Technical Document and User Guide for the Multi-Modal Passenger Simulation Model for Comparing Passenger Rail Energy Consumption with Competing Modes (2015)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

3 2 Performance Evaluation Framework 2.1 Methodology Overview The background section of the terms of reference for this project noted the limitations of using averages for the energy performance of rail and other competing modes. In particular it noted that:  Passenger rail fuel consumption data may not fully represent impacts, since they are based on broad averages that include many different variations in distance traveled, amenities provided, speeds, operating environment, type of train operated, and form of propulsion. Similarly, energy consumption estimates for competing modes usually represent broad averages that do not necessarily reflect the energy profiles of comparable trips on modes that compete with passenger rail service accurately.  Using disaggregated data, linked more directly to where and how the fuel and energy attributable to specific trips is consumed, can provide a greater understanding of what is actually occurring. In addition, significant variations in fuel and energy consumption can occur by regions of the country and by individual states and metropolitan areas, and these variations should also be taken into account when analyzing comparable modes of travel, along with specific characteristics of available technologies and operating environments. The key objective of this project, as stated in the terms of reference, is: to provide like-for-like comparisons of energy consumption and greenhouse gas emissions for commuter and intercity passenger rail operations and for competing travel modes. In order to accomplish the objective the terms of reference required development of:  A quantitative decision-support tool for evaluating and comparing fuel and energy consumption and GHG emissions by commuter and intercity passenger rail operations and by competing modes of transportation for comparable trips; and  An evaluation of opportunities to improve fuel and energy efficiency and reduce GHG emissions for intercity and commuter passenger rail. To meet the project objective, common energy intensity metrics are used across modes and fuel types. The focus is on the principal modal leg of a full door-to-door trip and each mode is simulated in detail for this principal leg. Overcoming the shortcomings of working with readily available averages for modal performance requires development of relatively detailed simulation models that reflect the variations in performance across regions, seasons and equipment types. Each mode is modeled, within the limitations of publicly available data, such that seasonal, regional and equipment-specific characteristics are reflected in the modal energy and emissions performance. Examples of the limitations of publicly available data in the rail mode are auxiliary power requirements for rail, resistance coefficients for specific types of rail equipment, and engine performance for recent vintage locomotives. Intercity scheduled bus operations also do not have load factors by region or service. This document and the case study analysis indicate where estimates were used to characterize modal characteristics. A common element of the mode-specific simulation modules used in the spreadsheet-based model is the ability to specify the equipment and route characteristics involved in making the specific trip of interest. Default choices are available for the user to select in a quick

4 simulation comparison; however, the user also has the ability to modify the default characteristics or define new characteristics for each mode if desired. The requirement for an “evaluation of opportunities to improve fuel and energy efficiency and reduce GHG emissions for intercity and commuter passenger rail” means that the rail simulation module requires additional details on the technological attributes of equipment such that the energy/emissions performance of alternative technologies can be assessed. Therefore, in addition to a comparison with other modes, the rail mode can be simulated in isolation and in comparison with other rail mode technologies for the same trip. The energy/emissions intensities of modes used in access and egress legs are included and separately identified; however, simple averages are used rather than detailed simulations. Indirect ‘well-to-pump’ energy and emissions intensities are included as a separately identified metric for each leg of a door-to-door trip. The principal leg of each modal trip is simulated in detail such that variations from the average performance are discernable. For example, when an automobile is used as the main leg of a trip, it is simulated in detail for the specified equipment, route and traffic congestion characteristics specified; however, when used as an access/egress leg to other modes, pre-processed averages for a composite vehicle can be used. The default average performances for access/egress by highway modes have been pre-processed for varying levels of traffic congestion. If the user wishes to simulate a new access/egress scenario in detail, it can be done as a separate one-time simulation of the relevant route and equipment used for the access/egress trip. 2.2 Mode Specific Data Constraints and Methodology Influences The types and level of details available to characterize modal energy and GHG intensities vary across the modes. The approach taken within each mode reflects the types of data available. In general terms:  The light-duty-vehicle (LDV) mode has very detailed performance data at the individual vehicle level but has poor in-service performance data. Traffic congestion influences are significant and the LDV module supports local commute and intercity traffic characteristics for five time-of-day periods. Seven generic LDV classes are characterized in conventional and hybrid configurations.  The bus mode has less detail than LDVs for individual vehicle performance but has much less variability in types of vehicles available and better aggregate energy performance data. Intercity and local commuter bus services are characterized by road type and traffic congestion in the same way as the LDV module.  Simulation of individual air mode trips on a second-by-second aircraft movement basis is a significant undertaking and beyond the scope of this project. The air mode has good in-service performance data which allows differentiation of equipment types by trip length as well as assessment of congestion effects. The equipment mix and equipment-specific energy performance are characterized by trip length for five representative types of aircraft.  The data available for the rail mode vary across services — detailed characterization data exist for some types of equipment but not all and the level of aggregation in total-fleet performance reports does not provide enough detail to accurately calibrate/validate individual services or equipment types. Intercity and commuter services are characterized for several generic conventional equipment types and a high speed equipment type.

5 The mode-specific data constraints and associated methodological details for rail, bus, LDV and air are each discussed in the following subsections. 2.3 Highway Modes (LDV and Bus) Methodology The highway modes’ energy and emissions performances are derived via simulation modules which simulate second-by-second movement of a specified vehicle over a set of pre-defined speed profiles (drive schedules) in urban areas and use cruise performance for rural intercity segments. The LDV drive schedules contain a high-acceleration performance drive schedule that is not included in the bus drive schedules; however, the simulation process is the same for both modules. Vehicle resistance to motion is based on rolling and aerodynamic resistance coefficients, vehicle power and energy performance is governed by specified engine and drivetrain characteristics and auxiliary loads are associated with regional climates by three seasons of the year (summer/winter/other). The engine efficiency of the highway modules are based on representative fuel maps from the literature – the bus module uses 2010 vintage fuel maps for 350 hp and 455 hp engines in the EPA’s Greenhouse Gas Emissions Model (GEM) [EPA, 2010] while the LDV vehicles are based on 2004 vintage fuel maps noted on Appendix C to Volume 1 of this report. The fuel maps are normalized to the engine’s minimum brake specific fuel consumption (bsfc) to allow different engine powers and fuel efficiency data to be used. 2.3.1 Bus Characterization The default characterization data are generic composites rather than any one specific bus. A typical bus is characterized using publicly available data for resistance coefficients, drivetrain efficiency, auxiliary loads and diesel engine efficiency. Bus rolling resistance is based on Australian test work {Biggs, 1987] and calibrated with published test track data for trucks as reported in Appendix B of the freight mode comparison study undertaken by English and Hackston [English, G., and D.C. Hackston, 2013]. The bus aerodynamic drag coefficient is set at 0.5 based on the observations of work of Patten et al. who indicate: “it is not unrealistic to expect their [North American intercity buses] drag coefficient to be in excess of 0.50” [Patten et al. 2012, p52]. Patten also cites an advanced European bus that is in service and has achieved a Cd of 0.35. The bus module is calibrated to a typical U.S. operating environment using two main data sources. Commuter bus operating characteristics and energy intensity performance are calibrated with data reported for commuter buses by municipal operators in the Federal Transit Administration’s (FTA) National Transit Database (NTD) [Federal Transit Administration, 2011]. Intercity buses are calibrated to data reported in the NTD-Service Table for short-distance intercity commuter bus travel (total-miles/revenue-miles X revenue car-miles/gallon for commuter bus (CB) operations of the Maryland Transit Administration, 2011) combined with data reported in a 2013 survey of major N.A. bus operators undertaken for the American Bus Association (ABA) [John Dunham & Associates, for the American Bus Association Foundation, 2013, Table 2-5, pg. 13]. The average fuel economy for the two sources is 42.3 L/100-km or 5.59 mpg (average of 6.09 and 5.08 from the ABA and NTD sources respectively). In calibrating bus performance to be in this fuel efficiency range, we used a Cd of 0.5 for a standard 45 ft bus and increased the fuel consumption of the EPA’s 2010 engine by 5% to reflect the mix of older less-efficient buses in service. Someone with more detailed data on specific bus performance might find a different combination of Cd and minimum bsfc that produces the same average results. Occupancy and non-revenue travel data are reported for commuter buses in the National Transit Database; while only occupancy is reported in the ABA’s 2013 survey of operators. We believe that the intercity commuter bus operations would understate the occupancy

