicate data points that contributed to the discrepancy between the measured and predicted emission rates in OpModes 12-16. ........ 225 Figure C.18 - Scatter plots of NO X emission rates vs. OpMode (a), engine speed vs. vehicle speed (b), latitude vs. longitude (c), and second-by-second vehicle speed (d) for Bus 9755 (Test T2 and T3 from the Flat Route) from Phase 3. The green points in panel a) indicate predicted emission rates from MOVES. The points marked in red indicate data points that contributed to the discrepancy between the measured and predicted emission rates in OpModes 12-16. ........ 226 Figure C.19 – OpMode distributions of All tests (a), WVU tests (b), TransLink Phase 1 tests (c), and TransLink Phase 2 tests (d). ................................................................................................. 227 Figure C.20 – Real-world OpMode distributions of TransLink Phase 3 Hilly route (a), TransLink Phase 3 Flat route (b), and 99 B-Line route. ............................................................. 227  xviii  Acknowledgements This dissertation would not have been possible without the support of family, friends, and colleagues. I’m especially grateful to Dr. Hadi Dowlatabadi for his mentorship and support throughout the process that has led to this dissertation and in the first place, for the opportunity and encouragement to pursue what has been one of my richest learning experiences. I would also like to thank my committee members Dr. Milind Kandlikar and Dr. Steven Rogak for their valuable insights and support in navigating the process. Further, I would like to thank all my friends and colleagues (who are numerous and know who they are) at the Institute for Resources, Environment and Sustainability. In particular, I would like to thank Laura DeVries for all her support in the process and Francis Ries whose collaborations helped make this dissertation possible. Finally, this acknowledgement would not be complete without acknowledging my parents, whose support and encouragement in any and all directions I have taken in life has been unconditional.  This research was supported by Scholarships from the Natural Sciences and Engineering Research Council of Canada, the Robert B. Caton Memorial Scholarship Fund, a Doctoral Fellowship from the University of British Columbia and grants from the Climate Decision Making Center at Carnegie Mellon University, Auto21 and TransLink.  xix  Dedication  To my parents and grandfather Cyril  xx  Chapter 1: Introduction Urban transportation has featured prominently in society’s social, political, public health and environmental debates (Divall, et al.,2003, McShane,1994, 1997, Melosi,1993, Tarr,1996). It has dramatically changed the social and physical landscape of urban areas and influenced cultural values (Foster,1981, Jackson,1985, Marland, et al.,1988, McShane,1994, Warner,1962). In so doing it has become an integral part of modern society.  The original mode of urban (passenger) transportation was walking (Jones,2008). Domestication and technological innovations allowed horses to be become the dominant motive power source in North America between 1830 and 1890, culminating in the omnibus and then the horse-pulled street railway, the first public transit 1 systems. After 1890, electricity emerged as the dominate power source for the street railways, replacing the horse. In 1920 the automobile began its rise to prominence and since 1945 it has been the dominate mode of urban transportation in many Canadian and American cities, supplanting public transit.  Public transit ridership in the United States (U.S.) peaked during World War II (Figure 1.1). Following the war, ridership collapsed for a complex and interlinked set of social, economic, political, and technological reasons (Jones,2008, McShane,1994). However, public transportation ridership has been gradually increasing since the 1970s and in 2009 reached its highest level in over 5 decades, 10.4 billion unlinked passenger trips 2 (APTA,2011).  Public transit is once again at the forefront of urban development discourse due to increased awareness of the adverse impacts of current urban transportation systems (e.g., private vehicles), a shift in values, and awareness of sustainability3. It is a core feature of emerging urban planning philosophies such as smart growth, new urbanism and transit-oriented development that focus on integrated transportation and land-use planning approaches, increased density and reduced private vehicle use (Cervero,2004, Newman, et al.,1996, Tiwari, et al.,2011). Public transit has also been widely promulgated as a policy measure to reduce emissions of pollutants 1  Also referred to as public transportation, transit, or mass transit: transportation by a conveyance that provides regular and continuing general or special transportation to the public, but not including school buses, charter or sightseeing service. (APTA,2011). 2 Unlinked Passenger Trips is the number of times passengers board public transportation vehicles (APTA,2011). 3 Definitions of sustainability vary (Costanza, et al.,1995, Robinson,2004), but it has historically been defined in terms of meeting the needs of the present without compromising the ability of future generations to meet their own needs (World Commission on Environment and Development,1987.).  1  that impact both human health and the climate from the transportation sector (APTA,2011, Kahn Ribeiro,2007). In such contexts, transit systems are often viewed as intrinsically sustainable, having significant social, economic, and environmental benefits. However, these systems are not inherently benign. As with other modes of urban transportation, constructing and operating  25 20 15 10  350  Other Trolley Bus Light Rail Heavy Rail Paratransit Commuter Rail Bus Population  300 250 200 150 100  5 0 1900  Population (Millions)  Unlinked Passenger Trips (Billions)  these systems consumes energy and resources that result in direct and indirect adverse impacts.  50 1910  1920  1930  1940  1950  World War I and Great World Growth of Street Railways Post War Boom Depression War II  1960  1970  Cheap Energy and Growth of Suburbs  1980  1990  2000  0 2010  Intergovernmental Partnership  Figure 1.1 - Public transportation ridership in the United States after (APTA,2006, 2011, United States Census Bureau,2000, 2012).  Over the course of its history, urban transportation has been framed as both a cause and a solution to many of society’s social, public health, and environmental problems. These once disconnected debates and discourses are increasingly being integrated under the banner of sustainability (Banister,2008). While many important questions must be answered if we are to define and achieve a collective vision of sustainable urban transportation (Robinson,2004), this dissertation takes a more focused and immediate perspective with the aim of fostering sustainability in transit systems.  Specifically, this dissertation focuses on: (a) quantifying the climate and health impacts of the emissions from North American, public transit heavy-duty 4 bus fleets 5 and, to a lesser extent, the cost of operating these fleets and (b) developing new strategies of reducing these impacts and costs. Climate and health impacts and operating costs align with three fundamental dimensions of sustainability (environmental, social, and economic) and have been identified by transit agencies as being key sustainability indicators (APTA,2012, CUTA,2012). Other equally important dimensions could also be considered, but were beyond the scope of this dissertation. 4  Gross Vehicle Weight (GVW) is greater than 8500 lbs. Buses represent the dominate mode employed by transit agencies, both in terms of passenger trips and passenger kilometers (Figure 1.1) (APTA,2011).  5  2  Transit agencies have historically faced significant fiscal challenges (Jones,2008). This has limited their ability to invest in new capital (e.g., modern, lower emitting buses) to not only maintain and expand service but to reduce the climate and health impacts of the operations of their bus fleets. These challenges have been exacerbated by the recent economic downturn. A survey by the American Public Transportation Association (APTA) showed that in 2010 almost 70% of United States (U.S.) transit agencies surveyed were facing budget shortfalls that had forced them to reduce service and delay or curtail capital investments (APTA,2010b). Almost 50% of the agencies surveyed reported transferring funds from capital to operations budgets. In light of these fiscal challenges, this dissertation focuses on strategies of reducing the climate and health impacts of emissions from bus fleets that target operations, as these strategies are likely to require significantly less capital investment to implement. 1.1  Climate and Health Impacts of Transit Bus Emissions  Most transit buses in North America are classified as heavy-duty vehicles and powered by either diesel or compressed natural gas (CNG) internal combustion engines (Figure 1.2). They are on average, 40 ft in length with a maximum capacity of approximately 70 passengers; however, there is a large diversity of transit buses 6 (APTA,2011). The combustion of hydrocarbon fuels (e.g., diesel and CNG) in internal combustion engines produce emissions that have adverse climate and human health impacts. These impacts occur through a complex series of processes that vary over a wide range of spatial and temporal scales. As a result, quantifying both emissions and impacts are difficult and current approaches often rely on simplifying assumptions that may not be valid in all contexts. 0%  1% Diesel  8%  Hybrid  18% 7%  CNG 66%  Electricity Gasoline Other  Figure 1.2 - Transit bus distribution by energy source (APTA,2011).  6  For example, many transit agencies also operate larger buses (≥60 ft) that can carry upwards of 120 passengers.  3  1.1.1  Heavy-Duty Vehicle Emissions  The emissions produced by heavy-duty vehicles, including transit buses, results from combustion as well as several other processes. The former are referred to as exhaust emissions, while the latter are referred to as non-exhaust emissions. Further, some primary exhaust emissions participate in photochemical reactions in the atmosphere that lead to the formation of secondary pollutants.  1.1.1.1  Exhaust  The most widely used internal combustion engines in heavy-duty vehicle and transit bus applications are compression ignition (CI) diesel engines and spark ignition (SI) CNG engines. Both engine types are typically operated using lean air-fuel mixtures (i.e., lean-burn) to increase thermal efficiency and reduce emissions (Heywood,1988, Manivannan, et al.,2003, McTaggartCowan, et al.,2006).  Under ideal stoichiometric conditions, hydrocarbon fuels such as diesel and CNG are completely oxidized into carbon dioxide (CO 2 ) and water (H 2 O) in the combustion process. However, as a result of rapid changes in temperature and pressure, variations in the local 7 fuel-air mixture, and the presence of other compounds in the fuel and air, a range of other compounds are also formed, including carbon monoxide (CO), hydrocarbons (HC), particulate matter (PM), nitrogen oxides (NO X ), and sulfur oxides (SO X ), as well as a group of compounds that are referred to as air toxics or mobile source air toxics (MSAT) (Heywood,1988).  CO and HC are products of incomplete combustion. CO is typically formed due to insufficient oxygen (i.e., locally rich air-fuel mixtures) whereas HC emissions primarily result from unburned fuel and lubricating oil. Both CO and Non-Methane Hydrocarbon (NMHC) emissions from lean-burn engines are normally low (Heywood,1988); however, methane (CH 4 ) emissions from CNG engines can be significant and contribute to higher total HC emissions (Cho, et al.,2007, Hesterberg, et al.,2008).  7  Local refers to the air-fuel mixture within the engine cylinders which can vary substantially especially in compression ignition engines.  4  NO X and SO X emissions are indirect 8 products of combustion resulting from the oxidation of atmospheric nitrogen and sulfur in the fuel and lubricating oil (Heywood,1988). NO X is composed of nitric oxide (NO) and a smaller fraction, typically 10-30% in diesel engine-out emissions, of nitrogen dioxide (NO 2 ). The formation of NO X is governed in part by the combustion temperature. As result of higher combustion temperatures, NO X emissions from heavy-duty diesel engines are typically greater than those from CNG engines 9 (McTaggartCowan, et al.,2006). Both conventional hydrocarbon fuels and lubricating oils commonly contain sulfur, although new fuel standards have significantly reduced sulfur levels in fuels. During combustion, most sulfur is oxidized to sulfur dioxide (SO 2 ), but a small fraction of the sulfur leads to the formation of sulphuric acid and sulfate particles (Kittelson,1998, Seigneur,2009).  PM is a complex mixture of particles in solid and liquid phase that are suspended in air. Unlike compounds such as CO 2 which are defined by their molecular composition, PM is operationally defined; its definition and the definition of its constituents are co-evolving with measurement methods and procedures (Athanasios, et al.,2004, Kittelson,1998, Tree, et al.,2007). For example, the gas-particle partitioning of the semi-volatile constituents of PM are sensitive to both the temperature and dilution ratios at which measurements are made (Shrivastava, et al.,2006). Thus, different measurement procedures may result in different definitions. As a result, there is some ambiguity in the terminology used to describe PM (Gelencsér,2004).  PM is commonly classified based on the aerodynamic diameter of the particles and measured in terms of total particle mass as well as particle numbers (Table 1.1) (Kittelson,1998). The mass and size distributions of diesel exhaust particles commonly exhibit bi- or tri- modal distributions with peaks in the nuclei, accumulation, and/or course mode particle size ranges. Accumulation mode particles typically make up the largest fraction by mass, whereas nuclei mode particles make up the largest fraction of particles by number (Kittelson,1998, Kittelson, et al.,2006).  8  Don’t participate in the primary combustion reaction. NOX emissions from light-duty SI engines are usually significantly lower than from either heavy-duty CI diesel or SI CNG engines. 9  5  Table 1.1 - Particulate Matter (PM) classifications by particle diameter (Kittelson,1998). Definition PM 10 or Coarse Particles PM 2.5 or Fine Particles Ultrafine Particles Nanoparticles Nuclei Mode Particles Accumulation Mode Particles Coarse Mode Particles  Particle Diameter < 10 µm < 2.5 µm < 0.10 µm < 0.05 µm or 50 nm 0.005 – 0.05 µm 0.1 – 0.3 µm > 1 µm  PM emissions from diesel engines are primarily composed of carbonaceous material, unburned hydrocarbons 10, and a smaller fraction of ash 11 and sulfur compounds (Heywood,1988, Kittelson,1998). The main constituent of PM is carbon, which occurs either in the form of Elemental Carbon (EC) or organic compounds referred to as Organic Carbon (OC) (Zielinska,2005). Although technically distinct, EC is also widely referred to as Black Carbon (BC) and the two terms are considered synonymous for the purposes of this dissertation (Gelencsér,2004). PM emissions from CNG engines are generally significantly lower than diesel engines and are primarily comprised of semi-volatile organic nanoparticles (Jayaratne, et al.,2008, Jayaratne, et al.,2010).  1.1.1.2  Non-Exhaust  Vehicle emissions can result from several other processes in addition to combustion including: fuel leakage and evaporation, brake and tire wear, and the re-suspension of road dust (Boulter,2005, Lloyd, et al.,2001, USEPA,2002a). The non-exhaust emissions that result from these processes have garnered significantly less attention than exhaust emissions, and studies and understanding of these emissions are still limited. Generally, evaporative emissions from diesel vehicles are low due to the low volatility of the fuel and closed fuel systems (Lloyd, et al.,2001). Evaporative and leakage emissions of methane from CNG vehicles may be significant and some researchers have attempted to account for them (Reynolds, et al.,2008). Resuspended road dust and, to a lesser degree, brake and tire wear, have received renewed attention and will likely become an increasingly significant source of PM as stricter emissions standards reduce PM exhaust emissions (Health Effects Institute,2010, Thorpe, et al.,2008). Further, because of greater turbulence induced by heavy-duty vehicles, they typically generate significantly greater amounts of resuspended PM than light-duty vehicles (Abu-Allaban, et al.,2003). 10  Associated with both the unburned fuel and lubricating oil. Unburned lubricating oil is believed to play a significant role in the formation of particles in the ultrafine range (Sakurai, et al.,2003, Seigneur,2009). 11 Associated with metal compounds in both the fuel and lubricating oil (Kittelson,1998, Seigneur,2009).  6  1.1.1.3  Secondary Pollutants  Once primary vehicle emissions enter the atmosphere, they participate in complex chemical and physical processes that transform some primary pollutants and produce secondary pollutants. Two secondary pollutants are particularly relevant in the context of heavy-duty vehicle emissions: ground-level ozone and secondary PM 12. Ozone is produced through a complex, nonlinear, photochemical reaction involving Volatile Organic Compounds (VOC) 13, NO X and sunlight (USEPA,2006). Due to the nonlinear chemistry, increased NO X emissions can lead to either increases or decreases in ozone depending on the concentrations of the compounds participating in the reaction. Secondary PM forms as a result of a number of complex chemical and photochemical process involving both inorganic and organic compounds (Ning, et al.,2010, Robinson, et al.,2007).  1.1.2  Health and Climate Impacts  The emissions produced by heavy-duty vehicles degrade air quality and adversely affect human health (Colvile, et al.,2001, Health Effects Institute,2010, IPCC,2007b, Lloyd, et al.,2001). In addition, they contribute to global climate change (IPCC,2007b, Lloyd, et al.,2001). Historically, their impacts on air quality and health have received greater attention. Many jurisdictions including Canada and the U.S. regulate pollutants believed to impact health (Bachmann,2007). However, rising concerns about climate change have brought about a renewed focus on the climate impacts of vehicle emissions (also commonly referred to as mobile source emissions). This dissertation considers both the climate and health impacts of transit bus emissions.  1.1.2.1  Health  Human health impacts are typically assessed in terms of mortality (death) and morbidity (a state of ill health). Epidemiological and toxicological studies have associated vehicle emissions with both short-term (minutes-days) and long-term (years-lifetimes) respiratory and cardiovascular morbidity and mortality as well as all-cause mortality and adverse neurological and developmental health effects (Health Effects Institute,2010). Pollutants including PM, NO X , 12  In the field of atmospheric chemistry, particulate matter is often referred to as aerosols. Volatile organic compounds are all carbon-containing gas-phase compounds in the atmosphere except CO and CO2. 13  7  CO, HC and MSAT as well as secondary pollutants such as ground-level ozone are believed to play important roles in causing these effects; however, the mechanisms (e.g., oxidative stress and pulmonary or systemic inflammation) by which these pollutants lead to adverse health outcomes are not fully understood. This makes it difficult to establish causation and attribute health outcomes (Brunekreef, et al.,2002, Health Effects Institute,2007, 2010, Hill,1965, Pope, et al.,2006). A recent review by the Health Effects Institute (HEI), found there was sufficient evidence to establish causality for the exacerbation of asthma, but for other outcomes, including all-cause mortality as well as cardiovascular mortality and morbidity, evidence was suggestive but not sufficient to establish causal associations (Health Effects Institute,2010). The wide range of health outcomes, particularly those related to morbidity, the different temporal scales of the evidence (i.e., short- and long-term) and scientific uncertainties regarding the causal mechanisms and the relationship between vehicle emissions and health effects makes it difficult to quantify the overall net health impact of vehicle emissions.  Despite these challenges, much attention has been devoted to the health impacts of PM (Dockery, et al.,1993, Lloyd, et al.,2001, McCarthy, et al.,2006, Pope, et al.,1995, Pope, et al.,2006, USEPA,2009c). The complex composition and structure of PM and the lack of a comprehensive measure that captures these properties have to some degree confounded efforts to understand its effects. For example, the most common measure of PM is the mass of particles in a specific size classification (e.g., grams of PM 2.5 ); however, other properties including the number of particles and the chemical composition of the particles are also believed to mediate health impacts. Nevertheless, the United States Environmental Protection Agency (USEPA) has concluded that there is sufficient evidence to infer a causal relationship between PM 2.5 and both short- and long- term mortality (USEPA,2009c). As a result of this body of evidence, PM has been widely used as an indicator of health impacts in a diverse range of studies, including studies of transit buses (Grieshop, et al.,2011, Levy, et al.,2010, Levy, et al.,2009, Stevens, et al.,2005, Tainio, et al.,2005).  Of the other pollutants implicated in causing health effects, NO X and ozone have garnered the most attention. This is especially true in the context of heavy-duty vehicle emission (Brunekreef, et al.,2002). Ozone is a powerful oxidant and has been linked to short-term respiratory morbidity and there is suggestive evidence of other effects including mortality (USEPA,2006). NO X has been connected to short-term respiratory and cardiovascular morbidity as well as other 8  health effects (USEPA,2008), although its indirect effect, as a precursor to ozone, has typically been of greater concern.  1.1.2.2  Climate  The climate is a complex global system consisting of the atmosphere, land surface, snow and ice (cryosphere), oceans and other bodies of water (hydrosphere), and living things (biosphere) (Le Treut,2007). It is often described as average weather and is usually characterised in terms of the mean and variability in properties including temperature and precipitation over long periods of time (e.g., months to millions of years). The scale at which the climate system operates makes detection and attribution of changes challenging (Le Treut,2007). Controlled experiments of the climate system are not possible. Instead, understanding of the system must be developed from past observations and theoretical computer models. Further, the potential impacts of these changes, which are ultimately what is of importance, must also be established. The challenges and uncertainties in understanding the climate system and the impacts of changes to it are significant and have stimulated substantial debate and controversy (Hulme,2009).  The Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) provides a comprehensive discussion of the properties used to quantify climate change and the potential impacts of changes in the climate (IPCC,2007b). In general, climate change is described in terms of the change in the radiative energy balance of the earth’s atmosphere and the global, annual average surface temperature. Several metrics, including Radiative Forcing (RF) and Global Warming Potential (GWP), have been developed to quantify the effect that compounds have on these properties (Forster,2007). Potential impacts of increased surface temperature and global warming include harm to ecosystems, sea level rise, increased ocean acidity, increased frequency and severity of storms, and adverse human health effects (IPCC,2007a). The range, time scales, and uncertainty of these impacts makes estimating the net impact and cost of climate change challenging and controversial (Hulme,2009, Nordhaus,2007, Stern,2007).  Numerous compounds that affect the radiative balance of the earth’s atmosphere have been identified. Both the magnitude of the forcing and the atmospheric lifetime of these compounds are important in quantifying their effects on the climate. Traditionally, the focus has been on long-lived, globally well-mixed compounds including CO 2 , CH 4 , and nitrous oxide (N 2 O). For 9  example, these compounds were defined as greenhouse gases (GHG) under the United Nations Framework Convention on Climate Change (UNFCCC) and Kyoto Protocol and used to establish reporting guidelines and national reduction targets (Environment Canada,2010). More recently, attention has shifted to short-lived compounds such as Black Carbon (BC), Organic Carbon (OC), and Sulfate (SO 4 ) (constituents of PM) as well as ground level ozone. (Bond, et al.,2005, Forster,2007, Grieshop, et al.,2009, Penner, et al.,2010, Ramanathan, et al.,2008). However, efforts to estimate the impacts of these compounds have been controversial due to their short life spans, globally heterogeneous distribution in the atmosphere and greater uncertainties in the mechanisms that lead to radiative forcing (Bond,2007, Forster,2007, Reynolds, et al.,2008). Despite these challenges, BC is now believed to be the second most important compound driving global warming after CO 2 (Jacobson,2009, Ramanathan, et al.,2008). Other PM constituents including OC and SO 4 lead to cooling (i.e., negative radiative forcing) and have been estimated to offset about half of the global warming to date (Jacobson,2009). Consequentially, assessing the climate impacts of short-lived compounds such as PM is more complex than assessing the climate impacts of long-lived compounds such as greenhouse gases.  1.1.3  Health and Climate Impact Pathways  The health and climate impacts of vehicle emissions occur as a result of a series of causal processes described by the health and climate impact pathways (Figure 1.3) (Fuglestvedt, et al.,2003, Marshall, et al.,2006, Smith,1993). Many aspects of the impact pathways are associated with significant scientific uncertainties. These uncertainties typically increase from emissions to effect. The first several processes in the impact pathways are common. Emissions arise from the operations or activity of vehicles, then disperse and may undergo transformation in the atmosphere resulting in pollutant concentrations both at ground-level (troposphere) and at higher layers of the atmosphere (e.g. stratosphere). At this point the impact pathways diverge. Through activity 14, humans are exposed 15 to ground-level pollutants (Monn,2001). A fraction of these pollutants (i.e., the intake fraction) are inhaled (intake) as a result of respiration and remain in the body (dose) leading to adverse health effects. Pollutants may also be ingested or absorbed 14  Population activity refers to the location of people over time and space. Note that this does not have to change for exposure to occur. 15 Exposure is formally defined as the contact of a stressor (i.e. pollutant) and a receptor (i.e., human) over space and time.  10  through the skin, but the primary focus of air pollution related health impact research has been the respiratory pathway.  As the pollutants migrate throughout the atmosphere (from troposphere to stratosphere) they can change the radiative balance of the atmosphere through several processes including the greenhouse effect, which leads to radiative forcing and either warming or cooling. Other mechanisms may also contribute to warming and cooling, such as changes in the albedo of the earth’s surface (Forster,2007). The resulting changes to the earth’s climate have the potential to adversely impact the environment and human welfare.  11  Increasing Uncertainty Increasing Relevance Scale (Sociopolitical)  Scale (Physical): Spatial and Temporal Variability  Drivers  Activity (Source & Receptor)  Geography Behaviour Built Environment Environment Population  Population (Receptor) Activity  Transit Planning Timetables Routes  Vehicle (Source) Activity  Emissions  Air Quality (Dispersion and Transformation)  Impacts (Health)  Exposure Intake Dose  Exhaust Emissions  Health Impacts Valuation & Trade-offs  Exposure  Life Cycle Dispersion and Transformation  Non-Exhaust Emissions  Climate Change Radiative Forcing  Decision Making  Climate Change  Impacts (Climate) Climate Impacts  Costs  Control Strategies Other Impact Pathways (e.g., noise) Not Shown  Feedback Pathways Not shown Impact Pathways Integrated Assessment  Figure 1.3 – Climate and health impact pathways and integrated assessment framework (Fuglestvedt, et al.,2010, Marshall, et al.,2006, Smith,1993). Grey indicates non-focus areas.  12  1.1.4  Scale and Variability  The spatial and temporal scales of the processes involved in the climate and health impact pathways differ significantly. Some fast processes vary rapidly over time and/or space (e.g., combustion) while other slow processes vary more gradually (e.g., the earth’s surface temperature). These processes form a complex system whose dynamics ultimately determine the climate and health impacts (Holling,2001). Understanding the dynamics and modeling these systems present significant challenges, in large part due to the wide range of scales involved.  There is considerable variability and ambiguity in the language used to describe spatial and temporal scales, particularly across disciplines. In the context of quantitative analyses, scale is often referred to as resolution, and the two terms are considered synonymous in this dissertation. Scale and resolution are related to a third concept, level of detail. Level of detail is more difficult to define, but refers to aspects such as the number of parameters of a model, and may also be referred to as complexity. The level of detail typically increases as the scale decreases.  In this dissertation, scale is described using three principal terms: macro, meso, and micro (Table 1.2) (National Research Council,2000). A certain latitude in interpreting these terms is required, as definitions vary between disciplines and context. In addition, several other terms are used to describe specific spatial (e.g., region) and temporal (e.g., hours) scales (Table 1.2). In general, the spatial and temporal classifications of scale are the same. For example, a process that is classified as macro on a spatial scale is also macro on a temporal scale. Further, it is important to distinguish between the scale used to describe a method or modelling approach and the scale used to describe its application. For example, it is possible to apply a micro scale modelling approach on a macro or meso scale.  13  Table 1.2 - Definitions of scales. Pathway Process Emissions and Vehicle Exposure, Dispersion, and Activity1 Population Activity2  microenvironments – regional; hours-seasons; census blocks; microenvironments concentrations  10s meters; seconds – minutes; individual vehicles - disaggregated vehicle groups  microenvironments; minuteshours; personal exposure  (Intra-Regional5)  100s meters – 10s kilometers (e.g. roadway links); hours-months; disaggregated vehicle groups (e.g., transit buses by model year)  Micro4  Macro  regional; years; total population; ambient concentrations  regional - national; seasons -years; coarse vehicle groups (e.g. transit buses)  Meso  Scale  Impacts Climate  3  Health3  regional – global; months – millennia microenviron ments regional; minutes – lifetimes  1  Defined in terms of: spatial scale; temporal scale; vehicle aggregation. 2 Defined in terms of: spatial scale; temporal scale; population aggregation. 3 Defined in terms of: spatial scale; temporal scale. 4 Also commonly referred to as local scale in the context of air quality and health. 5Any scale below the regional scale. Also referred to as intra-urban.  The different scales of the impact pathway processes mean that the choice of scale or resolution in an analysis or model has important implications. Generally, there is a trade-off between the level of detail, which, as discussed, is related to scale, and cost in terms of effort and data requirements. Averaging and aggregation techniques are typically employed to make models and analyses feasible for application on large scales. For example, it is not currently feasible to model individual vehicles when estimating the total emissions from all vehicles in a region. These averaging and aggregation techniques often have the advantage of reducing uncertainty. For example, the uncertainty in predicting the emissions of a single vehicle at specific point in time is much greater than predicting the emissions of a group of vehicles over a long period of time. However, depending on the assumptions made, these techniques may also introduce errors and biases.  1.1.5  Traditional Management and Quantification Approaches and Limitations  Climate and health impacts have traditionally been analysed and managed at the macro scale (i.e., regional-national scale for climate and regional scale for health) in independent decisionmaking frameworks based on emissions inventories. An emissions inventory is an accounting or estimate of the total mass of pollutants discharged into the atmosphere from a region, province, 14  state, nation, etc. over a specific period of time. Inventories of vehicle emissions are typically estimated using either a combination of distance travelled data and distance-based emissions factors, or fuel consumption data and fuel-based emission factors: 𝐸𝑐,𝜏 = � 𝐴𝛼,𝜏 × 𝐸𝐹𝛼,𝑐  Equation 1.1  𝛼  where E is the total emissions of pollutant c over time τ; A is the activity of vehicle classification  α over time τ; and EF is the fuel- or distance- based emission factor for vehicle classification α and pollutant c (Section 1.6). Distance-based approaches are commonly used when estimating inventories of emissions that impact air quality and health, but have been criticised for their inability to account for the effects that vehicle operations (e.g., acceleration, cruise, etc.) have on emissions (Barth, et al.,1996a, National Research Council,2000). Fuel-based approaches are primarily used when estimating inventories of pollutants that impact the climate (Environment Canada,2010, USEPA,2010a). These inventories are then used to formulate independent control strategies and policies that target emission sources based on the total mass of emissions produced (Nazaroff,2008).  A limitation of the macro/regional scale management approach is that it effectively ignores exposure (Nazaroff,2008). Within this management paradigm, health impacts are related to regional, ambient pollutant concentrations (Dockery, et al.,1993) and exposure is assumed to be proportional to the total mass of emissions represented in the emissions inventory (Nazaroff,2008). However, health impacts are realized by people, and personal exposure among individuals or groups of individuals may vary as a result of different time-activity patterns and microenvironments with different pollutant concentrations (Greco, et al.,2007a, Kaur, et al.,2007, Stieb, et al.,2008, Zhou, et al.,2007). When exposure is assessed at the regional scale, variations in personal exposure are effectively averaged out and the spatial and temporal relationships between individuals and pollutant concentrations within a region are lost. If there is a correlation in these relationships, then the resulting exposure and effects estimate will be biased. A number of studies have shown that there is in fact a positive correlation that leads to negative bias and underestimates (Health Effects Institute,2010, Jerrett, et al.,2005b, Monn,2001, Setton, et al.,2011). An important implication of these findings with regards to control strategies is that they open a fundamentally new avenue of controlling impacts based on manipulating this relationship and exposure on an intra-regional scale. 15  Until recently, the interactions of control strategies and policies aimed at mitigating climate and health impacts have not been considered (Dowlatabadi,2007, Mazzi, et al.,2007, Rotmans, et al.,2003, Smith, et al.,2008). Recent studies have shown that this can lead to detrimental and unintended consequences; therefore, it cannot be assumed that strategies aimed at mitigating climate and health impacts are inherently mutually beneficial (Mazzi, et al.,2007). Fortunately, growing awareness of the climate impacts of pollutants such as PM and ozone, previously only considered because of their health effects, have encouraged the adoption of integrated assessment frameworks in an effort to identify strategies with climate and health co-benefits (Grieshop, et al.,2011, Shindell, et al.,2011, Smith, et al.,2008, Woodcock, et al.,2009). These pollutants have created an important linkage between once disparate regional air quality and health issues and global climate change. In the process, they are changing the way these impacts and strategies to control them are being evaluated.  In summary, there are at least three limitations with the traditional paradigm used to quantify and manage the impacts of vehicle emissions. First, emissions estimated using distance-based approaches poorly account for the influence of vehicle activity (e.g., speed, acceleration, and road grade) on emissions and as a result may be biased (Barth, et al.,1996a, National Research Council,2000). Second, exposure is not accounted for and as a result estimates of health impacts may be biased (Health Effects Institute,2010, Nazaroff,2008). Third, failure to evaluate control strategies in an integrated framework may lead to detrimental and unintended impacts (Mazzi, et al.,2007).  1.2  Mitigating Health and Climate Impacts: Capital and Operational Control Strategies  In principle, it is possible to influence the climate and health impacts of vehicle emissions at any stage in the impact pathways (Smith,1993). However, the processes of the impact pathway within the purview of transit agencies are limited to the activity and emissions processes (Figure 1.3). These processes and the subsequent climate and health impacts can be influenced by transit agencies in two fundamental ways: a) capital investments in technology that reduce emissions from buses, and b) operational changes that influence activity (e.g., reduce activity or affect where activity is occurring in time and space) and/or reduce emissions from buses. These two  16  mitigation approaches are herein referred to as capital and operational control strategies, respectively. Transit agencies have traditionally relied on capital control strategies such as the acquisition of new lower emitting buses and the retrofitting or repowering of existing buses to reduce the emissions and climate and health impacts of their fleets. There is an extensive body of literature, reviewed in Section 1.6.2.2, examining the application of engine, powertrain, and aftertreatment technologies as well as fuels to control emissions from heavy-duty vehicles, including transit buses. The current range of heavy-duty bus technologies means that transit agencies have several alternatives and inter-related choices, including the energy source or fuel type (e.g., electricity, diesel, or CNG), the powertrain type (e.g., conventional, hybrid-electric, electric), and the aftertreatment type. Transit agencies may also be able to retrofit existing buses with aftertreatment devices such as diesel particulate filters. Each technology is associated with different levels of emissions reduction and capital costs (Table 1.3), which have risen significantly16. Table 1.3 – Average capital costs of bus technologies in 2010 (APTA,2010a).  Standard1 Diesel  Average Cost (2010 USD) 377,000  Standard1 CNG  456,000  Bus Category  1  Standard Diesel Hybrid 2  Articulated Diesel  562,000 659,000  2  804,000 Articulated Diesel Hybrid 1 Classified by the APTA as a transit bus and > 27’6” ft and < 55’ ft in length. 2 Classified by the APTA as a transit bus and > 55’ ft in length.  Operational control strategies comprise a wide range of approaches related to the operations and operational efficiency of transit bus fleets including but not limited to: vehicle maintenance, vehicle operations and vehicle scheduling. In general, the costs of implementing operational control strategies are likely to be less than capital control strategies and have the potential to reduce operating costs. However, there is limited academic research on the effectiveness of operational control strategies to reduce the climate and health impacts of emissions from transit bus fleets. For example, the effects of driver behavior on emissions are often noted in studies 16  The cost of a standard transit bus (> 27’6” ft and < 55’ ft in length) rose by 62% between 2001 and 2010 (APTA,2011).  17  (Clark, et al.,1999a, Hayes, et al.,2006), but only one study has attempted to quantify the potential benefits of modifying the behaviour of transit bus drivers (Zarkadoula, et al.,2007) 17. Further, while there is an extensive body of literature on the application of operations research methods to optimise the operations of transit systems (Ceder,2011), few studies explicity addressed climate and health impacts. Existing research efforts have focused on how to optimally utilise a bus fleet to service a set of routes and a timetable 18. This is widely referred to as the vehicle scheduling problem 19. Traditionally, the objective of the vehicle scheduling problem has been to minimise the total number of vehicles required and the non-revenue (i.e., without paying passengers) distance or time travelled (e.g., between a depot and the start of a route). Although it is likely that vehicle scheduling optimisation has also reduced climate and health impacts, this remains a largely unexplored question, and very few studies have attempted to directly incorporate climate and health impacts as objectives in the vehicle scheduling problem (Figliozzi,2010, Li, et al.,2009a, Stasko, et al.,2010).  1.3  Integrated Assessment and Decision Making  The capital and operational control strategies available to transit agencies present them with a challenging decision problem (Clemen, et al.,2004, Keeney,1982). Transit agencies are forced to make complex planning and operating decisions involving multiple objectives and often, multiple stakeholders. Quantifying these objectives is often associated with significant uncertainties, as is the case, for example, with climate and health impacts. Until very recently, transit agencies have had few tools to aid in the decision making process. The tools that are currently available are still largely under development (Clark, et al.,2007b, Golub, et al.,2010, Wayne, et al.,2011b) and are emissions (inventory) models that do not address the full impact pathways (e.g., exposure). Further, there is limited guidance on how transit agencies should handle and evaluate trade-offs between multiple objectives. A challenge to developing these tools and decision aids is that the regulatory, institutional, physical, and economic contexts in which these decisions are made, as well as the values and priorities of individual agencies, vary.  Integrated assessment offers a potential interdisciplinary approach to developing a decision making framework to assist transit agencies. Although integrated assessment has traditionally 17  The study suggests that eco-driving could reduce fuel consumption and CO2 emissions by approximately 10%. Routes and timetable are typically determined in the planning process. 19 The vehicle scheduling problem is often solved using linear programming techniques. 18  18  been applied to global environmental problems such as climate change, it is equally suited to regional issues (Dowlatabadi, et al.,1993, Rotmans, et al.,2003). Integrated assessment is an iterative and “… structured process of dealing with complex issues, using knowledge from various scientific disciplines and/or stakeholders, such that integrated insights are made available to decision-makers” (Figure 1.4) (Rotmans, et al.,2003). However, it does not espouse specific methods or even a unifying theory (Risbey, et al.,1996, Rotmans, et al.,2003). Instead, integrated assessment provides a philosophical approach to structuring, integrating, and creating knowledge to inform decision making that embodies several key concepts (Dowlatabadi, et al.,1993, Morgan, et al.,1996, Morgan, et al.,1999, Risbey, et al.,1996, Rotmans, et al.,2003): •  decision oriented – its purpose is to engage decision makers and inform decisions  •  integrated – it is an interdisciplinary, systems approach with a focus on relationships and interactions and the integration of knowledge and values from multiple disciplines and stakeholders  •  learning – it is an iterative process of learning that provides insights rather than answers  •  scales – it acknowledges and treats a wide range of spatial, temporal and sociopolitical scales  •  limited knowledge – it includes explicit treatment of uncertainties; areas of limited knowledge should not be ignored Problem Definition Decision Making Modeling Assessment  Uncertainty and Sensitivity Analysis  Experiments  Figure 1.4 - Integrated assessment process after (Rotmans, et al.,2003).  19  Incorporating these concepts in practice is not trivial. The lack of a clearly defined methodology makes performing integrated assessment challenging and evaluating it even more difficult (Morgan, et al.,1996, Risbey, et al.,1996). Despite this, several methods of performing integrated assessments have evolved. The focus here is on quantitative integrated assessment modeling, which has been the dominate method of integrated assessment to date (Rotmans, et al.,2003).  Integrated assessment models are computer models that quantitatively characterise the causeeffect relationships (e.g., the climate and health impact pathways) and either quantitatively or qualitatively incorporate decision making processes (e.g., how to sustainably operate a public transit bus fleet) (Figure 1.3). They are typically developed by leveraging and integrating existing disciplinary methods and models and improving upon them (Morgan, et al.,1999). Rotmans et al. identify three key methodological issues related to integrated assessment modeling: level of detail, treatment of uncertainty, and the integration of qualitative and quantitative knowledge (Rotmans, et al.,2003). Further, Risbey et al. raise the issue of the need for more rigorous quality control and evaluation (Risbey, et al.,1996). Issues related to the level of detail were previously discussed in Section 1.1.4.  A hallmark of integrated assessment is the quantification of uncertainties. Uncertainty is defined as the lack of knowledge of the true value. There are numerous sources of uncertainty; however, they are commonly classified in two categories: model and parameter uncertainty (Morgan, et al.,1990). Uncertainties can be treated quantitatively using stochastic modeling approaches. For example, parameter uncertainties can be described using probability distributions. A current barrier to quantifying uncertainties is that many disciplinary models are deterministic and do not provide estimates of uncertainties.  Several methods can be used to characterise uncertainty, including uncertainty analysis, local and global sensitivity analysis (also referred to as importance analysis) and scenario analysis (Helton, et al.,2006, Morgan, et al.,1990, Saltelli, et al.,2008). Uncertainty analysis involves the propagation of uncertainties from model inputs to outputs. It is commonly performed using sampling methods such as Monte Carlo simulation. Both sensitivity and scenario analysis involve characterising the response of the model to variability in model inputs.  20  Evaluation and validation are crucial in ensuring quality and establishing the trust of decision makers and stakeholders. Unfortunately, they are often only performed on a limited basis (National Research Council,2000, Risbey, et al.,1996). There is some ambiguity in the two terms and they are often used interchangeably. Here, evaluation is defined as assessing the ability of a model to accurately represent the phenomenon being modelled, whereas validation is defined as assessing the mathematical derivation and structure of a model (National Research Council,2000).  