6 attained by longer distance intercity scheduled carriers and the average occupancy reported for all operators surveyed in the ABA would be higher than that of scheduled carriers (since charter carriers realize very high load factors and account for over 50% of the operators surveyed). In the absence of route-specific data, the default occupancy used in the present modal comparisons is the average of that realized by the intercity commuter bus travel (passenger-miles / vehicle-revenue-miles for commuter bus (CB) operations of the Maryland Transit Administration, 2011) and that reported for all bus services in the ABA survey (i.e. an average of 32.6 riders). With an estimated distribution of buses having the following seating capacity: 90% with 56 seats, 5% with 81 seats and 5% with 48 seats, the average load factor is 57%. The model also supports simulation of hybrid buses in commuter service. Hybrid performance is characterized with lithium ion batteries having charge/discharge efficiencies based on data from Hofman et al, [2008,Table 6]. 2.3.2 LDV Characterization from Test Data The Light duty vehicle module is similar to the bus module but since there is a vast amount of certification data for LDVs, it is possible to calibrate a number of classes of vehicles. Our calibration/validation of LDVs was undertaken by modifying the minimum bsfc of our generic fuel map and the average weight of vehicles in each of the classes. The approach is similar to that taken by Ates in modeling light and medium duty vesicles and we used Ates’s data for LDV transmission and differential efficiency coefficients [Ates, M, 2009, p.72]. This calibration process is described in the section. In the subsequent section we describe the steps we then took to present real-driving conditions/environment to be used in the simulations. The EPA requires LDV manufacturers to provide coast-down data coefficients for each LDV- model sold in the U.S. In addition, EPA requires that each LDV’s energy performance be measured or calculated in the generation of the EPA’s city/highway fuel economy label. These publicly available characterization data are more detailed than all the other passenger modes. However, details of the in-service performance of LDVs are survey- based and have much more uncertainty than is typically available for the other passenger modes. Details of engine and transmission characteristics are not as publicly available for LDV as for the other modes. Also, the range of technologies deployed and the range of equipment types are much wider than for the other modes. Each year the EPA publishes the sales-weighted class-average performance of LDV for several classes of new vehicle. The class-averages reflect the actual mix of vehicles sold within each class in the U.S. during that year. We grouped LDVs for MY-2011 on the basis of exhibiting similar coast-down resistance coefficients. Default parameters are developed for the following six representative vehicle classes from 2011:  Small cars (Sm-Car)  Midsize cars and all station wagons (Mid-car/SW)  Minivans and non-truck SUVs (Mini-V/sm-tSUV)  Large cars, medium truck-SUVs and small pickup trucks (Lg-car/cSUV/smPU)  Large pickup trucks (PU)  Large truck SUVs (Lg-tSUV) The six individual class-average vehicles are provided as user-selectable default vehicles. In addition, four composite vehicles are provided to typify the performance of the following specific mixes of vehicles:

7  the EPA’s sales-weighted composite vehicle,  a composite based on the estimated mix of personal LDVs used in local trips,  a composite based on the estimated mix of personal LDVs used for intercity trips, and  a composite based on the estimated LDV mix for taxis. In relation to the EPA sales-weighted vehicle-mix, the vehicle composite for local commuting assumes a shift from large to smaller vehicles, the intercity travel mix assumes a shift from smaller to larger vehicles and the taxi mix assumes a larger proportion of midsize vehicles, no pickup trucks and a higher proportion of hybrid vehicles (Note: these alternates are illustrative estimates at this point – to be refined as data is located). In each case one must recognize that the default vehicles are representative of the composite-class of LDVs being simulated and that individual vehicle types within the class could provide significantly better or worse performance. As noted above with respect to the higher-efficiency side of the range, the model does separately simulate hybrid vehicles within each class. As discussed later, driver-behavior and other factors can also lead to variations and decreases in operational fuel economy (FE) of any one specific vehicle. The relative 2011 MY sales distribution and the estimated distribution of derived composite vehicles are summarized in Table 1. Table 1. LDV Class Distributions used for Composite Vehicles Class 2011 MY Proportion of Sales a) Local Travel b) Intercity Travel b) Taxi b) Sm-car 17.70% 22.70% 14.70% 0.00% Mid-car/SW 25.40% 28.40% 23.40% 40.00% Mini-V/sm-tSUV 5.30% 5.30% 5.30% 50.00% Lg-car/cSUV/smPU 28.00% 25.00% 30.00% 10.00% PU 14.10% 11.10% 14.10% 0.00% Lg-tSUV 9.50% 7.50% 12.50% 0.00% Sources: a) derived from EPA data, b) TranSys Research Ltd illustrative assignments (to be refined as data is located). The actual market shares of individual vehicles are not published and the variation of class- specific performance across manufacturers can be significant. Our analyses are based on the 2011 model year (MY) fuel economy trends report [EPA, March, 2012]. In 2011 two standard deviations of manufacturers’ averages per class ranged from a low of +/-22% for midsize cars to a high of +/-27% for large cars. In all classes, the sales-weighted average demonstrated better fuel economy than the simple average of all manufacturers. Our LDV module provides class average performance characteristics which are developed by applying class-specific modifiers to a generic engine fuel-map and a generic six-speed transmission such that the EPA’s sales-weighted average performance is exhibited when simulated on the EPA’s underlying certification drive schedules. The variation can be due to differences in vehicle resistance parameters and/or engine/drivetrain losses. A somewhat arbitrary assignment was made to adjust the resistance coefficients and vehicle mass by 50% of the initial error and the engine/drivetrain losses by 50% of the initial error. As the model’s initial average parameters were based on the simple average of published LDVs rather than the sales-weighted average, most modifiers reduce resistance parameters and loss-factors to attain the sales-weighted average performance.

8 A similar process is used to characterize the 2011 “driven fleet” to reflect the relative performance of older vehicles with an appropriate vehicle age distribution. The algorithm processor assumes that 50% of the difference in fleet-average fuel economy of the new model year relative to the 2011 sales-weighted fleet is due to drive-train efficiency (engine and/or transmission efficiency is scaled by 50% of the fuel economy difference) and 50% is due to vehicle body design or fleet composition (the 2011 fleet average weight and resistance coefficients are scaled by 50% of the fuel economy difference). The LDV characterization data for the model year 2011 are summarized in Table 2.

9 Table 2. MY 2011 LDV Sales-Weighted Characterization Data Model Development Phase Class Derived Averages for EPA Classes of MY 2011 LDVs Derived Composites small mid/S W MV/ Sm- tSUV Lg/cSUV / smPU PU- truck Lg- tSUV Local d) Inter- city d) Taxi d) 2011- Sales- weighted 2011 Driven Fleet Initial Base Parameters a) a (N) 141.77 171.21 164.86 211.70 225.63 238.72 185.42 194.81 172.1 191.09 N.A. b (Nsm -1 ) 2.407 2.416 3.256 4.462 5.963 6.360 3.659 4.066 3.04 3.907 N.A. c (Ns 2 m -2 ) 0.418 0.455 0.567 0.628 0.670 0.672 0.536 0.565 0.53 0.554 N.A. Mass (kg) 1,496 1,590 1,828 2,040 2,397 2,554 1,856 1,958 1,75 1,917 N.A. Power (kW) b) 121 133 163 189 233 253 166 179 153 174 N.A. Hybrids c) 8.00% 1.88% 0.40% 1.88% 0.40% 0.40% 3.86% 3.86% 10% 2.2% 1.3% non-Hyb-CVTs 4.00% 9.99% 6.90% 9.99% 6.90% 6.90% 6.74% 6.74% 8.00% 7.8% 2.9% Initial FE Difference -2.8% 8.2% -2.4% 9.8% -9.6% 2.7% N.A. N.A. N.A. 4.5% -7.9% Adjusted Model Parameters to get Sales- weighted FE loss reduction -1.4% 4.1% -1.2% 4.9% -4.8% 1.4% N.A. N.A. 1.9% 2.3% -4.0% e) a (N) 143.78 164.19 166.87 201.31 236.46 235.47 182.35 191.57 169.2 187.16 194.51 b (Nsm -1 ) 2.44 2.32 3.30 4.24 6.25 6.27 3.61 4.01 3.00 3.83 3.98 c (Ns 2 m -2 ) 0.42 0.44 0.57 0.60 0.70 0.66 0.53 0.56 0.52 0.54 0.56 Mass (kg) 1,517 1,524 1,851 1,940 2,512 2,520 1,828 1,929 1,729 1,878 1,952 Final FE Difference -0.7% 0.1% -0.6% 0.3% 1.0% 0.0% N.A. N.A. N.A. 0.3% N.A. Source: TranSys Research Ltd. analysis of EPA fuel economy data. Notes: a) Resistance Parameter Units are: N – Newton; Nsm -1 – Newton / (meter/second); Ns 2 m -2 – Newton / (meter/second) 2 b) Average power for each group is based on a derived power/weight equation for the MY 2011 data: P= -86.933 + 0758W; with P (hp) and W (lb). c) Conventional vehicles and hybrid vehicles are separately characterized and simulated in the model and default proportions of hybrid vehicles are included within each class’s characterization data. If one wishes to only simulate a hybrid (or only a conventional) LDV the default proportion- of-hybrids can be set to one or zero accordingly. d) These composites are illustrative estimates pending better data for user updates. e) The value used in LDV-Resist sheet of the model is a different number and includes an offset by the loss-reduction already included in the Sales-weighted composite (i.e. -0.017 = -0.079/2 + 0.023) which is the reference vehicle used for 2011 and future year composite vehicles. Source: TranSys Research Ltd., derived from data in: EPA, Fuel Economy Guide for DOE-2013; and Light-Duty Automotive Technology, Carbon Dioxide Emissions, and Fuel Economy Trends: 1975 through 2012, EPA-datasheets, 2013.

10 2.3.3 Adjustment of LDV Characteristics to Reflect Driving Conditions The characteristics developed above are for the running performance of LDVs. All LDVs experience worse fuel economy on start-up and this initial fuel penalty increases with colder temperatures. The increase has a significant impact on short commuter trips but is less consequential for intercity travel. We estimated the cold-start fuel increment on the basis of dynamometer tests of hot and cold starts for a 2009 Jetta diesel published by DOE [Argonne National Laboratory, D3 website. http://webapps.anl.gov/D3/index.html]. The 13.5% incremental fuel consumed over the 7.46 mi long UDDS test cycle when cold versus hot was adopted as representative of all LDVs. The value is pre-processed by simulating each default vehicle over the same UDDS drive schedule and applying the 13.5% factor to get the fuel/start value. The resulting ‘cold start’ fuel increment for the Sales-weighted 2011 MY at 22ºC is 0.1136 kg. The influence of temperature on the cold-start fuel increment follows the EPA’s derived equations assuming a 12-hr or greater soak time for the forward trip and 9 hr. for the reverse trip (100% and 87.5% of the cold-start fuel increment respectively). The ratio ‘R’ of cold-start fuel at ambient temperature (Ta) relative to the cold-start fuel at 75 ºF is [EPA, Final Technical Support Document Fuel Economy Labeling of Motor Vehicle Revisions to Improve Calculation of Fuel Economy Estimates, EPA-420-R-06-017, December, 2006]: 𝑹 = 𝟏 + 𝒂(𝑻𝒂 − 𝟕𝟓) + 𝒃(𝑻𝒂 − 𝟕𝟓) 𝟐 Equation 1 where: Ta = ambient temperature of interest in ºF R is the ratio of cold-start fuel increment at ambient temperature Ta divided by the cold-start fuel increment at an ambient temperature of 75ºF. a and b are estimated coefficients; which for gasoline engines are: a = -0.01971 b =0.000219 and for diesel engines are: a = - 0.00867 b = 0.000096 The running performance of LDVs is derived by the EPA and manufacturers via dynamometer and coast down tests. The test conditions are ideal in relation to actual driving conditions and the drive schedules are not necessarily representative of any one particular journey made on a specific time-of-day and season. The EPA discusses a wide range of factors which influence the translation of lab test results into real-world experience [EPA-420-R-06-017, December, 2006]. The assessment (Table III.A-28. of EPA-420-R-06-017) indicates about a 12%-to-15% decrement to predicted 5-cycle fuel economy due to various non-test factors; most of which EPA believes relate to wind ~ 6%, road surface ~ 1.4% – 3.2%, road gradient ~ 1.9% and fuel quality ~ 1.1 – 1.5%. We specifically model many of these factors and others via specified trip input data. Table 3 summarizes how these FE-influencing factors are treated in the model. Still, we note that driver-behavior, auxiliary power usage and vehicle maintenance can all affect the FE of a specific vehicle. There will always be a range of FE performance around any derived average. Our objective is to inherently model most parameters and provide an average in- service FE that is representative of real-world experience.

11 Table 3. Adjustments Made to Test-based Characteristics to Reflect Actual Driving Conditions and the Driven Fleet Factor Assumption / Treatment in the Model Source Reference Road/tire condition resistance factors An estimated 10/30/50/10 mix of concrete/smooth- /medium-/rough-asphalt pavement is adopted relative to an estimated 30/70 mix of concrete/smooth-asphalt for test conditions. The corresponding increase in rolling resistance coefficients (a and b) is 12%. TranSys Research Ltd estimates of road type usage. Biggs, 1987 for relative road type influence. Aerodynamic drag (wind effect) Wind increases aerodynamic drag for most yaw angles except those approaching 180º (a pure tailwind) where it reduces drag. We adopt an average 3.5% increase in aerodynamic drag for LDV (and use the same factor in the bus and rail modules). TranSys Research estimate. Seasonal aerodynamic drag (temperature effect) Aerodynamic drag varies directly with air density and the density of air is negatively correlated to its absolute temperature (º Kelvin or º Rankine). This temperature dependence is applied to LDV (and also bus and rail mode) simulation modules. Marks’ Handbook for Mechanical Engineers, 8 th Edition, McGraw- Hill. Seasonal cold-start fuel consumption (temperature effect on tires, and engine/ drivetrain friction losses). EPA has developed an equation to estimate the incremental impact of colder temperatures on the ‘cold start’ fuel consumption. The basis of the dynamometer data used does not include the aerodynamic impact noted above so the effects are additive. We apply the EPA equations for incremental cold start fuel consumption on the basis that all our rail-competitive trips occur after 12 hr. of sitting (soak time) for the forward trip and 9 hr. for the return trip. EPA-420-R-06-017. Seasonal auxiliary power loads Average base auxiliary power is set at 0.75 kW. The air conditioning compressor, when on, is estimated to add 2.67 kW when running and 1.1 kW at idle. Actual AC power varies with technology and can be higher or lower than the average default values. Usage is based on regional temperature and humidity profiles. The blower is estimated to add another 0.5 kW 100% of the time in winter and summer. [Rugh, John P. (NREL) et al.]. for AC compressor power and usage. Gradient Specific road gradients are not included in the time-step drive schedules. The gradient influence on the FE of LDVs is assessed as the probability that potential energy recovery on downgrades is lost because the vehicle is braking for speed reduction purposes. The same procedure is used for the rail and bus modules. Distributions of gradient severity are estimated by region. The default vehicles are based on —and are thus representative of— 2011 technology and class proportions. If one wishes to consider an age distribution, the model assumes the fleet average performance enhancement that was reported by the EPA for the intervals 1990-2004, 2004-2011 [EPA-420-R-1 3-001, 2013 - Table 1 (cars and trucks) 2012 Fuel Economy Trends Report]. For the interval 1990-2004 the average trend was a 0.55% annual decrease in fuel economy (from 21.2 to 19.3 mpg), while for 2004 – 2011 the average trend was a 2.15% annual increase in fuel economy (from 19.3 to 22.4 mpg). The EPA estimates the relative usage of various calendar years (CY) in generating its annual GHG emissions inventory. The effects of using the CY usage distribution (or the ‘driven-fleet’) on fuel economy and GHG emissions are summarized in Table 4. The EPA’s