Attempting to fully describe a complex problem in a quantitative, mathematical model is fraught with difficulties and unlikely to provide meaningful results without further interpretation (Rotmans, et al.,2003). The complexities of the problems, particularly decision making processes, often preclude them from formal mathematical treatment. This does not mean that modeling exercises are without value, but it is important to understand the limits of such models.  Decision makers are often faced with conflicting objectives and complex trade-offs arising from uncertainties as well as the distribution of costs and benefits over time and space. For example, costs typically occur in the present while benefits accrue over time. Further, costs and benefits may be unequally distributed across populations both in space and time. Several formal decision making methods have been developed including decision analysis, multi-attribute utility theory, and economic social cost-benefit analysis (Arrow, et al.,1996, Atkinson, et al.,2008, Clemen, et al.,2004). Of these methods, cost-benefit analysis has been widely applied in the assessment of the impacts of vehicle emissions, including transit buses (Levy, et al.,2010, USEPA,2011a) (Krupnick, et al.,1991, Stevens, et al.,2005).  21  1.4  Research Objectives and Scope  This dissertation is interdisciplinary in nature and strives to both be relevant to the real-world (i.e., transit agencies) as well as to contribute rigorous and original research to the academic community. The research in this dissertation draws from multiple fields including environmental science, public health and epidemiology, operations research, systems engineering, and integrated assessment, in an effort to gain a more comprehensive and holistic perspective of the challenges faced by public transit agencies and develop solutions that contribute to multiple objectives. To increase the real-world relevance of this research and engage decision makers, it was conducted in partnership with the transit agency 20 in Vancouver, Canada, TransLink, as well as GIRO Inc., a world leader in transit scheduling software. These partnerships allowed the research to capture the concerns and operational experiences of one of Canada's largest transit agencies and offered the potential to disseminate the benefits of the research, through GIRO, to other transit agencies.  Broadly stated, the objectives of this research were to develop tools and control strategies to aid public transit agencies in creating more sustainable public transportation systems. More specifically, the objectives were:  a) to explore, develop, and evaluate methods of quantifying the climate and public health impacts of the emissions from heavy-duty transit bus fleets (i.e., model the impact pathways), and b) to develop the first iteration of an integrated assessment model of the operations of transit bus fleets that accounts for climate, health, and operating costs and impacts and to employ it to quantify the potential of vehicle scheduling optimisation to mitigate these impacts and costs.  To develop models of the impact pathways, specific disciplinary methodologies and models were reviewed and linked together (Figure 1.3). In the process, particular attention was devoted to understanding the implications of scale and linkages between models. Although this dissertation touches on many aspects of the impact pathways, it was not possible to address all of them in detail. A detailed review and evaluation of methodologies and models associated with 20  Also referred to as authority.  22  the vehicle activity and emissions processes of the impact pathways was conducted. In order to manage scope, this dissertation draws heavily on previous work and the methodology developed by Greco et al. and a related project to quantify and model the dispersion and exposure processes (Appendix A) (Greco,2007, Greco, et al.,2007a, Greco, et al.,2007b).  Specific objectives  addressed in this dissertation related to developing a model of the impact pathways included:  a) evaluating Global Positioning System (GPS) -based methods of measuring vehicle activity (Chapter 3), b) identify vehicle activity factors that influence emissions and their spatial distribution (Chapter 3), c) reviewing and evaluating different emissions modeling approaches and specific models (Chapters 1-4), d) quantifying the spatial relationship between emissions and people and the implications of this relationship on exposure and health impacts (Chapter 2, Chapter 3), and e) exploring the implications of different spatial scales and levels of detail (Chapter 3).  Capital control strategies have traditionally dominated the set of alternatives considered by transit agencies to reduce the climate and health impacts of their bus fleets. However, operational strategies may also offer alternatives to reducing these impacts (Chapter 2). To quantify this potential, a first iteration of an integrated assessment model was developed by incorporating models of the impact pathways into a vehicle scheduling optimisation framework. By including climate, health and operating cost and impacts as objectives in the vehicle scheduling problem it was possible to identify scheduling solutions with co-benefits and to quantify the trade-offs between conflicting objectives.  In situations where trade-offs existed, social cost-benefit analysis was explored as a method of evaluating the trade-offs and identifying an optimal social cost solution. However, the application of cost-benefit analysis to public health and environmental decision making is not without controversy. It was not the intent of this research to wade into this debate. Cost-benefit analysis was explored as one possible decision making framework, one that because of its prevalence may be more amenable to transit agencies. Regardless of the framework employed or not employed, transit agencies are explicitly or implicitly making trade-offs between objectives that impact climate and health. Thus, despite the possible limitations of the 23  framework, it can provide useful insights into the decision making process and bring awareness to these trade-offs.  The scope of the integrated assessment model and optimisation was limited to operational decision making (Figure 1.5). It does not address the decision making and planning processes that lead to the construction and development of bus routes, timetables, and bus fleets. These were assumed to exist a priori. Thus the objective of this component of the research was to determine how a given bus fleet should be operated to minimise climate, health, and operating costs and impacts for a given fleet, timetable and set of routes.  National - Global  Transit Operations (Vehicle Scheduling)  Regional Organisational  Transit Planning (Routes, Timetables, Fleets)  Transit Transit Operations Operations  Urban Transportation Planning  Urban Planning and Air Quality Management Climate Change Agreements (Kyoto Protocol)  Figure 1.5 - Decision making context. Highlighted ring indicates the scope of this dissertation.  Throughout the dissertation efforts were made to quantify uncertainties and identify limitations and areas of limited knowledge in order to understand the value of additional information and knowledge. Further, the research in this dissertation was shaped by several key hypotheses listed in Table 1.4. These hypotheses guided and influenced the decisions of the specific research topics addressed in the dissertation, which are discussed in the following section.  24  Table 1.4 - Hypotheses. #  Hypothesis  Chapters  H1  Macro scale emissions modeling approaches, specifically those based on distance-based emission factors, are biased because they do not fully account for the effects of vehicle activity on emissions.  3  H2  Regional/macro scale assessments of exposure and health impacts, specifically those based on emissions inventories, are biased because they do not account for the intra-regional spatial relationship between emissions and populations.  2,3  H3  Operational control strategies such as vehicle scheduling optimisation can reduce the climate and health impacts of emissions from transit bus fleets as well as operating costs when incorporated in an integrated assessment framework.  2  H4  Models that have not been thoroughly evaluated may be biased.  4  25  1.5  Overview and Structure (Road-Map) of Dissertation  This dissertation follows the University of British Columbia’s single format thesis structure which merges the previous traditional monograph and manuscript formats. The dissertation is comprised of an introductory chapter followed by three research chapters and a concluding chapter. As previously discussed, versions of the research chapters were developed as manuscripts for publication in academic journals. Each research chapter includes an introduction, methods, and results and discussion section as specified by the formatting guidelines. Additional details describing the data and methods are provided in corresponding appendices for each research chapter.  The structure of the dissertation is framed around the climate and health impact pathways and scale (Figure 1.6). Progressively increasing levels of detail were addressed in the research chapters in order to gain insights into developing an integrated assessment model and decision making.  Impact Activity Pathways  Emissions  Radiative Forcing  Dispersion and Transformation Exposure  Scale  Macro Meso  Chapter 3  Intake  Dose  Climate Change & Impacts Health Impacts  Decision Making  Chapter 4  Chapter 2  Micro  Figure 1.6 - Dissertation overview and structure (road-map).  Chapter 1. The present introductory chapter provides the overall context, motivation and structure of the dissertation. It describes the research objectives and scope as well as the hypotheses examined in the dissertation. Further, it includes a literature review, emphasising those areas that are the primary focus of the dissertation.  Chapter 2. This chapter is based on a case study of the transit system in Vancouver, Canada. Meso scale models of the impact pathways are developed and incorporated into an integrated 26  assessment optimisation model. The distance travelled (i.e., vehicle activity) along each route was estimated using a Geographic Information System (GIS) and timetables obtained from TransLink. A basic emission factor model was developed using emissions test data collected by TransLink as well as chassis dynamometer data from the literature (Section 1.6.2). Exposure along each route was estimated using the intake fraction metric. The models were combined with indicators of the impacts to quantify the climate and health impacts of emissions from TransLink’s bus fleet. These impacts as well as operating costs were then incorporated as objectives in the vehicle scheduling problem. This optimisation model was used to quantify the implications of different optimisation objectives (e.g., minimising health impacts). Cost-benefit analysis was explored as a method of evaluating trade-offs between conflicting objectives. The results indicate that vehicle scheduling optimisation has the potential to reduce climate, health, and operating costs and impacts; however, the results are contingent upon heterogeneity in the emission factors of the buses as well as the exposure potential of the bus routes. This demonstrates the importance and value of intra-regional scale exposure and health assessments and the need for accurate characterisation of vehicle emissions, including uncertainties. Further, it provides the motivation for further study of these topics, which are addressed in Chapter 3 and Chapter 4.  Chapter 3. This chapter is based on a case-study of the 99 B-Line bus route in Vancouver, Canada, one of the busiest bus routes in Metro Vancouver. In it, a micro scale model of the activity-to-emissions impact pathway was developed in order to compare macro/meso scale modeling approaches to micro scale approaches (Figure 1.3). Second-by-second speed, acceleration, and position vehicle activity data were collected using a GPS receiver. The GPS data were evaluated against data collected from the buses speed sensor. Road grade was estimated using a digital elevation model. The relative importance of the different vehicle activity measures were quantified. The vehicle activity data was used as an input to a micro scale emissions model to predict emissions and their uncertainties in 50 m intervals along the entire bus route (i.e., the spatial distribution of emissions). The predicted emissions were compared to those from a macro scale emissions model (MOBILE). Further, the spatial distribution of the population, including pedestrians, was estimated along the route. The implications of the spatial relationship between emissions and the population on exposure were quantified. The results of this study indicate that macro scale modeling approaches underestimate exposure and total emissions due to poor characterization of the influence of 27  vehicle activity. Although it is likely not feasible to model an entire transit system at the level of detail described in this chapter, the micro scale analysis provides the insights necessary to make judgements related to the appropriate level of detail and to assess the value of information. In particular, this chapter highlights the need for an emissions model that better accounts for the effects of vehicle activity, such as the MOVES emissions model. However, the MOVES model was not considered in this study because it was still under development at the time.  Chapter 4. In this chapter, the MOVES2010a emissions model was evaluated. MOVES is a multi-scale, modal model that incorporates a micro scale modeling approach similar to the model employed in Chapter 3 (Section 1.6.2.3). The model was evaluated against a large dataset consisting of over 2500 individual emissions tests. For each test, the measured emissions were compared to the emissions predicted by the MOVES model. The results suggest that the model is biased and in general underestimates emissions from CNG buses and overestimates emissions from diesel buses. Further, the results provide an estimate of the uncertainty in the MOVES model.  Chapter 5. In the concluding chapter, the overall conclusions of the research are summarised, integrated and synthesised. Further, the strengths and limitations of the research, the applications of the research, specifically with respect to transit agencies, and possible future research are discussed.  28  1.6  Literature Review  This section reviews in detail methods of quantifying (measuring and modeling) the vehicle activity and emissions processes of the impact pathway (Figure 1.3). Specific methods and models are examined in further detail and evaluated in the research chapters (Chapter 3 and Chapter 4). An overview of methods of quantifying the air quality process (i.e., dispersion and transformation) is also provided. Other processes of the impact pathway are addressed as necessary in the individual research chapters.  Quantification can take two forms: empirical measurement or analytical models. Empirical measurement refers to physically measuring the phenomenon of interest whereas modeling refers to developing a mathematical or statistical description of the phenomenon, typically in the form of a computer model. In many cases it is not feasible, for example due to costs, to measure a phenomenon, especially on a continuous basis. In these situation models can be used to augment and generalise the available empirical data. Models are also required when future values are of interest, for example to evaluate the effectiveness of a control strategy. Thus models are used to both predict future values and past values in cases where empirical data is not available or limited.  There are several key emerging themes in the literature reviewed in this section that are shaping the trajectory of research in the fields. First, the representativeness of laboratory data has been questioned and in response there has been a renewed focus on real-world, in-use measurement methods and data. Second, there has been a move towards higher spatial and temporal resolutions in efforts to improve the accuracy of models. Third, many of the developments in the fields are being made possible by technological advances in computing, sensors, and information and communications technology (ICT).  1.6.1  Vehicle Activity  Vehicle activity is a generic term that refers to a wide range of measures (e.g., distance, time, speed and acceleration) used to describe a vehicle’s movement and operation in space and time. The most common measure of activity is distance, typically referred to as vehicle kilometers or miles travelled (VKT or VMT) (National Research Council,2000). Fuel consumption is also used (Dreher, et al.,1998, Singer, et al.,1996). Increasingly, speed, acceleration, and road grade 29  measurements sampled at 1.0 Hz (i.e., second-by-second data) or higher are being used to characterize vehicle activity (Barth, et al.,1996a, National Research Council,2000, Yoon, et al.,2005a). These measures are often used to characterise the mode of operation (e.g., idle, cruise, acceleration, and deceleration), which describes how a vehicle is operated over time and space (Barth, et al.,1996a).  The most commonly used methods of measuring vehicle activity are outlined in Table 1.5. Odometers provide a simple means of measuring distance and are standard equipment on all vehicles. Increasingly sophisticated engine control units (ECU) and sensors are being used in heavy-duty vehicles. These systems monitor and control a wide range of vehicle functions that influence emissions including vehicle speed and fuel consumption (Zhen, et al.,2009). Data from these systems are available at relatively high sampling frequencies (1-10 Hz). Fuel consumption is also typically tracked for accounting purposes and is available at various levels of aggregation, although not always at the vehicle level.  Attribution 2  Vehicle n1  Real-World  Table 1.5 – Vehicle activity measurement methods.  Resolution Method Variables Temporal Spatial  Odometer distance n/a 2 1 ≥ day  ECU3 speed, engine speed, temperature, load, etc. n/a 3 1 ≤ second  Fuel Consumption volume of fuel n/a 1 2 ≥ day  GPS4 position, time, speed 3 1 ≥ second ≤ 5.0 m  Drive Cycle time, speed n/a n/a 4 1 1 Qualitative ranking of the number of vehicles that can be feasibly sampled in a campaign where 1 indicates the method associated with the highest number of vehicles and 4 indicates the lowest number of vehicles. 2 Qualitative ranking of the ability to attribute the activity measured to a specific vehicle where 1 indicates perfect correspondence. 3 Engine Control Unit. 4 Global Positioning System  GPS systems are increasingly used to measure and estimate vehicle activity including speed, acceleration, and road grade (Bae, et al.,2001, Davis, et al.,2005, Farrell, et al.,1999, Yoon, et al.,2005a). Over 60% of buses in the United States are now equipped with GPS or Automatic Vehicle Location (AVL) systems (APTA,2011). Modern GPS receivers are able to provide accuracies of ± 5 m and ±1 m·s-1 at 1.0 Hz sampling frequencies under optimal, steady state conditions. However, real-world conditions have been reported to degrade accuracy and performance, particularly in urban areas (Jackson, et al.,2005, Yoon, et al.,2005a). For example, errors in velocity have been found to be greatest at low speeds and during acceleration (Jackson, 30  et al.,2005, Yoon, et al.,2005a). Critically, GPS measurements are the only method that also measures position which is essential for assessing health impacts.  In laboratory emissions testing, vehicle activity is typically specified in the form of a chassis drive cycle 21, which defines the vehicle speed as a function of time. For regulatory certification testing of heavy-duty vehicles, engines are tested independently (CFR,2012). For these tests, engine drive cycles that define the engine speed and torque as a function of time are used (DieselNet,2007b). Drive cycles are not so much a measurement method as a sample of vehicle activity meant to capture real-world driving conditions. However, their representativeness has been called into question and identified as a source of bias in emissions inventories and models (National Research Council,2000). Numerous test cycles have been developed specifically for heavy-duty vehicles and transit buses (Gautam, et al.,2002) (Appendix C). The Central Business District (CBD) cycle is widely used for chassis dynamometer emissions tests of transit buses.  In addition to vehicle activity, accurate identification of vehicles and characterization of vehicle fleets is necessary (National Research Council,2000). This can be a significant challenge and a source of uncertainty when considering large, diverse vehicle fleets, such as when estimating total emissions from a region or nation; however, in the context of analyzing transit fleets, the vehicles are known.  Because transit agencies operate relatively small and known fleets of vehicles, for the purposes of estimating past activity, it is in many cases feasible to directly measure vehicle activity rather than model it. However, for the purposes of predicting future activity and considering different control strategies, activity must be modeled. In the case of distance, GIS that include the geographical layout of roads and bus routes combined with timetables can be used to estimate vehicle activity and provides a powerful computational modeling framework that incorporates position information (Bachman, et al.,2000). GIS can also be used in conjunction with Digital Elevation Models (DEM) to estimate road grade. Modelling other vehicle activity measures such as speed and acceleration at high temporal resolution requires more complex simulation models such as VISSIM and was beyond the scope of this dissertation (Ishaque, et al.,2008).  21  Also referred to as a drive schedule.  31  1.6.2  Emissions  This section reviews emission measurement methods, factors affecting vehicle emissions, and emissions models. The review primarily focuses on North American models and research of primary exhaust emissions from heavy-duty vehicles and transit buses. Those emissions models that specifically support transit buses, or that had the potential to, were reviewed in greater detail in order to determine what model would be most appropriate in the context of modeling the climate and health impacts of transit bus emissions. Given the large and diverse body of (English language) literature from both North America and Europe, there are surprisingly few previous reviews to draw from (Esteves-Booth, et al.,2002, Yanowitz, et al.,2002).  1.6.2.1  Emissions Measurement and Sampling Methods  Several methods have been developed to measure and sample vehicle exhaust emissions (Heywood,1988, Ropkins, et al.,2009, Shorter, et al.,2005, Stone,1992, Yanowitz, et al.,2000). Measurement refers to the technique or device used to physically identify and quantify a pollutant, whereas sampling refers to how these techniques and devices are employed.  Gaseous pollutants are typically measured in terms of concentrations (e.g., ppm), which can be combined with the total volume of emissions produced and dilution ratio to estimate the total mass of emissions or the exhaust flow rate and dilution ratio to estimate the mass of emissions per unit time. These measurements are commonly made using optical techniques such as nondispersive infra-red (NDIR) and chemiluminescence, although in the case of HC, flame ionization techniques are used (Heywood,1988, Stone,1992). Methods of measuring gaseous pollutants are relatively mature and support sampling rates in the range of 1.0 Hz.  Measurement of PM, particularly at high sampling rates, is more challenging and methods are less mature (Abdul-Khalek, et al.,1998, Athanasios, et al.,2004). Several techniques and devices have been developed to measure the various properties of PM (Abdul-Khalek, et al.,1998, Kittelson, et al.,2002, McMurry,2000, Moosmuller, et al.,2001). Gravimetric methods in which particles are captured on a heated filter and weighed are the most common, but provide only low temporal resolution estimates (e.g., total emissions produced over a drive cycle) (Heywood,1988). More recently, several instruments including the Tapered Element Oscillating Microbalance (TEOM), DustTrak nephelometer, Scanning Mobility Particle Sizer (SMPS), and 32  Engine Exhaust Particle Sizer (EEPS) have been developed to measure PM at higher sampling rates (~1.0 Hz); however data from these instruments is limited (Kittelson, et al.,2006, Moosmuller, et al.,2001, Seigneur,2009).  The increased temporal resolution of emissions measurement has been an important achievement and has made it possible to quantify transient vehicle emissions. However, processing these high resolution measurements present new challenges. At higher resolutions (~1.0 Hz) the instrument response time and time alignment of the vehicle activity data and emissions measurement data is critical and can have a significant impact on results. Several studies have examined this problem and proposed various solutions (Kamarianakis, et al.,2010, Ramamurthy, et al.,1998, Weilenmann, et al.,2003, Zhang, et al.,2008). Despite some limitations, the most common approaches are to: (a) visually align the signals based on a common event (e.g., an acceleration event that is detectable across signals) or (b) shift the signals such that their correlation coefficient is maximised (e.g., maximise the correlation between CO 2 and power) (North, et al.,2006, USEPA, et al.,2002).  There are six principal sampling methods used to study vehicle emissions, each with certain advantages and disadvantages (Table 1.6). For example, there is often a trade-off between the number of vehicles that can feasibly be sampled in a campaign and the ability to attribute a measurement to a specific vehicle or type of activity. Further, the sampling method effects what measurement methods can be employed.  Attribution 2  Vehicle n1  Real-World  Table 1.6 - Emission sampling methods.  Sampling Method Studies  Dynamometer (Chassis & Engine) 4 1 (Prucz, et al.,2001, Ramamurthy, et al.,1999)  On-Board 5 1 (Younglove, et al.,2005, Zhai, et al.,2008)  Chase 3 2 (Shorter, et al.,2005)  Remote Sensing 2 2 (Bishop, et al.,2001, Burgard, et al.,2006)  Tunnel 1 3 (Rogak, et al.,1998)  Inverse Dispersion 2 3 1 Qualitative ranking of the number of vehicles that can be realistically sampled in a campaign where 1 indicates the method associated with the highest number of vehicles and 5 indicates the lowest number of vehicles. 2 Qualitative ranking of the ability to attribute the emissions measured to a specific vehicle where 1 indicates perfect correspondence.  33  Chassis and engine dynamometer studies have been the primary source of emissions data for emissions modeling and regulatory certification purposes. These studies involve measuring vehicle emissions over a drive cycle. Engine dynamometer tests based on the Federal Test Procedure (FTP) are used to certify heavy-duty engines (CFR,2012, Ramamurthy, et al.,1999). Results of these tests are reported in terms of mass of emissions per brake horsepower hour (g⋅bhp-1⋅h-1). For emissions modeling purposes the results are often converted to mass emissions per unit distance (i.e., distance-based emission factors in units of g⋅km-1) based on the brake specific fuel consumption (L⋅bhp-1⋅h-1) and average fuel economy (L⋅km-1) (Clark, et al.,2003). Chassis dynamometer studies are commonly reported directly in terms of mass of emissions per unit distance (g⋅km-1). Due to concerns regarding the real-world representativeness of the results of these studies, particularly engine dynamometer studies, several other sampling methods have been developed and employed (Clark, et al.,2003).  The use of on-board or Portable Emission Measurement Systems (PEMS) has increased significantly in recent years. These systems can be installed on vehicles and directly measure exhaust emissions as vehicles operate in the real-world. By either measuring or estimating exhaust flow these systems are able to estimate total emissions.  In chase studies, the measurement equipment is installed on a secondary vehicle which follows the vehicle of interest and samples its exhaust plume. Chase studies are better able to capture the real-world effects of environmental conditions such dilution and temperature, but are only able to measure pollutant concentrations, not total emissions, as the volume of exhaust is not known. Thus, chase study results are typically expressed in terms of mass of emissions per mass of CO 2 (g⋅g-1 CO 2 ), or fuel consumption as a fuel-based emissions factor (g⋅kg-1 fuel).  In a remote sensing study, the measurement equipment is setup at a specific location near a roadway. As vehicles pass by, the equipment measures the concentration of pollutants in the exhaust plume. Like chase studies, the results of remote sensing studies are expressed in terms relative to CO 2 or fuel consumption. A limitation of remote sensing studies is that they only sample a very limited range of vehicle activity.  Both tunnel and inverse dispersion sampling methods involve measuring the aggregate emissions of all vehicles passing through a tunnel or along a road. In tunnel studies, the concentration of 34  pollutants as well as the volume of air entering and exiting the tunnel are measured and used to estimate the total mass of pollutants emitted by the vehicles in the tunnel. In inverse dispersion studies, a rearrangement of the Gaussian line source dispersion model equation is used to estimate on-road emissions based on road-side measurements of the pollutant concentrations. A limitation of these sampling methods is that it is difficult to attribute the measurements to specific vehicles or vehicle types; however, they better capture the effects of environmental conditions.  1.6.2.2  Factors Affecting Heavy-Duty Vehicle Emissions  A wide range of factors varying over multiple temporal and spatial scales have been identified as contributing to both the inter- and intra- vehicle variability in heavy-duty vehicle emissions, which can range from 10s of percent to over an order of magnitude (Clark, et al.,2002, Heywood,1988, Maricq,2007, Tree, et al.,2007, Yanowitz, et al.,2000). These factors can be grouped into three principal categories: vehicle and energy source, environment, and vehicle activity (Table 1.7). These categories also generally correspond to the scale over which these factors vary. In general, vehicle activity factors vary at higher rates than environmental factors, which vary at higher rates than vehicle and energy source factors. For example, vehicle speed varies at a higher rate than ambient temperature, which varies at a higher rate than technological change. Thus, the variability in vehicle emissions can be thought of as being driven by the superposition of these (fast and slow) factors. It is possible to examine factors related to engine operation and combustion at even higher resolution (Frey, et al.,2010), but it is typically not practical to apply this level of resolution to analyses of vehicle fleets. In addition to the primary factors outlined in Table 1.7, a wide range of secondary factors (e.g., facility 22 type, traffic volumes, level of congestion) have been identified and are used in some emissions models. The focus here, however, is on primary factors.  Determining the relative importance of the factors is difficult as it depends on both scale and context. For example, grade would not be important in regions where the terrain is flat. Of the few studies that have attempted to rank factors, vehicle activity factors, specifically mode, typically rank as the most important (Clark, et al.,2002).  22  Road.  35  Table 1.7 – Factors affecting heavy-duty vehicle emissions. Factors Slow  Category  Scale Temporal Spatial n/a ≥ years – decades  vehicle class, mass1, model year, age2, mileage, engine, powertrain, aftertreatment, fuel  Environment  ambient temperature, humidity, altitude  ≥ hours  ≥ meso  Vehicle Activity  distance, time, speed, acceleration, road grade, power, mode, accessory loads, mass of load3  ≥ seconds  n/a  Fast  Vehicle and Energy Source  1  Unloaded mass of the vehicle. 2 Includes effects related to deterioration and maintenance. 3 Includes mass of passengers and/or freight.  Factors in the vehicle and energy source category are related to the physical and technological characteristics of a vehicle. They have been shaped by a range of regulatory and economic factors that have driven technological advancements in engine design (e.g., turbocharging, fuel injection, and exhaust gas recirculation) and aftertreatment devices. These advancements have led to significant reductions in HC, CO and PM emissions from heavy-duty vehicles, but not in NO X and CO 2 emissions or fuel economy (Burgard, et al.,2006, McGeehan, et al.,2005, Prucz, et al.,2001, Yanowitz, et al.,1999, Yanowitz, et al.,2000). However, new technologies such as hybrid-electric powertrains and DeNO X aftertreatment devices show promise in reducing CO 2 and NO X emissions as well as fuel consumption (Clark, et al.,2006, Johnson,2010, McKain, et al.,2000, Wayne, et al.,2004). Because many of the factors in the vehicle and energy source category are related to technological change driven by regulatory emission standards, model year is commonly used as an explanatory variable for these factors. The use of aftertreatment devices on heavy-duty vehicles 23 has only become common in the last decade as engine manufactures have historically been able to meet stricter emissions standards through advances in engine design (Clark, et al.,2002, McGeehan, et al.,2005, van Setten, et al.,2001). However, it is now widely accepted that emissions reductions through engine design have reached their limit and that aftertreatment devices will be required to meet future emissions standards (Herner, et al.,2009, McGeehan, et al.,2005). As a result, aftertreatment devices including Oxidation Catalysts (OxCat), Diesel Particulate Filters (DPFs) and DeNO X devices have been developed to control emissions of HC and CO, PM, and NO X respectively (Johnson,2010).  23  Three-way catalysts commonly employed on light-duty gasoline vehicles do not function in the oxidizing conditions of exhaust from the lean-burn engines used in heavy-duty vehicles.  36  Oxidation catalysts and Diesel Oxidation Catalysts (DOC) are used in both CNG and diesel engine applications to control CO and HC emissions. Oxidation catalysts have been shown to reduce CO emissions by 45-93%, HC emissions by 50-93%, and PM emissions by 24-60% (USEPA,1999). However, in CNG engine applications, CH 4 emissions can make up a significant fraction of HC emissions and are difficult to oxidize in conventional catalysts at normal exhaust temperatures (Cho, et al.,2007, McTaggart-Cowan, et al.,2006).  Diesel particulate filters and traps filter and collect PM from the exhaust (van Setten, et al.,2001). The collected PM is then oxidized through either a passive or active regeneration process (Clark, et al.,2002, van Setten, et al.,2001). In-use studies have shown DPFs can reduce PM mass emissions by 95-99% (Herner, et al.,2009, Shorter, et al.,2005), but some studies have also shown they may increase the particle number emissions (Grose, et al.,2006). Further, DPFs may incur a fuel economy penalty as a result of increased back pressure due to filter clogging and typically require the use of low sulphur diesel fuel (Herner, et al.,2009).  DeNO X devices are used to reduce NO X emissions. While this is a challenge in lean-burn engine applications, DeNO X devices are considered key in meeting increasingly stringent heavy duty emissions standards (Clark, et al.,2002, McGeehan, et al.,2005). Several different DeNO X technologies have been developed, including the Selective Reduction Catalyst (SCR), which currently appears to be the leading DeNO X technology (Johnson,2010). Studies have reported that SCR devices reduce NO X emissions by 76-90% (Conway, et al.,2005, McGeehan, et al.,2005); however, they have not been found to be effective at low exhaust gas temperature (< ~ 200 C), which occur at idle and low power demand (Herner, et al.,2009, McGeehan, et al.,2005).  Fuel type also has a significant impact on emissions from heavy-duty vehicles. In general, PM emissions from diesel vehicles without PM aftertreatment devices are substantially greater (by an order of magnitude) than from CNG vehicles (Ayala, et al.,2002, Clark, et al.,1999a, Hesterberg, et al.,2008). However, with aftertreatment devices, PM emissions from diesel and CNG vehicles are comparable. CNG vehicles have been shown to have moderately lower NO X emissions, but are typically less efficient than diesel vehicles, and therefore consume more fuel on a per unit energy basis. However, because CNG has a higher hydrogen-carbon ratio than diesel (3.6-4.0 vs. 1.8), CO 2 emissions per unit fuel energy are lower (Cho, et al.,2007, McTaggart-Cowan, et al.,2006). Thus, despite being less efficient, CO 2 emissions from CNG vehicles are typically 37  lower than from diesel vehicles. Biodiesel has been shown to reduce emissions of PM, CO, and HC (Clark, et al.,2002, Frey, et al.,2006, Graboski, et al.,1996).  Environmental factors such as altitude, temperature and humidity change the composition of the intake air and affect the combustion chemistry. For example, increased humidity reduces NO X emissions (Yanowitz, et al.,2000). As a result, when reporting NO X emissions, the values are commonly corrected for environmental conditions (CFR,2012). At higher altitudes the concentration of oxygen is decreased, which may increase emissions of HC, CO and PM (Yanowitz, et al.,2000).  Vehicle activity factors such as the distance travelled or the time in operation have obvious implications for emissions. However, higher order vehicle dynamics such speed and acceleration (i.e., mode) also significantly affect many pollutants (Clark, et al.,2002). Over the past two decades, significant efforts have been made to identify variables to explain these effects (Barth, et al.,1996b, Clark, et al.,2003, Jiménez,1999, North, et al.,2006, Ramamurthy, et al.,1999, USEPA,2002b, Yoon, et al.,2005a, Zhai, et al.,2008). Engine power demand and other surrogates have emerged as some of the most common explanatory variables used to account for these effects.  Fuel consumption and CO 2 emissions have been shown to be strongly correlated with power (Ross,1994, 1997, Zhai, et al.,2008). Emissions of NO X have also been found to be correlated with power but less strongly than CO 2 emissions. Further, in some studies, variability in injection timing strategies has been shown to distort the relationship between NO X and power (Ramamurthy, et al.,1999, Yanowitz, et al.,2002). PM emissions have shown some correlation with power, but appear to be particularly sensitive to transients and rapid changes in power (Ericson, et al.,2005, Gajendran, et al.,2003, Rakopoulos, et al.,2009, Wang, et al.,2006, Zhai, et al.,2008). In addition, the composition and size distributions of PM may vary with power and mode (Khalek, et al.,2003, Liu, et al.,2007, Ristovski, et al.,2006, Vaaraslahti, et al.,2004).  38  Power from a vehicle’s engine is required to overcome inertia (i.e., increase kinetic energy) and the resistive forces acting on the vehicle including gravity (i.e., increase potential energy), rolling resistance, and drag (Ross,1997). This can be expressed as 𝑃𝑡𝑟𝑎𝑐𝑡 = 𝑃𝑖𝑛𝑒𝑟𝑡𝑖𝑎 + 𝑃𝑔𝑟𝑎𝑑𝑒 + 𝑃𝑟𝑜𝑙𝑙 + 𝑃𝑑𝑟𝑎𝑔 𝑃𝑡𝑟𝑎𝑐𝑡 = 𝑚𝑎𝑣 + 𝑚𝑔 sin(𝜃)𝑣 + 𝑚𝑔𝑣𝐶𝑅 + 1�2 𝐶𝑑 𝜌𝑎 𝐴𝑣 3  Equation 1.2  Equation 1.3  where P tract is the tractive power 24 at the wheels (W); v is the velocity (m·s-1); a is the acceleration (m·s-2); θ is the grade (degrees); g is the acceleration due to gravity (m·s-2); C R is the rolling resistance (m); C d is the aerodynamic drag (dimensionless); ρ a is the air density (kg·m-3); A is the frontal area (m2); and m is the vehicle mass (kg) (Barth, et al.,1996a). Further, power is required to overcome resistive losses associated with inefficiencies in the powertrain and operate accessories (e.g., air conditioners). Thus, engine power demand can be expressed as 𝑃=  𝑃𝑡𝑟𝑎𝑐𝑡 + 𝑃𝑎𝑐𝑐 𝜖  Equation 1.4  where P is the engine power demand (W); ε is the powertrain efficiency; and P acc is the accessory power (W) (Barth, et al.,1996a). Tractive power is often used as a surrogate for engine power demand. Another common surrogate is Vehicle Specific Power (VSP), which is defined as 𝑉𝑆𝑃 =  𝑃𝑡𝑟𝑎𝑐𝑡 𝑚  Equation 1.5  where VSP is the vehicle specific power (W·kg-1); P tract is the tractive power (W); and m is the mass of the vehicle (Jiménez,1999).  24  Also referred to as axle power.  39  1.6.2.3  Emissions Modeling  Due to the cost and complexity of measuring vehicle emissions, continuous measurement is currently not feasible and it is only practical to conduct small sampling campaigns on a limited number of vehicles. As a result, emissions models are needed to generalise the available empirical data to predict past as well as future emissions.  To date, attempts to model vehicle emissions have been only modestly successful and even stateof-the-art models are associated with significant uncertainties (National Research Council,2000, Smit, et al.,2010) (Chapter 4). For example, a comprehensive review by Smit et al. found that in general, prediction errors were within a factor of 1.3 of the measured values for CO 2 , a factor of 2 for HC and NO X , and a factor of 3 for CO and PM (Smit, et al.,2010). These challenges stem from the wide range of factors that affect vehicle emissions, spatial and temporal scales, and model applications as well as the limited availability of data (National Research Council,2000). The process of developing a good emissions model is not trivial and involves: (a) determining an appropriate scale and level of detail; (b) identifying a comprehensive set of explanatory variables that explain a significant portion of the variability; and (c) obtaining a representative data sample. Such processes can be costly and time consuming, as reflected by the USEPAs recent experience developing the MOVES emissions model, which took nearly a decade (CFR,2010, USEPA,2001).  Emissions models are typically grouped into one or more of the following scales: macro, meso, and micro (National Research Council,2000). Micro scale models typically predict second-bysecond emissions of specific vehicles or vehicle categories. This level of detail is not feasible when estimating regional and national emission inventories on a macro scale. Averaging and/or aggregation 25 techniques are necessary to make modeling feasible at this scale. Further simplifications may be required because data describing the explanatory variables (e.g., vehicle activity data) is often limited or only available at lower resolution. Meso scale models are typically employed to predict emissions along transportation corridors and roadways (i.e., links) within a region, thus averaging and aggregation is performed at the roadway level.  25  A common aggregation approach is to express the explanatory variables as probability distributions.  40  Most emissions models incorporate or make assumptions regarding at least one explanatory variable from each category in Table 1.7. Approaches used to account for slow varying factors in the Vehicle and Energy Source and Environment categories are relatively similar between models but do depend somewhat on scale. There are a variety of vehicle classification schemes; however, vehicles are typically stratified based on fuel type, vehicle class (e.g., light-duty and heavy-duty), and model year. Some emissions models support only specific vehicle classifications (e.g., light-duty gasoline). Environmental factors are either stratified (e.g., low and high altitude) or modeled as an additive or multiplicative correction factor.  The resolution and way in which fast varying factors in the Vehicle Activity category are modeled differs the most between models. Physical, regression and binning based approaches are the most commonly used methods to model the relationship between vehicle activity and emissions 26 (Barth, et al.,1996a, Clark, et al.,2003, National Research Council,2000, Rakha, et al.,2004). Physical based approaches attempt to describe mathematically the causal series of processes that lead to emissions 27. An advantage of this approach is that it provides insights into why emissions occur, opposed to the purely descriptive characterisation offered by statistical approaches such as regression and binning. Binning based approaches produce a multidimensional lookup table or matrix indexed by the explanatory variables, for example speed and acceleration (Grieshop, et al.,2012, Yoon, et al.,2005a). In most cases the bins can be interpreted as representing modes. Regression based approaches express emissions as a mathematical function of the explanatory variables. In practice, it is not uncommon for more than one of these approaches to be applied.  Thus the general, overall modeling approach involves stratifying vehicles into groups, describing the relationship between vehicle activity and vehicle emissions on a scale appropriate to the application, and developing additive or multiplicative correction factors to account for other significant explanatory variables such as environmental factors. This results in the general model form for exhaust emissions: 𝐸𝑐,𝛼 = 𝐴𝛼 × 𝐸𝑋𝛼,𝑐 (𝑥𝑖 ) +�× 𝐶𝐹𝑘 (𝑥𝑘 ) 26 27  Equation 1.6  For a specific vehicle and environmental conditions stratification. Similar to the overall modeling approach of the impact pathway.  41  where E is the total emissions of pollutant c for vehicle classification α; A is the primary vehicle activity explanatory variable (e.g. distance, fuel, or time) for vehicle classification α; EX is the base emission factor (EF) or emission rate (ER) that is a function 28 of the secondary vehicle activity explanatory variables x i (e.g., average speed, speed, power, mode); and CF are the additive or multiplicative correction factors for explanatory variable x k (e.g., humidity). A diverse range of emissions models have been developed for a variety of research, regulatory and air quality management applications (Table 1.8). Few models support heavy-duty transit buses (Table 1.9). The taxonomy and terminology used in the literature to describe emissions models and the different modeling approaches are diverse and in some cases inconsistent (Boulter, et al.,2007, Esteves-Booth, et al.,2002, Smit, et al.,2010). Traditionally, models have been grouped into one of three general categories: emission factor, average speed, and modal (Esteves-Booth, et al.,2002, Smit, et al.,2010). However, average speed models are in fact a variant of emission factor models. Further, the term modal has been applied inconsistently. The term has conventionally been used in reference to micro scale models; however, modal modeling approaches are now being applied on other scales. To address the inconsistency in the use of these terms, the categories were further partitioned. The list of models in Table 1.8 is not exhaustive and focuses on the most prevalent North American models (Yu, et al.,2009).  28  In some cases this is also modeled as a simple additive or multiplicative correction factor. For example MOBILE uses a speed correction factor.  42  Table 1.8 – Emissions models and categories. General Model Form1  Category  Emission Factor  Average Speed  Level of Detail  Basic  Average Activity  Modal  Multi scale Meso scale Micro scale4  𝐸𝐹𝛼,𝑐  𝐸𝐹𝛼,𝑐 = 𝐸𝐹𝛼,𝑐 (𝑣̅ ) +�× 𝐶𝐹𝑘 (𝑥𝑘 )  𝐸𝐹𝛼,𝑐 = 𝐸𝐹𝛼,𝑐 (𝑥̅𝑖 ) +�× 𝐶𝐹𝑘 (𝑥𝑘 )  𝐸𝛼,𝑐 = � 𝑡𝛼,𝑚 × 𝐸𝑅𝑚,𝛼,𝑐 +�× 𝐶𝐹𝑘 (𝑥𝑘 ) 𝑚  𝐸𝛼,𝑐 = � 𝐸𝑅𝑚(𝑡),𝛼,𝑐 +�× 𝐶𝐹𝑘 (𝑥𝑘 )  Scale2 (Application)  Models3  macro, meso  handbooks; emissions measurement data  macro, meso  MOBILE*a; EMFAC*b  meso, macro  IBIS*c  macro, meso, micro  MOVES***i  meso, macro  MEASUREh  micro, meso  CMEM**d; PERE**e; VT-Micro*f; EMIT*g  E is the total emissions of pollutant c for vehicle class α; EF is the fuel- or distance- based emission factor for vehicle class α and pollutant c; CF are the additive or multiplicative correction factors for explanatory variable x k (e.g., humidity); ¯v is the average speed of vehicle class α; x i are the secondary explanatory variables (e.g., average speed); t α,m is the time vehicle class α spent in mode m; and ER is the emission rate in mode m for vehicle class α and pollutant c. 2 Bold italics indicates intended scale of application. 3 Not all models listed; focus is on North American models. 4 Also referred to as instantaneous, microscale, microscopic, continuous, real-time. * Regression; ** Physical; *** Binning based model approach. a (USEPA,2003); b(CARB,2002); c(Wayne, et al.,2011b); d(Barth, et al.,1999); e(Giannelli, et al.,2005); f(Rakha, et al.,2004); g(Cappiello, et al.,2002); h(Bachman, et al.,2000, Randall Guensler,2005); i(USEPA,2009d) 1  Emission factor models do not estimate total emissions, but express emissions in terms relative to a primary measure of vehicle activity, typically distance or fuel (i.e. g⋅km-1 or g⋅kg-1). To estimate total emissions, emission factors are combined with the activity measures (Equation 1.1). Distance-based emission factors are commonly derived from chassis and engine dynamometer emissions tests over a specific driving cycle. Basic emission factor models provide a simple approach of estimating vehicle emissions. Both the average speed and average activity models are extensions of the basic distance-based emission factor model. These models attempt to account for vehicle activity effects using average speed and other average measures of vehicle activity.  Modal models were originally developed for micro scale applications and have since been adapted for meso and macro scale applications. As the name implies, these models attempt to account for vehicle activity effects by using mode as an explanatory variable. There are many different definitions of mode. For example, modes may be described on a continuous basis as mathematical functions or on a discrete basis as a lookup table (e.g., idle, cruise, acceleration, 43  and deceleration). In all cases, the basic approach is to estimate emissions rates expressed in terms of mass of emissions per unit time (g⋅s-1) for each mode from second-by-second emissions test data. At the meso and macro scale, total emissions are estimated by combining the emission rates of the modes with the time spent in the modes (i.e., the mode is described in terms of a probability distribution). At the micro scale, emission rates are combined with second-bysecond vehicle activity data describing the mode (e.g., speed, acceleration, and road grade) to estimate total second-by-second emissions.  The primary purported advantage of modal models over emission factor models is that they better account for the effects of vehicle activity on emissions, although few studies have empirically verified this (e.g., Chapter 3). The disadvantage is that they are more computationally complex and require high resolution or disaggregated (by mode) emissions test data as well as vehicle activity data which may be more difficult and costly to collect.  1.6.2.4  Emissions Models  This section reviews in greater detail emissions models that support or could feasibly be made to support transit buses (Table 1.9). These include two emission factor models, MOBILE and IBIS, as well as two modal models MOVES and CMEM.  Table 1.9 - Heavy-duty vehicle emissions models. Model  Vehicle Activity Explanatory Variables  Pollutants  Level of Detail  Data Source  Transit Bus Class  MOBILEa  distance, average speed  CO 2 , NO X , CO, HC, SO 2 , SO 4 , PM, NH 3  low  engine dyno2  Yes  MOVESb  distance, time1, power2 (speed, acceleration, grade)  energy, fuel, CO 2 , N 2 0, NO X , CO, HC, SO 2 , SO 4 , PM, CH 4  medium  on-board and chassis dyno2  Yes  IBISd  distance, average speed, percentage idle, stops per mile, standard deviation of speed, and kinetic intensity  fuel, CO 2 , CO, HC, NO X  medium  chassis dyno2  Yes  CMEMc  time, power (speed, acceleration, grade), accessory load  fuel, CO 2 , CO, HC, NO X  high  on-board  Possible  1  Time is estimate from average speed and distance. 2 MOVES uses Vehicle Specific Power (VSP) as a proxy for power for light-duty vehicles and Scaled Tractive Power (STP) as a proxy for power for heavy-duty vehicles. 2 Dynamometer. a(USEPA,2003); b(USEPA,2009d); c(Barth, et al.,2004); d(Wayne, et al.,2011b)  44  1.6.2.4.1  MOBILE  The MOBILE emissions model was first developed in 1978 by the USEPA (National Research Council,2000). It is widely used by consultants, academics, and government agencies in a diverse range of applications, including many that it was not originally intended for (National Research Council,2000, Yu, et al.,2009). Initially designed to estimate emissions inventories, it was also used for regulatory and transportation conformity purposes in the U.S. until 2010 when it was succeeded by the MOVES model (USEPA,2009d). The last version of the model, MOBILE6.2, was released in 2004 (USEPA,2003). A Canadian version, MOBILE6.2c, has also been developed (Taylor,2005).  MOBILE is an average speed emissions factor model. It estimates distance-based emission factors (g·mile-1) for both light-duty and heavy-duty diesel and CNG vehicles with conventional powertrains and has a specific class for transit buses. A limitation of MOBILE is that the model for heavy-duty vehicles is relatively simplistic, particularly with regards to its treatment of PM emissions (National Research Council,2000). With respect to heavy-duty vehicles, the model contains a database of base emission factors developed from engine dynamometer emissions tests collected for certification purposes (Section 1.6.2.1), which are combined with correction factors to account for average speed, altitude, vehicle age and mileage (i.e., deterioration). Average speed correction factors are only applied to NO X , HC, and CO emissions; other pollutants are not corrected.  Due to its regulatory and policy role, MOBILE has received intense scrutiny and many criticisms over its lifetime. These criticisms have included allegations of inadequate peer review, documentation, validation and evaluation, characterisation of uncertainties, characterisation of heavy-duty vehicle emissions, and characterisation of PM emissions (GOA,1997, National Research Council,2000). In addition, concerns have been raised with regards to the ability of average speed to adequately account for the effects of vehicle activity and mode. These deficiencies limit its usefulness with regards to estimating the impacts of heavy-duty transit bus emissions. However, its low level of detail means that MOBILE is less computationally intensive than other models.  45  1.6.2.4.2  Integrated Bus Information System (IBIS)  The Integrated Bus Information System (IBIS) emissions model is a relatively new meso scale emissions factor model still under development at West Virginia University (WVU) (Wayne, et al.,2011b). A related life-cycle cost model was also developed as part of the overall project (Clark, et al.,2009, Golub, et al.,2011). These models were specifically developed as tools to assist transit agencies in capital procurement decisions and are being made available through a website (ibis.wvu.edu).  Buses are classified by the powertrain type (conventional or hybrid), fuel type (diesel or CNG), and model year (1988-2010+). The effects of vehicle activity were modeled using a polynomial function developed from a regression of up to five metrics describing a driving pattering including: average speed, percentage idle, stops per mile, standard deviation of speed, and kinetic intensity (Clark, et al.,2010). These functions are used to estimate distance-based emission factors that reflect the effects of vehicle activity. The effects of grade, accessory loads (e.g., air conditioning) and environmental conditions are not currently accounted for in the model but are under development.  The IBIS model has the potential to provide transit agencies with a valuable and easy-to-use tool to model fleet emissions. However, as it was still under development it was not considered in this dissertation. As well, the current lack of support for PM emissions limits its usefulness with regards to estimating health impacts.  1.6.2.4.3  Comprehensive Modal Emissions Model (CMEM)  The Comprehensive Modal Emissions Model (CMEM) is a micro scale modal model developed using a physical modeling approach (An, et al.,1997, Barth, et al.,1996a, Barth, et al.,2004, Barth, et al.,2005). CMEM supports both light- and heavy- duty vehicles. The fundamental strategy employed by the model is to predict fuel consumption rates which are then related to exhaust emissions through a linear function. This is accomplished by modeling the physical phenomenon associated with vehicle operations and emissions using six modules (Figure 1.7). The primary vehicle activity inputs are second-by-second speed, acceleration, grade, and accessory use. An additional 6 operating parameters (e.g., wind speed and ambient temperature) are used to account for other factors. A further 31 static model parameters are used to 46  characterize the vehicle (e.g., number of gears). Although CMEM is a micro scale model (i.e., it predicts second-by-second emissions), it was intended to be applied at higher scales by employing averaging and/or aggregation techniques (Barth, et al.,2004). For example, the parameters could be modeled using statistical distributions.  Engine Speed  Input Operating Variables Power Demand Model Parameters  Fuel Rate  EngineOut Emissions  Exhaust Aftertreatmetn  Exhaust Emissions and Fuel Use  Engine Control Unit Operating Time  Figure 1.7 – Comprehensive Modal Emissions Model (CMEM) structure for heavy-duty vehicles after (Barth, et al.,2004).  The CMEM model provides a sophisticated and informative framework for modeling vehicle emissions. However, in the context of estimating the climate and health impacts of transit bus emissions, the current CMEM model has several limitations including: (a) it does not predict PM emissions, (b) it does not currently have a transit bus vehicle category, (c) it does not support hybrid powertrains and (d) it is has not been previously used to model emissions from CNG vehicles. The lack of support for transit buses and PM emissions is a significant drawback. If emission testing data was available, it would be possible to calibrate the model and develop a transit bus category. Unfortunately, the large number of parameters required to be calibrated would likely make this a significant undertaking. As a result, it was not considered in this dissertation.  47  1.6.2.4.4  Motor Vehicle Emissions Simulator (MOVES)  The Motor Vehicle Emissions Simulator (MOVES) model is the most recent mobile source emissions model developed by the USEPA (USEPA,2002c). MOVES is a multi-scale modal model that supports transit buses and a wide range of pollutants and emission processes (Appendix C). It was implemented using a modern software architecture and a relational database. The database contains emission rates (g⋅hr-1) for each vehicle classification and mode (USEPA,2005, 2009a). Modes are defined based on surrogates of power (e.g., VSP) and vehicle speed. To improve the prediction of real-world, in-use vehicle emissions, attempts were made to develop emission rates from on-board and chassis dynamometer emissions tests of in-use vehicles (USEPA,2009a). Further details regarding the MOVES model can be found in Chapter 4.  1.6.3  Air Quality (Dispersion and Transformation)  Air quality refers to the processes that describe the dispersion, transformation, fate, and ultimately the concentrations of pollutants resulting from emissions released into the atmosphere. It is feasible to directly measure ambient pollutant concentrations and in many urban areas, particularly in developed countries, mobile and stationary air quality monitoring networks are used to support air quality management efforts (Griffin,2006). However, the cost of the monitoring stations dictates the number that can be deployed and as a result the spatial resolution of the monitoring networks. Further, it can be difficult to attribute measured concentrations to specific sources, although statistical source apportionment techniques have been developed for this purpose (Watson, et al.,2008). Air quality models are often employed to augment and address limitations of air quality monitoring networks as well as to predict future pollutant concentrations in order to evaluate control measures.  A large number of air quality models have been developed for a range of applications and scales. A complete discussion of these models is beyond the scope of this dissertation; however, there is an important distinction between how different classes of pollutants are modeled, specifically with regard to the transformation process. Some pollutants are non-reactive or can reasonably be modeled as non-reactive, such as CO and primary PM 2.5 mass (Marshall, et al.,2005, Zhou, et al.,2007). These pollutants can be modeled independently of other pollutants and their 48  atmospheric concentrations are primarily governed by dispersion. For reactive pollutants such as NO X and HC and secondary pollutants such as ozone, physical and chemical atmospheric transformation processes must be considered and their atmospheric concentrations are governed by dispersion, transformation, and the concentration of other pollutants in the atmosphere. As a result, when modeling reactive pollutants other emission sources in a region must be considered. The principal focus of this dissertation was the impact of primary, non-reactive pollutants emitted from buses along bus routes. Line source dispersion models such as CALINE are commonly used for this application (Benson,1992, USEPA,2010b).  49  Chapter 2: Integrated Assessment: Minimising the Climate and Health Impacts of Emissions from Heavy-Duty Public Transportation Bus Fleets through Vehicle Scheduling 2.1  Introduction  Public transit systems are widely considered to have significant social and environmental benefits in comparison to private automobiles in the provision of passenger transportation services. However, the vehicles (primarily buses) that make up transit fleets are typically powered by heavy-duty internal combustion engines that emit pollutants which adversely impact both public health and the climate. While the social and environmental benefits of public transportation have been widely touted (APTA,2011, Kahn Ribeiro,2007), surprisingly few studies have investigated how to minimise the impacts of their operations by incorporating social and environmental objectives into the operational planning of public transportation systems (Figliozzi,2010, Li, et al.,2009a, Stasko, et al.,2010).  The majority of transit fleets in North America are powered by heavy-duty diesel or compressed natural gas (CNG) engines (APTA,2011). The combustion of these fuels produces a wide range of gaseous and particulate emissions that directly and/or indirectly adversely affect both human health and the climate (Health Effects Institute,2010, IPCC,2007b, Lloyd, et al.,2001). These impacts occur over multiple spatial and temporal scales through a series of processes characterized by the health and climate impact pathways (Fuglestvedt, et al.,2003, Smith,1993).  Carbon dioxide (CO 2 ), methane (CH 4 ) and nitrous oxide (N 2 O) emissions have been identified as greenhouse gases that affect the radiative balance of the earth’s atmosphere and contribute to global climate change (Forster,2007). The Global Warming Potential (GWP) indicator has traditionally been used to estimate the net climate impact of these long-lived, globally wellmixed compounds. More recently, short-lived compounds such as Black Carbon (BC), Organic Carbon (OC), Sulfate (SO 4 ), all constituents of PM, and ground level ozone have been identified as having potentially significant climate impacts (Bond, et al.,2005, Forster,2007, Grieshop, et al.,2009, Penner, et al.,2010, Ramanathan, et al.,2008). However, efforts to estimate the GWP of these compounds have been controversial because of their shorter life spans and greater uncertainties in their net impact on climate forcing (Bond,2007, Forster,2007, Reynolds, et al.,2008). 50  Collectively, vehicle emissions have been associated with short- and long- term respiratory and cardiovascular morbidity and mortality as well as adverse neurological and developmental health effects (Health Effects Institute,2010). Particulate Matter (PM), Nitrogen Oxides (NO X ), Carbon Monoxide (CO), Hydrocarbons (HC) and Air Toxics (AT) as well as secondary pollutants including ground level (tropospheric) ozone (O 3 ) are believed to play significant roles in causing adverse health effects (Health Effects Institute,2007, 2010, Pope, et al.,2006).  PM and NO X are typically the primary concerns when assessing the health impacts of emissions from heavy-duty vehicles, but other pollutants and synergistic effects cannot be ruled out. NO X is significant primarily as a precursor to ground level O 3 , whereas PM has direct health impacts. Significant focus has been placed upon PM by health and epidemiological researchers, suggesting it may be disproportionately responsible for the net health impact of heavy-duty diesel vehicle emissions (Dockery, et al.,1993, Lloyd, et al.,2001, Pope, et al.,1995, Pope, et al.,2006, USEPA,2009c). As a result, studies of the health impacts of transit bus emissions have also focused on PM emissions (Levy, et al.,2009, Stevens, et al.,2005, Tainio, et al.,2005).  Uncertainties in the health impact pathways make quantification and development of a comprehensive indicator of public health impact challenging. As a result, there is no wellestablished public health impact indicator analogous to GWP/GWC. Previous studies examining the health impacts of transit bus emissions have used Quality Adjusted Life Years (QALY) due to the intake of PM and O 3 (Cohen, et al.,2003), mortality due to the intake of PM 2.5 (Levy, et al.,2009, Tainio, et al.,2005) and total PM 2.5 and NO X emissions (Li, et al.,2009a, Stasko, et al.,2010) as health indicators. Disability Adjusted Life Years (DALY) is also commonly used (Smith, et al.,2008).  Public health and climate impacts have traditionally been analysed in independent frameworks. However, studies have shown that failing to employ an integrated assessment framework that combines consideration of health and climate impacts can result in detrimental and unintended outcomes (Mazzi, et al.,2007). As a result of these studies and wider acceptance of the role that short-lived compounds such as PM play in climate and health impacts, researchers are turning to integrated, multi-objective frameworks with the aim of identifying win-win solutions with co-  51  benefits across multiple impact dimensions (Grieshop, et al.,2011, Smith, et al.,2008, Woodcock, et al.,2009).  There is a fundamental difference between the climate and health impact pathways. Unlike climate impacts, health impacts result from exposure - the contact of pollutants and people in space and time. As a result, health impacts are more sensitive to the choice of scale. For example, studies have shown that regional scale analyses focusing on the relationship between ambient pollutant concentrations and health impacts underestimate exposure and health effects because they do not account for variability in the intra-regional spatial and temporal distribution of pollutants and people and the relationships between them (Greco, et al.,2007a, Health Effects Institute,2010, Jerrett, et al.,2005b, Nazaroff,2008). Thus, there is potential to reduce exposure and health impacts by altering the spatial distribution of bus emissions at the intra-regional scale.  The aim of this study was to evaluate the potential of operational (in contrast to capital) control strategies to reduce the public health and climate impacts of transit systems by including these impacts as well as operating costs as objectives in the vehicle scheduling problem (Ceder,2002). Specifically, this study explores how heterogeneity in the emissions of different bus technologies and the exposure potential along bus routes can be exploited to reduce impacts. Due to the complexity and computational challenges associated with solving these problems (Ceder,2011), only one dimension of the vehicle scheduling problem was considered, the assignment of vehicles to routes. The vehicle assignment problem provides a lower bound on the potential benefits of vehicle scheduling optimisation.  Several studies have explored various aspects of incorporating climate and health impacts as objectives in the vehicle scheduling problem (Li, et al.,2009a, Stasko, et al.,2010), but none have previously considered exposure, the impacts of emerging pollutants of concern such as BC, or the trade-offs between health, climate and operating costs in detail. To that end, the goals of this study were to develop comprehensive indicators of climate and public health impacts as well as estimates of operating costs in order to evaluate different vehicle assignment solutions with respect to these objectives (e.g., minimising health impacts) in a case study of a real-world transit system.  52  2.2  Materials and Methods  This analysis is based on a real-world case study of the transit system in Vancouver, Canada. Distance-based emissions factors, fuel consumption, and fuel and maintenance costs were estimated for diesel and CNG buses. Climate impacts of both long-lived and short-lived compounds were quantified using their estimated GWP. The total intake of PM 2.5 emissions, calculated using the estimated intake fraction of the bus routes, was used as the public health impact indicator. The estimated health, climate, and operating costs and impacts were incorporated into a vehicle assignment optimisation model and scenarios were developed to explore the implication of different optimisation objectives (e.g., minimising climate impacts).  2.2.1  Transit System  The South Coast British Columbia Transportation Authority, commonly known as TransLink, and its operating subsidiary Coast Mountain Bus Company (CMBC) operate a fleet of approximately 1300 buses on 150 routes throughout the Metro Vancouver area (Figure 2.4). The fleet consists predominately of diesel buses but also includes approximately 50 CNG buses and 240 electric trolley buses (not considered here because they produce no primary emissions) (Table 2.1 and Table A.1). The schedule for the Fall of 2009 and the corresponding geographic locations of the bus routes were obtained from TransLink.  2.2.2  Emissions Model and Operating Costs  Distance-based emissions factors and operating costs were developed for 9 bus categories (Table 2.1 and Table A.3). Bus categories were developed based on the bus size, the powertrain type, the emissions control technology, and the model year and associated emissions certification level (Table A.2). All emissions factors were based on the Central Business District (CBD) drive cycle and developed from emissions tests of actual buses from TransLink’s fleet (TransLink,2006) as well as published studies (Hayes, et al.,2006, Wayne, et al.,2011b). Operating costs were developed from studies by TransLink and Clark et al. (Table A.4) (Clark, et al.,2007b, M.J. Bradley & Associates,2006). Two bus fleets were considered (Table 2.1). Fleet A was representative of TransLink’s fleet in 2009 and Fleet B was a hypothetical fleet with more modern buses. Additional details are provided in the Appendix A.  53  Table 2.1 - Bus categories, operating costs, emission factors and Global Warming Commitment. Fleets (#) Costsa PM 2.5 NO X GWC 100 b Description -1 -1 -1 ($·km ) (g·km ) (g·km ) (gCO 2 e·km-1) A B 40DO 0.889 0.662 17 2516 95 0 40 ft Old (2-Stroke) Diesel OxCatb 40DB2 0.656 0.212 22 1732 294 0 40 ft Baseline Diesel with OxCat 40DB 0.594 0.109 12.3 1546 54 170 40 ft Baseline Diesel with OxCat 40DA 0.625 0.0244 6.83 1590 193 170 40 ft Advanced Diesel with DPFc 40DH 0.587 0.0125 5.2 1190 1 170 40 ft Hybrid Diesel with DPF 60DB 1.042 0.196 16.3 2984 76 37 60 ft Baseline Diesel with OxCat 60DA 1.017 0.0628 12.2 2850 10 37 60 ft Advanced Diesel with DPF 60DH 0.821 0.0125 8.97 1860 26 37 60 ft Hybrid Diesel with DPF 40CG 0.600 0.0168 12.4 1635 43 170 40 ft CNG with OxCat a Operating costs include fuel, propulsion-related maintenance, battery replacement for hybrids, facility maintenance, and electricity costs associated with compressing CNG (2007 U.S. dollars). b 100 year global warming commitment due to CO 2 , CH 4 , and PM 2.5 (i.e., BC and OC). c Oxidation Catalyst. d Diesel Particulate Filter. Category  2.2.3  Climate Impact Indicator  The global warming commitment (GWC) of the exhaust emissions resulting from the operation of the bus fleet over a period of one day (tCO 2 e·day-1); was used as the indicator of climate impact (Smith, et al.,2000). GWC approximates the net climate impact of a group of compounds and is equal to the emissions weighted sum of their global warming potentials (GWP). GWP is the time-integrated change in the radiative forcing of a unit impulse of a compound relative to that of CO 2 over a given time horizon (typically 20 or 100 years) (Forster,2007). GWPs for both long- and short- lived, gaseous and particulate compounds were taken from the literature (Table A.5) (Forster,2007, Grieshop, et al.,2011, Reynolds, et al.,2008). The GWC due to the emissions from a bus traveling one kilometer (Table 2.1) was estimated as 𝐺𝑊𝐶100,𝑏 = � 𝐸𝐹𝑏,𝑐 ∙ 𝐺𝑊𝑃100,𝑐  Equation 2.1  𝑐  where GWC is the global warming commitment of bus b over a 100 year time horizon (tCO 2 e·km-1); c are the climate forcing compounds CO 2 , CH 4 , BC, and OC; EF is the emission factor of bus category b and compound c (g·km-1); and GWP is the global warming potential for compound c over a 100 year time horizon (tCO 2 e). Climate impacts were valued at $25 per tCO 2 e based on the offset price set by the Pacific Carbon Trust, which is a crown corporation  54  that was created by the Province of British Columbia in an effort to establish a carbon price signal (Pacific Carbon Trust,2011).  2.2.4  Public Health Impact Indicator  The intake or mass of primary PM 2.5 exhaust emissions inhaled by the population within 5000 m of the bus routes and resulting from the operation of the bus fleet over a period of one day was used as the measure of public health impact (g·day-1). Total primary emissions of PM 2.5 and NO X as well as the premature mortality attributable to the intake of PM 2.5 were also considered. To calculate intake, the intake fraction (iF) was estimated along each bus route. The intake fraction is a measure of exposure potential and estimates the proportion of pollutants emitted on a route that are inhaled by the population around the route. Using an approach similar to that of Greco et al. (Greco, et al.,2007a), the average annual intake fraction was estimated for each route as 6  𝐶𝑧 𝑖𝐹𝑟 = � 𝑃𝑧 ∙ � � ∙ 𝑄 𝐸�  Equation 2.2  𝑧=1  where iF r is the intake fraction for route r (dimensionless); P z is the population within each zone z; C·Ê-1 is the average annual concentration to emissions ratio relating the concentration C in zone z to the unit on-route emission Ê (m-3); and Q is the population average breathing rate equal to 14.5 m3·day-1 (Layton,1993, Marty, et al.,2002). The average annual concentration to emissions ratio (C·Ê-1) was estimated using the CALINE4 line-source dispersion model (USEPA,1998) and one year of meteorological data. Additional details are provided in Appendix A. The ratio was estimated in six concentric zones at distances of 0-50 m, 50-100 m, 100-200 m, 200-500 m, 500-1000 m and 1000-5000 m from the routes in order to sample the concentration gradient (Greco, et al.,2007a) (Figure 3.1, Figure A.2, and Figure A.3). Concentrations were assumed to be constant in each zone and zero beyond 5000 m. The population in each zone was determined using ESRI ArcGIS 9.3 (ESRI,2009) and blocklevel census data from the 2006 Census of Canada (Statistics Canada,2006). Following Greco et al., dispersion processes were assumed to be constant across the region and therefore constant along and between routes on an annualized basis.  55  Residential population data for Metro Vancouver was obtained from the 2006 Census of Canada (Statistics Canada,2006) at dissemination block resolution (14815 blocks, 2116581 people in total). Within Canadian urban areas, each dissemination block typically corresponds to a single city block, and is the smallest census unit for which population counts are available (Statistics Canada,2007a, b, c). Using ArcGIS, the six zones were spatially intersected with the dissemination blocks and the total population within each zone was estimated as 𝑄  𝑃𝑧 = � 𝐷𝑞 × 𝐴𝑞,𝑧  Equation 2.3  𝑞=1  where P z is the population in zone z; q is the census dissemination block index; Q is the total number of census dissemination blocks; D q is the population density of dissemination block q (people⋅m-2); and A q,z is the fraction of the area of zone z that interests dissemination block c (m2), that is dissemination blocks falling partially within a given zone had total population prorated based on the block area contained within the zone.  The total intake of PM 2.5 emissions resulting from a bus traveling one kilometre on a route was estimated as 𝐼𝑟,𝑏 = 𝑖𝐹𝑟 ∙ 𝐸𝐹𝑏  Equation 2.4  where I r,b is the intake of PM 2.5 emission from a bus of category b on route r (g·km-1·day-1); iF r is the intake fraction of route r (dimensionless); and EF b is the PM 2.5 emission factor of bus category b (g·km-1). Note that the terms intake and exposure are used interchangeably in this study as a constant breathing rate was assumed.  To estimate the premature mortality attributable to PM 2.5 and monetise health impacts the methods and assumptions employed by Levy et al. and Stevens et al. were adopted (Levy, et al.,2010, Levy, et al.,2009, Stevens, et al.,2005). The premature morality attributable to the intake of PM 2.5 was estimated as ∆𝑀 =  𝐼 ∙ 𝐵𝑀𝑅 ∙ 𝐶𝑅 𝑄  Equation 2.5  where ∆M is the change in mortality (deaths·year-1); I is the intake of PM 2.5 (g·day-1); Q is the population average breathing rate equal to 14.5 m3·day-1; BMR is the annual background all56  cause mortality rate equal to 730 deaths per 100000 people (deaths·year-1) (Statistics Canada,2008); and CR is the concentration response function equal to 1.0% increase in all-cause mortality per µg·m-3 increase in the annual average PM 2.5 concentration (Levy, et al.,2010, Levy, et al.,2009). The value of a statistical life (VSL) was defined as $7.7 million (2007 U.S. dollars) (Levy, et al.,2010).  2.2.5  Vehicle Scheduling and Assignment Optimisation  The goal of vehicle scheduling optimisation is to find the series of trips made by a fleet of buses that minimises an objective function subject to a set of spatial (e.g., geographical layout of the routes), temporal (e.g., timetable), and operation and capital constraints (e.g., number of buses) determined in the planning process (Ceder,2002). The optimisation solutions typically consist of a set of blocks with start and end times and locations that define the series of trips made by a single physical bus. Trips occur over a single route; however, blocks may consist of trips on different routes.  In a traditional vehicle scheduling problem the objective is to minimise the number of buses (i.e., rolling stock capital) and operating inefficiencies (typically defined as the non-revenue service time or total distance travelled) (Ceder,2002). In this study, TransLink’s weekday vehicle scheduling solution for the Fall of 2009 was used. The solution fixes the value of the traditional objectives. However, within the solution there is some flexibility in terms of which bus category is assigned to which block and thus route. The effects of this flexibility on climate, health, and operating costs and impacts were explored using vehicle assignment optimisation. TransLink’s actual bus assignments were not considered.  The optimisation problem was formulated as a binary programming problem: 𝑚𝑖𝑛/𝑚𝑎𝑥 � Ψ𝑐 = � � � 𝑥𝑘,𝑏 ∙ Υ𝑟,𝑏,𝑐 ∙ 𝑑𝑟,𝑘 � 𝑥 ∈ {0,1} s.t.  � 𝑥𝑘,𝑏 = 1 ∀𝑘  𝑏  𝑘  Equation 2.6  𝑟  (one physical bus per block)  Equation 2.7  𝑏  57  � 𝑥𝑘,𝑏 ≤ 𝑛𝑏 ∀𝑏, ∀𝑡 = �𝑡𝑖=0…23/𝑤 = 𝑖 ∙ 𝑤�  (maximum number of buses)  Equation 2.8  𝑘∈𝐾𝑡  where Ψ is the objective function equal to the total intake of PM 2.5 (g·day-1), the total GWC (tCO 2 e ·day-1); or the total emissions of compound c (g·day-1); b is the bus category; k is the block; r is the route; Υ is either the intake (I), GWC or emission factor (EF) of bus category b and compound c on route r (g·km-1); x is the decision variable indicating if bus category b is assigned to block k; d is the total distance traveled on route r for block k; n is the total number of buses of category b in the fleet (Table 2.1); K is the set of blocks that are active between time t and t + w; and w is the time window equal to 0.5 hours. As blocks are designated for either 40 ft or 60 ft buses, the optimisation problem was solved independently for each bus size. Further, to reduce complexity, all buses were assumed to be dispatched from a single depot, although in actuality they are dispatched from seven depots. The optimisation problem was solved using IBM ILOG CPLEX V12.4. Note that for presentation purposes, in some cases the results have been expressed over a one year period by multiplying by 365.  Twelve scenarios (A-L) were developed to characterise the best and worst case solutions and the trade-offs between objectives for both bus Fleets A and B (Table 2.1). In each scenario a specific indicator (e.g., GWC) was minimised and the percent change in the other indicators relative to the scenario in which the indicator was minimised (i.e., the best case) were estimated (Table 2.2). The worst case scenarios were also estimated by maximizing each indicator in order to establish an upper bound.  2.3  Results and Discussion  The optimisation results show that vehicle assignment has a significant effect on exposure to PM 2.5 emissions (Table 2.2). For the worst case vehicle assignments, PM 2.5 intake was more than double (107-123%) that of the best case assignments. Vehicle assignments had a similar effect on total PM 2.5 emissions (96-108%) and a smaller but still significant effect on NO X emissions (35-28%). In relative terms, the effect on operating costs, fuel consumption and GWC (i.e., climate impacts) was much smaller and the difference between the best and worst case solutions ranged from 3-15%. However, in absolute terms, the effects may be significant to transit agencies. For example, in this case study, a 1% increase in operating cost would translate into an annual increase of approximately $0.5 million. 58  Although TransLink’s actual vehicle assignments were not considered (they are not typically included in vehicle scheduling solutions), it is unlikely that transit agencies operate their fleets at the extreme, worst case scenario. Therefore the real-world benefits would likely be smaller than expressed by the relative difference between the best and worst case scenario. However, these scenarios provide a useful characterization of the potential benefits that may be derived from vehicle assignment optimisation in reducing health, climate, and operating costs and impacts.  Table 2.2 - Vehicle assignment optimisation results.  #  Fleet  Scenarioa Objective (Minimise)  Percent increase relative to the best case (%) Costb  Fuel  GWCc  Costb 1.01 0.06 Best Fuel 0.33 0.16 Best GWCc 0.08 0.51 Best 3.02 4.06 3.62 A NO X PM 2.5 0.14 1.59 0.21 PM 2.5 Intaked 1.08 2.30 1.50 8.79 11.7 12 Worst (Maximise) G Costb 2.15 0.25 Best H Fuel 0.34 0.00 Best I GWCc 0.34 0.00 Best J 0.66 0.41 0.19 B NO X K PM 2.5 0.47 4.77 0.88 L PMI 2.5 Intaked 0.93 5.82 2.31 2.88 15.3 9.93 Worst (Maximise) a Each scenario (A-L) involved minimising the listed objective. d Health impacts and exposure. A B C D E F  NO X  PM 2.5  PM 2.5 Intaked (Exposure)  5.64 3.04 14.1 6.90 7.86 18.7 4.69 3.17 14.2 29.7 41.9 Best 4.47 12.5 Best 9.63 12.9 Best 35 108 123 9.89 41.3 47.1 4.29 43.2 48.9 4.29 43.2 48.9 25.7 34.1 Best 4.47 10.3 Best 6.06 9.99 Best 28.1 96.2 107 b Operating costs.c Climate impacts.  Operational optimisation strategies fundamentally rely on heterogeneity. In the case of the vehicle scheduling and assignment problem, variability must exist in the bus category emission factors and either the total distance all buses in a category travel or the intake fraction (i.e., exposure potential) of the bus routes for any change in impacts or costs to be possible. For example, if a bus fleet was composed of a single bus category, impacts and costs would be constant regardless of changes in vehicle assignments.  Visualisations of the vehicle assignment solutions are shown in Figure 2.1 and Figure 2.2. These figures show graphically how different optimisation objectives and the emissions factors, distance travelled, and intake fractions impact the assignment of bus categories to blocks (e.g. routes). For example, a comparison of Figure 2.1b and Figure 2.1d shows that the routes served by 40CG and 40DB buses were swapped in order to meet the two different objectives (i.e., 59  climate and health). The distinct bands and clustering show how the blocks are prioritised with respect to the objectives. The shift from vertical bands to angled bands in the case of optimising for PM 2.5 intake demonstrates graphically why the intake fractions impact the optimisation results. a)  b)  Minimise Cost (Scenario A)  40CG 40DA  Minimise GWC (Scenario C)  45  40DB  PM2.5 Intake Fraction (×10-6)  40  40DB2 40DH  35  40DO  30  60DA  25  60DB 60DH  20 15 10 5 0  0  100  200  300  400  500  600 0  100  Distance (km ⋅ day -1)  c)  200  300  400  500  600  500  600  Distance (km ⋅ day -1)  d)  Minimise PM2.5 Intake (Scenario F)  Minimise PM2.5 (Scenario E)  45  PM2.5 Intake Fraction (×10-6)  40 35 30 25 20 15 10 5 0  0  100  200  300  400  Distance (km ⋅ day ) -1  500  600 0  100  200  300  400  Distance (km ⋅ day ) -1  Figure 2.1 – Scatter plots of the total distance traveled on a block and the intake fraction of the block. Each point represents a specific block and is color coded by bus category. The scatter plots show the bus-block assignments for Fleet A and the scenarios that minimised operating costs (a), climate impacts (b), health impacts (c), and total emissions of PM 2.5 (d).  60  a)  b)  Minimise Cost (Scenario G)  Minimise. GWC (Scenario I)  45 40CG 40DA  PM2.5 Intake Fraction (×10-6)  40  40DB  35  40DH  30  60DA  25  60DB 60DH  20 15 10 5 0  0  100  200  300  400  500  600 0  100  Distance (km ⋅ day -1)  c)  200  300  400  500  600  500  600  Distance (km ⋅ day -1)  d)  Minimise PM2.5 Intake (Scenario L)  Minimise PM2.5 (Scenario K)  45  PM2.5 Intake Fraction (×10-6)  40 35 30 25 20 15 10 5 0  0  100  200  300  400  500  600 0  Distance (km ⋅ day -1)  100  200  300  400  Distance (km ⋅ day -1)  Figure 2.2 - Scatter plots of the total distance traveled on a block and the intake fraction of the block. Each point represents a specific block and is color coded by bus category. The scatter plots show the bus-block assignments for Fleet B and the scenarios that minimised operating costs (a), climate impacts (b), health impacts (c), and total emissions of PM 2.5 (d).  2.3.1  Indicators and Optimisation Objectives  The climate and health impacts of bus emissions were the primary focus of this study. Operating costs were also estimated, but may vary between agencies. Labour costs were assumed to be unaffected by changes in bus assignments and were therefore not considered. Further, given the operational scope of this study, capital financing costs were assumed to be sunk costs and were therefore also not considered.  61  2.3.2  Climate  Of the climate forcing compounds considered, only CO 2 , CH 4 , and BC where found to be significant contributors (Table 2.3). All other species combined, accounted for less than 0.5% of the total GWC on a per kilometer basis. For all bus categories, CO 2 was responsible for a minimum of 90% of the total GWC on per kilometer basis. However, if a shorter time horizon was used to estimate the GWP (e.g., 20 instead of 100 years), BC would make up a more significant fraction of the total GWC. Table 2.3 - Percent contribution of climate forcing compounds to the total GWC·km-1 . Category  CO2 (%)  CO (%)  NMHCa (%)  CH4 (%)  BC (%)  OC (%)  N2O (%)  40DO  92.4  0.467  0.285  0  7.09  -0.342  0.0471  40DB2  96.6  0.387  0.0303  0  3.08  -0.177  0.0702  40DB  98.0  0.181  0.0967  0  1.73  -0.105  0.0787  40DA  99.9  0.0402  0.00747  0  0.0582  -0.0476  0.0767  40DH  99.8  0.037  0.00499  0  0.0399  -0.0325  0.102  60DB  97.8  0.345  0.0486  0  1.83  -0.0812  0.0408  60DA  99.9  0.0732  0.0133  0  0.0836  -0.0683  0.0428  60DH  99.9  0.0584  0.00340  0  0.0255  -0.0208  0.0655  40CG  89.8  0.0621  0  10.1  0.0318  -0.0335  0.0293  a  Non-Methane Hydrocarbons.  2.3.3  Health and Exposure  Public health impacts (i.e., intake of PM2.5 emissions) were quantified on an intra-regional scale in order to account for the effects of changes in the spatial distribution of PM2.5 emissions due to different vehicle assignment solutions. To do this, the PM2.5 intake fractions of the bus routes were estimated (Equation 2.2). The intake fractions ranged from 5.98 to 41.4 grams inhaled per million grams emitted and had a mean value of 19.2 per million (Figure 2.3a). These values are consistent with those reported by Greco et al. for road segments in the Boston, MA area, which ranged from 0.81 to 53 per million and had a mean of 12 per million (Greco, et al.,2007a). The smaller range in the intake fractions found in this study likely reflects the fact that bus routes are aggregations of individual road segments.  Although intake fractions provide a measure of exposure potential, they do not account for the frequency of bus service on a route and thus provide an incomplete description of the potential for health impacts along a route. To provide a more accurate measure of the health impact 62  potential, the distance-weighted PM2.5 intake fractions (i.e., distance travelled on the route per day × intake fraction) were calculated (Figure 2.3b). The distance-weighted intake fractions exhibited a significantly more skewed distribution than the intake fractions and there were a small number of very high potential impact routes including the 99 B-Line route, which is explored in detail in Chapter 3. Further, Figure 2.4 shows that the routes with the highest health impact potential were located in the City of Vancouver, in and around the downtown core. These routes would be given the highest priority in the vehicle assignment optimisation and assigned to the lowest emitting buses. a)  b)  50  Number of Routes  20  15  10  5  0  99 B-Line  25  Number of Routes  60  99 B-Line  30  40  30  20  10  0  10  20  30  PM2.5 Intake Fraction (×10-6)  40  0  0  0.1 0.05 0.15 Distance Weighted PM2.5Intake Fraction (km)  0.2  Figure 2.3 - Histograms of the primary PM2.5 intake fractions (exposure potential) (a) and distanceweighted intake fractions (b) for bus routes in Vancouver, Canada.  These results indicate that there is significant variability in the exposure and health impact potential between bus routes and therefore the potential to influence health impacts through different vehicle assignments solutions.  Scenarios E, F, K and L show that there were differences between solutions that minimise exposure to PM2.5 and solutions that minimise total PM2.5 emissions (Table 2.2). The counterintuitive finding that exposure to PM2.5 emissions were minimised by increasing total PM2.5 emissions on the order of 10% is a direct result of the differences in the intake fractions and the distance travelled between routes (Section 2.3.5). It implies that minimising total PM2.5 emissions does not guarantee minimum exposure to PM2.5. These results quantify the difference between regional and intra-regional scale approaches to estimating exposure (Nazaroff,2008). 63  Figure 2.4 – Map of population density and bus routes in Vancouver, Canada. Routes are color coded and grouped based on the distance-weighted primary PM2.5 intake fraction.  Scenarios E, F, K and L also show that total NOX emissions were increased by between 6-10% when either total PM2.5 emissions or exposure to them were minimised. This indicates a tradeoff between these pollutants 29. However, as the impact of PM2.5 emissions likely dominates the net health impact and the relative increases in NOX are small, this trade-off is unlikely to be a significant concern.  2.3.3.1  Concentration-Response Function  Studies of the relationship between PM2.5 concentrations and premature mortality in North America have shown a linear no threshold (i.e., no safe level) concentration-response function (Pope, et al.,2006). A linear relationship implies that in relative terms, the change in health impacts (e.g., mortality) is equivalent to the change in PM2.5 intake. Therefore, the results 29  This is a well-known trade-off in heavy-duty engine design (Clark, et al.,2002). Higher combustion temperatures, due for example to advanced injection timing, increase efficiency and decrease PM emissions but increase NOX emissions.  64  presented in Table 2.2 are not sensitive to the concentration-response function or other factors related to estimating mortality. However, in order to evaluate the trade-offs between health, climate, and operating costs and impacts, the absolute value of the change in mortality is important and is sensitive to the value of the concentration-response function. In this study, the assumptions and concentration-response function developed by Levy et al. were adopted (Levy, et al.,2009). They estimated a central value of 1.0% increase in all-cause mortality per µg⋅m-3 increase in annual average PM2.5 and a 90% confidence interval of 0.3 to 2.0 % based on a review of major epidemiological cohort studies. The large confidence interval is indicative of the significant uncertainties associated with estimating actual health effects and is one reason for using intake instead of mortality as the primary indicator in this study. Intake provides a better basis for exploring the implications of different values of the concentration-response function as well as monetisation.  2.3.4  Co-benefits and Trade-offs of Multi-Objective Optimisation  The results in Table 2.2 show that there were trade-offs between minimising operating costs, climate impacts and health impacts. For example, in Scenario C, minimising climate impacts (GWC) resulted in a 14% increase in health impacts (PM2.5 intake) and in Scenario I, it resulted in a substantial 49% increase. Similarly, in Scenarios A and G, minimising operating costs resulted in 14% and 47% increase in health impacts respectively. Trade-offs between operating costs and climate impacts were much smaller because fuel consumption significantly influences both of these objectives.  2.3.4.1  Climate and Health Impact Trade-offs  Figure 2.5 shows the change in the minimum health impact (PM2.5 intake) as a function of the change in the climate impact (GWC). Pareto optimal solutions, where there are trade-offs between climate and health impacts, occur on the segment of the curve between the cross and asterisk (i.e., the Pareto frontier). For both fleets, trade-offs only occurred as climate impacts approached their minimum value, at which point health impacts increased exponentially. Vehicle assignment optimisation would result in co-benefits (i.e., both health and climate impacts would decrease) if a transit agency’s actual vehicle assignment solution were above the curve or on the curve and to the right of the asterisk. Note that solutions do not exist below the  65  curve. Figure 2.5 helps provide a more complete picture of the results in Table 2.2, which may be deceiving given the non-linear relationships shown in the figure.  150 Δ PM2.5 Intake (g • year-1)  b)  Fleet A 0%  3.9%  7.9%  12%  Min. GWC Max. GWC Min. PM2.5 Intake  100  65%  33%  50  0%  0 0  5000  10000  Δ GWC (tCO2e • year-1)  Fleet B 25  98% Δ PM2.5 Intake (g • year-1)  a)  15000  0%  1.8%  3.6%  5.4%  7.2%  9.1%  Min. GWC Max. GWC Min. PM2.5 Intake  20  11% 62% 50%  15  37%  10  25%  5  12% 0%  0 0  2000  4000  6000  8000  Δ GWC (tCO2e • year-1)  10000  12000  Figure 2.5 - Relationship between climate and health impacts for Fleet A (a) and Fleet B (b). The change in the minimum health impact is expressed as a function of the change in the climate impact. Bottom and left axes show the change in absolute value and top and right axes show the corresponding percent change.  