12 CY usage distribution in 2011 as adopted in its GHG inventory [EPA, 430-R-12-001, April, 2012] reduces the fleet efficiency to 91.32% of the sales-weighted 2011 MY’s efficiency (or 1/.9132 = 1.095 times more fuel is consumed by the 2011 driven-fleet). The CO2e emissions intensity of LDVs also varies with age. While the CO2 ratio is constant with fuel type, the N2O and CH4 emissions ratios vary with the regulatory period. The EPA’s emissions factors for cars were constant from 2009 to 2011 at 3.6 mg/mi and 17.6 mg/mi for N2O and CH4 respectively [EPA, Emission Factors for Greenhouse Gas Inventories, 2011]. Even with the higher CO2-equivalency factors for these gases, the impacts are minimal — CO2 emissions are 3.172 kg/kg-fuel and with the fuel intensity in 2011 at 0.126 kg/mi (22.4 mpg) is 0.3997 kg/mi. The CO2-equivalent emissions were: CO2e (kg/mi) = 0.3997 +298 * 3.6/10 6 + 25 * 17.6/106 = 0.40121 Equation 2 Thus, the impact of CH4 and N2O emissions post 2009 are a 0.37% increase over CO2 emissions. The N2O and CH4 emissions rates are based on the ftp certification cycle and include a g/start factor allocated over the 7.4 mi ftp route distance. The running emission rates are an even smaller proportion than the 0.37% shown above. Scaling the impacts for the fleet average MY composition by usage leads to a greater impact but it is still relatively small. The above cited EPA fuel economy trends report was used to estimate the 2011 fleet average emissions rate by applying MY fuel economy and MY emissions factors in the same way as the fleet-average fuel economy was derived. The resulting fleet average scale factors relative to the modelled 2011 sales mix were: 4.13 and 1.11 for N2O and CH4 respectively. We estimate the 2011 LDV driven-fleet to have a GHG emission intensity of: CO2e = 0.375 kg/start + 3.19 kg/kg-fuel-running. Equation 3 Table 4. Performance of the 2011 Sales-weighted and 2011Driven Fleets Fleet composition Fuel Economy Cold Start* GHG emissions Intensity Running Starting CO2 N2O CH4 CO2e CO2e mpg kg/km kg/start kg/kg mg/kg mg/kg kg/kg kg/start 2011 sales weighted 22.37 0.126 0.114 3.172 14.19 59.85 3.178 0.363 2011 driven fleet 20.67 0.137 0.123 3.172 54.470 71.71 3.19 0.373 scale factors applied 0.913 1.095 1.095 N.A. 3.84 1.20 N.A. N.A. * Cold start is the incremental fuel consumed over the initial few miles of travel (before the engine, drivetrain and tires warm up) following 12 hr of sitting stopped, versus the fuel consumed over the same trip with a fully warmed vehicle. The base value shown is for an ambient temperature of 22ºC. Source: TranSys Research; derived from EPA emissions sales-weighted and MY usage data (see text).

13 The EPA has developed a range of speed profiles to characterize the influence of traffic congestion on individual vehicle speed variations on freeway, urban and arterial streets. However, no single drive schedule provides a realistic characterization of a specific commuter or intercity trip. In the LDV and Bus simulation modules, the user specifies the proportion of each of eight individual drive schedules encountered in making the trip being simulated. Different congestion performance can be specified for 5 specific time-of-day congestion intervals. The user is guided with feedback of the total delay encountered when making the specified trip during each time of day. Default values for the matrix are provided for a typical large urban city and a smaller urban city as developed for the case studies undertaken within the project. Table 5 illustrates the matrix involved and the illustrative data for origin and destination (O and D) cities involved in an intercity trip. Table 5. Illustrative Proportional Allocation of Congestion Encountered by Time- of-Day (% of route) Location Time Period Creep LOS-F City Streets Urban Arterial Urban FW LOS-E FW- cruise FW LOS-F FW- access US06 Delay (min / 10- km) ~0.9 km/h ~14 km/h ~25 km/h ~40 km/h ~75 km/h ~119 km/h ~33 km/h ~100 km/h Arterial (O and D) a.m. pk 3% 7% 25% 65% 30.2 p.m. pk 3% 12% 10% 75% 25.1 midday 15% 85% 9.0 shoulders 5% 95% 2.9 overnight 100% 0.0 Urban FW (O) a.m. pk 3% 12% 30% 20% 5% 15% 15% 56.3 p.m. pk 3% 5% 25% 27% 5% 10% 25% 38.8 midday 20% 25% 25% 5% 25% 19.1 shoulders 30% 50% 20% 5.8 overnight 5% 75% 20% 0.0 Urban FW (D) a.m. pk 3% 30% 25% 5% 20% 17% 47.0 p.m. pk 3% 20% 35% 12% 10% 20% 31.4 midday 20% 60% 20% 4.3 shoulders 10% 70% 20% 2.0 overnight 5% 85% 10% 0.0 Source: TranSys Research Ltd., for illustration only. 2.4 Air Mode Methodology 2.4.1 Air Made Analytic Framework Simulation of individual air mode trips on a second-by-second time-step simulation is complex and requires knowledge of many parameters that are not readily available. There are a few complex simulation models that take this approach (e.g. the SAGE program sponsored by the US FAA [Federal Aviation Administration, 2005] and the BADA model in Europe [European Organisation for the Safety Of Air Navigation, 2009]. Such detail is not required for this project as the US Bureau of Transportation Statistics publishes good in- service performance data for air mode operations in the U.S. and it is the final in-service performance that is relevant for a modal comparison. The published data [Research and Innovation Technology Administration, Bureau of Transportation Statistics (BTS), 2013] are used to define congestion effects on energy performance by equipment type and trip length for typical U.S. domestic scheduled service operations.