2.3.4.2  Valuation and Optimal Assignments  As a result of the trade-offs, in order to identify an optimal, multi-objective vehicle assignment solution, the impacts must be expressed in a common measure. There are a number of potential methods of doing this. Social cost-benefit analysis (i.e., monetising costs and benefits) is widely used in the assessment of impacts of vehicle emissions, including transit buses (Krupnick, et al.,1991, Levy, et al.,2010, Stevens, et al.,2005, USEPA,2011a) and was therefore used here. Further, this method may be more amenable to the current economic based decision making practices of transit agencies. However, the application of social cost-benefit analysis to environmental and public health decision making is controversial. To provide a bound on the effect of different methods of evaluating trade-offs, the Pareto frontiers were estimated (i.e., the curve segment between the cross and asterisk in Figure 2.6). Thus, regardless of the methodology used to evaluate the trade-offs, the optimal solution lies at some point on the Pareto frontier. Figure 2.6 shows the change in the minimum health impact (mortality and PM2.5 intake) as a function of the change in the sum of the operating and climate costs. Because operating costs and climate impacts are highly correlated, the shape and features of the curves in Figure 2.6 and Figure 2.5 are very similar. As in Figure 2.5, vehicle assignment optimisation would result in 66  co-benefits (i.e., health, climate, and operating costs would decrease) if a transit agency’s actual vehicle assignment solution were above the curve or on the curve and to the right of the asterisk. The circles indicate the optimal, minimum social cost vehicle assignment solutions given the monetization assumptions. If either the VSL or mortality estimates increased, the optimal solution would shift right along the curve, approaching the minimum health impact solution (asterisk). If the value of the climate impacts increased, the shape of the curve would approach the shape in Figure 2.5 and the optimal solution would shift left along the curve, approaching the minimum climate impact solution (cross). Note however, that the climate costs represent less than 7% of the operating cost (Table 2.1). Therefore, the magnitude or value of the climate impacts would have to increase substantially before climate impacts would significantly affect the optimal solution. Thus, the primary trade-off in vehicle scheduling optimisation is likely between health impacts and operating costs.  1 0.5 0  1.9%  Fleet A 3.9% 5.8%  b) 7.7%  Min. Operating + Climate Cost Max. Operating + Climate Cost Min. PM2.5 Intake Optimal (VSL = $7.7M)  100  65%  33%  50  25  0%  0 0  1  2  3  4  Fleet B  9.7% 98%  5  Δ Operating + Climate Cost (Million $ • year-1)  0.3 0.25 0.2 0.15 0.1 0.05 0  Δ PM2.5 Intake (g • year-1)  1.5  150 Δ PM2.5 Intake (g • year-1)  Δ Mortality (Deaths • year-1)  2  0%  Δ Mortality (Deaths • year-1)  a)  0%  1%  2.1%  3.1%  Min. Operating + Climate Cost Max. Operating + Climate Cost Min. PM2.5 Intake  20  62% 50%  Optimal (VSL = $7.7M)  15  37%  10  25%  5  12% 0%  0 0  0.5  1  1.5  Δ Operating + Climate Cost (Million $ • year-1)  Figure 2.6 - Relationship between health and operating and climate costs for Fleet A (a) and Fleet B (b). The change in the minimum health impact is expressed as a function of the sum of the operating and climate costs assuming a value of $25 per tCO2e. Bottom and left axes show the change in absolute value and top and right axes show the corresponding percent change.  Operating, health and climate impacts make up 71%, 24 %, and 5%, respectively, of the total social cost of the optimal solution for Fleet A and 87%, 8%, and 5%, respectively, for Fleet B. Although the operating costs dominate the total social cost, it is the rate of change of the costs with respect to one another that influences the optimisation results. Thus, for the monetisation assumptions made in this study, the exponential increase in health impacts as operating and climate costs approached their minimum value meant that the optimal social cost solutions were close to the minimum health impact solutions. 67  The difference between the two fleets highlights the implications of fleet modernization. In general, the pollutants that impact health decrease more rapidly as a result of technological change than operating costs and pollutants that impact climate. Fleet B represents a cleaner, more modern, fleet. Its health impacts are much smaller than Fleet A and represent a smaller fraction of the total social costs. Nevertheless, health impacts play a significant role in determining the optimal solution. Therefore, operational optimisation remains significant for modern fleets, although the benefits in absolute terms are reduced. 2.3.5  Factors Affecting Vehicle Scheduling  The optimisation results are sensitive to the characteristics of the specific transit system, and in the case of health impacts, the population density and meteorology of a region. These factors are summarised in the influence diagram in Figure 2.7. onal Planning  Capital Planning  Time Table  Routes  Fleet  Vehicle Scheduling  Bus Categories  Max. # of Buses  # of Buses  Popu on Density Blocks Meteorology  Emission Factor  Intake Fra on  Distance  Opera onal Rel onships  Physical Rel onships  Figure 2.7 - Influence diagram of factors affecting vehicle scheduling optimisation that incorporates climate, health, and operating costs and impacts.  The bus categories and their associated emission factors are influenced by the capital purchasing decisions of transit agencies. Distinct categories arise because transit agencies commonly  68  purchase multiple buses of a single type from a specific manufacturer, rather than individual buses from different manufacturers.  In this study, PM2.5 emission factors were found to vary by over an order of magnitude between bus categories. This variation was responsible for the significant effect of vehicle assignment on health impacts. Although regulatory PM emissions standards for heavy-duty engines are likely approaching their lower limit and PM2.5 emission factors of bus fleets may become more homogenous, low turnover due to the longevity of these engines and the limited capital budgets of transit agencies will likely mean that heterogeneity in PM2.5 emissions factors will persist in the near future.  A number of other operational factors also influence the optimisation results, including: (a) the distance travelled by each bus category; (b) the construction of blocks; (c) operational constraints that impose a relationship between bus categories and routes; and (d) physical factors that impose a relationship between the total distance travelled on a route and the route intake fraction (Figure 2.7).  The total distance travelled by each bus category can vary. A transit agency must have enough buses to meet the schedule at peak demand. During non-peak periods fewer buses are required and operated. Which bus categories serve these periods impacts the distance they travel. Thus, the difference between peak and off-peak periods influences the distance travelled by the various bus categories. This influence would hold true in other regions with similar time-activity patterns.  The way that trips are linked together to form blocks (i.e., the set of trips made by a physical bus over the course of a day) can also influence the distance travelled and the net intake fraction, as blocks may cover multiple routes. This was not considered; however, the vehicle assignment optimisation performed in this study provides a lower bound of the potential benefits of vehicle scheduling optimisation, which would be able to exploit this.  There are almost certainly operational constraints that limit which bus categories can be assigned to a route. For example, in this case study, routes were constrained based on the bus size (either  69  40 ft or 60 ft). It is difficult to determine the effect of other operational constraints but they may limit the benefits of vehicle assignment optimisation.  The relationship between the total distance travelled on a route and the intake fraction has implications for the solutions that minimise total PM2.5 and exposure to PM2.5 (Figure 2.1 and Figure 2.2). If the ranking of the routes based on the total distance and the intake fraction were the same (e.g., positively correlated) then there would be no difference in the assignment solutions for the two objectives, but the relative change in exposure would be greater than the relative change in the total PM2.5 as a result of the correlation. If on the other hand the rankings were the opposite (e.g., negatively correlated), the trade-offs between these two objectives and regional versus intra-regional scale approaches to estimating exposure would be maximised. In this study, distances and intake fractions were weakly negatively correlated (r = -0.19), which contributed to the difference between scenarios based on total PM2.5 and those based on exposure to PM2.5. With respect to objectives involving total emissions or costs, trade-offs between objectives (e.g., minimising GWC and PM2.5) only occur if there are trade-offs in the emission factors or costs between bus categories (i.e., one bus category has a higher GWC emission factor than another bus category but a lower PM2.5 emission factor). If trade-offs between bus categories do not exist any indicators can be used to find the optimal assignment solution and optimisation would always result in co-benefits. However, this is not the case when considering exposure. As discussed, the relationship between the total distances travelled on a route and the intake fraction plays a role in determining whether there are trade-offs.  2.3.6  Assumptions and Limitations  Because the optimisation results are sensitive to variability in the emission factors, distance travelled, and intake fractions, it is important to characterize how various assumptions and limitations affect variability in these factors.  With respect to the optimisation of a single objective, because the objective function is the sum of the product of the emission factor, distance travelled, and in the case of health, the intake fractions, a proportional change in any of these factors would not change the optimisation solution, but an additive change would. For example, if a different breathing rate was used to 70  estimate the intake fractions it would not affect the optimisation solutions or the results in Table 2.2. However, if the intake fractions were increased by an equal amount (e.g., to correct a bias), this would affect the optimisation results. Thus, the optimisation results are sensitive to both the mean and variability and as a result, the coefficient of variation (CV) of these factors. For example, the difference between the best and worst case scenarios in Table 2.2 tend to zero as the CV of either the emission factors or distance travelled approaches zero.  The choice of spatial (and temporal) scale imposes certain assumptions regarding variability and the relationship between factors. The health impact pathway is particularly sensitive to these assumptions, as demonstrated by the differences in the scenarios that minimise total PM2.5 emissions and exposure to PM2.5. In general, if one factor is assumed to be homogeneous, it effectively averages out the variability in the other factors. For example, by using distancebased emission factors, emissions are assumed to be homogeneously spatially distributed over the length of a route. If combined with an intake fraction that accounted for variability along the route (e.g., due to variability in population density), it would be equivalent to using the average intake fraction along the route. Therefore, there is no value in having high spatial resolution in one factor but not another or if one factor does not exhibit variability. However, if in fact variability does exist and is lost due to the choice of scale, and the factors are positively correlated, then the health impacts calculated would be underestimated. In this study, only spatial variability was considered. Temporal variability would likely also be important but data (e.g. population time-activity) was not available to assess this (Greco, et al.,2007a). A more detailed analysis that accounted for the spatial and temporal relationships between factors at higher resolutions would likely have resulted in higher health impact estimates (Chapter 3) (Health Effects Institute,2010, Jerrett, et al.,2005b, Setton, et al.,2011).  2.3.6.1  Emissions Model  Macro scale emissions modeling approaches that employ distance-based emission factors have been criticised because they do not fully account for the effects of vehicle activity (i.e., speed, acceleration and road grade) on emissions (National Research Council,2000). As there are likely differences with respect to road grade, levels of congestion, number of stops, etc., between bus routes, the use of distance-based emission factors likely underestimate both the mean and variability in the total route emissions. However, the overall effect this would have on the optimisation results is unclear. 71  Emissions of pollutants that impact health (e.g., PM2.5 and NOX) are generally more sensitive to vehicle activity than those that impact climate (CO2), so it is likely that these pollutants would be underestimated to a greater degree. If this were the case, the optimal social cost vehicle assignment solution would be shifted to the right, towards the solution that minimised health impacts (Figure 2.6).  Regardless of the modeling approach, there will be uncertainties in the emission estimates. A key question, therefore, is what effect this uncertainty has on the categorization of buses and the optimisation results. Although this could have been explored using Monte Carlo simulation techniques, most emissions models (e.g., MOBILE and MOVES) do not estimate uncertainties and there are no widely accepted distributions for modeling the uncertainties in vehicle emissions. Further, this would have added significant computation complexity to the study. As a result, it was not explored, but warrants further consideration.  2.3.6.2  Intake Fraction  Given the methodological similarities used to estimate the intake fraction, this study shares many of the limitations of the study by Greco et al. (Greco, et al.,2007a). These limitations relate to simplifying assumptions employed to make the modeling of the population, dispersion, and transformation (which was not accounted for), quantitatively feasible over an urban area given the data sources available.  2.3.6.2.1  Population  Residential census data, which does not account for time-activity patterns, does not provide a complete accounting of the population that is actually exposed to transit bus emissions. Static, residential census data may provide a reasonable estimate of the spatial and temporal distribution of the population at night, but is unlikely to do so during the day when transit systems and transportation systems in general, are most active. Activity of the population during the day would be expected to increase, as for example people move to and from work locations, and change the spatial and temporal distribution of the population. Greco et al. single out the use of residential census data as the largest source of both bias and uncertainty in the intake fraction estimates. Specifically, Greco et al. concluded that this assumption leads to underestimates of both the magnitude and variability in the intake fractions. This is likely to be especially true in 72  the context of public transportation systems, which are inherently spatially and temporally correlated with the population movements. For example, a greater number of people are likely to be in closer proximity to bus routes during the day than is suggested by residential census data. Although it is difficult to determine the effect that underestimating both the magnitude and variability would have on the CV and the optimisation results, the underestimated magnitudes suggest that the value of the health impacts were understated. If this was true, the optimal social cost vehicle assignment solution would shift to the right, toward the solution that minimised health impacts (Figure 2.6).  2.3.6.2.2  Dispersion and Transformation  In this study and the study by Greco et al., only primary emissions were considered. Considering the impacts of secondary pollutants that form as a result of physical and chemical atmospheric transformation processes is also important but requires the application of more complex regional air quality models (e.g., the Community Multi-scale Air Quality (CMAQ) model) that account for the interaction of pollutants from multiple sources. Further, the mechanisms by which secondary PM are formed are not yet fully understood (Robinson, et al.,2007). By focusing on primary PM2.5 mass emissions, which can reasonably be modeled as non-reactive, the analysis is simplified while likely still capturing a significant proportion of the total health impacts.  Dispersion is a complex process that can be particularly challenging to model in urban areas, even if considering only non-reactive pollutants such as primary PM2.5. A number of dispersion models have been developed, including the CALINE series of line source dispersion models used by Greco et al. and in this study (Benson,1992, Holmes, et al.,2006, Nagendra, et al.,2002, Vardoulakis, et al.,2003). CALINE is a Gaussian plume line-source dispersion model that accounts for the advection and diffusion of pollutants. It has been widely used by both researchers and regulators to estimate the dispersion of vehicle emissions, in part because of its computational simplicity (Greco, et al.,2007a, Ishaque, et al.,2008, Singh, et al.,2006, USEPA,2010b). The primary inputs to the model are meteorological conditions and receptor locations as well as parameters related to the road configuration and pollutant properties (Appendix A).  73  The performance of CALINE, particularly in complex urban environments, has been questioned (Holmes, et al.,2006, Yura, et al.,2007), and Gaussian plume models suffer from several fundamental limitations. For example, they perform poorly at low wind speeds and only estimate steady-state conditions. Further, in both studies, average meteorological conditions and a single representative road configuration were used as input parameters because more detailed data along the bus routes was not available. This meant that dispersion was modeled as being constant over the region, but in reality it varies within regions and between bus routes. While the above limitations likely contributed significant uncertainty to the dispersion and intake fraction estimates, the application of more complex models (e.g., computational fluid dynamic models) were beyond the scope of this study. Further, the data necessary to employ these models was not available.  Many empirical studies of the spatial distribution of PM2.5 mass concentrations have shown that it is relatively homogenously spatially distributed throughout a region and its intra-regional variability is small (Health Effects Institute,2010, Karner, et al.,2010, Zhu, et al.,2002). Even near roadways, the gradient in PM2.5 mass concentrations has been found to be small (Karner, et al.,2010, Zhou, et al.,2007, Zhu, et al.,2002). This is not surprising given, that primary PM2.5 emissions from transportation and other local emission sources typically make up a small fraction of total PM2.5 emissions and secondary PM2.5 is a major contributor (Health Effects Institute,2010, USEPA,2012). While this may raise questions about the validity of line source dispersion models such as CALINE that predict significant gradients near roadways (e.g., an exponential decay in concentration), it is difficult to interpret and apply these findings in situations where only a single local emission source is considered (e.g., primary PM2.5 emissions from transit buses) and they do not preclude the validity of these models. In fact, significant gradients have been found for components of PM such as black carbon and the ultrafine fraction which are dominated by local transportation sources (Karner, et al.,2010, Zhou, et al.,2007, Zhu, et al.,2002). However, if the empirical findings were found to be true for local sources and the PM2.5 mass concentrations were homogenous in a region, this would mean that there would be no variability in the route intake fractions (i.e., the CV of the intake fractions would be zero) and there would be no differences in the exposure potential between routes for the optimisation to exploit. While this would reduce the potential benefits of the optimisation, it would not eliminate them.  74  The spatial extent of the dispersion modeling was limited to 5000 m. To evaluate the sensitivity of the results to different spatial extents, the intake fractions were estimated for populations within 500 m (iF500) of the route as well as for the population over the entire region (iFReg). The results showed that as the spatial extent decreased, the CV of the intake fraction increased, resulting in a greater difference between optimisation solutions that minimise exposure to PM2.5 emissions and total PM2.5 emissions (Table 2.4). Thus, the use of a 5000 m spatial extent would underestimate the benefits of the vehicle assignment optimisation with respect to health impacts if a 500 m spatial extent was deemed more appropriate. Further, they show that as the CV of the intake fractions increased, the difference between regional and intra-regional approaches increased. Table 2.4 - Optimisation results based on intake fractions calculated using different spatial extents Fleet  Scenarios Objective (Min.)  PM2.5 PMI2.5 Intake (500 m) A PM2.5 Intake (5000 m) PMI2.5 Intake (Regional) PM2.5 PMI2.5 Intake (500 m) B PMI2.5 Intake (5000 m) PMI2.5 Intake (Regional) Coefficient of Variation of Intake Fractions  2.3.6.3  Percent change relative to the best case (%) PM2.5 Best 17.8 12.9 8.87 Best 15.6 9.99 6 -  PM2.5 Intake Spatial Extent 500 m 19.5 Best 2.15 3.33 18 Best 2.41 3.98 0.49  5000 m 12.5 2.51 Best 0.484 10.3 2.28 Best 0.583 0.38  Regional 7.39 3.48 0.484 Best 6.29 3.31 0.373 Best 0.30  Indicators and Valuation  The greatest source of uncertainty, and the most difficult to quantify, in studies of this type are likely associated with the estimation of actual health and climate damages and the valuation of these damages. In terms of health impacts, many studies have used measures of PM, specifically the premature mortality attributable to long-term exposure to PM2.5 (measured in terms of mass), as indicators because PM is suspected to play a dominate causal role in air quality related health impacts (Grieshop, et al.,2011, Levy, et al.,2010, Levy, et al.,2009, Stevens, et al.,2005, Tainio, et al.,2005). For example, Levy et al. cite several studies by the USEPA which found that the premature mortality attributable to PM2.5 dominated the total monetised health impacts as justification for the use of PM2.5 mortality as an indicator. However, people are ultimately exposed to a complex mixture of pollutants from transit buses as well as other sources. Further, other properties of PM such as its composition and particle numbers likely mediate its health  75  impacts and vary by source and atmospheric residence time Thus, there remains uncertainty and debate over the health effects attributable to exposure to PM and vehicle emissions in general.  In terms of climate impact indicators, GWP (and by extension GWC) has been used almost exclusively and become institutionally embedded, in part because of its computational simplicity. However, numerous researchers have pointed out limitations of this indicator (Forster,2007, Fuglestvedt, et al.,2003, Kandlikar,1996). A complete discussion of these limitations is beyond the scope of this study, but in general these criticisms have focused on how impacts are aggregated and weighted over time (e.g., time horizons and discounting), the validity of GWP when applied to short-lived species, and the relationship between GWP and actual changes in the climate (e.g., surface temperature) and damages. Despite these limitations, GWP continues to be the most widely accepted and used indicator of climate impacts (Forster,2007, Fuglestvedt, et al.,2003, Smith, et al.,2008).  As previously mentioned, the application of social cost benefit analysis to environmental and public health decision making is controversial. The monetisation of both climate and health benefits has sparked fierce and divisive debates, and estimates vary widely (Applebaum,2011, Arrow, et al.,1996, Atkinson, et al.,2008, Hulme,2009, Morgan, et al.,1999, Nordhaus,2007, Robinson,2007, Stern,2007). For example, estimates of the VSL by the USEPA range from $0.95 to 21.4 million (2007 USD) (Robinson,2007) and estimates of the social cost of carbon range from $11 to 100 (2007 USD⋅ tCO2e-1) (Nordhaus,2007, Stern,2007). The values used in this study fall within these ranges.  2.3.6.4  Optimisation Solutions  In this study, all buses were assumed to be served from a single depot, although in actuality they are served from multiple depots. This simplification means that some vehicle assignment solutions may not be consistent with the ways in which the transit agency distributes buses across its depots and could violate constraints on the total number of buses at a specific depot or result in additional deadhead (non-revenue service) trips. Thus the vehicle assignment optimisation problem formulated here represents the ideal case; however, it provides a baseline from which to evaluate the implications and costs of how buses are actually allocated to depots.  76  2.3.7  Implications  Operational optimisation strategies such as vehicle assignment and scheduling optimisation have significant potential to reduce the climate and health impacts as well as operating costs of transit systems with minimal capital expenditure. However, the magnitude of these reductions are dependent on the heterogeneity of the bus fleet, the characteristics of the region the transit system serves and how trade-offs between objectives are evaluated.  Transit agencies that optimise the operations of their bus fleets based on operating costs and/or climate impacts alone may inadvertently increase the health impacts associated with the emissions of their bus fleets. This highlights the need for integrated assessment frameworks to ensure win-win vehicle scheduling solutions with co-benefits across multiple dimensions of impacts (e.g., climate, health, economics) (Mazzi, et al.,2007). In this study, a social cost-benefit analysis approach was employed to identify an optimal social cost solution; however, there are other approaches to evaluating trade-offs that could be explored (Keeney,1982). Regardless of the specific approach, this study illustrates the value of a multi-objective framework. Further, by quantifying the Pareto frontier, it also permits a more systematic quantification of different approaches with the potential to arrive at a solution acceptable to multiple stakeholders.  Unlike previous studies (Li, et al.,2009a, Stasko, et al.,2010), this study considered exposure at the intra-regional scale. This not only provides a more accurate assessment of the health impacts of transit fleet emissions (which were found to be underestimated by conventional regional scale approaches), but opens up the possibility of reducing health impacts by exploiting differences in the exposure potential along bus routes arising from variability in the population density and bus emission factors. Although the intake fraction calculation is computationally complex, a GISbased tool has been developed as part of this study to automate the calculation, thereby reducing the barriers to the real-world application of the methods developed in this study. A further advantage of intra-regional scale assessments is that equity implications can be considered (Levy, et al.,2009). A consequence of the strategy proposed here is that bus emissions are shifted from routes with high surrounding population densities to routes with low surrounding population densities, which changes the spatial distribution of health impacts over a region (Figure 2.4). While Levy et al. found that control strategies that reduced health impacts also reduced inequity, these environmental justice issues warrant further consideration (Levy, et al.,2009). 77  Chapter 3: Micro Scale Modeling: Spatial Distribution of Diesel Transit Bus Emissions and Urban Populations and the Implications of Coincidence and Scale on Exposure 3.1  Introduction  Numerous studies have shown a correlation between air pollution and adverse health effects (Dockery, et al.,1993, Health Effects Institute,2010). However, there remains significant uncertainty and debate surrounding the attribution of effects and the causal linkages of the emissions-to-effect health impact pathway (Health Effects Institute,2010, Marshall, et al.,2006, Pope, et al.,2006). Consequentially, the health impact pathway has not been incorporated into the prevailing air pollution management paradigm in North America (Nazaroff,2008). Better understanding of the health impact pathway will give decision and policy makers the ability to promulgate policies that more efficiently and equitably mitigate adverse health effects (Levy, et al.,2006, Nazaroff,2008).  The current management paradigm has focused on the relationship between ambient air pollution levels and health effects at the regional scale. However, impacts are realized by individuals and inequitably distributed across populations (Health Effects Institute,2010). As a result there is a need to move to intra-regional scales that better account for personal exposure and heterogeneity in the spatial and temporal processes of the impact pathway (Greco, et al.,2007a, Health Effects Institute,2010, Jerrett, et al.,2005b, Levy, et al.,2009, Nazaroff,2008).  Mobile source emissions are major contributors of air pollution (Health Effects Institute,2010, National Research Council,2000). Elevated pollutant concentrations have been found in transportation and public transportation microenvironments and linked to adverse health effects (Brugge, et al.,2007, Health Effects Institute,2010, Kaur, et al.,2007). These findings underscore the health significance of mobile source emissions and the need for increased spatial resolution to account for heterogeneity in the distribution of pollutants and impacted populations 30.  Macro scale emissions models such as the USEPA’s MOBILE model are based on emission factors that are estimates of pollutant mass emitted per unit distance traveled (Barth, et al.,1996a, National Research Council,2000). These modeling approaches have been widely applied in 30  Also referred to as exposed populations.  78  assessments of health and equity impacts of diesel transit bus emissions (Cohen, et al.,2003, Levy, et al.,2009, Tainio, et al.,2005). However, these approaches do not capture the change in total emissions or the spatial variability in emissions due to the influences of driverinfrastructure interactions on vehicle activity and mode (Barth, et al.,1996a, Clark, et al.,2002, Hallmark, et al.,2002, Jackson, et al.,2006, National Research Council,2000).  Models such as the USEPA’s MOVES model are being developed to address these limitations and capture the influences of vehicle activity (Barth, et al.,1996a, Koupal, et al.,2002, National Research Council,2000). These models employ micro scale modeling approaches, which are based on emission rates that are estimates of pollutant mass emitted per unit time. Engine power-demand and surrogates such as vehicle specific power (VSP) have been identified as key explanatory variables of the emission rates of pollutants sensitive to mode and have been used in numerous models including MOVES and the model developed by Zhai et al. and employed in this study (Clark, et al.,2003, North, et al.,2006, Zhai, et al.,2008).  The aim of this study was to integrate research by Zhai et al. and Greco et al. to investigate the implications of employing a micro scale emissions modeling approach to estimate exposure to diesel transit bus emissions at a fine spatial scale along one of the busiest transit corridors in Vancouver, Canada (Greco, et al.,2007a, Ishaque, et al.,2008, Zhai, et al.,2008). The objectives were to quantify: (a) the spatial distribution of emissions of carbon monoxide (CO), nitrogen oxides (NOX), and hydrocarbons (HC); (b) the relative importance of the explanatory variables of these emissions; (c) the spatial distribution of the impacted populations; and (d) the implications of the spatial relationship between emissions and impacted populations on exposure. Although studies have quantified the spatial distribution of emissions using micro scale approaches, I’m not aware of any previous studies that have quantified the exposure implications (Li, et al.,2009b).  3.2  Materials and Methods  Diesel transit bus emissions were estimated in 50 m intervals along a bus rapid transit (BRT) route using a VSP-based emission rate model and vehicle activity data collected using a GPS receiver. The populations impacted by the emissions were estimated from census, pedestrian, and transit ridership data in seven zones around the route. A metric was developed to quantify 79  the change in exposure due to the spatial relationship between the emissions and impacted populations. Additional details of the methods can be found in the Appendix B.  3.2.1  Route and Vehicle  This study was carried out on the 99 B-Line Bus Rapid Transit (BRT) route in Vancouver, Canada (Figure 3.1). The 13.5 km route is the main east-west connector within the City of Vancouver and one of the busiest transit corridors in Metro Vancouver. Just under half of the 100,000 person trips made daily along the corridor are made on the 99 B-Line (Owen, et al.,2010). Further, the route has the highest distance-weighted PM2.5 intake fraction in the region (Chapter 2). The route is operated between the hours of 6:00am and 1:00am by Coast Mountain Bus Company (CMBC), an operating subsidiary of TransLink. Service intervals are every 2 to 3 minutes during peak hours and every 15 minutes during off-peak hours. At the time of the study the route was predominately serviced by 1998-2000 model year 60 ft New Flyer D60LF buses powered by Detroit Diesel Series 50 engines with a tare weight of 18,900 kg.  Figure 3.1 - Map of the 99 B-Line Bus Rapid Transit (BRT) route from the University of British Columbia at the western terminus to Commercial Drive at the eastern Terminus in Vancouver, Canada.  3.2.2  Data Collection  Vehicle activity data (time, position, and velocity) was collected using a handheld GPS receiver that was carried on the bus. The GPS receiver was configured for a 1.0 s sampling period.  A  total of 61 traversals, 30 west-bound and 31 east-bound, were collected over a one week period. In order to assess the quality of the GPS data along the route, an additional one hour sample of  80  vehicle speed sensor (VSS) data and GPS data was collected. Additional details on the data collection and processing are provided in Appendix B.  3.2.3  Vehicle Activity Data Point  The GPS data points were transformed into vehicle activity data points for the analysis. The vehicle activity data point (Φk) was defined as: 𝛷𝑘=1…𝑁 �𝑡𝑘 , 𝑑𝑘 , 𝑔𝑘 , 𝑣𝑘 , 𝑎𝑘 , 𝑉𝑆𝑃,𝑘 , 𝐸𝑅𝑐,𝑘 �  Equation 3.1  where k is the time index of the vehicle activity data point; N is the total number of data points in one traversal of the route; tk is the time since the previous data point (s); dk is the linear distance along the route measured from the start point (m); gk is the grade (dimensionless); vk is the velocity of the vehicle (m s-1); ak is the acceleration of the vehicle (m·s-2); VSPk is the vehicle specific power (W·kg-1); c is the pollutant type (CO, NOX or HC); ERc,k is the emission rate (g·s1  ) of pollutant c.  3.2.3.1  Sampling Interval (tk)  The sampling interval was estimated by subtracting the time-stamp of the current GPS point (k) by the time-stamp of the previous GPS point (k-1).  3.2.3.2  Linear Distance (dk)  The west-bound and east-bound routes were encoded using the ESRI ArcGIS linear referencing function. The bus stop at the western terminus of the routes was defined as the start point (0.0 m). Using a shortest path map-matching function, each GPS data point was mapped to a specific location on the bus routes and the linear distance from the start point, dk, was calculated.  3.2.3.3  Road Grade (gk)  The road grade at each GPS data point was estimated using a Digital Elevation Model (DEM) with a vertical positional accuracy of ± 5 m 90% of the time in a four step process (GeoBase,2003). First, elevation profiles were created by overlaying the DEM and shapefiles of the routes in ArcGIS and sampling the elevation at 5 m intervals along the routes. Second, the elevation profile was smoothed over a 250 m window using a 51-point central running average 81  filter to remove discontinuities that results in unreasonably large estimates of grade (Figure 3.2 and Figure 3.3). Additional details on the filtering are provided in Appendix B. Third, the grade profile was calculated by differentiating the elevation profile using the central difference numerical method. Fourth, gk was estimated at each dk though linear interpolation of the grade profile. The resulting elevation and grade profiles of the west-bound and east-bound routes are shown in Figure 3.4 and Figure 3.5.  500  2000 a)  b)  Number of Points  Number of Points  400 300 200 100 0 -10%  -5%  0% Grade (%)  5%  1500  1000  500  0 -100%  10%  -50%  0% Grade (%)  50%  100%  Figure 3.2 - Grade distribution for the west-bound route after filtering (a) and before filtering (b) with a 51-point central averaging filter. 2500  400  b) 2000  300  Number of Points  Number of Points  a)  200  100  0 -10%  1500 1000 500  -5%  0% Grade (%)  5%  10%  0 -100%  -50%  0% Grade (%)  50%  100%  Figure 3.3 - Grade distribution for the west-bound route after filtering (a) and before filtering (b) with a 51-point central averaging filter.  82  10%  100  5%  60 0% 40 -5%  20 East  West 0  Grade (%)  Elevation (m)  80  2000  0  4000  8000 6000 Linear Distance (m)  10000  12000  -10% 14000  Figure 3.4 – Final elevation and grade profiles of the west-bound (a) and east-bound (b) route after the elevation profile was filtered with a 51-point central averaging filter. 10%  100  5%  60 0% 40 -5%  20 West 0  Grade (%)  Elevation (m)  80  0  East 2000  4000  8000 6000 Linear Distance (m)  10000  12000  -10% 14000  Figure 3.5 - Final elevation and grade profiles of the east-bound route after the elevation profile was filtered with a 51-point central averaging filter.  3.2.3.4  Vehicle Dynamics (vk, ak)  The velocity of each data point was estimated using unprocessed GPS data. The acceleration was estimated by differentiating velocity using the central difference numerical method (Jun, et al.,2006b, Yoon, et al.,2005b).  3.2.3.5  Vehicle Specific Power (VSPk)  Vehicle specific power is defined as the instantaneous power per unit mass generated by the engine (Zhai, et al.,2008). It was estimated as  83  𝑉𝑆𝑃𝑘 = 𝑣𝑘 × (𝑎𝑘 + 𝑔 × sin(𝑔𝑘 ) + 𝜓) + 𝜁 × 𝑣𝑘 3  Equation 3.2  where g is the acceleration due to gravity (m·s-2); ψ is the rolling resistance term; and ζ is the aerodynamic drag term. Typical values of the coefficients were taken from the literature: ψ = 0.092; ζ = 0.00011. A derivation of the VSP formulation developed by Zhai et al. (i.e., Equation 3.2) is provided in Appendix B.  3.2.3.6  Emission Rate Model (ERc,k)  Emission rates of CO, NOX and HC were estimated using the VSP-based emission rate model developed by Zhai et al.(Zhai, et al.,2008). The model defines emission rate distributions for 8 VSP modes or bins (Table B.4). The model was developed from a dataset collected using a portable emission measurement system. The fleet analyzed consisted of 1996-1995 40 ft New Flyer diesel transit buses powered by Detroit Diesel Series 50 engines equipped with oxidation catalysts. The MOVES model was under development at the time of this study and was therefore not used.  3.2.4  Evaluation of GPS Data  Modern GPS receivers can directly measure both position and velocity 31; however, acceleration must be estimated by differentiating velocity. Further, these receivers often employ filtering techniques such as Kalman filters that improve measurement accuracy (Jun, et al.,2006a). In this study, both position and velocity measurements were evaluated.  To evaluated the position measurements, map-matching techniques were used (Quddus, et al.,2007). Map-matching involves locating a GPS measurement on a digital road map. Map matching errors are defined as the perpendicular distance from the road centre line to the GPS data points and were calculated using ArcGIS and maps (i.e., GIS shapefiles) of the west-bound and east-bound routes obtained from TransLink. To evaluate the velocity measurements, the one hour sample of GPS and VSS data were compared.  31  Velocity is measured by calculating the Doppler shift of the GPS carrier signal.  84  The GPS estimates were found to lag behind the VSS estimates by 1.3 s. This was corrected for by shifting the GPS data in time by a corresponding amount. For both velocity and acceleration estimates, the errors between the GPS and VSS data were found to be small (Section 3.3.4). Furthermore, 97.5% of all GPS points were found to produce physically consistent results (Appendix B). As a result, no filtering was applied and the raw GPS data was used in the analysis. However, it was verified that all velocity (vk) measurements were less than or equal to 80 kph and that all acceleration (ak) estimates were between -15 kph·s-1 and 10 kph·s-1. Only one point violated these conditions.  3.2.5  Total Emissions (TE)  Total emissions were estimated as 𝑘=𝑁  𝑇𝐸𝑐 = � 𝑡𝑘 × 𝑚𝑒𝑎𝑛�𝐸𝑅𝑐,𝑘 , 𝐸𝑅𝑐,𝑘−1 �  Equation 3.3  𝑘=2  where TEc are the total emissions per traversal of pollutant c (g); k is the index of the vehicle activity data point; N is the total number of data points in the traversal; tk is the sampling interval (s); and ERc,k is the emission rate of pollutant c (g·s-1).  3.2.6  Total Interval Emissions (TIE) and Instantaneous Emission Factor  To model the spatial distribution of emissions, the east- and west- bound bus routes were partitioned into 50 m intervals and the total interval emissions were estimated as 𝐾𝑖  𝑇𝐼𝐸𝑐,𝑖 = � 𝑡𝐹𝑖 × 𝑚𝑒𝑎𝑛�𝐸𝑅𝑐,𝐾𝑖 , 𝐸𝑅𝑐,𝐾𝑖 −1 �  Equation 3.4  where TIEc,i are the total emissions per traversal of pollutant c in interval i (g); Ki is the set of vehicle activity data points that result in emissions in interval i; tF is the set of the fractions of times tk=K spent in interval i (s); and ERc,K are the set of emission rates of pollutant c (g·s-1). The times spent in the interval (tF) were estimated using standard laws of motion. The model depicted in Figure 3.6 was employed in the calculation. The algorithm is described in further detail in Appendix B.  85  Primary Input Variables  Secondary Input Variables  Output Variables  x vk  f(x,v,a)  Total Time to Cover Distance x  t  ak  X  gk ψ  VSPk  ζ  Emission Rate Model  Total Emissions Over Distance x  ERc,k  g  Figure 3.6 - Diagram of the model used to calculate emissions over an interval of length x. The function f(x,v,a) solves –x + v⋅t + ½⋅a⋅t2 for t.  Total interval emissions were also expressed as  ,  =  ,  ×  Equation 3.5  where EF is the instantaneous emission factor of pollutant c in interval i (g·km-1); U is a conversion factor equal to 1000 m·km-1; and L is the interval length equal to 50 m. Emission factors were also obtained from MOBILE6.2 (Appendix B). 3.2.7  Impacted Population (P)  Seven zones around the route were defined (Figure 3.1). Zone 1 represented near-road pedestrian populations. Zones 2-7 were defined following Greco et al. and represented residential populations within 0-50, 50-100, 100-200, 200-500, 500-1000 and 1000-5000 m of the route (Greco, et al.,2007a). 3.2.7.1  Zone 1  The pedestrian population within each interval was modeled using pedestrian count data collected by the City of Vancouver and transit ridership data obtained from TransLink. The pedestrian counts were conducted during peak morning and afternoon periods between 2000 and 2008 and were only available at major intersections between Blanca St. and Commercial Dr. (Figure 3.8). Additional details regarding the model including a flow chart are provided in Appendix B.  86  3.2.7.2  Zones 2-7  Data from the 2006 Canadian census aggregated at the dissemination block level was used to estimate the impacted population in zone 2-7 (Figure 3.1) (Statistics Canada,2006). The total impacted population within each interval and zone was estimated as 𝑞=𝑄  𝑃𝑖,𝑧=2…7 = � 𝐷𝑞 × 𝐴𝑞,𝑧  Equation 3.6  𝑞=1  where Pi,z is the impacted population associated with interval i and zone z = 2…7; q is the dissemination block index; Q is the total number of dissemination blocks; Dq is the population density of dissemination block q (people⋅m-2); Aq,z is the fraction of the area of the ring defining zone z that interests dissemination block q and is centered in interval i (m2).  3.2.8  Spatial Coincidence Factor (SCF)  The spatial coincidence factor (SCF) is a measure of the factor increase or decrease in exposure due to the spatial correlation or coincidence of emissions and the impacted populations. The change is measured relative to the estimate of exposure that assumes emissions and/or impacted populations are homogenously spatially distributed (i.e., modeled at the macro scale) and accounts for spatial heterogeneity characterized by micro scale modeling. It was estimated as 𝑖=𝑀  𝑆𝐶𝐹𝑐,𝑧 = � 𝑖=1  𝑇𝐼𝐸𝑐,𝑖 × 𝑃𝑖,𝑧  𝑚𝑒𝑎𝑛�𝑇𝐼𝐸𝑐,𝑖=1…𝑀 � × 𝑚𝑒𝑎𝑛�𝑃𝑧,𝑖=1…𝑀 �  Equation 3.7  where SCFc,z is the spatial coincidence factor per traversal for zone z and pollutant c; i is the interval number; M is the total number of intervals in the traversal; TIEc,i is the total interval emissions; and Pi,z is the impacted population. A derivation of the spatial coincidence factor is provided in Appendix B.  3.2.9  Uncertainty and Importance Analysis  Uncertainty analysis was used to characterize the influence of uncertainty and variability on the spatial distribution of emissions. Table 3.1 describes the input variable distributions and sampling methods. Each measured traversal was sampled 100 times. A sample traversal was generated by sampling the grade distribution, then calculating VSP, and finally sampling the emission rate distribution for each vehicle activity data point (Фk) in the measured traversal. 87  Results were based on simulated and measured traversals. Uncertainties in population estimates were not quantified.  Importance analysis was used to rank the influence of input variable uncertainty and variability on the spatial distribution of emissions. Each vehicle activity data point (Фk) where vk was greater than 5.0 kph was used to estimate the total emissions over a fixed interval (Table 3.1). Ranks were based on the Partial Rank Correlation Coefficient (PRCC). Both primary (velocity, acceleration, and grade) and secondary inputs (time and emission rate) were analyzed (Figure 3.6).  Table 3.1 - Uncertainty Analysis (UA) and Importance Analysis (IA) input variable distributions. Distribution Variability Uncertainty  Input Variables  Analysis  vk  UA/IA  Empirical  -  ak  UA/IA  Empirical  -  gk  UA/IA  DEM  Normal  Traversal level, non-parametric bootstrap with replacement Traversal level, non-parametric bootstrap with replacement Monte Carlo  ERc,k  UA/IA  VSP Model  Lognormal  Monte Carlo  Sampling Method  The uncertainty of the grade was estimated by modeling the uncertainty of the elevation using a normal distribution with mean 0.0 and standard deviation 3.04, derived from the uncertainty in the DEM (± 5 m 90% of the time). The uncertainty of the emission rates for each pollutant and VSP mode were modeled using a lognormal distribution with the parameters µ and σ estimated as 𝜇 = ln �  𝑚2  √𝑣 + 𝑚2  𝜎 = �ln �  �  𝑣 + 1� 𝑚2  Equation 3.8  Equation 3.9  where m is the sample mean and v is the sample variance obtained from Zhai et al. (Zhai, et al.,2008). The uncertainty in emissions rates reflects that of a fleet of buses not an individual bus.  88  3.3  Results  The spatial distribution of emissions from diesel transit buses and the populations impacted by those emissions were estimated along the 99 B-Line BRT route (Figure 3.7 and Figure 3.8).  3.3.1  Spatial Distributions of Emissions and Impacted Populations  Emissions along the route exhibited significant variability. However, there was a consistent pattern to the spatial distribution of emissions and vehicle activity over the study period (Figure 3.9). As a result of these patterns there was an increased probability of hot spots around major intersections and bus stops. Emissions and emission factors near bus stops and intersections were 1.6-3.0 times greater than the route mean (Table 3.2). Idle emissions made up 19-43% of the near bus stop emissions, whereas overall idle emissions made up 8-22% of total emissions (Table 3.2). MOBILE6.2 underestimated emissions of CO and NOX but not HC (Table 3.2, Figure B.18). The elevated emissions near Cambie St. were a result of traffic congestion and delays associated with road construction that occurred during the study period. Total emissions for west- bound traversals were higher as a result of the increased engine power-demand while traveling up the hill between Alma St. and Blanca St. (Table 3.2 and Figure 3.8).  The spatial distributions of the impacted populations in the seven zones around the route are shown in Figure 3.7, Figure 3.8, Figure 3.10 and Figure 3.11. The pedestrian population in zone 1 exhibited peaks around major intersections and bus stops. The largest peaks, at Granville St. and Commercial Dr., coincided with major transit interchanges.  89  Table 3.2 - Total, idle, near bus stop and near intersection emissions as well as emissions factors of CO, NOX, and HC for east- and west- bound traversals of the 99 B-Line bus route estimated using macro scale and micro scale modeling approaches.  Macro scale  CO MOBILE6.2 Total Emissions (g) MOBILE6.2 Emission Factors (g·km-1)  East 50.7  West 188  3.74  HC East 191  West 3.4  14.1  East 3.4 0.254  -30  -23  -24  -19  +2.5  +3.6  Total Emissions (g)  70.8 (0.059)  65.4 (0.064)  249 (0.064)  235 (0.073)  3.3 (0.089)  3.32 (0.11)  Mean Emission Factors (g·km-1)a  5.31 (0.059)  4.83 (0.064)  18.7 (0.064)  17.4 (0.074)  0.248 (0.089)  0.245 (0.11)  Near Bus Stop Emission Factors (g·km-1)b  11.4 (0.082)  11.9 (0.10)  42.2 (0.086)  45.2 (0.10)  0.651 (0.12)  0.743 (0.13)  Near Intersection Emission Factors b  8.61 (0.13)  8.14 (0.13)  30.9 (0.13)  29.8 (0.14)  0.434 (0.17)  0.441 (0.17)  Percent Idle Emissions (%)  8.44 (1.8)*  9.65 (2.2)*  11.0 (2.1)*  12.4 (2.6)*  19.9 (3.1)*  21.7 (3.8)*  19.2 24.4 29.4 23.9 Percent Idle Emissions Near Bus Stops 37.7 (4.7)* (5.7)* (6.0)* (5.1)* (%)b (6.2)* Coefficient of variation shown in brackets. * Standard deviation shown in brackets. a Mean instantaneous emission factor (EF) estimate over the route. b Near is defined as within 50 m.  43.2 (6.7)*  Micro scale  Error in MOBILE6.2 Estimates (%)  NOX  West 49.9  90  59265 50605  58023  Wesbrook Mall  52095  58239 58502 58501 58500 Granville Cambie Fraser Burrard Hemlock Main Fir Oak Kingsway  Arbutus  MacDonald  50916  Clark Commercial  0.4  20  0.2  10  59265 50605  58023  Wesbrook Mall  50357  Alma  Alma  Blanca  100  52095 MacDonald  58239 58502 58501 58500 Arbutus Granville Cambie Fraser Burrard Hemlock Main Fir Oak Kingsway  50916  52094  Clark Commercial  0  0 1 0.8 0.6  80 60  0.4  40 0.2  20 0  c)  59265 50605  58023  Wesbrook Mall  50357  Alma  Alma  Blanca  2  52095  58239 58502 58501 58500 Granville Cambie Fraser Burrard Hemlock Main Fir Oak Kingsway  Arbutus  MacDonald  50916  52094  Clark Commercial  0 120 9000  X  NO (g  -1  km )  0.8 0.6  30  0  b)  1  52094  120 9000  -1  km )  Alma  Blanca  40  CO (g  50357  Alma  120 9000  a)  0 1 0.8  HC (g  km-1 )  0.6 0.4  1  0.2 0 0  2000  4000  6000 Linear Distance (m)  8000  10000  12000  0  0  Figure 3.7 - Histograms of the spatial distributions of estimated emissions of CO, NOX, and HC for east-bound traversals of the 99 B-Line bus route. The probability of the estimated instantaneous emission factor along the route is indicated by the color bar. The direction of travel is indicated by the red arrows. Bus stops are indicated by vertical red lines. Major intersections are indicated by vertical black lines. Minor intersections are indicated by vertical dashed lines. The elevation is indicated by the grey profile. Populations in zone 1 (pedestrian) and zone 5 (200-500 m) are shown as indicated  91  Blanca  CO (g • km-1)  40  Clark Commercial  10  2000  4000  50276  Wesbrook Mall  Blanca  100  6000  58037 Alma Alma  8000  10000  0  12000  50319  58503 52099 52100 52101 Arbutus Granville Cambie Fraser Burrard Hemlock Main MacDonald Fir Oak Kingsw ay  58499 58491 Clark Commercial  80 60 40 20 0 0  2000  59266 50268  4000  50276  Wesbrook Mall  Blanca  2  6000  58037 Alma Alma  8000  10000  58503 52099 52100 52101 Cambie Fraser Arbutus Granville Main Burrard Hemlock Kingsw ay MacDonald Fir Oak  1  0 0  2000  4000  6000 8000 Linear Distance (m)  10000  0  12000  50319  12000  1  58499 58491  20  59266 50268  NOX (g • km-1)  58503 52099 52100 52101 Arbutus Granville Cambie Fraser Burrard Hemlock Main MacDonald Fir Oak Kingsw ay  30  0 0  HC (g • km-1)  50319  Population Zone 1 120 Population Zone 5 8000  Wesbrook Mall  58037 Alma Alma  Population Zone 1 120 Population Zone 5 8000  50276  Clark Commercial  0.8 0.6 0.4 0.2 0  1 0.8 0.6 0.4 0.2 0  1  58499 58491  Population Zone 1 120 Population Zone 5 8000  59266 50268  0  0.8 0.6 0.4 0.2 0  Figure 3.8 - Histograms of the spatial distributions of estimated emissions of CO, NOX, and HC for west-bound traversals of the 99 B-Line bus route. The probability of the estimated instantaneous emission factor along the route is indicated by the color bar. The direction of travel is indicated by the red arrows. Bus stops are indicated by vertical red lines. Major intersections are indicated by vertical black lines. Minor intersections are indicated by vertical dashed lines. The elevation is indicated by the gray profile. Populations in zone 1 (pedestrian) and zone 5 (200-500 m) are shown as indicated.  92  5926550605  58023  Wesbrook Mall  Blanca  Velocity (kph)  80  MacDonald  58239 58502 58501 58500 Fraser Cambie Granville Arbutus Main Burrard Hemlock Kingsw ay Oak Fir  50916 52094 Clark Commercial  0.2  20  4000  2000  0  58023  Wesbrook Mall  6000 Linear Distance (m) 50357 Alma Alma  Blanca  10  52095 MacDonald  8000  10000  0  12000  58239 58502 58501 58500 Arbutus Granville Cambie Fraser Burrard Hemlock Main Fir Oak Kingsw ay  50916 52094 Clark Commercial  1 0.8 0.6  5  0.4  0  0.2  -5  0  2000 5926550605  4000 58023  Wesbrook Mall  6000 Linear Distance (m) 50357 Alma Alma  Blanca  20  52095 MacDonald  8000  10000  0  12000  58239 58502 58501 58500 Arbutus Granville Cambie Fraser Burrard Hemlock Main Fir Oak Kingsw ay  50916 52094 Clark Commercial  1 0.8 0.6  10  0.4  0  0.2  -10 -20  0.8  0.4  40  -10  1  0.6  5926550605  Acceleration (kph • s -1)  52095  60  0  VSP (W • s kg-1)  50357 Alma Alma  0  2000  4000  6000 Linear Distance (m)  8000  10000  12000  0  Figure 3.9 - Histograms of the spatial distributions of vehicle activity for west-bound traversals showing the probability of vehicle activity in 50 m intervals. The direction of travel is indicated by the red arrows. Bus stops are indicated by vertical red lines. Major intersections are indicated by vertical black lines. Minor intersections are indicated by vertical dashed lines. The elevation is indicated by the grey profile.  93  Zone 1  Zone 2  Zone 3  Zone 4  Zone 5  Zone 6  Zone 7  6  10  5  10  4  Population  10  3  10  2  10  1  10  0  10  -1  10  0  2000  4000  8000 6000 Linear Distance (m)  10000  12000  14000  Figure 3.10 – Population within the 7 zones (Pedestrian, 0-50 m, 50-100 m, 100-200 m, 200-500 m, 500-1000 m, 1000-5000m) along the west-bound 99 B-Line bus route. Major intersections are indicated by solid vertical grey lines. Minor intersections are indicated by dashed vertical grey lines. Bus stops are indicated by vertical red lines. Zone 1  Zone 2  Zone 3  Zone 4  Zone 5  Zone 6  Zone 7  6  10  5  10  4  Population  10  3  10  2  10  1  10  0  10  -1  10  0  2000  4000  6000 8000 Linear Distance (m)  10000  12000  14000  Figure 3.11 - Population within the 7 zones (Pedestrian, 0-50 m, 50-100 m, 100-200 m, 200-500 m, 500-1000 m, 1000-5000m) along the east-bound 99 B-Line bus route. Minor intersections are indicated by dashed vertical grey lines. Bus stops are indicated by vertical red lines.  94  3.3.2  Spatial Coincidence and Exposure  The coincidence of peaks in pedestrian (zone 1) populations and emission hot spots increased exposure on average by 27-55% (Figure 3.12). For populations in zones 2-7 exposure increased on average by 0-12%. Generally the increase in exposure was greatest for HC emissions. The underestimate of CO and NOX emissions by MOBILE6.2 translated into an additional 23-42% increase in exposure across all zones (Table 3.2). a) 100%  Change in Exposure  b)  CO  c)  NOX  HC  2 West East  75%  1.75  50%  1.5  25%  1.25  0%  1 1  2  3  4 5 Zone  6  7  1  2  3  4 5 Zone  6  7  1  2  3  4 5 Zone  6  7  Figure 3.12 - Boxplots of the spatial coincidence factor (SCF) showing the change in exposure due to the spatial coincidence of the impacted populations in seven zones around the route and estimated emissions of CO, NOX, and HC for west-bound and east-bound traversals of the 99 BLine bus route. Median values are indicated by bars and mean values by dots. Changes in exposure due to MOBILE6.2 underestimates are not shown.  3.3.3  Importance  The relative importance of each of the primary input variables used to model the spatial distribution of emissions is shown in (Figure 3.13). The high importance of acceleration indicated the distribution was sensitive to mode (Figure 3.13a, b), whereas the high importance of velocity indicated sensitivity to the time required to traverse an interval (Figure 3.13a-c). Negative PRCC values for velocity indicated that total interval emissions decreased with increasing velocity. Overall, velocity, acceleration, and grade were important explanatory variables of CO and NOX distributions, whereas velocity was the dominant explanatory variable of HC distributions. Analysis of the secondary input variables showed that regardless of the predictive power of the emissions rate model (i.e., importance was based on combined variability and uncertainty), the time required to traverse an interval and the emission rate were equally 95  important in determining the spatial distribution of emissions. This was also reflected in the fact that acceleration and velocity had similar levels of importance for emissions of CO and NOX, which were well-predicted by the emissions rate model (Figure 3.13a, b). a)  b) v  2  a g  c) v  1  t Secondary  v  a  3  Primary  2  g  1  2  t Secondary 1 ERNO  ERCO  a  3  Primary  g 2 1  X  -1  -0.5  0  0.5  1  -1  -0.5  0  0.5  1  1  2 3  Primary  t Secondary  2  ERHC  1 -1  -0.5  0  0.5  1  Figure 3.13 - Importance analysis of the input variables used to estimate the spatial distribution of emissions of CO (a), NOX (b), and HC (c) along the 99 B-Line bus route. Black bars indicate the partial rank correlation coefficients (PRCC) values for primary input variables: velocity (v), acceleration (a), and grade (g) and secondary input variables: interval traversal time (t) and emission rate (ER). Gray bars indicate negative PRCC values. Numerical rankings of importance were based on the absolute values of the PRCC. Lower ranks indicated greater importance.  3.3.4  Evaluation of GPS Data  The map matching errors for all east- and west- bound GPS data are shown in Figure 3.14. There are several potential sources of error including: (a) buses travel in a lane but the digital road map represents the road centre line; (b) GPS measurement errors; and (c) digital road map errors. Assuming an average road width of 20 m (i.e., confidence interval), 86% of east-bound points and 90% of west-bound points fell on the road. The errors of largest magnitude were found between Fir St. and Cambie St. where the heights of the buildings adjacent to the route were greatest (Figure 3.1). These results suggest that the GPS position data collected along the 99 B-Line route accurately described the position of the buses and can be used to estimate the distributions of emissions at a spatial resolution of 50 m.  96  99 West Map-Matching Error Distributions (µ=-7.782,σ=6.829 CI=0.857)  99 East Map-Matching Error Distributions (µ=3.653,σ=6.112 CI=0.898)  b)  10000  10000  8000  8000  6000  Road (w idth=20 m)  Number of Points  Number of Points  a)  4000  6000  4000  2000  2000  0 -80  Road (w idth=20 m)  -60  -40  -20 0 Error (m)  20  40  0 -50  60  0 Error (m)  50  Figure 3.14 - Map-matching errors for west-bound (a) and east-bound (b) traversals. The road centre lines are indicated by the dotted yellow line and the outer edges of the roads are indicated by the vertical black lines.  The GPS and VSS velocity and acceleration data are compared in Figure 3.15. The scatterplots suggest that GPS measurements of velocity (R2 = 0.99) and the corresponding estimates of acceleration (R2 = 0.91) along the 99 B-Line route were accurate. However, as was expected there was greater noise (i.e., uncertainty) in acceleration estimates. Further, the comparison shows that there were no systematic bias in the data and that the errors were homoscedastic. Additional plots are provided in Appendix B. a)  b) 60  10  R2=0.99269  R2=0.90966  VSS Acceleration (kph ⋅s -1)  VSS Velocity (kph)  50  40  30  20  5  0  -5  10  0  0  10  20 30 40 GPS Velocity (kph)  50  60  -10 -10  -5  0  5  10  GPS Acceleration (kph⋅s -1)  Figure 3.15 – Scatter plots comparing GPS and VSS velocity measurements (a) and acceleration estimates (b). Points located on the identity line indicate agreement between the GPS and VSS data.  97  3.4  Discussion  The emissions-to-effect impact pathway describes the causal processes by which emissions disperse and transform to form concentrations that individuals come into contact with, resulting in exposure, intake, dose, and health effects (Marshall, et al.,2006). Regional scale assessments that do not account for spatial heterogeneity in these processes have been shown in this study and others to underestimate exposure and poorly characterize the distribution of impacts over the population (Greco, et al.,2007a, Health Effects Institute,2010, Jerrett, et al.,2005b, Levy, et al.,2009).  3.4.1  Emissions Modeling and Explanatory Variables  Macro scale emissions modeling approaches such as MOBILE that employ mass per unit distance emission factors are limited in their ability to characterize heterogeneity in the spatial distribution of emissions. Furthermore, these approaches have been found in this and previous studies to result in biased estimates of pollutants that are sensitive to mode, such as CO and NOX (Table 3.2) (Barth, et al.,1996a, National Research Council,2000). Micro scale modeling approaches that employ mass per unit time emission rates can be combined with activity data to address these limitations. Even if the emissions rate model has little predictive power (e.g., HC emissions), significant insights into the spatial distribution of emissions may be gained from the activity data alone.  Numerous factors have been shown to influence emissions from heavy-duty diesel vehicles (Clark, et al.,2002, Yanowitz, et al.,2000). For a given vehicle configuration, engine powerdemand and surrogates such as VSP have been found to be strong explanatory variables of emission rates for CO, NOX, and particulate matter (PM) but not HC (Clark, et al.,2003, North, et al.,2006, Zhai, et al.,2008). As a result, estimates of HC emissions were associated with greater uncertainty than CO and NOX (Table 3.2). Due to the significant health effects attributed to PM exposure, a drawback of current micro scale modeling approaches is the lack of PM emissions models and data (Clark, et al.,2003, Health Effects Institute,2010, North, et al.,2006, Pope, et al.,2006). However, there is some evidence that CO and PM mass emission rates are correlated (Clark, et al.,1999b). If this is the  98  case, the spatial distributions of the two pollutants would be similar, as well as the exposure implications and SCF estimates.  3.4.2  Importance of Explanatory Variables  The consistent spatial patterns of emissions modeled over the study period were a result of terrain and patterns in driver-infrastructure interactions that determined the time required for a bus to traverse an interval, the engine power-demand and the emission rate (Figure 3.13). However, the results of the importance analysis are contextual and quantify importance over the entire route. Analyses performed on specific intervals along the route or on different routes would potentially yield different results. For example, grade would not be an important input variable on intervals or routes without hills.  The importance analysis results applied only for buses in motion because the time required to traverse an interval could not be estimated from the primary input variables when the velocity was zero. To address this limitation, emissions within 50 m of bus stops and major intersections were analyzed in greater detail. Idle emissions made up a significant fraction of these emissions (Table 3.2). Thus, hot spots observed at these locations were not only the result of increased emission rates associated with acceleration and high engine power-demand, but also the time buses spent in these areas (Unal, et al.,2004). In fact, hot spots were more distinct for pollutants that are not sensitive to mode, such as HC. Pollutants sensitive to mode, such as CO and NOX, were “smeared” in the direction of travel, resulting in less distinct hot spots (Figure 3.7 and Figure 3.8).  Uncertainty and variability of input variables were not distinguished in this analysis. Because the GPS data were of high quality and uncertainties of the primary input variables were small relative to their variability this was not expected to have a significant impact on the results (Section 3.3.4).  3.4.3  Receptor and Population Modeling  There are significant challenges associated with modeling impacted populations, particularly at fine spatial scales. The limitations of census data have been previously discussed by Greco et al. (Greco, et al.,2007a). By incorporating pedestrian counts at intersections and transit ridership 99  data some of these limitations were addressed, but these data provided only a first-order characterization of the pedestrian population. The underlying assumption of the pedestrian model was that bus stops and major intersections introduce delays in pedestrian movements which cause pedestrians to concentrate in these areas (Ishaque, et al.,2008). However, there are other features in the urban environment that may have similar influences.  3.4.4  Exposure and Spatial Heterogeneity  Intake fraction (iF) (a metric which relates the amount of pollutant inhaled to the amount emitted) has been widely employed to characterize exposure to emissions from mobiles sources including diesel transit buses (Cohen, et al.,2003, Greco, et al.,2007a, Levy, et al.,2009, Marshall, et al.,2006, Stevens, et al.,2005, Tainio, et al.,2005). The SCF was derived from the spatially variant parameters of the iF to quantify the change in exposure due to the spatial coincidence of emissions and impacted populations (Appendix B). In this study only emissions and impacted populations were considered spatially variant. Meteorology and other parameters affecting dispersion were assumed to be spatially invariant (Greco, et al.,2007a). Although there are limitations associated with this approach, it allowed the SCF to be estimated for each zone without modeling dispersion.  The zones (modeled as areas of constant concentration) were defined so as to capture the pollutant concentration gradient across them (Greco, et al.,2007a). The change in total exposure (not quantified in this study) estimated across all zones would depend on this gradient. If concentrations proximate to the route were significantly elevated, as commonly reported, then the change in total exposure could be significant despite smaller populations proximate to the route (Greco, et al.,2007a, Health Effects Institute,2010, Zhou, et al.,2007, Zhu, et al.,2002).  The increase in exposure due to spatial coincidence (SCF > 1.0) experienced by pedestrian populations in zone 1 occurred because these populations were assumed to be under similar infrastructure influences (e.g., delays at intersections) as the buses and thus concentrated at locations of emission hotspots. It was more difficult to attribute changes in exposure (SCF) to specific terrain or infrastructure features in zones 2-7. At greater distances from the route the spatial variability in emissions observed on the route was reduced through the process of dispersion and the pollutant concentrations were more spatially homogeneous. As a result both 100  the variability in the change in exposure and the mean change in exposure were greater in zone 1 than zones 2-7 (Figure 3.12).  The change in exposure in zones 2-7 was due largely to the overall trend in emissions and impacted populations, which were generally lower over the western (left) section of the route than the eastern (right) section. As a result of this west-east spatial coincidence, there was approximately a 10% increase in exposure in zone 7. The west-east trend in population decreased over zones 6-2 (Figure 3.10 and Figure 3.11) and as a result the change in exposure over zones 6-2 tended to 0% (Figure 3.12). The decrease in the change in exposure over zones 6-2 may also be due to lower residential populations around major intersections and corresponding emission hot spots. Presumably this was due to the increased commercial space around major intersections, which would have populations that are not reflected in residential census data.  The greatest increase in exposure due to spatial coincidence occurred for HC emissions in zone 1 (Figure 3.12). The increase was smaller for emissions sensitive to mode, such as CO and NOX, because the smearing of these emissions had the counterintuitive effect of reducing the magnitude of hotspots. In addition to the increase in exposure due to spatial coincidence, exposure to CO and NOX emissions were further increased over all zones because macro scale modeling approaches underestimated total emissions of these pollutants.  3.4.5  Assumptions and Limitations  The model years of the buses differ between this study (2000-1998) and those used in the development of emissions rate model (1996-1995) (Zhai, et al.,2008). However, the engine families are the same (Detroit Diesel Corporation Series 50) and between 1995-2000 the regulatory emission standards for CO and HC did not change and decreased only slightly for NOX (Table A.2). As a result it is unlikely that there are large differences in the NOX, CO, and HC emission rates of these engines between 2000 and 1995. Furthermore, changes in emission characteristics over this period would more likely impact estimates of total emissions than the spatial distribution of emissions. Given there are currently no micro scale emissions models for 2000-1998 model year Detroit Diesel Corporation Series 50 engines, using the model developed  101  by Zhai et al. provides the most reasonable estimate of emissions for the buses in this study (Zhai, et al.,2008).  By definition VSP-based emission rate models are not sensitive to changes in vehicle weight. However, studies have shown empirically that emission rates of some pollutants are sensitive to changes in vehicle weight and passenger loads (Clark, et al.,2007a, Clark, et al.,2003, Frey, et al.,2007, Gajendran, et al.,2003). Although emissions of most pollutants generally appear to increase with vehicle weight, the empirical results are mixed and do not show a clear relationship between vehicle weight and emission rates. Because the buses tested in the development of the emissions rate model had 40 ft chassis and weighed 12,000 kg and the buses in this study had 60 ft chassis and were estimated to weigh 23,100 kg, it is likely that emissions of some pollutants were underestimated (Zhai, et al.,2008). The impact of vehicle weight on the spatial distribution of emissions is less certain. Given there are currently no micro scale emissions models for 60 ft buses, using the model developed by Zhai et al. provides the most reasonable estimate of emissions for the buses in this study (Zhai, et al.,2008).  People spend a significant portions of their time away from their residences (Greco, et al.,2007a). Although incorporating a pedestrian model addresses some of the limitations associated with the use of census data, workers are largely unaccounted for. Because transit facilities are often located in proximity to commercial and office space, relying on residential census data may underestimate both the total exposure and the change in exposure (i.e., SCF). However, the pedestrian model developed in this study provides only a first-order characterization of the spatial distribution of pedestrians along the route. The pedestrian count data was sparse and there were generally only one to two samples at each intersection. However, this data was consistent with the transit ridership data which was extensive. Boarding and alighting counts estimated over one hour periods were available at each stop over the study period. More advanced models of pedestrian movements have been studied but their application was beyond the scope of this study (Ishaque, et al.,2008, Ishaque, et al.,2009).  Parameters affecting dispersion such as meteorology were assumed to be spatially invariant along the route (Greco, et al.,2007a). In reality, factors such as vehicle induced turbulence, street canyon effects, and building heights would impact dispersion processes differently along the route; however, data were not available to analyze these effects at the micro scale. It is 102  difficult to determine the impact of this assumption on the results. For example, street canyon effects may increase the magnitude of hot spots, whereas depending on direction and intensity, wind could shift and/or reduce the magnitude of hot spots. Further study is required to characterize the impact of this assumption. The zones (regions) developed by Greco et al. were defined to capture significant gradients in concentration (Greco, et al.,2007a). In addition, zones should be defined to capture gradients in impacted populations. Given the block-level resolution of the census data and the corresponding lack of significant gradients in residential populations, it is unlikely that the definition of the zones had a large impact on the results of this study. 3.4.6  Implications and Recommendations  Compared to micro scale emissions modeling approaches, macro scale approaches underestimated exposure to diesel transit bus emissions because: (a) total emissions of pollutants sensitive to mode were underestimated, and (b) coincidence between the heterogeneous spatial distributions of emissions and impacted populations were not accounted for. As pedestrian populations proximate to the route were disproportionately impacted, these results may have implications for distributional equity. Furthermore, as similar patterns in emissions rates and vehicle activity have been found across a wide range of heavy-duty and light-duty vehicles, these results may have implications for mobile sources other than diesel transit buses (Ishaque, et al.,2008, Jackson, et al.,2006, North, et al.,2006, Unal, et al.,2004). Modeling on a micro scale increases the data requirements of an analysis. Consumer grade GPS receivers provide an accurate source of vehicle activity data including position, velocity, and acceleration data at sampling rates of up to 1.0 Hz; however further study is required to characterize the effects of interference due to surrounding buildings. Analyses at fine spatial scales are often associated with greater cost but only add value where heterogeneity exists, for example near roadways (Greco, et al.,2007a). Efficient and equitable mitigation of the health impacts associated with mobile source emissions requires improved characterization of the impact pathway that is sensitive to scale and heterogeneity. Micro scale emissions models have greatly improved characterization of mobile source emissions; however, uncertainties in the dispersion process and the time-activity patterns of impacted populations remain and warrant further study. 103  Chapter 4: Evaluating the MOVES Emissions Model using Heavy-Duty Transit Bus Data 4.1  Introduction  The Motor Vehicle Emissions Simulator (MOVES) model is the latest mobile source emissions model developed by the USEPA to estimate emissions from highway vehicles (USEPA,2009d). MOVES2010a was approved for regulatory (e.g., State Implementation Plan (SIP) development) and transportation conformity purposes in the US in March, 2010 and replaces the previous model, MOBILE6.2 (CFR,2010, USEPA,2009d, 2011b). In addition to their regulatory role, MOBILE and, increasingly, MOVES, have been widely used by consultants and researchers to quantify emissions from mobile sources and their environmental and health impacts (Gao, et al.,2009, Health Effects Institute,2010). The outputs of these models have significant economic, environmental, and public health implications that influence policy decisions affecting a wide range of stakeholders.  The MOVES model represents a significant evolution in mobile source emissions models. MOVES can estimate running exhaust emissions of carbon dioxide (CO2), nitrogen oxides (NOX), carbon monoxide (CO), total hydrocarbons (THC), methane (CH4), particulate matter (PM) as well as emissions from other processes (Table C.2) and of other pollutants (Table C.1) (USEPA,2009b, 2011b). It incorporates several advancements over its predecessor, including new methodologies that better account for the effects of vehicle activity, support for multiple analysis scales and in future versions, uncertainty estimation (National Research Council,2000, USEPA,2001, 2002c, 2009b, d, 2011b).  Numerous factors have been found to affect running exhaust emissions (e.g., Clark, et al.,2002). These explanatory variables can be grouped into three main categories: Vehicle Activity, Vehicle and Energy Source, and Environment (Table 4.2 and Table 1.7). In general, mobile source emissions models such as MOVES and MOBILE use or make assumptions about variables from each of these categories to predict emissions (USEPA,2005, USEPA, et al.,2011d). These models are generally constructed by developing base emission rates for a subset of explanatory variables and then using correction factors to account for other explanatory variables (Section 1.6.2.3).  104  To address criticisms of the MOBILE model, the US EPA introduced new explanatory variables to account for the effects of vehicle activity and mode (i.e., speed, acceleration, and road grade) on emission (National Research Council,2000). Originally Vehicle Specific Power (VSP) and speed were identified as the most suitable explanatory variables, but in MOVES2010, the USEPA replaced VSP with Scaled Tractive Power (STP) for heavy-duty vehicles including transit buses (USEPA,2002b, 2005, Zhai, et al.,2008). Both VSP and STP are proxies for engine power and are functions of speed, acceleration and road grade.  The MOVES model supports multiple analysis scales (National Research Council,2000, USEPA,2001, 2009b). At the micro scale, MOVES models emissions over road segments or links with no minimum length, thus there is no lower limit on the spatial resolution of the model (USEPA,2004). This spatial resolution is sufficient to model emission hot spots and near-road exposure (Chapter 3). However, MOVES does not currently have the ability to model secondby-second emissions which may also be useful for these purposes as well as for evaluation (Chapter 3) (Li, et al.,2009b, Unal, et al.,2004).  In developing the MOVES model, one of the USEPA’s objectives was to improve the accuracy of its mobile source emissions models. However, more complex models do not necessarily improve predictions and to date there have been very few evaluation studies of the MOVES model (Keepin, et al.,1984, Smit, et al.,2010, USEPA, et al.,2011c).  The goals of this study were to: (a) evaluate the MOVES2010a emissions model by comparing predicted running exhaust emissions to laboratory chassis dynamometer and on-board emissions tests results from heavy-duty diesel and Compressed Natural Gas (CNG) transit buses (Smit, et al.,2010); (b) identify sources of bias and uncertainty in the MOVES model using second-bysecond data; and (c) provide the reader with a better understanding of the MOVES model and its evolution. To the best of my knowledge, this is one of the first independent evaluation studies of the MOVES model and the first to examine heavy-duty, transit bus emissions.  4.2  Materials and Methods  A comprehensive dataset of laboratory chassis dynamometer and on-board emissions tests results of heavy-duty diesel and CNG transit buses was compiled from two sources (Table 4.1). For each test in the dataset, the measured emissions were compared to the predicted emissions 105  obtained from a custom implementation of the MOVES2010a model. The vehicle activity data inputs were obtained from either: (a) measured second-by-second vehicle speed in the case of the on-board emissions tests; or (b) the second-by-second drive cycle speed-time traces in the case of chassis dynamometer tests.  4.2.1  Emissions Measurement Dataset  Aggregate emissions measurements (i.e., total emissions measured over a test or drive cycle) of heavy-duty diesel and CNG transit buses made under laboratory conditions using standard drive cycles and chassis dynamometer systems were obtained from the Center for Alternative Fuels Engines and Emissions (CAFEE) at West Virginia University (WVU). In addition, a smaller dataset of second-by-second on-board data collected using a Portable Emission Measurement System (PEMS) was obtained from TransLink, the regional public transportation authority in Vancouver, Canada (Table 4.1). The test data was divided into two technology groups (Table 4.1). The conventional technology group represented conventional bus configurations within the intended scope of the MOVES model. Buses in this group were the primary focus. The advanced technology group represented bus configurations outside the scope of the MOVES model including all buses with hybrid powertrains and all buses before model year 2007 with diesel particulate filters (DPFs). With the exception of buses equipped with hybrid powertrains, alternative technologies and fuels were not included in the analysis. Biodiesel blends of up to B20 were aggregated into the Diesel fuel type. All data was processed and underwent quality assurance tests using a toolbox developed in MATLAB (Appendix C).  106  Table 4.1 - Emissions measurement data summary. Number of Tests and Vehicles Model Year Group1  Fuel2  3  Emissions Control  Powertrain  Technology Group  4  CO2  NOX  WVU PM  CH4  Veh.5  CO2  TransLink NOX PM  Veh. 5  1990 CNG None Conventional Conventional 12 12 11 0 2 0 0 0 0 1990 Diesel DPF Conventional Advanced 42 42 37 0 4 0 0 0 0 1990 Diesel None Conventional Conventional 98 98 89 0 16 0 0 0 0 1990 Diesel OxCat Conventional Conventional 16 16 16 0 2 0 0 0 0 1991-1993 CNG None Conventional Conventional 156 151 138 11 16 0 0 0 0 1991-1993 CNG OxCat Conventional Conventional 131 130 131 0 11 0 0 0 0 1991-1993 Diesel DPF Conventional Advanced 157 157 152 0 14 0 0 0 0 1991-1993 Diesel None Conventional Conventional 313 312 305 0 38 0 0 0 0 1991-1993 Diesel OxCat Conventional Conventional 47 47 47 0 4 0 0 0 0 1994-1997 CNG None Conventional Conventional 86 86 86 67 13 0 0 0 0 1994-1997 CNG OxCat Conventional Conventional 169 169 169 103 22 0 0 0 0 1994-1997 Diesel DPF Conventional Advanced 40 40 40 11 6 0 0 0 0 1994-1997 Diesel None Conventional Conventional 276 275 243 0 32 0 0 0 0 1994-1997 Diesel OxCat Conventional Conventional 141 140 141 0 21 0 0 0 0 1998 CNG None Conventional Conventional 30 30 30 29 3 0 0 0 0 1998 CNG OxCat Conventional Conventional 44 44 44 43 10 0 0 0 0 1998 Diesel DPF Conventional Advanced 6 6 6 0 1 8 8 0 1 1998 Diesel DPF Hybrid Advanced 18 18 18 0 1 0 0 0 0 1998 Diesel None Conventional Conventional 17 17 17 0 3 0 0 0 0 1998 Diesel OxCat Conventional Conventional 27 27 23 0 5 8 8 0 1 1999-2002 CNG None Conventional Conventional 49 49 49 40 6 0 0 0 0 1999-2002 CNG OxCat Conventional Conventional 37 37 36 33 12 0 0 0 0 1999-2002 Diesel DPF Conventional Advanced 47 47 44 4 11 0 0 0 0 1999-2002 Diesel None Conventional Conventional 81 81 81 8 11 14 14 14 4 1999-2002 Diesel OxCat Conventional Conventional 126 126 114 4 21 0 0 0 0 2003-2006 CNG None Conventional Conventional 49 49 49 48 9 0 0 0 0 2003-2006 CNG OxCat Conventional Conventional 46 46 46 31 7 12 12 12 4 2003-2006 Diesel DPF Conventional Advanced 33 33 33 0 5 23 23 23 5 2003-2006 Diesel DPF Hybrid Advanced 62 62 62 8 4 13 13 11 3 2003-2006 Diesel None Conventional Conventional 59 59 59 35 6 0 0 0 0 2003-2006 Diesel None Hybrid Advanced 4 4 4 4 1 0 0 0 0 2003-2006 Diesel OxCat Conventional Conventional 12 12 12 11 2 0 0 0 0 2007-2009 Diesel DPF Conventional Conventional 0 0 0 0 0 31 31 0 4 2007-2009 Diesel DPF Hybrid Advanced 0 0 0 0 0 8 8 0 1 Conventional 2007-2009 Diesel None Conventional 61 61 61 61 6 0 0 0 0 1 Model Year Groups are based on changes in regulatory emission standards (i.e., model years across which emission standards are the same for all regulated pollutants). 2 Diesel is any diesel fuel regardless of sulfur content or blend with less than 20% biodiesel (B20) and CNG is Compressed Natural Gas. 3 DPF is Passive or Active Diesel Particulate Filter and OxCat is Oxidation Catalyst. 4 Vehicles in the advanced technology group (red) were consider outside the intended scope of the MOVES emissions model and analysed independently of vehicles in the conventional technology group. 5 Number of vehicles.  107  4.2.1.1  West Virginia University Data  The majority (96%) of the emissions measurement data used in this study was chassis dynamometer data obtained from CAFEE/WVU through their Integrated Bus Information System (IBIS) (Wayne, et al.,2011a). CAFEE/WVU has developed an extensive database of emissions measurement tests from heavy-duty vehicles and specifically transit buses (Clark, et al.,2000, Prucz, et al.,2001, Wang, et al.,1997) using two Transportable Heavy Duty Emissions Testing Laboratories (Clark, et al.,1997, Clark, et al.,2003, Clark, et al.,1999a). Only aggregate test measurements of CO2, NOX, THC, CO and total PM emissions were used in this study, as second-by-second data was not generally available. Further, as second-by-second vehicle speed data was also not available, the ideal speed-time traces of the drive cycle were used as the vehicle activity data inputs to MOVES (Appendix C). Data processing and quality assurance tests are detailed in Appendix C. After processing, over 613 hours of data from 2486 tests of over 293 unique buses from CAFEE/WVU were used in this study (Table 4.1).  4.2.1.2  TransLink Data  The remaining emissions measurement data were on-board data obtained from TransLink. M.J. Bradley & Associates LLC (MJB&A) were contracted by TransLink to collect the data in a series of four test phases. Only data from Phases 1 to 3 were publicly available at the time of this study (M.J. Bradley & Associates,2006, 2008, 2009).  Emissions tests in Phase 1 and 2 were conducted on a flat, closed test track. Drivers were instructed to follow set driving cycles developed by MJB&A (Appendix C). Tests in phase 3 were conducted on-road on two bus routes, one flat and one hilly (Appendix C). For all tests in Phases 1 and 2 and for tests on the flat route in Phase 3 the road grade was 0. For the hilly route in Phase 3, GPS data and a Digital Elevation Model (DEM) were used to estimate road grade using the method developed in Chapter 3. Passenger loads were simulated using water barrels in all tests.  In all phases, second-by-second gaseous emissions of CO2, NOX, THC, and CO and vehicle speed were measured using the SEMTECH D/DS PEMS developed by Sensors Inc.. In addition, in Phases 1 and 2 the Dynamic Dilution On/Off-road Sampling System (DOES2) developed by Environment Canada was used to measure aggregate emissions of PM (Ainslie, et 108  al.,1999, Environment Canada,2008). The second-by-second speed measurements and road grade estimates were used as the vehicle activity data inputs to MOVES. Data processing and quality assurance tests are detailed in Appendix C. After processing, over 39 hours of data from 117 tests of 23 unique buses were used from TransLink in this study (Table 4.1 and Table C.8).  4.2.2  Emissions Modeling and MOVES  The predicted emissions were obtained from an implementation of the MOVES2010a emissions model developed in MATLAB. This implementation was developed for several reasons. First, MATLAB provided a convenient framework for automation, which was required in order to process the large amounts of data (over 2500 emissions tests). Second, the run time performance of MOVES is poor, particularly in the context of predicting emissions from specific emission tests 32. Third, the model is poorly documented and by developing and evaluating the MATLAB implementation, the actual algorithms used could be confirmed and validated. Fourth and finally, the ability to predict second-by-second emissions was deemed useful for diagnostic and evaluation purposes.  The explanatory variables used by the MOVES model are listed in Table 4.2. To create a response surface for the MOVES model, the USEPA defined bins based on STP (VSP for nonheavy-duty vehicles) and speed or Operating Modes (OpModes), Regulatory Class, Model Year Group, Fuel Type and for Criteria Air Contaminant (CAC) pollutants (e.g., NOX, PM, CO, and THC), Age Group, and developed base emission rate estimates for each bin and pollutant (Table 4.2, Table C.1) (USEPA,2009a). Some pollutants are “chained to” or derived from other pollutants (Table 4.2). Therefore, errors associated with these pollutants could either be due to errors in the estimate of the parent pollutant or the function (typically a multiplicative conversion factor) that relates the two pollutants.  32  The average run time of MOVES to simulate a single emissions test was on the order of 5 minutes, thus simulating over 2500 emissions tests would have taken in excess of 200 hours. The MATLAB implementation was able to process all emissions tests in less than 10 minutes.  109  Table 4.2 - MOVES explanatory variables for running exhaust emissions. Categories Vehicle Activity11 Vehicle and Energy Source  Total Emissions SOH1 ← VMT2 or Link Length, �������� Speed  Explanatory Variables Base Emission Rates  Correction Factors  OpMode ← STP or VSP ← Speed, Acceleration, Grade Regulatory Class, Model Year Group3, Fuel Type, Age Group4, (Weight5, Engine Size, Engine Technology)6  I/M7, Fuel8, AC9  Temperature10, Humidity10, Altitude10 Energy → CO2; Energy → PM2.5SO4; THC → CH4; PM2.5 OC/EC/ SO4 → PM10OC/EC/ SO4; Chained Pollutants12 PM2.5/10BC, + PM2.5/10EC + SO42.5/10 → PM2.5/10 1 Source Operating Hours (SOH). 2Vehicle Miles Travelled (VMT). 3Model year groups were based on changes in regulatory emission standards. 4Only applies to criteria air contaminants (e.g., NOX, THC, CO, PM) not greenhouses gases (e.g., CO2, N2O) and energy. 5In previous versions of MOVES weight was used as a binning variable but as of MOVES2010a it is accounted for in STP. 6Not supported in MOVES2010a. 7Not applicable to transit buses. 8Fuel composition including sulfur content and biodiesel ester volume. 9Air Conditioning (AC) correction factor is a function of both vehicle activity and the temperature. 10Only affects NOX emissions. 11Variables to the left of the arrows are estimated from variables to the right of the arrows. 12The pollutants to the right of the arrows are chained pollutants, derived from the pollutants to the left of the arrow. See Table C.1for pollutant names. Environment  In MOVES, the total operating time (i.e., SOH) is estimated from the average vehicle speed and depending on the application scale, either the link length or vehicle miles travelled (VMT). The fraction of the total operating time spent in each OpMode (i.e., the OpMode distribution) is estimated from the vehicle activity data. Total emissions for a given vehicle class (e.g., regulatory class, model year group, fuel type) are calculated by multiplying the time spent in each OpMode by the corresponding base emission rate, summing over all OpModes, and then applying any applicable correction factors.  The OpModes (STP - speed bins) are intended to account for the effects of vehicle activity on emission rates. In MOVES2010 the USEPA replaced VSP (tractive power normalized by vehicle weight) with STP (tractive power normalized by a constant scaling factor) (USEPA,2009a, USEPA, et al.,2011b). The advantage of STP is that it maintains the relationship between power and emission rates and can account for variations in vehicle mass, which have been shown to affect CO2, NOX, and PM emissions (Clark, et al.,2007a, Clark, et al.,2003, Frey, et al.,2007, Gajendran, et al.,2003).  Unfortunately, in the current version of MOVES, vehicle weight can only be input by advanced users (i.e., the sourceusetype database table must be modified) and only a single value (for each 110  vehicle use type such as transit buses) is supported per model execution. The USEPA typically adjusts the vehicle weight and several other parameters (in the sourceusetype table) when evaluating the MOVES model (USEPA, et al.,2011c). In this study, actual vehicle test weights were used. However, this may reduce the prediction errors encountered by typical users if actual vehicle weights deviate from the MOVES default value.  The Regulatory Class, Model Year Group, Fuel Type, and Age Group variables are intended to account for the effects of physical and technological differences in vehicles. Two regulatory classes and fuel types are used to model transit buses: the Urban Bus class (48) for diesel buses and the Medium Heavy-Duty (MHD) class for CNG buses. The Model Year Group variable is intended to capture the effect of changes in regulatory emission standards. The Age Group variable is intended to account only for changes in CAC emission rates associated with tampering and mal-maintenance, as emission deterioration rates of heavy-duty engines are typically low (USEPA,2009a). The majority of the tests (77%) in this study were of buses that fell into the first age group (i.e., < 4 years old). As a result, there was insufficient data to verify the tampering and mal-maintenance model developed by the USEPA and prediction errors associated with age effects were not explicitly considered in this study.  Correction factors are typically applied to derive the final estimate of emissions, but none were applicable in this study. Inspection and Maintenance (I/M) corrections are not applied to heavyduty diesel or CNG buses in MOVES (USEPA, et al.,2011a) and fuel correction factors were not applicable. Buses in the TransLink dataset were not equipped with Air Conditioners (AC) and it was assumed that if equipped, they were not operated during the WVU emissions tests, so no AC correction factors were applied. Further, all NOX emission measurements were corrected (e.g., for humidity) and reported for standard conditions and it was assumed that the NOX base emission rates in the MOVES model were also developed based on standard conditions (CFR,2012, McCormick, et al.,1997). Thus, no temperature and humidity correction was required. Therefore, the prediction errors reported in this study were solely associated with the base emission rates and not the correction factors. The MATLAB implementation of the MOVES2010a emissions model developed in this study directly accesses the underlying MySQL database of base emission rates and implements in MATLAB the algorithms necessary to estimate running exhaust emissions at the project level 111  (micro scale) for the pollutants listed in Table C.1 and the transit bus source type. Its development was simplified by the fact no correction factors had to be implemented. To extend the capability of the MOVES model and predict second-by-second emissions, the OpMode and emission rate associated with each second-by-second vehicle activity data point is determined. Total emissions are estimated by summing over all data points. The implementation was tested and evaluated by comparing its output to the output of the MOVES model.  