14 The air mode has some complexities that require flight segmentation in order to assess GHG intensity. The warming effects of emissions from aircraft at cruise altitudes are higher than emissions on the ground and low altitudes. The effects of aviation on the atmosphere and climate were considered comprehensively in the 1999 IPCC Report [Intergovernmental Panel on Climate Change, 1999]. The effects are complex, and poorly understood in some respects, due to the complexities of the chemical processes in the atmosphere. The prime contribution to global warming is expected to be through emissions of carbon dioxide, which bear a fixed relationship to aviation fuel use, and which are assumed to be “well-mixed” with emissions from other sources, and to have the same radiation-forcing1 potential as other emissions. In addition, emissions of nitrogen oxides from aircraft in the troposphere and lower stratosphere are expected to contribute to the formation of ozone more effectively than at ground level, further contributing to global warming. The effect is limited somewhat by the fact that in this process, nitrogen oxides reduce the atmospheric concentration of methane, which has the effect of reducing global warming. Aircraft also produce water vapor, which is eliminated rapidly in the troposphere, but in the upper atmosphere forms contrails (condensation trails) which can persist and even form cirrus clouds, and is expected to contribute further to global warming. The relative extents of these effects are uncertain, particularly the latter. The IPCC report estimated that in the mid-1990s aviation contributed about 2% of man-made CO2, and about 3.5% of man-made radiation-forcing, excluding the possible effects of water vapor in cirrus cloud. The report also cautioned about the uncertainty as follows: “The total radiative forcing due to aviation (without forcing from additional cirrus) is likely to lie within the range from 0.01 to 0.1 Wm-2 in 1992, with the largest uncertainties coming from contrails and methane. Hence the total radiative forcing may be about two times larger or five times smaller than the best estimate.” Later in the summary the IPCC concluded: “Over the period from 1992 to 2050, the overall radiative forcing by aircraft (excluding that from changes in cirrus clouds) for all scenarios in this report is a factor of 2 to 4 [times] larger than the forcing by aircraft carbon dioxide alone. The overall radiative forcing for the sum of all human activities is estimated to be at most a factor of 1.5 [times] larger than that of carbon dioxide alone.” That statement has proven somewhat ambiguous, with the implication of the latter sentence being overlooked, leading to interpretations that aviation emissions contribute 2 to 4 times as much to global warming as other emissions. The full statement that the ratio of radiation forcing to CO2 from aviation was 2-4 and for other sources “at most …1.5” meant in fact that the relative radiation forcing from aviation emissions versus the worst case in other emissions could be between (2/1.5) and (4/1.5), or approximately 1.33 to 2.67. The IPCC’s estimate that aviation contributed 2% of CO2 and 3.5% of radiation-forcing shows that the best (circa 1999) estimate was that its radiation-forcing proportion was 1.75 times that of the proportion of CO2 alone. In 2007 the Air Transport Bureau of the 1 Radiation forcing is the measure of heat flux associated with greenhouse gases (and other geophysical energy fluxes). Heat flux is defined as the amount of thermal energy transferred across a unit area over a time interval. Radiative heat flux is measured as Watts per square meter (W/m2 or Wm-2).

15 International Civil Aviation Organisation (ICAO) asked the Intergovernmental Panel on Climate Change (IPCC) to update its opinion to reflect that the impact is not as great as originally believed — its estimate of total radiative forcing was reduced from 3.5% to 3%, while its estimate of CO2 contribution remained at its prior estimate of 2%.2 Using this update, air mode GHG intensities for the cruise phase would be adjusted with a multiplier of 1.5 times CO2 emissions. The recent IPCC Fifth Assessment appears also to have endorsed the lower estimate of radiative forcing from persistent contrails but publication is scheduled for early 2014. The draft WGI report had included a combined RF for contrails and high cirrus in 2011 to be +0.05 (+0.02 to +0.15) W m-2, but the overall estimate for aviation CO2 in that year remained unpublished. 3 One can expect the CO2e/CO2 ratio to change for aviation with further scientific investigation; and the default factor of 1.5 adopted in the model should be modified as appropriate. The updated IPCC report was not available during this project and users should monitor and update the model if the IPCC’s final report modifies these values. Recognizing this higher-altitude GHG multiplier for aircraft emissions requires segmentation of the energy consumed at higher altitudes from that consumed in landing-and-takeoff (LTO), climb-out and descent phases of an air trip. Fortunately, data exist to support the segmentation of flights by equipment type. The U.S. DOT data noted above [Research and Innovation Technology Administration, Bureau of Transportation Statistics, 2013] provides information on the overall trip energy intensity by equipment type. Jet engine emissions certification data published by ICAO (ICAO Aircraft Engine Emissions Databank) includes data on fuel consumption for a typical landing and takeoff cycle and is required for all jet engines greater than 2,200 lb (9.8 kN) thrust. The ICAO engine database is maintained by and available from the European Aviation Safety Agency (http://easa.europa.eu). The two data sets (BTS and ICAO) are used to develop a model of air mode performance by equipment type. The data are assessed to provide the energy performance by flight segment for five representative equipment types: The default characterization data provided with the model are based on 2011-2012 operations of domestic US scheduled air carriers and can be updated by the user as desired in future years as air technology and operations practices change. The following five representative types of aircraft are assessed: 6. Turboprops (TP) 7. Small Regional Jets (SRJ) (defined here as jet aircraft with less than 50 seats) 8. Regional Jets (RJ) (defined here as short-range jet aircraft with 50 to 89 seats) 9. Narrow Body Jets (NBJ) (defined here as jet aircraft with greater than 89 seats in a single aisle configuration) 10. Wide Body Jets (WBJ) (defined here as jet aircraft with greater than 89 seats configured with more than one aisle) The Bureau of Transportation Statistics’ data (Table 254, Air Carrier Traffic and Capacity Statistics by Aircraft Type) for 2011 and 2012 is filtered to remove air cargo and air charter operators, operators with fewer than 65 flights per quarter and flights greater than 3,000 2 ICAO website March, 2010: http://www.icao.int/icao/en/env/aee.htm. 3 IPCC Working Group I Contribution to the IPCC Fifth Assessment Report Climate Change 2013: The Physical Science Basis, Final Draft Underlying Scientific-Technical Assessment, version September 26, 2013 (Unpublished but available at http://www.climatechange2013.org/images/uploads/WGIAR5_WGI-12Doc2b_FinalDraft_All.pdf.

16 great-circle miles (GC-mi) in length. The combined 2011-12 filtered dataset of US scheduled carriers is then analyzed to provide an indication of the mix of aircraft used in meeting the demand for different trip lengths. Each aircraft type is also analyzed to provide an indication of its average load factor and its per-seat fuel intensity (kg/seat-GC-mi). The ICAO engine emissions database is used to derive the LTO fuel consumption for a sample of engines used in each of the jet aircraft types and simulation-based data in the European Environmental Agency’s CORINAIR database (an inventory of air emissions) are used to derive LTO fuel consumption for a representative turboprop aircraft. The LTO data are used to identify fuel use during the landing and takeoff cycle (kg/seat-LTO) for each aircraft type thereby permitting the segmentation of fuel consumption into the LTO/climb-out/descent phase and the cruise phase of a trip. Climb-out and descent are estimated to occur at 3000 ft/minute for all aircraft and the average of the climb-out and descent fuel consumption rate is considered to be the same as the average rate during the cruise phase. All aircraft but turboprops are considered to cruise at altitudes above the 25,000 ft floor associated with exacerbated ‘high altitude’ impacts from emissions. Thus, emissions from all types but turboprops are assessed to have a higher effective impact at cruise altitude. 2.4.2 Air Mode Default Characterization Tables Five aircraft types were characterized using operating reports from US domestic scheduled airlines for the years 2011 and 2012 [http://www.transtats.bts.gov]. The reports include quarterly totals of key operating metrics by aircraft. We processed the data to group individual aircraft into the five classes (TP, SRJ, RJ, NBJ, WBJ). Not all of the reports included fuel consumption data. The full dataset was assessed in determining proportional usage of aircraft type by trip length and load factor, whereas the energy intensities were derived on the basis of those reports that included fuel consumption information. Fuel consumption was reported for 35% of the turboprop seat-miles, 44% of the small regional jet seat-miles and about 49% of the three other aircraft types. The mix of operators and specific aircraft involved in fuel-reported versus not-reported did not suggest any bias in the data for jet aircraft. However, the turboprops with fuel reported tended to be larger aircraft and thus, the fuel intensity of turboprops might not be representative of the overall turboprop fleet. Nonetheless, we believe the larger turboprops are likely to provide a better representation of the turboprop fleet used in larger centers where rail services are present. The resulting mix of aircraft usage (seat-GC-km) by seven trip length segments (GC-km) is summarized in Table 6. The upper boundary of each segment is shown in the second row (GC-mi) and the third row (GC-km). The seventh data-column indicates the upper limit of the data analyzed (i.e. 3,000 mi). For trips less than 250 GC-mi (402 GC-km) turboprops account for 81.5% of the seat-mi, small regional jets account for 16.5% and regional jets account for 1.9% of the seat-mi. For distances in segment 4 (750 to 1000 GC-mi) narrow body jets account for 94.9% and regional jets account for 4.5% of the seat-mi. This is the default distribution of aircraft used in performing a ‘representative’ air leg simulation. If desired, a user can override this distribution by indicating one specific aircraft type (or any alternate distribution) for the air leg of a simulated trip.