4.2.3  Evaluating the MOVES Model  Two evaluation analyses were carried out in this study. In both cases, the predicted emissions were obtained from the MATLAB implementation of the MOVES model. In the first analysis, aggregate emission test results from both WVU and TransLink were used to evaluate bias and uncertainty in the model. For each emissions test the prediction error of the MOVES model was estimated as 𝑃−𝑀  𝜀 = 100 × �  𝑀  �  Equation 4.1  where ε is the prediction error (%); P is the predicted aggregate test emissions of CO2, NOX, CO, THC, or PM (g); and M is the corresponding measured or observed emissions (g) (Smit, et al.,2010). Results were grouped by data source, pollutant, model year group, fuel type and technology group (Table 4.1). Bootstrap simulation was used to estimate the 95% confidence intervals (CI95) of the mean prediction error as well as the 95% prediction interval (PI95). In both cases the percentile method was used to estimate the confidence intervals from the resulting distributions. Two different sampling strategies were used: one for specific model year groups and one for the All model year group. For specific model years, a total of 10000 bootstrap samples were constructed by randomly selecting one emissions test result from each vehicle in the model year group. In this way each vehicle was weighted equally. For the All model year group, 10000 bootstrap samples were constructed by randomly selecting 100 emissions test results from each model year. In this way each model year was weighted equally. In the second analysis, second-by-second data from the TransLink dataset was used to compare measured and predicted emission rates (g⋅s-1) of CO2, NOX, CO and THC within OpMode bins, which are identified by OpModeIDs (Table C.3).  112  4.3  Results and Discussion  Prediction errors of the MOVES model were estimated for key pollutants from diesel (CO2, NOX, and PM) and CNG (CO2, NOX, PM, THC and CH4) transit buses that impact climate and health using aggregate emissions test data from two sources (WVU and TransLink). Second-bysecond data from the TransLink dataset was used to explore sources of prediction error. All plots are color coded by data source: blue indicates the data is from TransLink, black indicates the data is from WVU, red indicates the data is from both TransLink and WVU, and green indicates the data is from the MOVES model.  4.3.1  Prediction Errors  Prediction errors are shown in Figure 4.1 for diesel buses in the conventional technology group and Figure 4.2 for CNG buses. Prediction errors for buses in the advanced technology group are discussed in Appendix C. The mean prediction errors represent the bias, the 95% confidence intervals (CI95) represent the uncertainty in the prediction error when considering a fleet of buses (either from a single model year group or all model year groups), and the 95% prediction interval (PI95) represents the inter-vehicle variability and the uncertainty in the prediction error when considering a single bus (USEPA,2002c). In most applications, the goal is to predict emissions from a fleet of vehicles, thus the uncertainty described by the 95% confidence intervals was the primary focus here.  Over all model years of diesel buses in the conventional technology group, MOVES underestimated total fleet emissions of CO2 by 13% (±1.2%) and overestimated emissions of NOX and PM by 37% (+3.6/-3.7%) and 268% (+65/-36%) respectively (Figure 4.1). The mean prediction error varied across model year groups, particularly for PM emissions. Further, the mean prediction errors and confidence intervals were also significantly greater and more skewed for PM emissions than other pollutants. Upward skew in the confidence intervals of PM emissions suggests the presence of high emitting buses in the sample.  113  a)  b)  CO2  40  200  1500 100  0  1000  50 -20  0  -40  500  All 2  2007-Later 1  1998-20031 2004-2006 2007-Later  1993  1994-1997 1998-2003  1990  1991-1992  All 2  2007-Later 1  1998-20031 2004-2006 2007-Later  1993  1994-1997 1998-2003  0  1990  -100  1991-1992  All 2  2007-Later 1  1998-20031 2004-2006 2007-Later  1993  1994-1997 1998-2003  1991-1992  -50  1990  -60  PM  2000  150  20  Prediction Error (%)  c)  NOX  Figure 4.1 – Mean prediction errors for CO2 (a), NOX (b), and PM (c) emissions from diesel buses in the conventional technology group by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval.  Over all model years of CNG buses, MOVES underestimated total fleet emission of CO2, NOX, and THC by 22% (±1.2%), 56% (+3/-2.9%) and 87% (+1.5/-1.2%) respectively (Figure 4.2). Further, CH4 and PM emissions were underestimated by 99.7% (±0.0%) and 53% (+46/-15%) respectively (Figure C.11). Unlike diesel vehicles, THC emissions from CNG vehicles can be significant due to high CH4 emissions (Hesterberg, et al.,2008); however, MOVES grossly underestimated both THC and CH4 emissions (Figure 4.2c Figure C.11a). The reason for the error in THC emissions is not clear; however, there are at least two possible explanations for the error in CH4 emissions. First, CH4 emissions are derived from THC emissions based on a multiplicative factor: the CH4 to THC ratio. Thus, errors in THC emissions are transferred to CH4 emissions. Second, it appears that CH4 to THC ratios for diesel vehicles were used for CNG vehicles. Typical CH4 to THC ratios for CNG vehicles reported in the literature are on the order of 90% (Clark, et al.,1997, McTaggart-Cowan, et al.,2006), whereas the average ratio used in the MOVES model is on the order of 2%.  114  a)  b)  CO2  40  c)  NOX  150  THC  0 100 -20  -60  -100  -100  -40  All 2  2004-20061  2004-2006  1998-2003  1994-1997  1993  1991-1992  1990  All 2  2004-20061  2004-2006  1998-2003  1994-1997  1993  1991-1992  -60  1990  All 2  -80  2004-2006  -50  2004-20061  -40  1998-2003  0  1993  -20  1994-1997  50  1991-1992  0  1990  Prediction Error (%)  20  Figure 4.2 –Mean prediction errors for CO2 (a), NOX (b), and THC (c) emissions from CNG buses by model year group. 1TransLink data (blue), otherwise WVU data (black). 2All model year groups and datasets (red). Solid error bars indicate the 95% confidence interval and dashed error bars indicate the 95% prediction interval.  For both diesel and CNG buses, the results were generally consistent across the two datasets, with the exception of CO2 emissions from diesel buses in the 2007-Later model year group (Figure 4.1a) and NOX emission from CNG buses in the 2004-2006 model year group (Figure 4.2b). Although the magnitudes of the mean prediction errors for CO2 and NOX emissions were comparable between diesel and CNG buses, the MOVES model under predicted emissions of all pollutants from CNG buses but over predicted emissions of all pollutants from diesel buses except CO2. The mean prediction errors were comparable to those reported in a major review of emissions model evaluation studies (Smit, et al.,2010). In that study, the mean prediction errors for CO2, PM, and THC and NOX were reported to be within a factor of 1.3, 3-5, and 2 of the measured values, respectively. Thus with respect to transit bus emissions, MOVES does not appear to perform significantly better than other models, including less detailed models. In fact, in the case of THC and CH4 emissions from CNG buses, its performance is poor. Potential causes of the prediction errors are discussed further in the following sections.  115  4.3.2  OpMode and Second-by-Second Emission Rates  Measured and predicted emission rates of CO2 and NOX for each OpMode are shown in panels (a) and (d) in Figure 4.3 for 2007 model year diesel buses in the conventional technology group. These panels also show the time spent in each OpMode (i.e., the OpMode distribution). Panels (b) and (e) show the contributions from each OpMode to the prediction errors. The sum of the contributions from each OpMode is equal to the prediction error shown in Figure 4.1. In addition, the first 200 s of second-by-second data are shown in panels (c) and (f). Similar figures for 1998 and 2001 diesel buses in the conventional technology group as well as 2005-2006 CNG buses are provided in Appendix C (Figure C.14, Figure C.15, and Figure C.16). These figures provide valuable diagnostic insights into the causes of the predictions errors, but can only be created if second-by-second data is available (i.e., for the TransLink dataset).  The figures show that the prediction errors were a function of the time spent in each OpMode and the absolute error (i.e., the difference between predicted and measured) in emission rates associated with each OpMode, both of which varied across OpModes (e.g., Figure 4.3). For example, although the absolute error in OpMode 1 (idle) was generally small, the time spent in OpMode 1 was relatively long (Figure 4.3a). As a result, the contribution to the prediction error of OpMode 1 and idle emissions was significant (Figure 4.3b). Further, the figures show that the contributions from each OpMode to the prediction errors were not equal, and in some cases positive contributions in one OpMode cancelled out negative contributions from another OpMode (i.e., two wrongs made a right). This suggests that there may be greater value in reducing the bias and uncertainty of the emission rates of some OpModes (e.g., idle) compared to others.  In addition, the figures reveal that predicted and measured CO2 and NOX emission rates significantly deviated in OpModes 12-16 (cruise/acceleration) for 1998 and 2007 diesel buses, resulting in an unexpected arc-like shape to the response surface over these OpModes (e.g., Figure 4.3a). This did not occur for 2001 diesel buses and CNG buses and there was a more linear relationship between the measured and predicted emission rates over these OpModes (Appendix C). These deviations increased the magnitude of the prediction errors for CO2 emissions, for example diesel buses in the 2007-Later model year group from the TransLink dataset, but decreased the magnitude of the prediction errors for NOX emissions (Figure 4.1a, b). 116  These deviations were unexpected because: (a) both CO2 and NOX have been shown to be strongly linearly correlated with power (Clark, et al.,2003, Clark, et al.,2002, Zhai, et al.,2008) and (b) STP increases linearly over OpModes 12-16 and 22-25 (Table C.3). Thus, emissions rates would also be expected to follow a linear relationship over these OpModes. The non-linear relationship over OpModes 12-16 found for 1998 and 2007 diesel buses from the TransLink dataset suggests a problem with this data. The second-by-second data in panels (c) and (f) and the additional analysis discussed in Appendix C show that these inconsistencies occurred during acceleration and gear shifts. This suggests that the problem was likely caused by either erroneous or misaligned exhaust flow data. MJB&A noted several challenges with obtaining reliable exhaust flow data, which further supports this hypothesis. Note that these problems were only associated with TransLink data from Phase 3, which represents a very small fraction of the total data and thus would not affect the mean prediction errors of the All model year group. Also, these problems were only detected as a result of these figures (e.g., Figure 4.3a), demonstrating their diagnostic value.  117  b)  40 30 20 10 0 0  1 11 12 13 14 15 16 21 22 23 24 25 27 28 OpModeID  40  0  -5  -10  0 1 11 12131415 16212223 24252728 OpModeID  60 40  Predicted Speed  20 0  0  20  40  60  20 80  100 Time (s)  120  OpMode Distribution  0.4 0.3 0.2 0.1 0 0  160  180  NOX  10  Total = -1.94%  5  0  -5  NOX  0.15  60  Measured  0.1  40  Predicted Speed  0.05 0  20  40  60  0 200  0 1 11 12131415 16212223 24252728 OpModeID  1 11 12 13 14 15 16 21 22 23 24 25 27 28 OpModeID  f)  0  140  e)  NOX  0.5  Emission Rate (g⋅s -1)  Total = -28.3%  Measured  d)  Emission Rate (g⋅s -1)  5  CO2  60  Prediction Error Contribution (%)  Emission Rate (g⋅s -1)  c)  CO2  Speed (kph)  50  OpMode Distribution  Emission Rate (g⋅s -1)  60  Prediction Error Contribution (%)  CO2  20 80  100 Time (s)  120  140  160  180  Speed (kph)  a)  0 200  Figure 4.3 - (a,d) Measured (blue box plots) and predicted (green points) CO2 and NOX emissions rates by OpMode for 2007 diesel buses in the conventional technology group from the TransLink dataset. Upper, middle, and lower horizontal lines of the boxes indicate the 75th, 50th (median), and 25th percentiles. Whiskers extend to the furthest data point within 1.5 times the interquartile range. Diamonds indicate the mean. (b, e) Contributions to the total prediction error from each OpMode. (c, f) Sample of the first 200 s of second-by-second predicted and measured CO2 and NOX emission rates as well as speed. All other panels show all data.  118  4.3.3  Sources of Prediction Error Bias and Uncertainty  Biases (mean prediction errors) result from systematic differences between measured and predicted emissions, for example, as a result of an unrepresentative data sample. Uncertainties arise from random sampling error, model uncertainty, and uncertainty in the explanatory variables.  Further, both prediction error bias and uncertainty may arise if the errors in the OpMode emission rates are not proportional (i.e., the relative errors are not equal) across all OpModes. In such cases, the prediction errors would be affected by the vehicle activity and the OpMode distribution. For example, the greatest prediction error would occur when all time was spent in the OpMode with the largest relative error. The second-by-second data from the TransLink dataset suggested that the errors in emissions rates were not proportional across OpModes (e.g., OpMode 1 vs. OpMode 2). This likely contributed more to bias then to uncertainty because nearly 70% of the WVU data and the TransLink data from Phase 1 and 2 were based on the Central Business District (CBD) drive cycle or similar drive cycles (i.e., the OpMode distribution were similar across tests) (Appendix C).  4.3.3.1  Explanatory Variables  Both the uncertainty in the prediction errors (e.g., Figure 4.1) and the variability of the emission rates within OpModes (e.g., Figure 4.3a) means that the explanatory variables employed by the MOVES model did not explain all of the inter- and/or intra- vehicle variability in emission rates (Table 4.2). Although this was expected (explaining all of the variability in vehicle emissions is simply not realistic), the choice of explanatory variables may have contributed to this model uncertainty. For example, with regards to model year groups, the groups used by the USEPA in the development of the base emission rates were based on emission standards for heavy-duty vehicles (USEPA,2009a, USEPA, et al.,2011b), which are slightly different than standards for urban buses (Table A.2). Further, within some model year groups, emissions certification levels can vary significantly. For example, NOX emission standards for model years 2007-2010 were phased in and the certification levels vary by over an order of magnitude (2.5 – 0.2 g⋅bhp-1⋅h-1). Without an explanatory variable to account for this, significant uncertainty in the predictions of NOX emissions from this model year group would be expected.  119  OpModes and STP are functions of speed, acceleration and road grade. Errors in these explanatory variables could result in misclassification of the OpMode and prediction errors. For example, in this study, misclassification might have occurred because the ideal speed-time traces were used instead of actual measured speed for the chassis dynamometer tests. Although, given the size of the OpMode bins used in MOVES and the range of values that map to a single OpMode bin, this is unlikely to have significantly affected the results (Table C.3). However, this points to an additional source of model uncertainty: the size of the OpMode bins. Uncertainty arises because a range of operating conditions and emission rates can potentially map to a single OpMode bin. Increasing the number of bins would potentially reduce the uncertainty but would also increase the level of detail and cost associated with developing and running the model. Presumably, the choice of the OpMode bin size in MOVES reflected this trade-off.  The fact that the explanatory variables do not explain all of the variability in emissions means that the predictions errors would be affected by differences between the dataset used to develop MOVES and the one used here to evaluate it.  4.3.3.2  Datasets and Base Emission Rates  As no correction factors were applicable, the prediction errors reported were a result of: (a) errors in the factors used to derive chained pollutants and/or (b) differences between the base emission rates of the MOVES model and those reflected by the measured data.  In MOVES, CO2 emissions are estimated from energy consumption based on the carbon content and oxidation fraction of the fuel (USEPA,2005). As a result, prediction errors in CO2 emissions could have been due to either errors in these factors or in the base energy consumption rates. The latter is most likely. All PM species are derived from PM2.5 with the exception of sulfate (SO4) which is derived from energy (Table 4.2). In this study, measured total PM was compared to predicted PM10. Because the ratio of PM10 to PM2.5 is near unity (1.03) and SO4 is typically a small fraction of PM (<5%) for vehicles not equipped with DPFs, PM prediction errors were most likely associated with the PM2.5 base emission rates developed by the USEPA. Thus, for all pollutants the prediction errors were most likely a result of differences between the base emission rates of the MOVES model and those reflected by the measured data.  120  In general, it was neither possible nor feasible for the USEPA to derive base emission rates from test data for all explanatory variable combinations (i.e., bins). Several techniques and simplifying assumption were employed by the USEPA to address this problem; however, these assumptions and techniques likely contributed to the prediction errors. Unfortunately, with the exception of NOX, PM, CO, and THC emissions from heavy-duty diesel buses, documentation describing how the USEPA derived base emission rates was limited or not available (USEPA,2009a, USEPA, et al.,2011b). Thus, in many cases it was only possible to speculate on the cause of the prediction errors. For example, it appears that the base emission rates for pollutants classified as greenhouse gases (e.g., CO2) and those classified as CAC (e.g., NOX) were developed in independent processes (Table C.1) (USEPA,2005, 2009a, USEPA, et al.,2011d). This may explain why CO2 emissions for diesel vehicles are under predicted but other pollutants are over predicted (Figure 4.1a). It also appears that base emission rates for CNG buses were adopted from non-transit vehicles in the medium heavy-duty (MHD) regulatory class, which may explain the prediction error biases for CNG buses (Figure 4.2). Base emission rates for PM2.5 were also developed from non-transit vehicles due to the significant challenges in obtaining high resolution PM data (USEPA,2009a). No data was available for any heavy-duty diesel vehicles from model year 2007 onward. For these model years, the USEPA assumed that PM base emission rates were 96% lower than 2006 model year vehicles (USEPA, et al.,2011b). These simplifications, while likely necessary, may explain the large prediction error biases and uncertainties for PM emissions (Figure 4.1c). In contrast to other pollutants, NOX base emission rates were developed using on-board emissions test data from actual transit buses, but the number of vehicles was limited (less than 50 vehicles spanning only 3 model year groups) (USEPA,2009a). Thus, it appears there is limited empirical basis for the base emission factors of transit buses developed by the USEPA, which precludes any comparative discussion on the relative representativeness of the dataset used to develop MOVES and the one used here to evaluate it.  Although it was not possible to compare datasets, it is important to point out one potential limitation of the dataset used in this study. The dataset consists primarily of chassis dynamometer data. The real-world representativeness of this data has been questioned and it has been suggested that it may underestimate real-world emissions (National Research Council,2000). There was insufficient real-world, on-road data from the TransLink dataset to examine this hypothesis; however, even if it were true, it would only result in smaller prediction 121  errors for NOX and PM emission from diesel buses. The prediction errors for CO2 and all pollutants from CNG buses would increase. Further, any prediction error biases resulting from the use of chassis dynamometer data are likely to be small because: (a) the use of the OpMode explanatory variable mitigates the issue of representativeness and (b) the USEPA has also used chassis dynamometer data in the development of base emission rates for other vehicle types including those the transit bus base emission rates were developed from (USEPA,2009a, USEPA, et al.,2011b).  4.3.4  Correcting Prediction Error Bias and Uncertainty  It appears that the errors in the base emission rates may have stemmed from a lack of empirical data. This raises the question of whether the dataset developed in this study could be used to reduce the prediction errors associated with transit buses. It is both difficult to develop and to address errors in the base emission rates of the OpModes with only aggregate emission test data (USEPA,2002c). Generally, second-by-second data is required. However, it may be possible to use aggregate data to reduce the mean prediction error (bias) by applying a multiplicative and/or additive correction factor across all OpModes. This would be most effective in cases where the bias or a component of the bias in emission rates was equal or proportional across OpModes.  Uncertainties due to random sampling errors are functions of the sample size and the unexplained inter-vehicle variability. Thus the uncertainty decreases with increasing sample (or fleet) size. However, the unexplained inter-vehicle variability represents model uncertainty which cannot be reduced without modifying the MOVES model (e.g., adding additional explanatory variables) and can only be captured as uncertainty using a stochastic model. Although provisions have been made to support uncertainty estimation in MOVES, it is not currently supported in MOVES2010a (National Research Council,2000, USEPA,2009b, 2011b). However, the uncertainties estimated here, represent top-down estimates that could be used to describe the uncertainty in the model (USEPA,2002c).  In addition, several other actions could reduce prediction error bias and uncertainty, including: improved support for the input of vehicle mass; use of appropriate model year groups for urban transit bus; and use of an appropriate THC to CH4 ratios for CNG buses.  122  4.3.5  Implications and Recommendations  Significant biases in the predictions of key pollutants from transit buses that impact both climate and health were found. With respect to transit buses, the performance of MOVES does not appear to be substantially better or worse than other models (Smit, et al.,2010). While this may be viewed as disappointing, given the substantial assumptions made by the USEPA in the process of developing the base emission rates (i.e., the limited use of empirical data), the biases were not unexpected or necessarily unreasonable. In fact, the assumptions may have been justifiable given the costs involved in obtaining data and the fact that transit buses represent a small fraction of the total on-road vehicle population (i.e., from a value of information perspective). However, this means that with respect to transit buses, there is likely significant potential to reduce the biases and improve the model’s performance.  The biases found are unlikely to significantly affect regional or national emissions inventories estimated using the MOVES model because transit buses represent a small fraction of the total on-road vehicle population. However, the biases would likely have a significant impact on emission inventories of transit bus fleets and project level analyses where transit buses represented a significant fraction of the vehicle population. In these circumstances, the biases reported here should be taken into consideration when interpreting the results of the model. In particular, the biases should be carefully considered if the model is used to evaluate the relative performance of different bus technologies, specifically diesel and CNG buses, as the model currently presents a more favorable picture of CNG buses and a less favorable picture of diesel buses than is indicated by the empirical data used in this study. Thus, although addressing these biases may not currently be of value to the USEPA, it would be of significant value to transit agencies.  Evaluation and validation of the model was hindered by poor or limited documentation. In some cases the actual software had to be examined in order to ascertain how the model functioned. One challenge with the current documentation of the model is that it is spread over multiple documents and there is no master document describing what documents describe the various components of the model. Further, it was not always clear what version of the model the documentation applied to. Ideally, there would be a single document that is updated with each version of the model. 123  The methodology developed in this study demonstrates how both aggregate and second-bysecond emission test data can be applied to evaluate the MOVES model. It could be extended to evaluate other vehicle types. However, without the use of custom interfaces (e.g., the MATLAB implementation developed in this study), the runtime performance of the model may be a barrier to performing large scale evaluations. It is difficult to speculate on the prediction errors of other vehicle types. Transit buses may not have been a priority to the USEPA, thus the biases found in this study may not be representative of the biases associated with other vehicle types. Nevertheless, further evaluation studies are required to support the ongoing development and improvement of the MOVES model.  124  Chapter 5: Conclusion Public transportation has been widely promoted as a means of increasing the sustainability of urban transportation systems; however these systems also have adverse impacts. Further, although transit agencies are making efforts to address these impacts, the assessment tools and mitigation options available to them are limited. The goals of this dissertation were to explicitly address these impacts and develop tools and operational control strategies to aid public transit agencies in increasing the sustainability of the bus fleets they operate. Specifically, the objectives were: a) to explore, develop, and evaluate methods of quantifying the climate and public health impacts of the emissions from heavy-duty transit bus fleets (i.e., model the impact pathways); and b) to develop the first iteration of an integrated assessment model of the operations of transit bus fleets that accounts for climate, health, and operating costs and impacts and to employ it to quantify the potential of vehicle scheduling optimisation to mitigate these impacts and costs. 5.1  Contributions  This dissertation was motivated by the real-world challenges faced by transit agencies and four key hypotheses (Table 1.4). It expands and improves upon the tools, models, and control strategies available to transit agencies to quantify and mitigate the climate and health impacts of emissions from their bus fleets. The research in this dissertation not only identifies the limitations of traditional approaches and existing models, but in many cases quantifies the magnitude of the discrepancies resulting from the limitations. Throughout the research process, significant efforts were made to engage and collaborate with transit agencies (e.g., TransLink), industry (e.g., GIRO), and other researchers (e.g., West Virginia University) in order to ensure the research in this dissertation was current, relevant to stakeholders and implementable. The operational optimisation approach as formulated in Chapter 2 has not been previously considered and provides fiscally constrained transit agencies with a potentially low cost approach of mitigating the impacts of their fleets. It demonstrates that operational control strategies such as vehicle scheduling optimisation can reduce the climate and health impacts of emissions from transit bus fleets as well as operating costs when incorporated in an integrated 125  assessment framework (Hypothesis H3). This work highlights and quantifies the potential consequences of considering only single impacts or objectives when evaluating control strategies. Further, it demonstrates how cost-benefit analysis can be applied to evaluate the trade-offs between conflicting objectives and to identify an optimal control strategy. In addition, Chapter 2 quantitatively shows that regional/macro scale assessments of exposure and health impacts, specifically those based on emissions inventories, are biased because they do not account for the intra-regional spatial relationship between emissions and populations (Hypothesis H2). Thus, while regional/macro scale modeling approaches are likely suitable for assessing climate impacts, if there is intra-regional variability in the population density and pollutant concentrations, they are likely not suitable for assessing health impacts. Chapter 2 is one of a small number of studies that has actually quantified the implications of modeling exposure on an intra-regional scale and accounting for variability in emissions and population density. This finding has implications beyond the context of this dissertation and supports arguments made by many prominent researchers for a better accounting of the spatial and temporal relationships between people and emission sources (Jerrett, et al.,2005a, Nazaroff,2008).  In addition to the above academic contributions, the results and methodology developed in Chapter 2 are being considered by several transit agencies. The impact pathway models have been integrated into a commercial transit scheduling software package (HASTUS) developed by GIRO Inc. The HATUS software was used to perform a complete vehicle scheduling optimisation (i.e., beyond the assignment optimisation performed in Chapter 2) for one depot in TransLink’s system and the results are under review by TransLink.  Chapter 3 is one of the first studies to highlight the dual effects of vehicle activity on the spatial distribution of emissions, namely its effects on emission rates as well as its effects on the time in space of a vehicle. Most previous studies, such as Unal, et al. (2001), have only considered the effects on emission ratesUnal, et al. (2001)Unal, et al. (2001). Further, no previous studies have quantified the implications of the spatial coincidence of emissions and populations on exposure at the micro scale. Finally, Chapter 3 demonstrates that macro scale emissions modeling approaches, specifically those based on distance-based emission factors, are biased because they do not fully account for the effects of vehicle activity on emissions (Hypothesis H1). Chapter 3 also provides additional support for Hypothesis H2. 126  The results of Chapter 3 may be particularly important as models of population time-activity patterns improve, and are relevant to analyses of busy transportation corridors such as the one in which the 99 B-Line bus route operates. Further, the results can be used to understand the limitations of macro and meso scale models and provide insights into the value of information (e.g., vehicle activity data) and modeling at higher resolutions and levels of detail. Finally, the results of this study were considered by TransLink as part of an ongoing initiative to develop a new rapid transit solution along the 99 B-Line bus route, as the current system is at capacity.  Chapter 4 is one of the first evaluation studies of the MOVES model. It demonstrates that models that have not been thoroughly evaluated may be biased (Hypothesis H4) and reveals significant biases in the MOVES model. Although it only addressed predictions of transit bus emissions, it demonstrates a method of evaluating the model against a large dataset. Further, given the significant biases found and the economic and public health policy implications of the MOVES model, the results of Chapter 4 highlight the need for further evaluation studies. The MATLAB implementation of the MOVES model developed in Chapter 4 could be extended to support the evaluation of other vehicle types. In addition, the toolboxes developed to process the emission test data can be used to process other sources of emissions tests data necessary to evaluate the MOVES model.  5.2  Findings and Recommendations  This section reviews the major findings of this dissertation and provides recommendations with respect to modeling and mitigating the climate and health impact of emissions from public transportation bus fleets. With regards to mitigation, this dissertation primarily focused on operational control strategies. However, the findings of this dissertation can also be applied to evaluate capital control strategies.  5.2.1  Scale and Level of Detail  In developing a model, a compromise must be struck between the level of detail, determined in part by the scale, and the costs and feasibility (e.g., computational complexity and data requirements) of developing and running the model. From the perspective of a decision maker, 127  increasing the level of detail only has value if it has a bearing on the decision being made 33. For example, reducing the uncertainty in a parameter is only of value if it has the potential to influence a decision; although there may be value from other perspectives, such as learning. Further, as discussed in Chapter 2 and Chapter 3, there is generally no value in modeling at scales or resolutions where there is no variability in the factors of interest. For example, if either emissions or population density along a bus route were spatially homogenous, no further knowledge regarding exposure would be gained by modeling at a higher spatial resolution. Thus, the scale and level of detail of a model should reflect the scale of variability of the factors of interest as well as the value of the information.  With regards to modeling the health impact pathway and specifically exposure, traditional macro/regional scale emissions inventory based approaches effectively average the population and emissions over a region and assume that both are spatially and temporally homogeneously distributed. Thus, these approaches cannot account for the intra-regional spatial and temporal relationships between emissions and populations. This limitation has been shown to underestimate health effects and exposure estimates (Chapter 2 and Chapter 3) (Health Effects Institute,2010, Jerrett, et al.,2005b, Setton, et al.,2011). For example, in Chapter 2, macro scale estimates were found to underestimate exposure by 9-11% relative to meso scale estimates, and in Chapter 3, meso scale estimates were found to underestimate exposure by 0-11% for residential populations and by 21-35% for near-road pedestrian populations compared to micro scale estimates. While these differences were generally small, they did impact the vehicle scheduling/assignment optimisation solutions and may represent significant absolute costs when monetised. Thus, in the context of the operational decision making of transit agencies, there is likely value in considering exposure on an intra-regional scale, as differences in the exposure potentials between bus routes can be exploited by the vehicle scheduling optimisation to realise additional reductions in exposure and health impacts (Chapter 2). In general, traditional regional/macro scale emissions inventory based approaches are likely not appropriate for quantifying exposure and health impacts if there is significant intra-regional variability in the pollutant concentrations and population density; however, there would be no advantage to an intra-regional scale approach if the variability did not exist or could not be quantified.  33  Formally, this is defined by the value of information (Clemen, et al.,2004).  128  Although an intra-regional scale approach should be considered as outlined above, the precise resolution and level of detail is difficult to specify and is dependent on the specific application and data available. In Chapter 2, the modeling was done at the route level and only the variability in exposure between routes was considered. However, in Chapter 3 within route variability in emissions and population density was shown to effect exposure, particularly when considering near-road populations. Thus, although it is not practical to model an entire transit system at the micro scale, for example at a spatial resolution of 50 m as was done in Chapter 3, an intermediate level of detail where the analysis is conducted on segments of bus routes or road links, as was done by Greco et al., may be most appropriate (Greco, et al.,2007a). However, this would only be of value if the estimates of both emissions and population density exhibited variability over the route, which would not be the case if an emissions factor model was employed or if block level census data was not available. Ultimately, a judgement must be made that reflects the variability of the parameters, the costs and availability of data, and the decision being made when determining the scale and level of detail.  With regards to modeling the climate impact pathway, traditional macro scale emissions inventory based approaches are likely appropriate. The model of the climate impact pathway developed in this dissertation did not involve intra-regional scale processes. Further, although not specifically addressed in this dissertation, secondary pollutants that impact the climate such as ozone are generally associated with regional scale processes.  Finally, in the case of both impact pathways, the level of detail should be sufficient to account for the effects of vehicle activity on emissions. This is discussed in detail in the following two sections.  5.2.2  Vehicle Activity  In the context of modeling transit systems, the vehicles and the routes travelled are usually known. This greatly simplifies the analysis, as it is feasible to account for the activity of individual vehicles. Although the specific measures of vehicle activity used in an analysis depends on the application and emissions model employed, the findings of this dissertation have two important implications related to quantifying vehicle activity. First, vehicle activity measures describing mode are likely necessary to accurately estimate vehicle emissions. Second, the time and location of emissions and thus vehicle activity have the potential to affect 129  estimates of exposure. Thus, there is a need for vehicle activity data that can describe mode and a need to know when in time and where in space vehicle activity occurs.  Distance-based measures of activity can be obtained from vehicle odometers or estimated using GIS. Odometer measurements have no position information, so cannot describe where the activity occurred. As a result, odometer data can only be used to estimate emissions inventories on a regional/macro scale. However, GIS by definition provides position information and thus can be used to model emissions on intra-regional scale, as was done in both Chapter 2 and Chapter 3 (Bachman, et al.,2000).  For vehicle activity measures such as speed, acceleration, and road grade that are useful in characterising mode, both the vehicles’ on-board sensors (e.g., vehicle speed sensor) and GPS receivers can be used to measure activity at sampling rates of up to 1.0 Hz. As with odometer data, the measurements made by a vehicle’s on-board sensors have no position information. GPS is the only activity measurement method (reviewed here) that also measures position. Many bus fleets are being equipped with GPS in efforts to improve service and fleet management, and it may be possible to leverage these systems as a source of activity data, although this was not examined in this dissertation. GPS methods do, however, have limitations. Although the performance of the GPS receiver in this research was found to be excellent, performance is known to be degraded in urban areas due to interference from tall buildings (Chapter 3). Thus, more detailed evaluations should be conducted before these systems are used in such environments 34.  In order to make decisions, including evaluating control strategies, it is also necessary to predict future vehicle activity. Distance is relatively easy to predict based on the pre-determined bus routes and timetable, as was done in Chapter 2. However, other measures of vehicle activity, including speed, acceleration, and road grade, are necessary to describe the mode (Chapter 3). With the exception of road grade, which was estimated using a digital elevation model, predicting these measures was not specifically examined in this dissertation, but there are a number of possible approaches. In Chapter 3, vehicle speed and acceleration were shown to exhibit a consistent spatial pattern along the bus route studied. This suggests that physical 34  Inertial navigation systems and map-matching techniques may be used to improve performance (Farrell, et al.,1999, Quddus, et al.,2007).  130  features of the urban environment such as bus stops and intersections could potentially be used to predict these measures 35. It is also conceivable to collect a sample of vehicle activity data along a potential future route. As previously discussed, it is not necessary to predict second-bysecond vehicle activity, only to estimate the (joint) distributions of vehicle activity over the spatial and temporal scales of the analysis 36 (e.g., hourly over 500 m road segments).  5.2.3  Emissions  A number of emissions modeling approaches and models could be used to estimate emissions from transit bus fleets. This choice is influenced by the same factors that influence the choice of vehicle activity quantification approaches. First, the modeling approach should effectively account for the effects of vehicle activity. Second, the modeling approach should have the ability to model the temporal and spatial distribution of emissions on an intra-regional scale.  Numerous studies have shown that vehicle emissions are influenced by vehicle activity, and specifically, mode (Chapter 3) (Barth, et al.,1996a, Clark, et al.,2002, Zhai, et al.,2008). However, the technique employed by emission factor models to account for the effects of vehicle activity involves developing a single, representative vehicle activity pattern (e.g., drive cycle), which is difficult to accomplish in practice. Further, average speed, employed in models such as MOBILE, may not fully explain the change in emissions due to deviations from this activity pattern. As a result, the averaging techniques employed by emission factor models may introduce biases in the emissions predicted by these models. For example, in Chapter 3 the MOBILE emission factor model was found to underestimate emissions of NOX by 24-19% when compared to a modal emissions model. However, average activity emission factor model such as IBIS are new and have not yet been extensively evaluated. It is possible that this approach may have advantages over average speed emission factor models, but it was not possible to evaluate this as IBIS was still in development.  Modal emissions models such as MOVES were developed in part to address the limitations of emission factor models. Instead of averaging techniques, modal emissions models employ aggregation techniques that do not rely on assumptions regarding vehicle activity patterns. 35  For example, using techniques similar to land use regression modeling (Hoek, et al.,2008). For example, if the MOVES emissions model was used this would mean estimating the OpMode distribution based on speed, acceleration, and road grade. 36  131  Typically, this is accomplished by describing the vehicle activity data using distributions of the time spent in each mode. The trade-off is that in order to realise the benefits of modal models, detailed vehicle activity data is required (e.g., second-by-second speed, acceleration, and road grade). If no vehicle activity data is available to characterise mode, the advantages of employing a modal model over an emission factor model are limited (as was the case in Chapter 2). In both cases, assumptions regarding vehicle activity and mode would have to be made. One possible advantage of employing a modal model is that it would allow for the effects of different assumptions regarding vehicle activity and mode to be quantified using uncertainty and/or sensitivity analysis (similar to what was done in Chapter 3).  With regards to modeling emissions on an intra-regional scale, both modal and emission factor models could be used (e.g., Chapter 2 and Chapter 3). However, modal models offer more flexibility and support a wider range of spatial scales and resolutions. The spatial resolution of a modal model is typically only limited by the spatial resolution of the vehicle activity data, while it is only realistic to apply an emission factor model over a link several kilometers long. Thus, if modeling at a micro scale a modal model and high resolution vehicle activity data are essential (e.g., Chapter 3). At the meso scale (e.g., bus route level) the advantages of a modal model over an emission factor model are more nuanced and primarily depend on the model’s ability to account for the effects of vehicle activity.  In summary, despite the added level of detail and computational complexity of a modal model, these models offer a more flexible framework that can support multiple scales of analysis as well as incorporate new vehicle activity information as it becomes available and/or quantify the implications of assumptions regarding vehicle activity. Several emissions models could be used to predict exhaust emissions from heavy-duty transit buses, including the basic emission factor model developed in Chapter 2, MOBILE, MOVES, CMEM, and IBIS (Table 1.9). As discussed above, modal models such as MOVES and CMEM may have advantages over emission factor models. In particular, MOVES has several advantages over the other models considered including: (a) support for multiple scales of analysis; (b) support for a wide range of emission processes and pollutants, including PM; and (c) the potential to model uncertainties. Thus, as CMEM does not currently support transit buses, IBIS is still under development, and MOBILE has several significant limitations, MOVES 132  may be the most suitable model for assessing the impacts of transit bus emissions. However, the evaluation performed in Chapter 4 raises questions about the accuracy and performance of the model.  For diesel buses, MOVES underestimated emissions of CO2 by 13% and overestimated emissions of NOX and PM by 37% and 268% respectively. For CNG buses, MOVES underestimated total emission of CO2, NOX, THC, and PM by 22%, 56%, 87% and 53% respectively. Although the evaluation performed in Chapter 4 had some limitations (e.g., the use of chassis dynamometer data rather than real-world data), the biases in NOX, THC, and PM emissions were significant and suggest that with respect to transit buses, MOVES needs further calibration. Perhaps most concerning and relevant from a decision-making perspective is the fact that emissions were generally overestimated for diesel buses but underestimated for CNG buses. Thus, MOVES unfairly presents a more favorable picture of CNG buses and a less favorable picture of diesel buses. If MOVES had been used in Chapter 2, the biases might have affected the vehicle scheduling optimisation solutions, especially in the case of Fleet B. In addition, they might also affect decisions regarding capital control strategies. A further limitation of MOVES is that bus categories can only be defined based on model year and fuel type (Table 4.2). Other explanatory variables such as the powertrain (e.g., hybrid) and aftertreatment type are not currently supported by MOVES. However, it may be possible to extend the model to support these variables. While this is currently unlikely to be of value to the USEPA, it would likely be of value to transit agencies.  The biases in the MOVES model should be taken into account when selecting an emissions model. While MOVES is likely the more appropriate model to use in the context of estimating the climate and health impacts of transit bus emissions at this time, in some cases, particularly when emissions test data from the bus fleet of interest is available, a basic emissions factor model such as the one developed in Chapter 2 may be more appropriate. It may also be possible to use this emissions test data to calibrate MOVES or another emissions model, although this was not explored. If MOVES is used, then the biases should be taken into consideration when interpreting results from the model and using them in a decision making context. For example, the biases would have important implications when comparing diesel and CNG buses. Ultimately, the limitations of the various models, the data available, and the decision being made must be taken into consideration when selecting an emissions model. 133  5.2.4  Air Quality, Population, and Exposure  A detailed evaluation of methods and models used to quantify air quality (dispersion and transformation) population, and exposure was not possible within the scope of this dissertation. As discussed, methods previously developed by Greco et al. based on the intake fraction were employed (Greco,2007, Greco, et al.,2007a, Greco, et al.,2007b). An intake fraction based approach provides a convenient, flexible and computationally simple approach to estimate exposure on an intra-regional scale (e.g., along bus routes), although the estimation of the intake fraction itself can be complex. For example, the intake fraction for a bus route need only be calculated once for a given meteorological and population distribution scenario and can simply be multiplied by the estimated emissions along a route from one or more bus technology to estimate exposure. In this way it is easy to evaluate how different bus technologies affect health impacts. Further, as demonstrated in Chapter 3, it is easy to increase the spatial resolution of the intake fraction estimates (i.e., estimate the intake fraction at 50 m intervals along the route) if this is deemed of value.  There are a number of challenges associated with estimating the intake fraction. These include estimating the dispersion and transformation of pollutants after they have been emitted into the atmosphere and the resulting pollutant concentrations as well as estimating the population that comes into contact with these pollutants. In this study and the study by Greco et al. the focus was on primary PM2.5 mass emissions, which were assumed to be non-reactive. Thus, only dispersion was considered in estimating the resulting concentrations. Accounting for reactive or secondary pollutants would require models of the transformation process and would likely add considerable complexity to the analysis. In both studies, the CALINE line source dispersion model was used. This model has limitations and its performance in urban areas has been questioned (Section 2.3.6.2.2), but it is unclear whether other models offer more robust predictions. Modeling the population is also not trivial. Although the use of residential census data has limitations (Section 2.3.6.2.1), it is a convenient and readily available source of data (in developed countries). It may be possible to employ population time-activity models but this was beyond the scope of this dissertation. In summary, an approach based on the intake fraction provides a convenient and flexible method of modeling exposure and intake that can be revised as improved models of dispersion, transformation, and population time-activity become available. 134  5.2.5  Impact Indicators  Despite some criticisms, GWP (and by extension GWC) continues to be the most widely used indicator of climate impacts and is currently likely the most appropriate indicator to use in assessing the climate impacts of transit bus emissions (Section 2.3.6.3). This apparent certainty in indicators should not however distract from the fact that there are large uncertainties associated with GWP estimates and their relationship to changes in the climate and the associated damages. An important finding from Chapter 2 is that the net climate impact of transit bus emissions is likely dominated by long-lived species, primarily CO2 and CH4, for which the application of the GWP is less controversial.  As discussed throughout this dissertation, there is considerable uncertainty and debate over the health impacts of vehicle emissions and an appropriate indicator (Sections 1.1.2.1 and 2.3.6.3). There is yet no dominant health indicator. The use of primary PM2.5 emissions, and specifically the premature mortality attributable to the long-term exposure to PM2.5, is common and likely captures a significant fraction of the net health impact, but not all of it. Health-adjusted life year indicators such as the QALY and DALY are also increasingly being employed and are most analogous to GWP but do not share its dominance. Thus, there is significant opportunity for further development of health impact indicators and a judgement must be made with respect to which indicator is most appropriate in a given application.  5.2.6  Control Strategies and Decision Making  Capital control strategies have traditionally dominated the set of alternatives considered by transit agencies when making decision regarding how to reduce the climate and health impacts of emissions from their bus fleets. However, operational strategies also offer the potential to reduce these impacts as well as operating costs with limited capital expenditure and should also be considered. For example, as demonstrated in Chapter 2, vehicle scheduling/assignment optimisation that incorporates climate and health impacts and operating costs as objectives has the potential to reduce the impacts and operating costs of transit systems (i.e., result in cobenefits); although, the magnitude of the reductions are dependent on the characteristics and current operations (i.e., scheduling/assignment solution) of the transit system. It is important to emphasise that capital and operational control strategies are not mutually exclusive and that transit agencies can and should pursue both types of control strategies. Operational control 135  strategies offer transit agencies a way of maximising the benefits of their capital investments, which may be especially desirable to transit agencies faced with rising capital costs and budget shortfalls. Fuel costs, which dominated the operating costs estimated in Chapter 2, provide an additional incentive for transit agencies to pursue operational optimisation that extends the framework beyond traditional objectives (i.e., the number of buses and non-revenue service time).  Regardless of the type of control strategy considered, an integrated assessment decision making framework should be employed. Although developing these frameworks is more challenging, even if imperfect, they offer the only way of progressing towards more fundamental objectives typically framed in terms of sustainability, human welfare and environmental protection. Several studies, including Chapter 2, have shown that assessing impacts in independent frameworks leads to ill-considered trade-offs and detrimental outcomes (Mazzi, et al.,2007). For example, transit agencies that optimise the operations of their bus fleets based on operating costs and/or climate impacts alone may inadvertently increase the health impacts associated with the emissions of their bus fleets by as much as 47% (Chapter 2).  A challenge that arises when employing an integrated assessment framework is evaluating tradeoffs between a diverse set of impacts and objectives. Although a number of methods of evaluating these trade-offs exist, social cost-benefit analysis was explored in this dissertation, as it has been widely applied in the assessment of the impacts of vehicle emissions, including transit buses and may be more amenable to transit agencies (Krupnick, et al.,1991, Levy, et al.,2010, Stevens, et al.,2005, USEPA,2011a). However, as previously discussed, the application of cost-benefit analysis to public health and environmental decision making is not without controversy and it was explored as one possible decision making framework. Regardless of the framework employed or not employed, transit agencies are explicitly or implicitly making tradeoffs between objectives that impact climate and health. Thus, despite the limitations of the framework, it can provide useful insights into the decision making process and bring awareness to the trade-offs involved. Further, with regards to vehicle scheduling optimisation, by calculating the Pareto frontier, the impact of the decision making framework can be bounded.  136  5.3  Strengths and Limitations  This dissertation is subject to the very concerns it raises with regards to the need for integrated approaches to assessment and a broader perspective of the impacts of transit fleets. Within the context of transit systems, the operation of heavy-duty transit buses has a wide range of impacts beyond those to climate and human health, for example, congestion and passenger safety 37 (Lloyd, et al.,2001), although many of these would not be impacted by the assignment optimisation considered in Chapter 2. Even within the scope of climate and health impacts, it could be argued that there is a need to for a wider scope that accounts for the impacts of nonexhaust emissions, particularly resuspended road dust; the climate and health impacts of life cycle emissions associated with the production and distribution of the buses and fuel; and fuel system safety. In addition, the health impacts of noise may be of particular importance in the context of heavy-duty vehicles, which are typically noisier than light-duty vehicles (Babisch, et al.,2005).  A challenge in assessing these other impacts is that in many cases the science is still nascent. Thus the focus of this dissertation (as well as other studies) on the climate and health impacts of exhaust emissions is shaped by the evidence. However, it is important to acknowledge that the evidence may be incomplete and that evidence is not a measure of importance, and attempt to address gaps in knowledge (Dowlatabadi, et al.,1993). Failure to do so is detrimental to both decision making and learning.  Both integrated assessment and sustainability can be described as processes of learning (Dowlatabadi, et al.,1993, Risbey, et al.,1996, Robinson,2004). Thus, neither integrated assessment nor sustainability represent or define a static, tangible, or even achievable objective. Both integrated assessment and sustainability are continuous, iterative processes. As such, this dissertation represents a first iteration of a learning process. Despite its limitations, it can be used to help communicate both what is known and what is unknown and to help transit agencies make decisions based on the available evidence while also facilitating and guiding future research and learning (Morgan, et al.,1999).  37  Per kilometer, the fatality rate associated with motor vehicles is over 28 times greater than the rate associated with transit buses (APTA,2011).  137  Many of the specific limitations of this research have been previously discussed, therefore the focus here is on more general limitations that apply to the overall dissertation. There are at least three limitations or in the context of learning, future areas of research including: (a) the temporal dimension was not considered; (b) the case study approach meant sample sizes were small and findings were difficult to generalise; and (c) equity implications were not quantified.  In the context of the health impact pathway, only spatial variability was considered; however, temporal variability may also be significant and affect estimates of exposure and health impacts. For example, Greco et al. showed a diurnal pattern in dispersion and the intake fraction (Greco, et al.,2007a). Further, as discussed earlier, the activity of transit buses and the population is by nature correlated to a degree and also likely exhibit diurnal patterns. As with spatial variability, failing to account for the temporal relationship between population and emissions likely results in underestimates of exposure and health impacts.  For reasons of feasibility and scope, the results and conclusions of this research are based on single case studies (i.e., a small sample size), making generalisation of the results a challenge. In Chapter 2 the vehicle scheduling optimisation was performed on a single transit system. Given the number of factors that influence the optimisation it is difficult to extrapolate the results to other transit systems. On the other hand, there is no reason to believe that the transit system considered is unusual and that results for other similar or larger sized transit systems would not be comparable. In Chapter 3, the micro scale modeling exercise analysed a single route. This route was selected both for convenience and because it was the busiest route and from a value of information perspective, would likely have yielded the greatest benefits of a micro scale analysis. However, the route is likely an outlier and not necessarily representative. Thus, the conclusions regarding the quality of the GPS data, the importance of explanatory variables and the implications related to exposure are only suggestive. Nevertheless, the results provide valuable insights into the possible implications of micro scale analyses.  Finally, impacts related to equity and environmental justice were not considered in this dissertation but have potentially significant implications to decision makers. The vehicle scheduling optimisation in Chapter 2 is based in part on redistributing the emissions of transit buses and thus their health impacts. In general, emissions are shifted from areas of high population density to areas of low population density in order to minimise the net heath impact 138  without consideration of the burden to individuals or groups. The resulting redistribution may be considered inequitable. Thus there may be a trade-off between efficiency and equity, although at least one study has shown that reducing exposure also reduced inequity (Levy, et al.,2009). The results in Chapter 3 also have potential implications for equity, as they suggest that impacts on near-road populations may be underestimated by traditional approaches to a greater degree than other populations. Furthermore, on a global scale and thus beyond the purview of transit authorities, equity plays a fundamental role in shaping policies aimed at mitigating climate change (e.g., clean development mechanism). In summary, equity is a crucial measure of impact that should be considered in future analyses.  5.4  Future Research  This dissertation represents a first iteration of an integrated assessment model of the operations of public transit systems. As such there are numerous avenues of future research that could contribute to a next iteration. For example, fuel-based emissions modeling approaches may offer some advantages 38 over the models reviewed here, primarily with respect to estimating climate impacts, and should be explored in more detail. Further, in order to realise the benefits of models like MOVES, vehicle activity data that describes mode (i.e., speed, acceleration, and grade) are needed. This data could be obtained through a larger sampling campaign using the methods developed in Chapter 3, from a transit agency 39, and/or by developing a vehicle activity model. However, this research should be guided by both quantitative and qualitative sensitivity and uncertainty analyses of the model, as well as stakeholder engagement, in order to assess the value of future research (Figure 1.4). For example, given the importance of road grade in determining emissions, the method developed in Chapter 3 to estimate grade warrants evaluation. Also, if the biases in the MOVES model were addressed, the uncertainties in the model quantified in Chapter 4, as well as probability distributions of vehicle activity data and the uncertainty in exposure (i.e., SCF) estimated in Chapter 3, could be incorporated into a stochastic optimisation model and used to determine whether these uncertainties influence the optimisation results of Chapter 2. If they do, then there may be value in attempting to reduce  38  Because power and fuel consumption are highly correlated, fuel-based approaches likely have similar performance to power-based modal emissions models. Further, there are more numerous and in some cases less expensive methods of collecting data needed to estimate fuel-based emission factors. 39 The majority of TransLink’s bus fleet is now equipped with GPS.  139  these uncertainties. In addition, other impacts associated with non-exhaust emissions, secondary pollutant, and noise might also be considered.  The most significant uncertainties in the impact pathways are likely to be associated with processes after the emissions process, such as quantifying exposure and actual health and climate damages (Figure 1.3). For example, the health impact indicator employed in Chapter 2 only considered the effects of PM2.5 emissions. Thus, there will be a continued need to integrate new evidence into the model as it becomes available. The most significant and readily addressable source of uncertainty in the current model is likely associated with how the population was modeled (i.e., using static residential census data). Although this was explored to some degree in Chapter 3, there are significant opportunities to extend this work and to develop improved models of the spatial and temporal distributions of the population. The application of more sophisticated air quality models may also yield benefits. In general, efforts to reduce the uncertainty in the health impact estimates would likely be of greatest benefit as the primary trade-off in the vehicle scheduling optimisation performed in Chapter 2 was between health impacts and operating costs.  Finally, the integrated assessment model developed in Chapter 2 could be expanded beyond the operational decision making context to address capital decision making and planning (Figure 1.5). Further to this point, there is a need to consider the broader context in which this dissertation falls and the lessons afforded by history. If the goal is sustainability, then there is a need for a more holistic perspective that extends beyond public transit. Optimising the components (e.g., public transit) of a system (e.g., urban transportation) does not guarantee the overall system is optimal (Schelling,1978). This dissertation makes a first attempt to bring these perspectives to bear on transit operations and offers transit agencies practical and implementable solutions, but there are significant opportunities to expand the framework. In doing this it will be critical not to ignore the social dimensions of what are inherently socio-technical systems (Jasanoff,2005, Risbey, et al.,1996).  140  References Abdul-Khalek, I. S.; Kittelson, D. B.; Graskow, B. R.; Wei, Q.; Brear, F. Diesel exhaust particle size: measurement issues and trends. SAE Technical Paper 980525, 1998. Abu-Allaban, M.; Gillies, J. A.; Gertler, A. W.; Clayton, R.; Proffitt, D. Tailpipe, resuspended road dust, and brake-wear emission factors from on-road vehicles. Atmospheric Environment 2003, 37 (37), 5283-5293; DOI 10.1016/j.atmosenv.2003.05.005. Ainslie, B.; Rideout, G.; Couper, C.; McKinnon, D. The impact of Retrofit Exhaust Control Technologies on Emissions from Heavy-Duty Diesel Construction Equipment. SAE Technical Paper 1999-01-0110, 1999; DOI 10.4271/1999-01-0110. An, F.; Barth, M.; Norbeck, J.; Ross, M. Development of Comprehensive Modal Emissions Model: Operating Under Hot-Stabilized Conditions. Transportation Research Record: Journal of the Transportation Research Board 1997, 1587, 52-62; DOI 10.3141/1587-07. Andrei, P. Real-World Heavy-Duty Emissions Modeling. West Virginia University, 2001. Applebaum, B., As U.S. Agencies Put More Value on a Life, Businesses Fret. The New York Times February 16, 2011. APTA, Public Transportation Fact Book: 57th Edition; American Public Transportation Association: Washington, DC, 2006. APTA, 2010 Public Transportation Vehicle Database; American Public Transportation Association: Washington, DC, 2010a. APTA, Impacts of the Recession on Public Transportation Agencies: Survey Results; American Public Transportation Association: Washington, DC, 2010b. APTA, 2011 Public Transportation Fact Book: 62st Edition; American Public Transportation Association: Washington, DC, 2011. APTA, Quantifying and Reporting Transit Sustainability Metrics; APTA SUDS-CC-RP-003-12; American Public Transportation Association: Washington, DC, 2012. Arrow, K. J.; Cropper, M. L.; Eads, G. C.; Hahn, R. W.; Lave, L. B.; Noll, R. G.; Portney, P. R.; Russell, M.; Schmalensee, R.; Smith, V. K.; Stavins, R. N. Is There a Role for Benefit-Cost Analysis in Environmental, Health, and Safety Regulation? Science 1996, 272 (5259), 221-222; DOI 10.1126/science.272.5259.221. Athanasios, M.; et al. Comparability of particle emission measurements between vehicle testing laboratories: a long way to go. Measurement Science and Technology 2004, 15 (9), 1855. Atkinson, G.; Mourato, S. Environmental Cost-Benefit Analysis. Annual Review of Environment and Resources 2008, 33 (1), 317-344; DOI doi:10.1146/annurev.environ.33.020107.112927.  141  Ayala, A.; Kado, N.; Okamoto, R.; Holmen, B.; Kuzmicky, P.; Kobayashi, R. Diesel and CNG heavy-duty transit bus emissions over multiple driving schedules: Regulated pollutants and project overview. SAE Technical Paper 2002-01-1722, 2002. Babisch, W.; Beule, B.; Schust, M.; Kersten, N.; Ising, H. Traffic Noise and Risk of Myocardial Infarction. Epidemiology 2005, 16 (1), 33-40 10.1097/01.ede.0000147104.84424.24. Bachman, W.; Sarasua, W.; Hallmark, S.; Guensler, R. Modeling regional mobile source emissions in a geographic information system framework. Transportation Research Part C: Emerging Technologies 2000, 8 (1-6), 205-229. Bachmann, J. Will the Circle Be Unbroken: A History of the U.S. National Ambient Air Quality Standards. Journal of the Air & Waste Management Association (1995) 2007, 57 (6), 652-697. Bae, H. S.; Ryu, J.; Gerdes, J. C. Road Grade and Vehicle Parameter Estimation for Longitudinal Control Using GPS. IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC 2001, 166–171. Banister, D. The sustainable mobility paradigm. Transport Policy 2008, 15 (2), 73-80. Barnitt, R. A. In-Use Performance Comparison of Hybrid Electric, CNG, and Diesel Buses at New York City Transit. SAE Technical Paper 2008-01-1556, 2008; DOI 10.4271/2008-01-1556. Barth, M.; An, F.; Norbeck, J.; Ross, M. Modal Emissions Modeling: A Physical Approach. Transportation Research Record: Journal of the Transportation Research Board 1996a, 1520, 81-88. Barth, M.; An, F.; Younglove, T.; Levine, C.; Scora, G.; Ross, M.; Wenzel, T., The Development of a Comprehensive Modal Emissions Model. Final report submitted to the National Cooperative Highway Research Program; November, 1999; p 255. Barth, M.; Scora, G.; Younglove, T. Modal Emissions Model for Heavy-Duty Diesel Vehicles. Transportation Research Record: Journal of the Transportation Research Board 2004, 1880, 10-20; DOI 10.3141/1880-02. Barth, M.; Tadi, R. Emissions Comparison Between Truck and Rail: Case Study of California I40. Transportation Research Record: Journal of the Transportation Research Board 1996b, 1520, 44-52. Barth, M.; Younglove, T.; Scora, G., Development of a Heavy-duty Diesel Modal Emissions and Fuel Consumption Model; California PATH Program, Institute of Transportation Studies, University of California at Berkeley: Berkeley, August, 2005. BCMOE Smoke control forecast for British Columbia and Yukon. Victoria, British Columbia. http://www.env.gov.bc.ca/epd/epdpa/venting/polled_data/ (accessed Febuary 15, 2009). Benson, P. E. A review of the development and application of the CALINE3 and 4 models. Atmospheric Environment. Part B. Urban Atmosphere 1992, 26 (3), 379-390; DOI 10.1016/0957-1272(92)90013-i. 142  Bishop, G. A.; Morris, J. A.; Stedman, D. H.; Cohen, L. H.; Countess, R. J.; Countess, S. J.; Maly, P.; Scherer, S. The effects of altitude on heavy-duty diesel truck on-road emissions. Environmental Science & Technology 2001, 35 (8), 1574-1578. Bond, T. C. Can warming particles enter global climate discussions? Environmental Research Letters 2007, 2 (4), DOI 10.1088/1748-9326/2/4/045030. Bond, T. C.; Sun, H. L. Can reducing black carbon emissions counteract global warming? Environmental Science & Technology 2005, 39 (16), 5921-5926; DOI 10.1021/es0480421. Boulter, P. G., A review of emission factors and models for road vehicle non-exhaust particulate matter; PPR065; TRL Limited: Berkshire, UK, 2005. Boulter, P. G.; McCrae, I. S.; Barlow, T. J., A review of instantaneous emission models for road vehicles; PPR267; Transport Research Laboratory: Berkshire, UK, 2007. Brugge, D.; Durant, J. L.; Rioux, C. Near-highway pollutants in motor vehicle exhaust: A review of epidemiologic evidence of cardiac and pulmonary health risks. Environmental Health 2007, 6 (23), 12. Brunekreef, B.; Holgate, S. T. Air pollution and health. Lancet 2002, 360 (9341), 1233-1242. Burgard, D. A.; Bishop, G. A.; Stedman, D. H.; Gessner, V. H.; Daeschlein, C. Remote sensing of in-use heavy-duty diesel trucks. Environmental Science & Technology 2006, 40 (22), 69386942; DOI 10.1021/es060989a. Cappiello, A.; Chabini, I.; Nam, E. K.; Lue, A. A. L. A.; Abou Zeid, M. A. A. Z. M. In A statistical model of vehicle emissions and fuel consumption, Intelligent Transportation Systems, 2002. Proceedings. The IEEE 5th International Conference on, 2002; Chabini, I., 2002; pp 801809. CARB, Emfac2001 Emfac2002 User’s Guide: Calculating emission inventories for vehicles in California; California Air Resources Board: 2002; p 60. Ceder, A. Urban Transit Scheduling: Framework, Review and Examples. Journal of Urban Planning and Development 2002, 128 (4), 225-244. Ceder, A. Public-transport vehicle scheduling with multi vehicle type. Transportation Research Part C: Emerging Technologies 2011, 19 (3), 485-497; DOI 10.1016/j.trc.2010.07.007. Cervero, R., Transit-oriented development in the United States: experiences, challenges, and prospects. Transportation Research Board: 2004. Official Release of the MOVES2010 Motor Vehicle Emissions Model for Emissions Inventories in SIPs and Transportation Conformity; Code of Federal Regulations, Title 40, Part 75; United States Environmental Protection Agency. Subpart N - Emission Regulations for New Otto-Cycle and Diesel Heavy-Duty Engines; Gaseous and Particulate Exhaust Test Procedures; Code of Federal Regulations, Title 40, Part 86; United States Environmental Protection Agency. 143  Cho, H. M.; He, B.-Q. Spark ignition natural gas engines--A review. Energy Conversion and Management 2007, 48 (2), 608-618. Clark, N.; Bata, R.; Gautam, M.; Wang, W.; Chandler, K. L.; Norton, P. Natural Gas and Diesel Transit Bus Emissions: Review and Recent Data. SAE Technical Paper 973203, 1997; DOI 10.4271/973203. Clark, N.; Khan, A.; Wayne, W.; Gautam, M.; Thompson, G.; McKain, D.; Lyons, D.; Barnett, R. Weight Effect on Emissions and Fuel Consumption from Diesel and Lean-burn Natural Gas Transit Buses. SAE Technical Paper 2007-01-3626, 2007a; DOI 10.4271/2007-01-3626. Clark, N.; Prucz, J. C.; Gautam, M.; Lyons, D. W. The West Virginia University heavy duty vehicle emissions database as a resource for inventory and comparative studies. SAE Technical Paper 2000-01-2854, 2000; DOI 10.4271/2000-01-2854. Clark, N. N.; Gajendran, P.; Kern, J. M. A predictive tool for emissions from heavy-duty diesel vehicles. Environmental Science & Technology 2003, 37 (1), 7-15; DOI 10.1021/es0113192. Clark, N. N.; Gautam, M.; Rapp, B. L.; Lyons, D. W.; Graboski, M. S.; McCormick, R. L.; Alleman, T. L.; Norton, P. Diesel and CNG Transit Bus Emissions Characterization By Two Chassis Dynamometer Laboratories: Results and Issues. SAE Technical Paper 1999-01-1469, 1999a; DOI 10.4271/1999-01-1469. Clark, N. N.; Jarrett, R. P.; Atkinson, C. M. Field measurements of particulate matter emissions, carbon monoxide, and exhaust opacity from heavy-duty diesel vehicles. Journal of the Air & Waste Management Association 1999b, 49 (9), 76-84. Clark, N. N.; Kern, J. M.; Atkinson, C. M.; Nine, R. D. Factors affecting heavy-duty diesel vehicle emissions. Journal of the Air & Waste Management Association 2002, 52 (1), 84-94. Clark, N. N.; Vora, K. A.; Wang, L. J.; Gautam, M.; Wayne, W. S.; Thompson, G. J. Expressing Cycles and Their Emissions on the Basis of Properties and Results from Other Cycles. Environmental Science & Technology 2010, 44 (15), 5986-5992; DOI 10.1021/es100308q. Clark, N. N.; Wayne, W. S.; Zhen, F.; Khan, S. R.; Schiavone, J. J.; Chambers, C. A.; Chandler, K. L., A Review of the Performance of Hybrid-Electric Bus Technology on Fuel Economy and Emissions. In APTA 2006 Bus & Paratransit, 2006. Clark, N. N.; Zhen, F.; Wayne, W. S.; Lyons, D. W., Transit Bus Life Cycle Cost and Year 2007 Emissions Estimation; FTA-WV-26-7004.2007.1; Federal Transit Administration: Washington DC, 2007b. Clark, N. N.; Zhen, F.; Wayne, W. S.; Schiavone, J. J.; Chambers, C.; Golub, A. D.; Chandler, K. L., TCRP Report 132: Assessment of Hybrid-Electric Transit Bus Technology; Transportation Research Board: Washington, D.C., 2009. Clemen, R. T.; Reilly, T., Making Hard Decisions with DecisionTools: An Introduction to Decision Analysis. 2nd, 2004 Update ed.; Duxbury Press: Belmont, CA, 2004; p 733.  144  Cohen, J. T.; Hammitt, J. K.; Levy, J. I. Fuels for urban transit buses: A cost-effectiveness analysis. Environmental Science & Technology 2003, 37 (8), 1477-1484; DOI 10.1021/es0205030. Colvile, R. N.; Hutchinson, E. J.; Mindell, J. S.; Warren, R. F. The transport sector as a source of air pollution. Atmospheric Environment 2001, 35 (9), 1537-1565. Conway, R.; Chatterjee, S.; Beavan, A.; Lavenius, M.; Viswanathan, S.; Walker, A.; Rawson, S. Combined SCR and DPF Technology for Heavy-Duty Diesel Retrofit. SAE Technical Paper Series 2005, 2005-01-1862. Costanza, R.; Patten, B. C. Defining and predicting sustainability. Ecological Economics 1995, 15 (3), 193-196; DOI 10.1016/0921-8009(95)00048-8. CUTA, Steering Toward Sustainability: New Guidance for Canadian Transit Systems; Issue Paper 36; The Canadian Urban Transit Association: Toronto, 2012. Davis, N.; Lents, J.; Osses, M.; Nikkila, N.; Barth, M. Part 3: Developing Countries: Development and Application of an International Vehicle Emissions Model. Transportation Research Record: Journal of the Transportation Research Board 2005, 1939, 155-165. Detroit Diesel Corporation Detroit Diesel Series 50. http://www.detroitdiesel.com/public/specs/3SA432.pdf (accessed November 10, 2009). DieselNet Emission Standards United States: Heavy-Duty Truck and Bus Engines. http://www.dieselnet.com/standards/us/hd.php (accessed Sept. 28, 2011). DieselNet Heavy-Duty FTP Transient Cycle. http://www.dieselnet.com/standards/us/hd.php (accessed Sept. 28, 2011). Divall, C.; Bond, W., Suburbanizing the masses : public transport and urban development in historical perspective. Ashgate: Aldershot, Hants, England; Burlington, VT, USA, 2003. Dockery, D. W.; Pope, C. A.; Xu, X. P.; Spengler, J. D.; Ware, J. H.; Fay, M. E.; Ferris, B. G.; Speizer, F. E. An association between air-pollution and mortality in 6 united-states cities. New England Journal of Medicine 1993, 329 (24), 1753-1759. Dowlatabadi, H. On integration of policies for climate and global change. Mitigation and Adaptation Strategies for Global Change 2007, 12 (5), 651-663; DOI 10.1007/s11027-0079092-7. Dowlatabadi, H.; Morgan, M. G. Integrated Assessment of Climate Change. Science 1993, 259 (5103), 1813. Dreher, D. B.; Harley, R. A. A fuel-based inventory for heavy-duty diesel truck emissions. Journal of the Air & Waste Management Association 1998, 48 (4), 352-358. Environment Canada, SEPTA Conventional and Electric Hybrid Urban Bus Test Program Regulated Emissions and Fuel Economy Results; ERMD Report 03-38; Environmental Technology Center Emissions Research and Measurement Division: Ottawa, 2003. 145  Environment Canada DOES2. http://www.etccte.ec.gc.ca/organization/mbdo/mbdo_does2_1_e.html (accessed 26/09/2011, 2011). Environment Canada Climate Data Online - Canada's National Climate Archive. http://climate.weatheroffice.gc.ca/climateData/canada_e.html (accessed Febuary, 15, 2009). Environment Canada, National Inventory Report 1990–2008: Greenhouse Gas Sources and Sinks in Canada; Environment Canada: Ottawa, 2010. Ericson, C.; Westerberg, B.; Egnell, R. Transient Emission Predictions With Quasi-Stationary Models. SAE Technical Paper 2005-01-3852, 2005. ESRI ArcGIS Desktop (Release 9.3.1), Environmental Systems Research Institute: Redlands, CA, 2009. Esteves-Booth, A.; Muneer, T.; Kubie, J.; Kirby, H. A review of vehicular emission models and driving cycles. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science 2002, 216 (8), 777-797. Farrell, J.; Barth, M., The Global Positioning System and Inertial Navigation. McGraw-Hill Professional: New York, 1999. Figliozzi, M. Vehicle Routing Problem for Emissions Minimization. Transportation Research Record: Journal of the Transportation Research Board 2010, 2197, 1-7. Forster, P., V. Ramaswamy, P. Artaxo, T. Berntsen, R. Betts, D.W. Fahey, J. Haywood, J. Lean, D.C. Lowe, G. Myhre, J. Nganga, R. Prinn, G. Raga, M. Schulz, R. Van Dorland, Changes in Atmospheric Constituents and in Radiative Forcing. In Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M.Tignor and H.L. Miller, Ed. Cambridge University Press: Cambridge, United Kingdom and New York, NY, USA, 2007. Foster, M. S., From streetcar to superhighway : American city planners and urban transportation, 1900-1940. Temple University Press: Philadelphia, 1981. Frey, H.; Kim, K. Comparison of Real-World Fuel Use and Emissions for Dump Trucks Fueled with B20 Biodiesel Versus Petroleum Diesel. Transportation Research Record: Journal of the Transportation Research Board 2006, 1987, 110-117; DOI 10.3141/1987-12. Frey, H. C.; Rouphail, N. M.; Zhai, H.; Farias, T. L.; Goncalves, G. A. Comparing real-world fuel consumption for diesel- and hydrogen-fueled transit buses and implication for emissions. Transportation Research Part D: Transport and Environment 2007, 12 (4), 281-291; DOI 10.1016/j.trd.2007.03.003. Frey, H. C.; Zhang, K.; Rouphail, N. M. Vehicle-Specific Emissions Modeling Based upon onRoad Measurements. Environmental Science & Technology 2010, DOI 10.1021/es902835h.  146  Fuglestvedt, J. S.; Berntsen, T. K.; Godal, O.; Sausen, R.; Shine, K. P.; Skodvin, T. Metrics of Climate Change: Assessing Radiative Forcing and Emission Indices. Climatic Change 2003, 58 (3), 267-331; DOI 10.1023/a:1023905326842. Fuglestvedt, J. S.; Shine, K. P.; Berntsen, T.; Cook, J.; Lee, D. S.; Stenke, A.; Skeie, R. B.; Velders, G. J. M.; Waitz, I. A. Transport impacts on atmosphere and climate: Metrics. Atmospheric Environment 2010, 44 (37), 4648-4677; DOI DOI 10.1016/j.atmosenv.2009.04.044. Gajendran, P.; Clark, N. N. Effect of Truck Operating Weight on Heavy-Duty Diesel Emissions. Environmental Science & Technology 2003, 37 (18), 4309-4317; DOI 10.1021/es026299y. Gao, H. O.; Stasko, T. H. Cost-minimizing retrofit/replacement strategies for diesel emissions reduction. Transportation Research Part D: Transport and Environment 2009, 14 (2), 111-119; DOI 10.1016/j.trd.2008.11.006. Gautam, M.; Clark, N.; Riddle, W. C.; Wayne, W. S.; Maldonado, H.; Agrawal, A.; Carlock, M. A. Development and Initial Use of a Heavy-Duty Diesel Truck Test Schedule for Emissions Characterization. SAE Technical Paper 2002-01-1753, 2002. Gautam, M.; Clark, N. N.; Lyons, D. W.; Long, T. R. J.; Howell, A. D.; Loth, J. L.; Palmer, G. M.; Wang, W.; Bata, R. M., Design overview of a heavy duty mobile vehicle emissions testing laboratory. In ASME Winter Annual Meeting, Advanced Automotive Technology, ASME: Atlanta, 1991; Vol. 40. Gelencsér, A., Chapter 3 - Major Carbonaceous Particle Types and their Sources. In Carbonaceous Aerosol, Springer: Dordrecht, The Netherlands, 2004. GeoBase, 092G06. In Canadian Digital Elevation Data, Level 1 (CDED1), Dataset: 1.0 Metadata: 2 ed.; Natural Resources Canada, Centre for Topographic Information: Sherbrooke, Quebec, 2003. Giannelli, R. A.; Nam, E. K.; Helmer, K.; Younglove, T.; Scora, G.; Barth, M. Heavy-Duty Diesel Vehicle Fuel Consumption Modeling Based on Road Load and Power Train Parameters. SAE Technical Paper 05CV-3, 2005. GOA, Air Pollution: Limitations of EPA’s Motor Vehicle Emissions Model and Plans to Address Them: Report to the Chairman, Subcommittee on Oversight and Investigations,; GAO/RCED97-210.; U.S. General Accounting Office: Washington, DC, 1997. Golub, A.; Zhen, F.; Clark, N.; Wayne, W.; Schiavone, J.; Chambers, C.; Chandler, K. Creation of Life-Cycle Cost Tool for Transit Buses to Evaluate Hybrid Electric Bus Technologies in RealWorld Operation. Transportation Research Record: Journal of the Transportation Research Board 2011, 2218, 19-26; DOI 10.3141/2218-03. Golub, A.; Zhen, F.; Clark, N. N.; Wayne, W. S.; Schiavone, J. J.; Chambers, C.; Chandler, K. L., Creation of a Transit Bus Life-Cycle Cost Tool to Evaluate Hybrid-Electric Bus Technologies in Real World Operation. In Transportation Research Board 90th Annual Meeting, Washington, D.C., 2010. Graboski, M. S.; Ross, J. D.; McCormick, R. L. Transient Emissions From No. 2 Diesel and Biodiesel Blends in a Ddc Series 60 Engine. SAE Technical Paper 961166, 1996. 147  Greco, S. L. Characterizing Mobile Source Fine Particulate Matter Emissions-To-Exposure Relationships. Harvard, Boston, 2007. Greco, S. L.; Wilson, A. M.; Hanna, S. R.; Levy, J. I. Factors influencing mobile source particulate matter ernissions-to-exposure relationships in the Boston urban area. Environmental Science & Technology 2007a, 41 (22), 7675-7682; DOI 10.1021/es062213f. Greco, S. L.; Wilson, A. M.; Spengler, J. D.; Levy, J. I. Spatial patterns of mobile source particulate matter emissions-to-exposure relationships across the United States. Atmospheric Environment 2007b, 41 (5), 1011-1025; DOI 10.1016/j.atmosenv.2006.09.025. Grieshop, A. P.; Boland, D.; Reynolds, C. C. O.; Gouge, B.; Apte, J. S.; Rogak, S. N.; Kandlikar, M. Modeling air pollutant emissions from Indian auto-rickshaws: Model development and implications for fleet emission rate estimates. Atmospheric Environment 2012, 50 (0), 148-156; DOI 10.1016/j.atmosenv.2011.12.046. Grieshop, A. P.; Marshall, J. D.; Kandlikar, M. Health and climate benefits of cookstove replacement options. Energy Policy 2011, 39 (12), 7530-7542; DOI 10.1016/j.enpol.2011.03.024. Grieshop, A. P.; Reynolds, C. C. O.; Kandlikar, M.; Dowlatabadi, H. A black-carbon mitigation wedge. Nature Geoscience 2009, 2 (8), 533-534; DOI 10.1038/ngeo595. Griffin, R. D., Principles of Air Quality Management, Second Edition. CRC Press: 2006; p i-xix. Grose, M.; Sakurai, H.; Savstrom, J.; Stolzenburg, M. R.; Watts, W. F.; Morgan, C. G.; Murray, I. P.; Twigg, M. V.; Kittelson, D. B.; McMurry, P. H. Chemical and physical properties of ultrafine diesel exhaust particles sampled downstream of a catalytic trap. Environmental Science & Technology 2006, 40 (17), 5502-5507. Hallmark, S. L.; Guensler, R.; Fomunung, I. Characterizing on-road variables that affect passenger vehicle modal operation. Transportation Research Part D: Transport and Environment 2002, 7 (2), 81-98. Hayes, R. R.; Williams, A.; Ireland, J.; Walkowicz, K.; Black, S. King County Metro-Allison Hybrid Electric Transit Bus Testing. SAE Technical Paper 2006-01-3570, 2006; DOI 10.4271/2006-01-3570. Health Effects Institute, Mobile-source air toxics : a critical review of the literature on exposure and health effects; Health Effects Institute: Boston, MA, 2007. Health Effects Institute, Traffic-Related Air Polluiton: A Critical Review of the Literature on Emissions, Exposure, and Health Effects: Special Report 17; Health Effects Institute: Boston. Masasachusetts, 2010. Helton, J. C.; Johnson, J. D.; Sallaberry, C. J.; Storlie, C. B. Survey of sampling-based methods for uncertainty and sensitivity analysis. Reliability Engineering & System Safety 2006, 91 (1011), 1175-1209; DOI 10.1016/j.ress.2005.11.017.  148  Herner, J. D.; Hu, S. H.; Robertson, W. H.; Huai, T.; Collins, J. F.; Dwyer, H.; Ayala, A. Effect of Advanced Aftertreatment for PM and NOx Control on Heavy-Duty Diesel Truck Emissions. Environmental Science & Technology 2009, 43 (15), 5928-5933; DOI 10.1021/es9008294. Hesterberg, T. W.; Lapin, C. A.; Bunn, W. B. A comparison of emissions from vehicles fueled with diesel or compressed natural gas. Environmental Science & Technology 2008, 42 (17), 6437-6445; DOI 10.1021/es071718i. Heywood, J. B., Internal combustion engine fundamentals. McGraw-Hill: New York, 1988. Hill, A. B. The Environment and Disease: Association or Causation? Proceedings of the Royal Society of Medicine-London 1965, 58 (5), 295-300. Hoek, G.; Beelen, R.; de Hoogh, K.; Vienneau, D.; Gulliver, J.; Fischer, P.; Briggs, D. A review of land-use regression models to assess spatial variation of outdoor air pollution. Atmospheric Environment 2008, 42 (33), 7561-7578. Holling, C. S. Understanding the Complexity of Economic, Ecological, and Social Systems. Ecosystems 2001, 4 (5), 390-405. Holmes, N. S.; Morawska, L. A review of dispersion modelling and its application to the dispersion of particles: An overview of different dispersion models available. Atmospheric Environment 2006, 40 (30), 5902-5928; DOI 10.1016/j.atmosenv.2006.06.003. Hulme, M., Why we disagree about climate change : understanding controversy, inaction and opportunity. Cambridge University Press: Cambridge, UK; New York, 2009. IPCC, Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge, UK, 2007a. IPCC, Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge, UK and New York, NY, USA, 2007b. Ishaque, M. M.; Noland, R. B. Simulated pedestrian travel and exposure to vehicle emissions. Transportation Research Part D: Transport and Environment 2008, 13 (1), 27-46. Ishaque, M. M.; Noland, R. B. Pedestrian and Vehicle Flow Calibration in Multimodal Traffic Microsimulation. Journal of Transportation Engineering-Asce 2009, 135 (6), 338-348; DOI 10.1061/(asce)0733-947x(2009)135:6(338). Jackson, E.; Aultman-Hall, L.; Holmén, B.; Du, J. Evaluating the Ability of Global Positioning System Receivers to Measure a Real-World Operating Mode for Emissions Research. Transportation Research Record: Journal of the Transportation Research Board 2005, 1941 (1), 43-50. Jackson, E.; Qu, Y.; Holmén, B.; Aultman-Hall, L. Driver and Road Type Effects on Light-Duty Gas and Particulate Emissions. Transportation Research Record: Journal of the Transportation Research Board 2006, 1987, 118-127. 149  Jackson, K. T., Crabgrass frontier : the suburbanization of the United States. Oxford University Press: New York, 1985. Jacobson, M. Z. Review of solutions to global warming, air pollution, and energy security. Energy & Environmental Science 2009, 2 (2), 148-173; DOI 10.1039/b809990c. Jasanoff, S., Designs on Nature: Science and Democracy in Europe and the United States. Princeton University Press Princeton, NJ, 2005. Jayaratne, E. R.; He, C.; Ristovski, Z. D.; Morawska, L.; Johnson, G. R. A comparative investigation of ultrafine particle number and mass emissions from a fleet of on-road diesel and CNG buses. Environmental Science & Technology 2008, 42 (17), 6736-6742; DOI 10.1021/es800394x. Jayaratne, E. R.; Meyer, N. K.; Ristovski, Z. D.; Morawska, L.; Miljevic, B. Critical Analysis of High Particle Number Emissions from Accelerating Compressed Natural Gas Buses. Environmental Science & Technology 2010, 44 (10), 3724-3731; DOI 10.1021/es1003186. Jerrett, M.; Arain, A.; Kanaroglou, P.; Beckerman, B.; Potoglou, D.; Sahsuvaroglu, T.; Morrison, J.; Giovis, C. A review and evaluation of intraurban air pollution exposure models. Journal of Exposure Analysis and Environmental Epidemiology 2005a, 15 (2), 185-204; DOI 10.1038/sj.jea.7500388. Jerrett, M.; Burnett, R. T.; Ma, R. J.; Pope, C. A.; Krewski, D.; Newbold, K. B.; Thurston, G.; Shi, Y. L.; Finkelstein, N.; Calle, E. E.; Thun, M. J. Spatial analysis of air pollution and mortality in Los Angeles. Epidemiology 2005b, 16 (6), 727-736. Jiménez, J. L. Understanding and Quantifying Motor Vehicle Emissions with Vehicle Specific Power and TILDAS Remote Sensing. Massachusetts Institute of Technology, 1999. Johnson, T. V. Review of Diesel Emissions and Control. SAE Technical Paper 2010-01-0301, 2010. Jones, D. W., Mass motorization + mass transit : an American history and policy analysis. Indiana University Press: Bloomington, 2008. Jun, J.; Guensler, R.; Ogle, J. Smoothing Methods to Minimize Impact of Global Positioning System Random Error on Travel Distance, Speed, and Acceleration Profile Estimates. Transportation Research Record: Journal of the Transportation Research Board 2006a, 1972 (1), 141-150; DOI 10.3141/1972-19. Jun, J.; Ko, J.; Yoon, S.; Guensler, R., Impacts of Acceleration Calculation Methods on Onroad Vehicle Engine Power Estimation. In A&WMA 99th Annual Conference and Exhibition, 2006b. Kahn Ribeiro, S., S. Kobayashi, M. Beuthe, J. Gasca, D. Greene, D. S. Lee, Y. Muromachi, P. J. Newton, S. Plotkin, D. Sperling, R. Wit, P. J. Zhou, Transport and its infrastructure. In Climate Change 2007: Mitigation. Contribution of Working Group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, B. Metz, O. R. D., P.R. Bosch, R. Dave, L.A. Meyer, Ed. Cambridge University Press: Cambridge, United Kingdom and New York, NY, USA, 2007. 150  Kamarianakis, Y.; Gao, H. O. Accounting for Exhaust Gas Transport Dynamics in Instantaneous Emission Models via Smooth Transition Regression. Environmental Science & Technology 2010, 44 (4), 1320-1326; DOI 10.1021/es901913k. Kandlikar, M. Indices for comparing greenhouse gas emissions: Integrating science and economics. Energy Economics 1996, 18 (4), 265-281. Karner, A. A.; Eisinger, D. S.; Niemeier, D. A. Near-Roadway Air Quality: Synthesizing the Findings from Real-World Data. Environmental Science & Technology 2010, 44 (14), 53345344; DOI 10.1021/es100008x. Kaur, S.; Nieuwenhuijsen, M. J.; Colvile, R. N. Fine particulate matter and carbon monoxide exposure concentrations in urban street transport microenvironments. Atmospheric Environment 2007, 41 (23), 4781-4810. Keeney, R. L. Decision-Analysis - an Overview. Operations Research 1982, 30 (5), 803-838. Keepin, B.; Wynne, B. Technical analysis of IIASA energy scenarios. Nature 1984, 312 (5996), 691-695; DOI 10.1038/312691a0. Khalek, I. A.; Spears, M.; Charmley, W. Particle Size Distribution From a Heavy-Duty Diesel Engine: Steady-State and Transient Emission Measurement Using Two Dilution Systems and Two Fuels. SAE Technical Paper 2003-01-0285, 2003. Kittelson, D.; Watts, W.; Johnson, J., Diesel Aerosol Sampling Methodology - CRC E-43: Final Report; Coordinating Research Council: 2002. Kittelson, D. B. Engines and nanoparticles: A review. Journal of Aerosol Science 1998, 29 (5-6), 575-588. Kittelson, D. B.; Watts, W. F.; Johnson, J. P. On-road and laboratory evaluation of combustion aerosols—Part1: Summary of diesel engine results. Journal of Aerosol Science 2006, 37 (8), 913-930. Koupal, J.; Michaels, H.; Cumberworth, M.; Bailey, C.; Brzezinski, D., EPA's Plan for MOVES: A Comprehensive Mobile Source Emissions Model. In International Emission Inventory Conference, Atlanta, GA, 2002. Krupnick, A. J.; Portney, P. R. Controlling Urban Air-Pollution: A Benefit-Cost Assessment. Science 1991, 252 (5005), 522-528. Lanni, T.; Chatterjee, S.; Conway, R.; Windawi, H.; Rosenblatt, D.; Bush, C.; Lowell, D.; Evans, J.; McLean, R. Performance and Durability Evaluation of Continuously Regenerating Particulate Filters on Diesel Powered Urban Buses at NY City Transit. SAE Technical Paper 2001- 010511, 2001. Lanni, T.; Frank, B. P.; Tang, S.; Rosenblatt, D.; Lowell, D. Performance and Emissions Evaluation of Compressed Natural Gas and Clean Diesel Buses at New York City’s Metropolitan Transit Authority. SAE Technical Paper 2003-01-0300, 2003.  