17 Table 6. Proportional Aircraft Usage (%-seat-miles) by Trip Segment Length Segment No. 1 2 3 4 5 6 7 Min (GC-mi) 0 250 500 750 1000 1,500 2,000 Max (GC-mi) 250 500 750 1,000 1,500 2,000 3,000 Max (GC-km) 402 805 1,207 1,609 2,414 3,219 4,828 TP 81.5% 6.0% 0.0% 0.0% 0.0% 0.0% 0.0% SRJ 16.5% 6.1% 0.0% 0.0% 0.0% 0.0% 0.0% RJ 1.9% 80.0% 22.0% 4.5% 0.0% 0.0% 0.0% NBJ 0.1% 8.0% 78.0% 94.9% 92.1% 77.7% 0.0% WBJ 0.0% 0.0% 0.0% 0.6% 7.9% 22.3% 100.0% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% Source: TranSys Research Ltd. via analysis of Table 254 Air Traffic Data for years 2011-2012 [U.S. BTS, 2013 (http://www.transtats.bts.gov)]. The average load factor (as measured by total revenue-passenger-miles / total revenue- seat-miles) for each aircraft type for the 2011-12 interval is summarized in Table 7. Load factor is seen to increase with the size (and associated longer range) of the aircraft. Table 7. Average Load Factor by Aircraft Type Aircraft Type TP SRJ RJ NBJ WBJ Load Factor 69.93% 74.29% 78.16% 83.27% 86.76% Source: TranSys Research Ltd. via analysis of Table 254 Air Traffic Data for years 2011-2012 [US BTS, 2013 (http://www.transtats.bts.gov)] The energy intensity of each aircraft type was derived via regression analysis of those records in the filtered dataset that contained fuel consumption. As there is a wide range of individual aircraft sizes within each type/class, the fuel consumption data were normalized by number of seats. Regressions were performed (with a forced zero coefficient) for each aircraft type with fuel consumed (kg/seat) as the dependent variable and the number of LTO cycles and the total GC-km traveled for the aircraft/quarter as the independent variables. Two regressions were reassessed on different metrics. The SRJ regression did not provide a good fit when done on a kg/seat basis. Upon review it was determined that all SRJs had the same engine regardless of seating capacity. Thus, a regression based on fuel consumption per aircraft gave a better fit to the data than one based on kg/seat. The regression results in kg/aircraft were then adjusted to kg/seat on the basis of the average seating capacity of 40.5 seats per SRJ in the dataset. The initial regression of WBJ was not significant in the LTO term so that term was dropped from the final regression. Also, the WBJ data displayed a meaningful proportion of cargo being carried (11.9% by weight). To facilitate allocation of fuel between cargo and passengers, the WBJ regression was undertaken on the basis of kg-fuel/payload capacity. The results were then allocated to seats on the basis of 178.6 kg-payload/seat (with 11.6% of fuel going to cargo and 88.4% going to passenger seats). The final regression results are presented in Table 8. All regressions display a high explanation of data variation (adjusted R-square values all exceed 0.93). All final coefficients were significant (t-statistics > 2.0) though the significance was somewhat less for the SRJ and RJ categories than for the other categories.

18 Table 8. Direct Results from the Raw Regression Results AC Type Number of data points Adjusted R Square Coefficient Coefficient Values t Stat P-value TP 48 0.9674 kg/seat-LTO 10.62 11.30 7.16E-15 kg/GC-skm 0.011222 4.59 3.43E-05 SRJ* 18 0.9363 kg/seat-LTO* 11.457 6.38 0.000 kg/GC-skm* 0.045 2.53 0.022 RJ 228 0.9314 kg/seat-LTO 8.888 2.69 0.00766 kg/GC-skm 0.039004 9.25 1.72E-17 NBJ 374 0.9850 kg/seat-LTO 6.814 8.26 2.57E-15 kg/GC-skm 0.0224 35.06 6E-120 WBJ** 121 0.9887 kg/PL-km** 0.157325 201.02 1.4E-153 Source: TranSys Research Ltd, analysis of Table 254 Air Traffic Data for years 2011-2012 [US BTS, 2013 (http://www.transtats.bts.gov)]. Notes: * The SRJ regressions did not provide a good fit when done on a kg/seat basis (since most SRJs have the same engine regardless of seating capacity). The regression was done on the basis of kg/aircraft and adjusted to kg/seat on the basis of the average seating capacity of 40.5 seats. ** Because WBJs had a significant proportion of cargo, the WBJ regression was done on the basis of payload capacity (kg/PL-km) so that the cargo and passenger fuel could be allocated. The WBJ regression did not provide a statistically significant result for the LTO coefficient so that term was dropped from the final regression. While the regression results provide meaningful data, the confines of the data prevent a literal interpretation of the results (i.e. the kg-fuel / seat-LTO is not a direct measure of fuel consumed in a LTO cycle). In general the resulting coefficients had higher values for LTOs and lower values for distance traveled than would be expected from the engineering relationships. This is because the distance traveled in the data is simply the GC distance between each origin-destination (OD) pair rather than the actual distance flown. The actual distance flown will always be higher than the GC distance and involves incremental distance related to:  takeoff and landing headings that are constrained by runway layout and wind direction and are not the direct GC alignment headings for the OD,  following actual navigation points between the OD pair, and  landing delays (circling an airport) while awaiting a landing slot. These additional en route travel distances are somewhat independent of flight distance and show up as an increment in the LTO regression coefficient. While the interpretation of the regression result does not matter in generating the total fuel consumed, it does matter to the GHG emissions from aircraft since the effects of emissions at cruise altitudes are different than those emitted during the LTO cycle. Thus, the initial regression results were adjusted to reflect the actual fuel consumed in the LTO cycle, and the en route incremental distance quantity was transferred to the cruise segment of the flight. The difference between the regression-based LTO fuel and the