151  Lapple, C. E. Characteristics of Particles and Particle Dispersoids. Stanford Research Institute Journal 1961, 5 (Third Quarter), 95. Layton, D. W. METABOLICALLY CONSISTENT BREATHING RATES FOR USE IN DOSE ASSESSMENTS. Health Physics 1993, 64 (1), 23-36. Le Treut, H., R. Somerville, U. Cubasch, Y. Ding, C. Mauritzen, A. Mokssit, T. Peterson, M. Prather, Historical Overview of Climate Change Science. In Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor, H.L. Miller, Ed. Cambridge University Press: Cambridge, United Kingdom and New York, NY, USA, 2007. Levy, J.; Chemerynski, S.; Tuchmann, J. Incorporating concepts of inequality and inequity into health benefits analysis. International Journal for Equity in Health 2006, 5 (1), 2. Levy, J. I.; Buonocore, J. J.; von Stackelberg, K. Evaluation of the public health impacts of traffic congestion: a health risk assessment. Environmental Health 2010, 9, DOI 10.1186/1476069x-9-65. Levy, J. I.; Greco, S. L.; Melly, S. J.; Mukhi, N. Evaluating Efficiency-Equality Tradeoffs for Mobile Source Control Strategies in an Urban Area. Risk Analysis 2009, 29 (1), 34-47. Li, J.-Q.; Larry Head, K. Sustainability provisions in the bus-scheduling problem. Transportation Research Part D: Transport and Environment 2009a, 14 (1), 50-60. Li, J.-Q.; Zhang, W.-B.; Zhang, L. A web-based support system for estimating and visualizing the emissions of diesel transit buses. Transportation Research Part D: Transport and Environment 2009b, 14 (8), 533-540. Liu, Z. G.; Ford, D. C.; Vasys, V. N.; Chen, D. R.; Johnson, T. R. Influence of engine operating conditions on diesel particulate matter emissions in relation to transient and steady-state conditions. Environmental Science & Technology 2007, 41 (13), 4593-4599; DOI 10.1021/es0616229. Lloyd, A. C.; Cackette, T. A. Diesel engines: Environmental impact and control. Journal of the Air & Waste Management Association 2001, 51 (6), 809-847. M.J. Bradley & Associates, Bus Technology & Alternative Fuels Demonstration Project: Final Phase 1 - Test Program Report; TransLink: Burnaby, BC, 2006. M.J. Bradley & Associates, Bus Technology & Alternative Fuels Demonstration Project: Final Phase 2 - Test Program Report; TransLink: Burnaby, BC, 2008. M.J. Bradley & Associates, Bus Technology & Alternative Fuels Demonstration Project: Phase 3 Emissions Test Results Executive Summary; TransLink: Burnaby, BC, 2009. Manivannan, A.; Ramprabhu, R.; Tamilporai, P.; Chandrasekaran, S. Lean-Burn Natural Gas Spark Ignition Engine-An Overview. SAE Technical Paper Series 2003, 2003-01-0638.  152  Maricq, M. M. Chemical characterization of particulate emissions from diesel engines: A review. Journal of Aerosol Science 2007, 38 (11), 1079-1118. Marland, G.; Weinberg, A. M., Longevity of Infrastructure. In Cities and their Vital Systems: Infrastructure Past, Present, and Future, Ausubel, J. H.; Herman, R., Eds.; National Academy Press: Washington, DC, 1988. Marshall, J. D.; Nazaroff, W. W., Intake Fraction. In Exposure Analysis, Ott, W. R.; Steinemann, A. C.; Wallace, L. A., Eds.; CRC Press: Boca Raton, FL, 2006; pp 237 - 251. Marshall, J. D.; Teoh, S. K.; Nazaroff, W. W. Intake fraction of nonreactive vehicle emissions in US urban areas. Atmospheric Environment 2005, 39 (7), 1363-1371; DOI 10.1016/j.atmosenv.2004.11.006. Marty, M. A.; Blaisdell, R. J.; Broadwin, R.; Hill, M.; Shimer, D.; Jenkins, M. Distribution of daily breathing rates for use in California's Air Toxics Hot Spots program risk assessments. Human and Ecological Risk Assessment 2002, 8 (7), 1723-1737. Mazzi, E. A.; Dowlatabadi, H. Air quality impacts of climate mitigation: UK policy and passenger vehicle choice. Environmental Science & Technology 2007, 41 (2), 387-392; DOI 10.1021/es060517w. McCarthy, M. C.; Eisinger, D. S.; Hafner, H. R.; Chinkin, L. R.; Roberts, P. T.; Black, K. N.; Clark, N. N.; McMurry, P. H.; Winer, A. M. Particulate matter: A strategic vision for transportation-related research. Environmental Science & Technology 2006, 40 (18), 5593-5599. McCormick, R. L.; Graboski, M. S.; Newlin, A. W.; Ross, J. D. Effect of humidity on heavyduty transient emissions from diesel and natural gas engines at high altitude. Journal of the Air & Waste Management Association 1997, 47 (7), 784-791. McGeehan, J. A.; Yeh, S.; Couch, M.; Hinz, A.; Otterholm, B.; Walker, A.; Blakeman, P. On The Road to 2010 Emissions: Field Test Results and Analysis With DPF-SCR System and UltraLow-Sulfur Diesel Fuel. SAE Technical Paper Series 2005, 2005-01-3716. McKain, D. L.; Clark, N.; Balon, T.; Moynihan, P. J.; Lynch, S. A.; Webb, T. C. Characterization of Emissions From Hybrid-Electric and Conventional Transit Buses. SAE Technical Paper 2000-01-2011, 2000. McMurry, P. H. A review of atmospheric aerosol measurements. Atmospheric Environment 2000, 34 (12-14), 1959-1999. McRae, G. J.; Goodin, W. R.; Seinfeld, J. H. Development of a second-generation mathematical model for Urban air pollution - I. Model formulation. Atmospheric Environment 1982, 16 (4), 679-696; DOI 10.1016/0004-6981(82)90386-9. McShane, C., Down the asphalt path : the automobile and the American city. Columbia University Press: New York, 1994. McShane, C., The automobile : a chronology of its antecedents, development, and impact. Greenwood Press: Westport, Conn., 1997. 153  McTaggart-Cowan, G. P.; Reynolds, C. C. O.; Bushe, W. K. Natural gas fuelling for heavy-duty on-road use: current trends and future direction. International Journal of Environmental Studies 2006, 63 (4), 421 - 440; DOI 10.1080/00207230600802056. Melosi, M. V. The Place of the City in Environmental History. Environmental History Review 1993, 17 (1), 1-23. Monn, C. Exposure assessment of air pollutants: a review on spatial heterogeneity and indoor/outdoor/personal exposure to suspended particulate matter, nitrogen dioxide and ozone. Atmospheric Environment 2001, 35 (1), 1-32. Moosmuller, H.; Arnott, W. P.; Rogers, C. F.; Bowen, J. L.; Gillies, J. A.; Pierson, W. R.; Collins, J. F.; Durbin, T. D.; Norbeck, J. M. Time resolved characterization of diesel particulate emissions. 1. Instruments for particle mass measurements. Environmental Science & Technology 2001, 35 (4), 781-787. Morgan, M.; Henrion, M.; Small, M., Uncertainty: A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis. Cambridge University Press: Cambridge 1990. Morgan, M. G.; Dowlatabadi, H. Learning from integrated assessment of climate change. Climatic Change 1996, 34 (3-4), 337-368. Morgan, M. G.; Kandlikar, M.; Risbey, J.; Dowlatabadi, H. Why conventional tools for policy analysis are often inadequate for problems of global change - An editorial essay. Climatic Change 1999, 41 (3-4), 271-281. Nagendra, S. M. S.; Khare, M. Line source emission modelling. Atmospheric Environment 2002, 36 (13), 2083-2098. National Research Council, Modeling Mobile-Source Emissions; National Research Council: Washington, D.C., 2000. Nazaroff, W. W. New Directions: It's time to put the human receptor into air pollution control policy. Atmospheric Environment 2008, 42 (26), 6565-6566. Newman, P. W. G.; Kenworthy, J. R. The land use--transport connection : An overview. Land Use Policy 1996, 13 (1), 1-22; DOI 10.1016/0264-8377(95)00027-5. Ning, Z.; Sioutas, C. Atmospheric Processes Influencing Aerosols Generated by Combustion and the Inference of Their Impact on Public Exposure: A Review. Aerosol and Air Quality Research 2010, 10 (1), 43-58; DOI 10.4209/aaqr.2009.05.0036. Nordhaus, W. Economics. Critical assumptions in the Stern Review on climate change. Science 2007, 317 (5835), 201-2; DOI 10.1126/science.1137316. North, R. J.; Noland, R. B.; Ochieng, W. Y.; Polak, J. W. Modelling of particulate matter mass emissions from a light-duty diesel vehicle. Transportation Research Part D: Transport and Environment 2006, 11 (5), 344-357; DOI 10.1016/j.trd.2006.06.006.  154  Ogle, J.; Guensler, R.; Bachman, W.; Koutsak, M.; Wolf, J. Accuracy of Global Positioning System for Determining Driver Performance Parameters. Transportation Research Record: Journal of the Transportation Research Board 2002, 1818, 12-24; DOI 10.3141/1818-03. Owen, S.; Knight, N. UBC Line Update. http://www.bog.ubc.ca/__shared/assets/SUB10.04.08_3.3%20UBC%20Line%20Update.pdf (accessed July 6, 2010). Pacific Carbon Trust, Annual Report 2010/11; Province of British Columbia: Victoria, BC, 2011. Penner, J. E.; Prather, M. J.; Isaksen, I. S. A.; Fuglestvedt, J. S.; Klimont, Z.; Stevenson, D. S. Short-lived uncertainty? Nature Geoscience 2010, 3 (9), 587-588. Pope, C.; Bates, D. V.; Raizenne, M. E. Health effects of particulate air pollution: time for reassessment? Environmental Health Perspectives 1995, 103 (5), 472. Pope, C. A.; Dockery, D. W. Health effects of fine particulate air pollution: Lines that connect. Journal of the Air & Waste Management Association 2006, 56 (6), 709-742. Prucz, J. C.; Clark, N. N.; Gautam, M.; Lyons, D. W. Exhaust Emissions from Engines of the Detroit Diesel Corporation in Transit Buses: A Decade of Trends. Environmental Science & Technology 2001, 35 (9), 1755-1764; DOI 10.1021/es001416f. Quddus, M. A.; Ochieng, W. Y.; Noland, R. B. Current map-matching algorithms for transport applications: State-of-the art and future research directions. Transportation Research Part C: Emerging Technologies 2007, 15 (5), 312-328; DOI DOI 10.1016/j.trc.2007.05.002. Rakha, H.; Ahn, K.; Trani, A. Development of VT-Micro model for estimating hot stabilized light duty vehicle and truck emissions. Transportation Research Part D: Transport and Environment 2004, 9 (1), 49-74. Rakopoulos, C. D.; Giakoumis, E. G., Diesel Engine Transient Operation : Principles of Operation and Simulation Analysis. Springer: London, 2009. Ramamurthy, R.; Clark, N. N. Atmospheric Emissions Inventory Data for Heavy-Duty Vehicles. Environmental Science & Technology 1999, 33 (1), 55-62. Ramamurthy, R.; Clark, N. N.; Atkinson, C. M.; Lyons, D. W. Models for Predicting Transient Heavy Duty Vehicle Emissions. SAE Technical Paper 982652, 1998. Ramanathan, V.; Carmichael, G. Global and regional climate changes due to black carbon. Nature Geoscience 2008, 1 (4), 221-227; DOI 10.1038/ngeo156. Randall Guensler, S. Y., Chunxia Feng, Hainan Li, Jungwook Jun, Heavy-Duty Diesel Vehicle Modal Emission Model (HDDV-MEM) Volume I: Modal Emission; EPA-600/R-05/090a; U.S. Environmental Protection Agency: 2005. Reynolds, C. C. O.; Kandlikar, M. Climate impacts of air quality policy: Switching to a natural gas-fueled public transportation system in New Delhi. Environmental Science & Technology 2008, 42 (16), 5860-5865; DOI 10.1021/es702863p.  155  Risbey, J.; Kandlikar, M.; Patwardhan, A. Assessing integrated assessments. Climatic Change 1996, 34 (3-4), 369-395. Ristovski, Z. D.; Jayaratne, E. R.; Lim, M.; Ayoko, G. A.; Morawska, L. Influence of Diesel Fuel Sulfur on Nanoparticle Emissions from City Buses. Environmental Science & Technology 2006, 40 (4), 1314-1320; DOI 10.1021/es050094i. Robinson, A. L.; Donahue, N. M.; Shrivastava, M. K.; Weitkamp, E. A.; Sage, A. M.; Grieshop, A. P.; Lane, T. E.; Pierce, J. R.; Pandis, S. N. Rethinking organic aerosols: Semivolatile emissions and photochemical aging. Science 2007, 315 (5816), 1259-1262. Robinson, J. Squaring the circle? Some thoughts on the idea of sustainable development. Ecological Economics 2004, 48 (4), 369-384. Robinson, L. A. How US Government Agencies Value Mortality Risk Reductions. Review of Environmental Economics and Policy 2007, 1 (2), 283-299; DOI 10.1093/reep/rem018. Rogak, S. N.; Pott, U.; Dann, T.; Wang, D. Gaseous emissions from vehicles in a traffic tunnel in Vancouver, British Columbia. Journal of the Air & Waste Management Association 1998, 48 (7), 604-614. Ropkins, K.; Beebe, J.; Li, H.; Daham, B.; Tate, J.; Bell, M.; Andrews, G. Real-World Vehicle Exhaust Emissions Monitoring: Review and Critical Discussion. Critical Reviews in Environmental Science and Technology 2009, 39 (2), 79 - 152. Ross, M. Automobile Fuel Consumption and Emissions: Effects of Vehicle and Driving Characteristics. Annual Review of Energy and the Environment 1994, 19, 75-112. Ross, M. Fuel efficiency and the physics of automobiles. Contemporary Physics 1997, 38 (6), 381-394. Rotmans, J.; Asselt, M., Integrated Assessment Modelling Climate Change: An Integrated Perspective. In Martens, P.; Rotmans, J.; Jansen, D.; Vrieze, K., Eds.; Springer Netherlands: 2003; Vol. 1, pp 239-275. Sakurai, H.; Tobias, H. J.; Park, K.; Zarling, D.; Docherty, K. S.; Kittelson, D. B.; McMurry, P. H.; Ziemann, P. J. On-line measurements of diesel nanoparticle composition and volatility. Atmospheric Environment 2003, 37 (9-10), 1199-1210; DOI 10.1016/s1352-2310(02)01017-8. Saltelli, A.; Ratto, M.; Andres, T.; Campolongo, F.; Cariboni, J.; Gatelli, D.; Saisana, M.; Tarantola, S., Global Sensitivity Analysis: The Primer. Wiley-Interscience: 2008. Schelling, T. C., Micromotives and macrobehavior. W. W. Norton & Company, Inc.: New York, 1978. Seigneur, C. Current Understanding of Ultrafine Particulate Matter Emitted from Mobile Sources. Journal of the Air & Waste Management Association 2009, 59 (1), 3-17; DOI 10.3155/1047-3289.59.1.3. Setton, E.; Marshall, J. D.; Brauer, M.; Lundquist, K. R.; Hystad, P.; Keller, P.; Cloutier-Fisher, D. The impact of daily mobility on exposure to traffic-related air pollution and health effect 156  estimates. Journal of Exposure Science and Environmental Epidemiology 2011, 21 (1), 42-48; DOI 10.1038/jes.2010.14. Shah, S. D.; Cocker, D. R.; Miller, J. W.; Norbeck, J. M. Emission rates of particulate matter and elemental and organic carbon from in-use diesel engines. Environmental Science & Technology 2004, 38 (9), 2544-2550; DOI 10.1021/es0350583. Shindell, D.; Faluvegi, G.; Walsh, M.; Anenberg, S. C.; Van Dingenen, R.; Muller, N. Z.; Austin, J.; Koch, D.; Milly, G. Climate, health, agricultural and economic impacts of tighter vehicle-emission standards. Nature Climate Change 2011, 1 (1), 59-66; DOI 10.1038/nclimate1066. Shorter, J. H.; Herndon, S.; Zahniser, M. S.; Nelson, D. D.; Wormhoudt, J.; Demerjian, K. L.; Kolb, C. E. Real-time measurements of nitrogen oxide emissions from in-use New York City transit buses using a chase vehicle. Environmental Science & Technology 2005, 39 (20), 79918000. Shrivastava, M. K.; Lipsky, E. M.; Stanier, C. O.; Robinson, A. L. Modeling Semivolatile Organic Aerosol Mass Emissions from Combustion Systems. Environmental Science & Technology 2006, 40 (8), 2671-2677; DOI 10.1021/es0522231. Singer, B. C.; Harley, R. A. A fuel-based motor vehicle emission inventory. Journal of the Air & Waste Management Association 1996, 46 (6), 581-593. Singh, R. B.; Desloges, C.; Sloan, J. J. Application of a microscale emission factor model for particulate matter to calculate vehicle-generated contributions to fine particulate emissions. Journal of the Air & Waste Management Association 2006, 56 (1), 37-47. Smit, R.; Ntziachristos, L.; Boulter, P. Validation of road vehicle and traffic emission models - A review and meta-analysis. Atmospheric Environment 2010, 44 (25), 2943-2953; DOI 10.1016/j.atmosenv.2010.05.022. Smith, K. R. Fuel Combustion, Air-Pollution Exposure, and Health - The Situation in Developing countries. Annual Review of Energy and the Environment 1993, 18, 529-566. Smith, K. R.; Haigler, E. Co-Benefits of Climate Mitigation and Health Protection in Energy Systems: Scoping Methods. Annual Review of Public Health 2008, 29 (1), 11-25; DOI doi:10.1146/annurev.publhealth.29.020907.090759. Smith, K. R.; Uma, R.; Kishore, V. V. N.; Zhang, J.; Joshi, V.; Khalil, M. A. K. Greenhouse Implications of Household Stoves: An Analysis for India. Annual Review of Energy and the Environment 2000, 25 (1), 741-763; DOI doi:10.1146/annurev.energy.25.1.741. Stasko, T. H.; Gao, H. O. Reducing transit fleet emissions through vehicle retrofits, replacements, and usage changes over multiple time periods. Transportation Research Part D: Transport and Environment 2010, 15 (5), 254-262; DOI 10.1016/j.trd.2010.03.004. Statistics Canada, 2006 Census of Canada. 2006. Statistics Canada, 2006 Census Dictionary. Ottawa, Ontario, Catalogue no. 92-566-XWE. February 15, 2009, 2007a. 157  Statistics Canada, Dissemination Block Boundary Files, British Columbia for the 2006 Census of Canada. Ottawa, Ontario, Catalogue no. 92-163-XWE2006011. February 15, 2009, 2007b. Statistics Canada, GeoSuite, 2006 Census (Geography Products: Geographic Data Products). Ottawa, Ontario, Catalogue no. 92-150-XBB. March 13, 2009, Downloadable database. 2007c. Statistics Canada, Deaths and mortality rates, by age group and sex, Canada, provinces and territories, annual. Ottawa, Ontario, Catalog no. CANSIM Table 102-0504. 2008. Stern, N., The Economics of Climate Change: The Stern Review. Cambridge University Press: Cambridge and New York, 2007. Stevens, G.; Wilson, A.; Hammitt, J. K. A benefit-cost analysis of retrofitting diesel vehicles with particulate filters in the Mexico City metropolitan area. Risk Analysis 2005, 25 (4), 883899. Stieb, D. M.; Evans, G. J.; Sabaliauskas, K.; Chen, L. I.; Campbell, M. E.; Wheeler, A. J.; Brook, J. R.; Guay, M. A scripted activity study of the impact of protective advice on personal exposure to ultra-fine and fine particulate matter and volatile organic compounds. Journal of Exposure Science and Environmental Epidemiology 2008, 18 (5), 495-502. Stone, R., Introduction to internal combustion engines. Macmillan: Basingstoke, 1992. Tainio, M.; Tuomisto, J. T.; Hanninen, O.; Aarnio, P.; Koistinen, K. J.; Jantunen, M. J.; Pekkanen, J. Health effects caused by primary fine particulate matter (PM2.5) emitted from buses in the Helsinki metropolitan area, Finland. Risk Analysis 2005, 25 (1), 151-160. Tarr, J. A., The search for the ultimate sink : urban pollution in historical perspective. University of Akron Press: Akron, Ohio, 1996. Taylor, B., The development of the MOBILE6.2C model; M6C-06-E; Environment Canada: 2005; p 39. Thorpe, A.; Harrison, R. M. Sources and properties of non-exhaust particulate matter from road traffic: A review. Science of the Total Environment 2008, 400 (1-3), 270-282; DOI 10.1016/j.scitotenv.2008.06.007. Tiwari, R.; Cervero, R.; Schipper, L. Driving CO2 reduction by Integrating Transport and Urban Design strategies. Cities 2011, 28 (5), 394-405; DOI 10.1016/j.cities.2011.05.005. TransLink, Bus Technology & Alternative Fuels Demonstration Project: Phase 1 - Test Program Report; TransLink: Burnaby, BC, 2006. Tree, D. R.; Svensson, K. I. Soot processes in compression ignition engines. Progress in Energy and Combustion Science 2007, 33 (3), 272-309. Unal, A.; Frey, C. H.; Rouphail, N. M.; Colyar, J. D. Hot Spot Analysis of Real World Vehicle Emissions Based upon a Portable On-Board Measurement System. Proc., Annual Meeting of the Air & Waste Management Association 2001.  158  Unal, A.; Frey, H. C.; Rouphail, N. M. Quantification of highway vehicle emissions hot spots based upon on-board measurements. Journal of the Air & Waste Management Association 2004, 54 (2), 130-140. United States Census Bureau, Historical National Population Estimates: July 1, 1900 to July 1, 1999. Washington, DC, 2000. United States Census Bureau, Table 1. Monthly Population Estimates for the United States: April 1, 2010 to June 1, 2012 (NA-EST2011-01-0612) Washington, DC, 2012. USEPA, CALINE4 - A Dispersion Model For Predicting Air Pollution Concentrations Near Roadways; FHWA/CA.lTL-84/15; United States Environmental Protection Agency, Office of Air Quality Planning and Standards: Washington, DC, 1998. USEPA, Heavy-Duty Diesel Emission Reduction Project Retrofit/Rebuild Component; EPA420R-99-014; United States Environmental Protection Agency: June, 1999. USEPA, EPA’s New Generation Mobile Source Emissions Model: Initial Proposal and Issues; EPA420-R-01-007; United States Environmental Protection Agency, Office of Air and Radiation, Office of Research and Development, Region 4: 2001. USEPA, Draft Design and Implementation Plan for EPA’s Multi-Scale Motor Vehicle and Equipment Emission System (MOVES); EPA420-P-02-006; US Environmental Protection Agency: 2002a. USEPA, EPA’s Onboard Analysis Shootout Overview and Results; EPA420-R-02-026; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: 2002b. USEPA, Methodology for Developing Modal Emission Rates for EPA’s Multi-Scale Motor Vehicle and Equipment Emission System; EPA420-R-02-027; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: 2002c. USEPA, User’s Guide to MOBILE6.1 and MOBILE6.2 Mobile Source Emission Factor Model; EPA420-R-03-010; Office of Transportation and Air Quality, US Environmental Protection Agency: 2003. USEPA, MOVES2004 Software Design Reference Manual: Draft; EPA420-P-04-017; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: March, 2004. USEPA, MOVES2004 Energy and Emission Inputs Draft Report; EPA420-P-05-003; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: 2005. USEPA, Air Quality Criteria for Ozone and Related Photochemical Oxidants Volume I of III; EPA 600/R-05/004aF; U.S. Environmental Protection Agency, National Center for Environmental Assessment-RTP Division Office of Research and Development: 2006.  159  USEPA, Integrated Science Assessment for Oxides of Nitrogen - Health Criteria; EPA/600/R08/07; U.S. Environmental Protection Agency, National Center for Environmental AssessmentRTP Division Office of Research and Development: 2008. USEPA, Development of Emission Rates for Heavy-Duty Vehicles in the Motor Vehicle Emissions Simulator (Draft MOVES2009) EPA-420-P-09-005; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: 2009a. USEPA, Draft Motor Vehicle Emission Simulator (MOVES) 2009 Software Design and Reference Manual; EPA-420-B-09-007; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: March, 2009b. USEPA, Integrated Science Assessment for Particulate Matter; EPA/600/R-08/139F; U.S. Environmental Protection Agency, National Center for Environmental Assessment-RTP Division Office of Research and Development: 2009c. USEPA, Policy Guidance on the Use of MOVES2010 for State Implementation Plan Development, Transportation Conformity, and Other Purposes; EPA-420-B-09-046; United States Environmental Protection Agency, Office of Transportation and Air Quality, Transportation and Regional Programs Division: 2009d. USEPA, INVENTORY OF U.S. GREENHOUSE GAS EMISSIONS AND SINKS: 1990 - 2008 United States Environmental Protection Agency Washington, DC, 2010a. USEPA, Transportation Conformity Guidance for Quantitative Hot-spot Analyses in PM2.5 and PM10 Nonattainment and Maintenance Areas; EPA-420-B-10-040; United States Environmental Protection Agency, Transportation and Regional Programs Division Office of Transportation and Air Quality: 2010b. USEPA, The Benefits and Costs of the Clean Air Act from 1990 to 2020: Final Report; United States Environmental Protection Agency, Office of Air and Radiation: 2011a. USEPA, Motor Vehicle Emission Simulator (MOVES): User Guide for MOVES2010a; EPA420-B-10-036; United States Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: August, 2011b. USEPA, Our Nation’s Air: Status and Trends Through 2010; EPA-454/R-12-001; United States Environmental Protection Agency, Office of Air Quality Planning and Standards: 2012. USEPA; Brzezinski, D., Exhaust Temperature, Air Conditioning and Inspection and Maintenance Adjustments in MOVES2010a. In MOVES Workshop, United States Environmental Protection Agency, Office of Transportation and Air Quality: Ann Arbor, Michigan, 2011a. USEPA; Choi, D.; Koupal, J., Development of Heavy-Duty Emission Rates for MOVES2010. In MOVES Workshop, Office of Transportation & Air Quality, USEPA: Ann Arbor, Michigan, 2011b. USEPA; Choi, D.; Koupal, J., MOVES Validation. In MOVES Workshop, United States Environmental Protection Agency, Office of Transportation and Air Quality: Ann Arbor, Michigan, 2011c. 160  USEPA; Choi, D.; Koupal, J., Updates to Energy Rates and Sourcebin Distribution in MOVES 2010a. In MOVES Workshop, Office of Transportation & Air Quality, USEPA: Ann Arbor, Michigan, 2011d. USEPA; Ensfield, C., On-Road Emissions Testing of 18 Tier 1 Passenger Cars and 17 Diesel Powered Public Transport Buses; EPA420-R-02-030; United Statest Environmental Protection Agency, Office of Transportation and Air Quality, Assessment and Standards Division: October, 2002. Vaaraslahti, K.; Virtanen, A.; Ristimaki, J.; Keskinen, J. Nucleation mode formation in heavyduty diesel exhaust with and without a particulate filter. Environmental Science & Technology 2004, 38 (18), 4884-4890; DOI 10.1021/es0353255. van Setten, B.; Makkee, M.; Moulijn, J. A. Science and technology of catalytic diesel particulate filters. Catalysis Reviews-Science and Engineering 2001, 43 (4), 489-564. Vardoulakis, S.; Fisher, B. E. A.; Pericleous, K.; Gonzalez-Flesca, N. Modelling air quality in street canyons: a review. Atmospheric Environment 2003, 37 (2), 155-182. Wang, J.; Storey, J.; Domingo, N.; Huff, S.; Thomas, J.; West, B. Studies of Diesel Engine Particle Emissions During Transient Operations Using an Engine Exhaust Particle Sizer. Aerosol Science and Technology 2006, 40, 1002-1015. Wang, W. G.; Clark, N. N.; Lyons, D. W.; Yang, R. M.; Gautam, M.; Bata, R. M.; Loth, J. L. Emissions comparisons from alternative fuel buses and diesel buses with a chassis dynamometer testing facility. Environmental Science & Technology 1997, 31 (11), 3132-3137; DOI 10.1021/es9701063. Warner, S. B., Streetcar suburbs: the process of growth in Boston, 1870-1900. Harvard University Press: Cambridge, 1962. Watson, J. G.; Chen, L. W. A.; Chow, J. C.; Doraiswamy, P.; Lowenthal, D. H. Source apportionment: Findings from the US Supersites program. Journal of the Air & Waste Management Association 2008, 58 (2), 265-288; DOI 10.3155/1047-3289.58.2.265. Wayne, W. S.; Clark, N. N.; Nine, R. D.; Elefante, D. A Comparison of Emissions and Fuel Economy from Hybrid-Electric and Conventional-Drive Transit Buses. Energy & Fuels 2004, 18 (1), 257-270; DOI doi:10.1021/ef030096t. Wayne, W. S.; Perhinschi, M.; Clark, N.; Tamayo, S.; Tu, J., Integrated Bus Information System (IBIS) – A Vehicle Procurement Resource for Transit. In Transportation Research Board 90th Annual Meeting, Washington, D.C., 2011a. Wayne, W. S.; Perhinschi, M.; Clark, N.; Tamayo, S.; Tu, J. Integrated Bus Information System Vehicle Procurement Resource for Transit. Transportation Research Record: Journal of the Transportation Research Board 2011b, (2233), 1-10; DOI 10.3141/2233-01. Weilenmann, M.; Soltic, P.; Ajtay, D. Describing and compensating gas transport dynamics for accurate instantaneous emission measurement. Atmospheric Environment 2003, 37 (37), 51375145; DOI 10.1016/j.atmosenv.2003.05.004. 161  Willis, A.; Gjersoe, N.; Havard, C.; Kerridge, J.; Kukla, R. Human movement behaviour in urban spaces: implications for the design and modelling of effective pedestrian environments. Environment and Planning B-Planning & Design 2004, 31 (6), 805-828; DOI 10.1068/b3060. Woodcock, J.; Edwards, P.; Tonne, C.; Armstrong, B. G.; Ashiru, O.; Banister, D.; Beevers, S.; Chalabi, Z.; Chowdhury, Z.; Cohen, A.; Franco, O. H.; Haines, A.; Hickman, R.; Lindsay, G.; Mittal, I.; Mohan, D.; Tiwari, G.; Woodward, A.; Roberts, I. Health and Climate Change 2 Public health benefits of strategies to reduce greenhouse-gas emissions: urban land transport. Lancet 2009, 374 (9705), 1930-1943; DOI 10.1016/s0140-6736(09)61714-1. World Commission on Environment and Development, Our Common Future. Oxford University Press: Oxford, 1987. Yanowitz, J.; Graboski, M. S.; McCormick, R. L. Prediction of in-use emissions of heavy-duty diesel vehicles from engine testing. Environmental Science & Technology 2002, 36 (2), 270-275. Yanowitz, J.; Graboski, M. S.; Ryan, L. B. A.; Alleman, T. L.; McCormick, R. L. Chassis dynamometer study of emissions from 21 in-use heavy duty diesel vehicles. Environmental Science & Technology 1999, 33 (2), 209-216. Yanowitz, J.; McCormick, R. L.; Graboski, M. S. In-use emissions from heavy-duty diesel vehicles. Environmental Science & Technology 2000, 34 (5), 729-740. Yoon, S.; Li, H.; Jun, J.; Ogle, J.; Guensler, R.; Rodgers, M. Methodology for Developing Transit Bus Speed-Acceleration Matrices for Load-Based Mobile Source Emissions Models. Transportation Research Record: Journal of the Transportation Research Board 2005a, 1941, 26-33. Yoon, S.; Li, H.; Jun, J.; Ogle, J. H.; Guensler, R. L.; Rodgers, M. O., Methodology for developing transit bus speed-acceleration matrices for load-based mobile source emissions models. In Energy and Environmental Concerns 2005, Transportation Research Board Natl Research Council: Washington, 2005b; pp 26-33. Younglove, T.; Scora, G.; Barth, M. Designing On-Road Vehicle Test Programs for the Development of Effective Vehicle Emission Models. Transportation Research Record: Journal of the Transportation Research Board 2005, 1941 (-1), 51-59. Yu, L.; Jia, S. C.; Shi, O. Y. Research on Transportation-Related Emissions: Current Status and Future Directions. Journal of the Air & Waste Management Association 2009, 59 (2), 183-195; DOI 10.3155/1047-3289.59.2.183. Yura, E. A.; Kear, T.; Niemeier, D. Using CALINE dispersion to assess vehicular PM2.5 emissions. Atmospheric Environment 2007, 41 (38), 8747-8757; DOI 10.1016/j.atmosenv.2007.07.045. Zarkadoula, M.; Zoidis, G.; Tritopoulou, E. Training urban bus drivers to promote smart driving: A note on a Greek eco-driving pilot program. Transportation Research Part D: Transport and Environment 2007, 12 (6), 449-451.  162  Zhai, H. B.; Frey, H. C.; Rouphail, N. M. A Vehicle-Specific Power Approach to Speed- and Facility-Specific Emissions Estimates for Diesel Transit Buses. Environmental Science & Technology 2008, 42 (21), 7985-7991; DOI 10.1021/es800208d. Zhang, K.; Frey, C. Evaluation of response time of a portable system for in-use vehicle tailpipe emissions measurement. Environmental Science & Technology 2008, 42 (1), 221-227. Zhen, F.; Clark, N. N.; Bedick, C. R.; Gautam, M.; Wayne, W. S.; Thompson, G. J.; Lyons, D. W. Development of a Heavy Heavy-Duty Diesel Engine Schedule for Representative Measurement of Emissions. Journal of the Air & Waste Management Association 2009, 59 (8), 950-959; DOI 10.3155/1047-3289.59.8.950. Zhou, Y.; Levy, J. I. Factors influencing the spatial extent of mobile source air pollution impacts: a meta-analysis. Bmc Public Health 2007, 7 (89), 11; DOI 10.1186/1471-2458-7-89. Zhu, Y. F.; Hinds, W. C.; Kim, S.; Shen, S.; Sioutas, C. Study of ultrafine particles near a major highway with heavy-duty diesel traffic. Atmospheric Environment 2002, 36 (27), 4323-4335. Zielinska, B. Atmospheric transformation of diesel emissions. Experimental and Toxicologic Pathology 2005, 57, Supplement 1 (0), 31-42; DOI 10.1016/j.etp.2005.05.006.  163  Appendices Appendix A: Integrated Assessment A.1  Vehicle Fleet and Categories  Table A.1 – TransLink vehicle fleet and categories. Vehicle Fleets Emissions Testsd  Category  Fuel  Emissions Control  40DO  40  Diesel  40DB2  40  40DB  Engine Make (Year)  None / OxCata  Detroit Diesel 6V92TA (1991,1992)  Diesel  None / OxCat  40  Diesel  OxCat  40DA  40  Diesel  DPFb  40DH  40  Diesel  DPF  60DB  60  Diesel  OxCat  Detroit Diesel S50 (1995-2001) Cummins ISC (2000,2001) Cummins ISL/ISM (2006-2008) Cummins ISB (1998) Detroit Diesel S50 (1998-2000) Cummins ISM (2007) Cummins ISL (2009)  Total  TransLink Vehicle Length (ft)  A  B  125  95  0  387  294  0  71  54  170  254  193  170  2  1  170  99  76  37  Engine Make (Year) Detroit Diesel 6V92TA (1991,1992) Detroit Diesel S50 (1995-2001) Cummins ISC (2001) Cummins ISC (2005) Cummins ISB (1998) -  n  Source  20  (Wayne, et al.,2011b)  22  (Wayne, et al.,2011b)  4  (TransLink,2006)  2  (TransLink,2006)  2  (TransLink,2006)  0  Estimated  Caterpillar C9 (Hayes, et 1 (2004) al.,2006) Caterpillar C9 (Hayes, et 60DH 60 Diesel DPF 33 26 37 1 (2004) al.,2006) Cummins Westport Cummins Westport C Gas + 40CG 40 CNG OxCat C Gas + 2 (TransLink,2006) 56 43 170 (2005,2006) (2005) a Oxidation Catalyst; b Diesel Particulate Filter; c Burnaby Transit Centre Depot; d Vehicles and studies used to develop the emission factors. 60DA  60  Diesel  DPF  13  10  37  164  Table A.2 - USEPA regulatory emission standards for urban transit buses (Barnitt,2008, DieselNet,2007a). THC NOX PM (g⋅bhp-1⋅h-1) (g⋅bhp-1⋅h-1) (g⋅bhp-1⋅h-1) 1990 1.3 6.0 0.60 1991-1992 1.3 5.0 0.25 1993 1.3 5.0 0.10 1994-1995 1.3 5.0 0.07 1996-1997 1.3 5.0 0.05a 1998-2003 1.3 4.0 0.05a 2.4 combined or 2004-2006 15.5 2.5 combined and a limit of 0.05a 0.5 NMHC 2007-2010 15.5 0.14 NMHC 0.2b 0.01 These standards apply to both diesel and lean-burn spark ignited (SI) compressed natural gas engines as the current federal definition of compression ignition (CI) engine is based on the engine cycle. a In-use PM standard 0.07 b NOX and NMHC emissions standards were phased in between 2007 and 2010. Very few vehicles meeting the 0.2 g⋅bhp-1⋅h-1 were expected to be on-road before 2010. Model Years  CO (g⋅bhp-1⋅h-1) 15.5 15.5 15.5 15.5 15.5 15.5  165  A.2  Emissions Model  Distance based emission factors (g·km-1) were developed for each bus category based on the CBD drive cycle (Table A.3). The specific vehicles tested and the studies used to develop the emission factors are listed in Table A.1. BC and EC were considered to be synonymous. Sulfur emissions (e.g., SO2 and SO4) were assumed to be negligible and were not considered because Ultra Low Sulfur Diesel (ULSD) fuel was used.  Table A.3 - Emission factors. Category 40DO  FE (DEL· km-1) 0.918  CO2 (g·km-1)  CO (g·km-1)  NOX (g·km-1)  THC (g·km-1)  PM2.5 (g·km-1)  CH4 (g·km-1)  N 2O (g·km-1)  BC/EC (g·km-1)  OC  GWCb (gCO2e·km-1)  GWC – Totalc (gCO2e·km-1)  2340  6.22  17  2.12  0.662  0  0.004001  0.407  0.255  2516  2531  40DB2  0.64  1680  3.54  22  0.155  0.212  0  0.004097  0.121  0.0908  1732  1739  40DB  0.566  1520  1.48  12.3  0.441  0.109  0  0.004097  0.0609  0.0481  1546  1551  40DA  0.603  1590  0.337  6.83  0.035  0.0244  0  0.004097  0.00210  0.0223  1590  1592  40DH  0.445  1190  0.235  5.2  0.0175  0.0125  0  0.004097  0.00108  0.0114  1190  1192  60DB  1.1  2930  5.44  16.3  0.428  0.196  0  0.004097  0.124  0.0716  2984  2995  60DA  1.07  2850  1.1  12.2  0.112  0.0628  0  0.004097  0.00541  0.0574  2850  2854  60DH  0.724  1860  0.572  8.97  0.0186  0.0125  0  0.004097  0.00108  0.0114  1860  1862  0.001611  0.00114  0.0157  1635  1637  40CG 0.754 1470 0.535 12.4 6.62 0.0168 6.62 Fuel Economy (FE) in Diesel Equivalent Litres (DEL) b GWC resulting from CO2, CH4, and PM (BC and OC). These values were used in the analysis. c GWC resulting from CO2, CH4, PM (EC and OC), CO, NOX, NMHC, and N2O. a  166  A.2.1 60xx Category Emission Factors Emissions measurement data from 60 ft articulated buses is very limited. I am aware of only one published study (Hayes, et al.,2006) that covers this vehicle type. However, the significantly increased mass and different configuration of these vehicles is extremely likely to impact their emission factors, which warrants creating distinct vehicle categories for these vehicles. Although the 60 ft buses tested by (Hayes, et al.,2006) differ from the 60 ft buses in TransLink’s fleet, this is the only source of data for this class of vehicles. No data was available for the 60DB category. To develop emission factor for this category, the average relative difference in the emission factors of 40 ft DPF equipped buses compared to 40 ft OxCat equipped buses was estimated from the following studies: (Ayala, et al.,2002, Clark, et al.,2002, Environment Canada,2003, Lanni, et al.,2001, Lanni, et al.,2003, McKain, et al.,2000, Ramamurthy, et al.,1999, TransLink,2006). This was then multiplied by the emission factors for the 60DA category to derive emission factors for the 60DB category.  A.2.2 Black Carbon (BC), Organic Carbon (OC), N2O Emission Factors BC and EC were considered to be equivalent for the purposes of this study. BC, EC, OC, and N2O are not commonly measured and were not measured in any of the studies used to develop the emission factors in this study. Further, the ratio of EC to OC has been shown to be effected by engine operation, thus the ratio is effected by vehicle activity and the drive cycle (Shah, et al.,2004). To estimate BC/EC and OC, the EC fraction was estimated using MOVES2010a. Each vehicle category was simulated in MOVES using the CBD drive cycle and the EC fraction was estimated. OC was estimated by subtracting the EC component from PM2.5. N2O emission factors were also estimated using this method.  A.3  Operating Costs  Operating costs were developed from studies by TransLink and Clark et al. (Table A.4) (Clark, et al.,2007b, M.J. Bradley & Associates,2006). Maintenance costs include propulsion-related maintenance, battery replacement for hybrids, facility maintenance, and electricity costs associated with compressing CNG. Based on both studies the propulsion-related maintenance were estimated to be 0.09 $·km-1 for all bus categories. Note that TransLink found maintenance costs of the CNG buses they tested to be higher, 0.14 $·km-1; however, as the other studies did not find this, a cost of 0.09 $·km-1 was deemed more representative. All other maintenance 167  costs were taken from Clark et al.. All costs considered in the study were reported in 2007 U.S. dollars with exception of those reported by TransLink. However, as the value of the Canadian and U.S. was approximately on par in 2007 the cost data obtained from TransLink was not adjusted. Further, it was assumed that the cost did not change between 2006 (the year they were reported) and 2007.  Table A.4 - Bus operating costs. Category  Operating Cost ($·km-1) Fuela  Maintenance  Total  40DO  0.770  0.119  0.889  40DB2  0.537  0.119  0.656  40DB  0.475  0.119  0.594  40DA  0.506  0.119  0.625  40DH  0.373  0.214  0.587  60DB  0.921  0.119  1.042  60DA  0.898  0.119  1.017  60DH  0.607  0.214  0.821  40CG 0.449 0.151 0.600 a Based on a diesel fuel price of 0.839 ($·L-1) and a CNG fuel price of 0.595 ($·DEL-1) (M.J. Bradley & Associates,2006)  A.4  Climate Metric (GWP)  GWP estimates based on a 100 year time horizon were taken from the literature (Table A.5). The contribution of CO, Non Methane Hydrocarbons (NMHC) and N2O to the net climate impact were negligible. Further, the impacts of CO and NMHC are indirect and complex. Thus the impacts of CO, NMHC, and N2O were not included in the GWC estimate used in the analysis (Table A.3). Table A.5 – Global Warming Potentials (GWP)  a  Pollutant  GWP100  Source  CO2  1  (Forster,2007)  CH4  25  (Forster,2007)  NMHCa  3.4  (Forster,2007)  CO  1.9  BC  455  OC  -35  N20  298  (Forster,2007) (Reynolds, et al.,2008) (Reynolds, et al.,2008) (Forster,2007)  Non-Methane Hydrocarbons  168  A.5  PM2.5 Intake Calculation Workflow  Figure A.1 – Route Intake (I) and Intake Fraction (iF) calculation workflow and associated data sources.  169  A.6  Concentration to Emissions Ratio (C·Ê-1) and Dispersion Modeling  The mean annual emissions to concentration ratio (C·Ê-1) was estimated using the CALINE4 line-source dispersion model (USEPA,1998). A single annual average modeling period was used and the mean annual dispersion conditions were assumed to be unidirectional and the same across the region and therefore the same at all points on all routes. Bus routes were represented in CALINE4 as an east-west road link 100 m in length. Receptors were located at the link midpoint as well as radially at the midpoints of each of the zones (i.e., at 25 m, 75 m, 150 m, 350 m, 750 m, 3000 m) and 8 directional groupings (North, Northeast, East, Southeast, South, Southwest, West, Northwest). The concentration was estimated at each receptor assuming a unit on-route (link) emission of 112 g⋅hr-1 (equivalent to 1 vehicle per second, with each vehicle having a PM2.5 emission factor of 1 g⋅mile-1)  Table A.6 provides a detailed listing of the CALINE4 input parameters used in this study. The non-metrological input parameters used were consistent with those developed by Greco et al. for their work with CAL3QHCR, which is based on CALINE3, the predecessor to CALINE4 (Greco, et al.,2007a). A barrier to using CALINE4 is that unlike CAL3QHCR, it only accepts a single set of meteorological data. Therefore to estimate average annual dispersion, the metrological data must be averaged and/or the model must be run for each set of meteorological data. A hybrid approach was used in this study.  In this study, the metrological conditions (i.e., wind speed and stability class) in the 8 directional groupings were estimated from one year of hourly weather data (July 2004 – June 2005) from the Vancouver International Airport, White Rock and Pitt Meadows meteorological stations (Environment Canada,2009). Atmospheric stability classes were assigned to the 8 directional groups, with all groupings assigned class C except for North, which was assigned class B due to its low average wind speed. Mixing height data was derived from historical British Columbia Smoke Control Forecast data (BCMOE,2009). The CALINE4 model was run for each of the 8 directional groupings to produce receptor concentrations associated with the metrological conditions in each of the directional groups. These concentrations were weighted by the frequency of the occurrence of each wind direction, and summed into a single concentration for each receptor location. Finally, the means of the receptor concentrations in each zone were estimated to provide the average annual concentration within each of the 6 zones resulting from 170  a unit on-route emission (Figure A.2 and Figure A.3). As the concentration is a linear function of the total on-route emissions, the C·Ê-1 ratio is constant at all distances or points on the dispersion curve. Thus the C·Ê-1 ratio was estimated by dividing the average annual concentration in each zone by the unit on-route emission.  171  Table A.6 - CALINE4 Configuration Parameters Parameter Pollutant Parameters Pollutant Type (PTYP) Settling Velocity (VS) Deposition Velocity (VD) Aerodynamic Roughness Coefficient (Zo) Altitude above sea level (ALT) Meteorology Wind Direction Bearing (BRG) Wind Speed (U) Atmospheric stability class (CLAS) Mixing Height (MIXH) Wind direction standard deviation (SIGTH) Temperature (TEMP) Ambient concentration (AMB) Link Details Link Type (TYP) Coordinates of link endpoint 1 (XL1,YL1) Coordinates of link endpoint 2 (XL2,YL2) Roadway height (HL) Mixing zone width (WL) Mixing width (right) (MIXWR) Mixing width (left) (MIXWL) Hourly traffic volumes by link (VPHL) Composite emission factors by link (EFL) Receptor Location X receptor coordinate (XR) Y receptor coordinate (YR) Z receptor coordinate (ZR)  Value 4 (Particulate) 0.021 cm/s 0.021 cm/s 500 cm 0m  Most common annual wind bearing: ~90° Mean annual wind speed: ~3.6 m⋅s-1 Most frequent stability class: C Mean regional daytime mixing height: ~500 m Wind direction standard deviation at mean wind speed: ~10° Mean annual temperature: ~10° C 0 1 = At-Grade -50, 0 m 50, 0 m 1m 30 0m 0m 3600 vehicles⋅hr-1 1 g⋅vehicle-1⋅mile-1  1.8 m  Discussion Because the analysis is for PM2.5, set Pollutant Type to 4 (particulate). Settling and Deposition velocities from values for 2.5 µm particles in (Lapple,1961). Roughness coefficient based on median value for Central Business District from (McRae, et al.,1982). An altitude of 0 m was chosen because Metro Vancouver is a coastal community with most areas having an altitude of less than 150 m above sea level. Summarized one year of hourly weather data (July 2004 – June 2005) from the Vancouver International Airport, White Rock and Pitt Meadows meteorological stations into the mean wind speeds for each of 8 the directional groupings (0, 45, 90, 135, 180, 225, 270, and 315°) and the number of hours that the wind blew in each direction. Mixing height data was derived from historical British Columbia Smoke Control Forecast data. Our modeling unit for bus routes in CALINE4 was an at-grade road link 100 m in length oriented from east to west. A mixing zone width of 30 m was chosen to correspond to 24 m road (6 4 m lanes, with 3 m setbacks on either side). Because CALINE4 output concentrations scale linearly with input emissions, we chose a large input emissions value E (1 vehicle/s at 1g/mile PM2.5 emission) to ensure good resolution in our output concentrations, rather than a value representative of Vancouver bus route emissions. Because the ratio between C to E is constant the absolute values of each are unimportant.  We located 1 receptor at the link midpoint, and other receptors around this link at the midpoints of each of our concentration buffer rings (25 m, 75 m, 150 m, 350 m, 750 m, 3000 m), with 8 receptors in each ring located to correspond to 8 directional groupings.  172  Figure A.2 – Mean unidirectional dispersion curve resulting from on-route emissions of 112 g⋅hr-1 developed using CALINE4. The data points correspond to the concentration at the midpoint of each of the six zones. Error bars represent the range of variation in the 8 directional dispersion curves (0, 45, 90, 135, 180, 225, 270, and 315°).  10  2  10  8  10  6  10  4  10 5000  2  Zone 1  Zone 3 Zone 4 10  0  Population  Concentration ( µg • m-3 )  Zone 2  Zone 5 Zone 6  10  -2  0  1000  2000  3000  4000  Distance from Road (m)  Figure A.3 – Estimated concentration resulting from 112 g⋅hr-1 on-route emissions and the population in the six zones around the 99 B-Line bus route in Vancouver, Canada.  173  A.6.1 Limitations and Assumptions Meteorological conditions are likely to vary both spatially and temporally over a region, which would differentially affect dispersion over the region (Greco, et al.,2007a). However, meteorological data is typically only available from a small number of fixed monitoring sites making it difficult to model meteorological conditions at an intra-regional spatial scale. Further, because distance-based emission factors were used, emissions along the bus route were homogenously spatially distributed. As previously discussed, this has the effect of averaging the intake fraction over the length of the route, and as bus routes typically traverse wide segments of the region, the effects of spatial variability in meteorological conditions on route intake fractions would be attenuated. A similar argument extends to the assumption of homogeneous population densities within zones, which negates the influence of wind direction and implies dispersion would effectively occur uniformly in all directions (Greco, et al.,2007a). On a temporal basis, if significant seasonal or diurnal variability in the meteorological conditions existed in a region, the vehicle scheduling solution could be recalculated with route intake fractions that reflected those meteorological conditions, resulting in unique time of day or year vehicle scheduling solutions. Nevertheless, in the context of this study using annual average, regional meteorological conditions was deemed to be a reasonable simplifying assumption and would most likely underestimate the variability in the route intake fractions.  Like meteorology, there is likely to be variability in the road configurations of routes (e.g., topography, proximate building height, and road width) which would differentially affect dispersion. For example, routes segments in the urban core would be subject to street-canyon effects and as a result, in comparison to the road configuration used in this study, near-road concentrations would be under-predicted. However, because the intake fraction is also a function of the population, the net effect of different road configurations on the route intake fractions is unclear but it is likely to be small and underestimate the variability between routes.  174  Min. Cost  0 d)  8  6  4  2  0  Min. Cost  0  10  Min. PM2.5  5  Min. PM2.5 Intake  10  Min. GWC  15  PM2.5 Intake (g ⋅ day -1)  20  Min. PM2.5  100  Min. PM2.5 Intake  25  Min. PM2.5 Intake  200  Min. PM2.5  30  Min. Cost  400  PM2.5 Intake (g ⋅ day -1)  c)  Min. Cost  0  Min. GWC  100  PM2.5 (kg ⋅ day -1)  d)  Min. GWC  0  200  Min. PM2.5  0.5 300  Min. PM2.5 Intake  5 300  PM2.5 (kg ⋅ day -1)  400  Min. Cost  10  Min. PM2.5 Intake  c)  Min. GWC  1.5  Min. PM2.5 Intake  4  x 10  Min. PM2.5  0  Min. PM2.5  5  Min. GWC  0.5  GWC (tCO2e ⋅ day -1)  10  Min. Cost  Min. PM2.5 Intake  1.5  Min. Cost  0 4  x 10  Min. GWC  15  GWC (tCO2e ⋅ day -1)  b)  Min. PM2.5 Intake  5  x 10  Min. PM2.5  0  Min. PM2.5  15  Min. GWC  Cost ($ ⋅ day -1)  1  Min. Cost  Min. PM2.5 Intake  Min. PM2.5  b)  Min. GWC  1  Cost ($ ⋅ day -1)  Min. Cost  5  x 10  Min. Cost  Min. PM2.5 Intake  2  Min. PM2.5  a)  Min. GWC  Distance (km⋅ day -1) 2  Min. GWC  Distance (km⋅ day -1) a) e) 0.5 40DO  0.4 40DB2 40DB  0.3 40DA  40DH  0.2 40CG  60DB 60DA  0.1 60DH  0  e) 0.2  0.15 40DB2 40DB  40DO  40DA  0.1 40DH  40CG  0.05 60DB 60DA  60DH  0  Figure A.4 - The distributions of the total distance travelled (a,f), operating costs (b,g), climate impacts (c,h), total PM2.5 emissions (d,h), and the health impacts (e,i) by bus category for optimisation scenarios A,B,E, and F and both of Fleet A and B.  175  Appendix B: Micro Scale Modeling B.1  Route Description  B.1.1 Data Collection Activity data was collected using a Garmin