19 ICAO-based LTO fuel was then allocated to the cruise portion of the flight. The resulting coefficient (kg/GC-km) can be interpreted as the product of two parameters: 1) the kg/km that one would get from a simulation model or direct measurement of fuel consumed and distance traveled applied to the GC-distance, and 2) the adjustment factor that scales the GC-distance to the actual miles flown. The above interpretation is not required to make use of the data, but does allow one to assess the reasonableness of the coefficients by assessing the reasonableness of the underlying parameters. We went through this exercise for each of the aircraft types, using published SAGE simulation model predictions [FAA-EE, (Appendix C), 2005] for a sample of aircraft in each aircraft-type group. In addition to the transfer of fuel from LTO to cruise segments, the forced zero nature of the regressions led to varying levels of bias in the total results. This bias was factored out of the final coefficients such that the application of the coefficients produced no bias in the predicted total fuel consumption. The resulting coefficients and the scale-factor inherent to the underlying kg/GC-km values are presented in Table 9. As can be seen, the largest proportional impacts are on the short-range aircraft – turboprops and small regional jets experience scale factors of 1.335 and 1.444 respectively. The long-range aircraft have decreasing multipliers — values of 1.114 for NBJ with an average GC-trip distance of 1,436 km and 1.061 for WBJ with an average GC-trip distance of 2,591 km. The results are a reflection of the impact of a relatively fixed extra distance having a greater impact on shorter trips than on longer trips. It could also reflect a lower priority being allocated to smaller aircraft when vying with larger aircraft for landing slots during times of congestion and/or the short range aircraft being used more for peak-period commuter travel and thus more frequently exposed to congestion. The higher multiple for SRJs than TPs could reflect the fact that the SRJs are mainly used in the east serving major airports, whereas the turboprops have higher usage in the Midwest and Pacific regions and service smaller airports. The average scale factors (Implied GC Multiplier in Table 9) are composites of all flights reported for each aircraft type. One can expect the multiplier to vary between peak and off-peak travel times. Reynolds found in-flight delays lead to average extra distances flown of 14% for intra-European flights and 12% for intra-U.S. flights [Reynolds, 2008]. He also found the delays to increase with traffic density. We provide a peak versus off-peak delay calculation in the model and make estimates of the relative impact as initial default values; however, research is required to refine these estimates. In generating the parameters, we distribute the delay component such that peak-period flights receive a 25% increment to the average excess-distance and additionally estimate that 45% of all domestic seat-arrivals occur during peak periods. Based on these estimates, weekend and off-peak periods receive a 5.1% decrement from the average excess-distance while week-day peak periods receive a 6.3% increment. The resulting average, peak and off- peak fuel intensities of each aircraft type are shown in Table 10.

20 Table 9. Derived Fuel Intensity Coefficients by Aircraft Type AC Type Regression Bias due to Forced-Zero Origin Coefficient Units Original Coefficient Values Derived Values 1 Average Trip Distance (km) Implied GC Multiplier Implied Extra Travel (km/trip) TP -5.9% kg/seat-LTO 10.62 4.70 375 1.335 126 kg/GC-skm 0.011222 0.0294 SRJ 2 16.4% kg/seat-LTO 11.457 8.34 631 1.444 281 kg/GC-skm 0.045 0.0514 RJ -1.8% kg/seat-LTO 8.888 7.50 789 1.102 80 kg/GC-skm 0.039004 0.0325 NBJ -1.2% kg/seat-LTO 6.814 6.88 1,436 1.114 164 kg/GC-skm 0.0224 0.0228 WBJ 3 -2% kg/seat-LTO Not Significant 8.26 2,591 1.061 158 kg/PL-km 0.157325 kg/GC-skm Not Applicable 0.0219 Source: TranSys Research Ltd, analysis of Table 254 Air Traffic Data for years 2011-2012 [US BTS, 2013 (http://www.transtats.bts.gov). Notes: 1. Derived values adjust for regression bias and force the LTO fuel consumption to ICAO certification data (see text). 2. SRJ regressions are per-aircraft and average seats/aircraft used since per-seat values provided poor regression results (see text). 3. WBJ regressions are based on payload capacity (kg/PL-km) rather than seats since WBJ aircraft had a significant proportion of cargo. Fuel is split between cargo and seat-payload- capacity to get per-seat km fuel intensity (see text). Table 10. Cruise-phase Fuel Intensities by Aircraft Type and Traffic Congestion. AC Type Trip GC-Distance Multiplier Fuel Rate (kg/GC-skm) Implied GC Multiplier Estimated WD-peak Multiplier Estimated non-peak Multiplier Average Peak Period Off-Peak Period TP 1.335 1.419 1.267 0.0294 0.031219 0.027866 SRJ 1.444 1.556 1.354 0.0514 0.055317 0.048133 RJ 1.102 1.127 1.081 0.0325 0.033252 0.031887 NBJ 1.114 1.143 1.091 0.0228 0.023429 0.022364 WBJ 1.061 1.076 1.048 0.0219 0.022167 0.021597 Source: TranSys Research Ltd: Average values based on analysis of BTS data. Peak and off-peak values are preliminary best-estimates requiring further research. GHG intensities during the LTO phase are calculated on the basis of factors used in EPA’s inventory model, which adopts zero emissions of CH4 from aircraft and 0.1 g/kg of N2O [EPA, Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2010]. The CO2- equivalent emission during the high-altitude cruise phase of flight is based on the 1.5 multiplier factor discussed above. Thus, the equations for air-mode GHG emissions are:

21 GHG (kg-CO2e) = GHG(LTO) * Fuel(LTO) + GHG(cruise) * Fuel(cruise) Equation 4 where: GHG(LTO) = 3.158 + 25 * 0 + 298 * 0.0001 = 3.188 kg-CO2e/kg-fuel GHG(Cruise) = 3.158 * 1.5 = 4.737 kg-CO2e/kg-fuel LTO includes fuel allocated to climb out and descent to/from 25,000 ft. An alternate measure of Air-mode GHG emissions is provided with the high altitude impacts ignored to facilitate comparison of the impact of the high-altitude multiplier on air- mode GHG intensities. 2.5 Rail Mode Methodology 2.5.1 Analytic Overview Since the focus of the model is on energy intensity rather than overall train performance, some simplifications can be made in the simulation model over what would be required by a detailed train performance calculator. Also, passenger services have characteristics that allow additional simplifying assumptions. Specifically, passenger consists are short relative to freight trains, which make a lumped mass approach more realistic and the power-to-weight ratio of passenger trains is much higher than freight trains, which makes the influence of gradients much less important for passenger than for freight trains. Gradient and train length influences on energy intensity are included in the model; however, due to the simplifying assumptions (which are reasonable for passenger services), the model will not provide accurate results for freight trains. It is a rail passenger service simulation model, not a general purpose railway simulation model. The model is based on characterization data that will usually be available to a rail-agency and rail system operators; however, the data will not necessarily be publicly available. Detailed track gradient and curvature profiles that are required by most railway train performance calculators are not used as inputs; however, condensed route characteristics are included. The model has been developed with some default region-and service- specific characteristics (as developed for the case studies) built in. Nonetheless, simulation of a specific service will benefit from development of data specific to that service. One of the key influences of passenger train performance is the number of speed changes involved on a route. Permanent speed limits shown in railway timetables are more generally available than are track gradient profiles and we recommend that actual speed limit tables be used in simulating a service wherever possible. Gradients have less impact on passenger train performance and the regional gradient characteristics developed as default regional tables may be more generally applied without significant impact on the accuracy of the results. The gradient characteristics developed for the default tables were developed from a range of actual track gradient profiles and do not represent any one track subdivision or any one railway. We caution again that the regional characteristics and the model itself are not applicable to freight trains. As with the other ground modes, energy intensity and associated emissions of GHG involve similar calculations. However, since the model supports comparisons of rail mode technologies, additional breakout details of energy dissipation components are provided for the rail mode. The rail simulation module is configured to separate the individual components of energy dissipated in overcoming inherent resistance. In addition to inherent resistance, power is required to:

22 1) accelerate the mass of the vehicle, its rotating elements and the load it is carrying to a desired speed, and 2) climb uphill grades encountered. These two additional power requirements do not directly translate into energy as they are essentially stored energy. The potential energy gained in climbing grades can be partially or fully recovered to overcome inherent resistance on downgrades. Similarly, the kinetic energy/inertia gained in acceleration can be partially recovered in deceleration. It is only through braking (and to a lesser extent, drivetrain-drag during coasting) that these stored energy components are lost/consumed. The rail module accumulates the energy associated with each of the sub-categories of energy dissipation such that the effectiveness of alternative technologies can be gauged from the model output for a single train run. 2.5.2 Rail Equipment Characterization Table A-1 of Volume I of this report summarizes the publicly available passenger locomotive characterization data, and Table A-2 of Volume I of this report summarizes the publicly available passenger rail coach and trainset characterization data that have been located for use in the model case study simulations. 2.6 Access Egress Modes Characterization Only the primary modes being compared are simulated in detail in the model; the performance attributes of access and egress modes are simple averages provided in default lookup tables located in the Regional-Properties worksheet. The attributes of public transport modes have been derived from the 2011 National Transit Database’s Service and Energy Tables. The electricity supply is region-dependent but all other performance metrics are based on one average applied to all regions. The properties included in the calculation of the various modal averages are shown in Table 11. Table 11. Transit Properties Included in Modal Averages Property MB CB CR (D2) CR (el) HR (el) LR (el) Central Puget Sound Regional Transit Authority X X X Maryland Transit Administration X X X X Georgia Regional Transportation Authority X City of Los Angeles Department of Transportation X Massachusetts Bay Transportation Authority X X X X Southeastern Pennsylvania Transportation Authority X X X Washington Metropolitan Area Transit Authority X X Utah Transit Authority X X X South Florida Regional Transportation Authority X Metro Transit (Minneapolis/St. Paul) X Denton County Transportation Authority X Peninsula Corridor Joint Powers Board dba: Caltrain X Southern California Regional Rail Authority dba: Metrolink X Legend: MB = Municipal Bus, CB = Commuter Bus, Cr = Commuter Rail, HR = Heavy Rail (defined as dedicated commuter tracks), LR = Light Rail, D2 = diesel, el = electricity

23 Attributes for personal automobiles and taxis were derived with the simulation model in a one-time simulation of the 2011 driven fleet composite vehicle. The following assumptions were made for the highway access/egress modes:  Taxis were assumed to travel 1.5 km for every km of passenger carrying travel;  drop-off/pick-up was assumed to have 60% return-to-origin travel and 40% being part of a 2-person trip that incurs 10% extra travel distance.  Carpools are assumed to involve 3 persons and the trip length is 15% longer than any one-person trip. The resulting characteristics (for the Continental U.S. electricity generating fuel mix for upstream energy and emissions) are presented in Table 12. Table 12. Access/Egress Modes’ Default Performance Data Mode Average Speed (mph) Fuel Source Direct Travel Upstream Fuel Intensity Energy Intensity CO2-e Emission Intensity Energy Intensity CO2-e Emission Intensity (kg/p-mi) (kJ/p-mi) (g/p-mi) (kJ/p-mi) (g/p-mi) (kWh/p-mi) Walk 3.1 N.A. 0 0 0 0 0 Bicycle 10 N.A. 0 0 0 0 0 Walk/Bicycle 10 N.A. 0 0 0 0 0 Auto: Drive alone & park 25 Conv. gasoline 0.125 5,431 399 1,091 99.5 Auto: Drop off / Pick up 25 Conv. gasoline 0.236 10,251 753 2,060 187.7 Carpool, Van, Shuttle 25 Conv. gasoline 0.048 2,082 153 418 38.1 Taxi 25 Conv. gasoline 0.188 8,147 598 1,637 149.2 City Bus 12.4 U.S. conv. diesel 0.089 3,801 310 761 69.6 Commuter Bus 24.4 U.S. conv. Diesel 0.055 2,357 192 472 43.2 Subway 21.2 Electricity Continental U.S. Mix 0.396 3,322 228 337 23.0 Streetcar/Light Rail 13.8 Electricity Continental U.S. Mix 0.338 2,840 195 288 19.7 Commuter Rail (elec) 26.9 Electricity Continental U.S. Mix 0.373 3,129 215 317 21.7 Commuter Rail (diesel) 33.4 U.S. conv. Diesel 0.064 2,743 205 549 50.2 Source: TranSys Research analysis: public modes derived from the National Transit Database and LDV derived via commuter-trip simulation. The fuel intensity of highway modes is adapted to congestion conditions via peak and off- peak multipliers, which can be specified for three city sizes (large, small and rural municipality).

24 2.7 Regional Characterization 2.7.1 Region and Season Definitions The model provides default regional characterization data for four regions of the continental U.S. and electricity generation characteristics are further disaggregated into 9 sub-regions. Table 13 defines the state composition of each region and sub-region. Three seasons are defined – summer, winter and ‘other’ being spring and fall. Summer and winter each have 3 months and ‘other’ has 6 months. Summer is comprised of June, July and August; winter is comprised of December, January and February; and ‘other’ is the remaining months. Table 13. State Composition of Regions and Sub-regions Region Sub-region States Northeast Middle Atlantic (MA) NY, CT, PA, NJ New England (NE) NH, VT, ME, MA, RI South West South Central (WSC) OK, AR, LA, TX East South Central (ESC) KY, TN, MS, AL South Atlantic (SA) WV, VA, DE, MD, DC, NC, SC, GA, FL Midwest West North Central (WNC) ND, SD, MN, NE, IA, KS, MO East North Central (ENC) WI, MI, IL, IN, OH West Pacific WA, OR, CA Mountain (MTN) MT, ID, WY, NV, UT, CO, AZ, NM Source: U.S. Energy Information Administration (Weather Data Regional Composition) 2.7.2 Emissions Intensity of Electricity Generation by Region The distribution of fuels used in generating electricity by region was derived from the Energy Information Administration’s data for fuels used in electricity generation by state in 2011 [Energy Information Administration, 2012]. The upstream fuels consumed in providing the fuels for electricity generation was derived from the GREET model [Argonne National Laboratory, GREET1_2012]. Table 14 indicates region breakdown in 2011of the energy content of carbon fuels usage in electricity generation (Direct) and the incremental upstream carbon fuels consumed in getting fuels to the electricity generation stations.

25 Table 14. Direct and Upstream Carbon Fuels Usage in Electricity Generation by Region in 2011 Region Electricity Generation Upstream Fuel Increment Carbon fuels (BTU/kWh) CO2e (kg/kWh) Carbon fuels (BTU/BTU) CO2e (kg/kWh) Northeast 6,976 0.397 16.7% 0.066 South 8,297 0.614 12.2% 0.075 Midwest 8,623 0.730 7.0% 0.051 West 6,865 0.421 13.1% 0.055 Continental U.S. 7,938 0.577 11.2% 0.065 Source: TranSys Research Ltd, derived from Energy Information Administration’s state data for fuels used in generating electricity and GREET1_2012 data for fuel properties and upstream fuel intensities. 2.7.3 Climate-Influences by Region Seasonal climate properties influence auxiliary power usage and aerodynamic drag for all modes. Seasonal daytime temperatures for each region were derived from the National Climate Data Center’s (NCDC) hourly readings data for 1981-2010 [National Climate Data Center, 2013]. Air conditioning usage by LDVs was derived from previous work undertaken by Rugh et al. [Rugh, 2004], who derived estimates of annual air conditioning usage by state for automobiles and light duty trucks (LDTs). Their analysis derived average air conditioning usage weighted by the total light duty vehicle registrations in each state. The analysis concluded that 7 billion gallons of fuel (about 5.5% of total LDV consumption) was consumed annually by LDVs for air conditioning usage. We applied the same vehicle registration weighting by state to get weighted average values of daytime temperatures for each of our four defined regions and for the continental U.S. We derived seasonal variations of air conditioning usage on the basis of these temperatures and Rugh’s data on average temperature while air conditioning was on for each state. The resulting seasonal values are shown by region in Table 15. Table 15. Regional Climate Related Characteristics Region Measure Season Winter Summer Other Northeast Temperature C -0.1 23.3 11.9 Temperature F 31.8 74.0 53.3 LDV Air conditioning time on 0.0% 82.0% 2.3% South Temperature C 9.5 27.9 19.4 Temperature F 49.2 82.2 67.0 LDV Air conditioning time on 0.0% 90.9% 39.6% Midwest Temperature C -2.4 23.9 11.6 Temperature F 27.6 75.0 52.9 LDV Air conditioning time on 0.0% 92.7% 2.2% West Temperature C 7.5 24.5 16.0 Temperature F 45.6 76.1 60.8 LDV Air conditioning time on 0.0% 77.8% 19.5%