Integrating Aviation and Passenger Rail Planning (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.

156 C H A P T E R 1 1 Introduction and Structure Highlights The project has created a model to help improve understand- ing of the relationship among several key factors in the explana- tion of the choice of mode between air and rail, which, when seen in a scenario of improved rail services, is associated with the diversion from air to rail. The Air/Rail diversion model: • Is a simplified forecasting tool strategically designed to address specific questions around consumer preferences and trade-offs in response to conventional rail/HSR and air service modifications. • Is designed to be a quick-response tool, includes conven- tional rail/HSR and air service trade-offs. • Is accessible and relatively easy for planners and analysts to use. • Incorporates data on ground origin to ground destination air travel patterns, merging ground access patterns with airport-to-airport flows. • Includes accurate data on existing intercity rail passenger flows (from Amtrak). • Incorporates an airline response model to estimate the effects of rail/HSR services on flight volumes. • Is built on an open-source code base in a scripted (non- compiled) programming language. Structure Chapter 11 is presented in two parts. Part One explores how the model fits into the context of the policy analysis, and the need for better tools discussed in Chapter 10. The Chap- ter explores the context and setting in which the model was developed and summarizes the purposes of the model. Sev- eral early examples of how the model could be used in policy analysis are presented. In the East Coast, the role of several conditions and service attributes are examined for both their incremental and cumulative effects on diversion to rail. In the West Coast, a different case study was developed, reflect- ing the very low speeds of the base case rail system, and the incremental steps that might be taken to improve the full sys- tem; the emphasis on the exercise is the understanding of the interaction of the separate policy variables, not the prediction of flows on an actual system. The scenario tests also serve as an introduction to how to use the model. In Part Two, the Chapter reviews how the model works, and how it was developed, and summarizes the use of eight steps in model application. It presents a summary of the model at a level of detail appropriate for transportation managers, plan- ners and analysts. At the same time, the project has created a Technical Appendix with documentation designed for those who are interested in demand models and the model develop- ment process. In addition, a free standing “Users Guide” to the application of the model has been created in PowerPoint format and can be found at the end of this report as well as on the accompanying CD. The Technical Appendix is available on the accompanying CD and could be utilized by all who want to apply the model in one way or other. In short, the first half of this Chapter examines what the model does, and the second half examines how it does it. All of this is presented with the understanding that the math- ematical model cannot indeed predict the future. No one can do that. As expressed to the Research Team by one expert, these models are, best used as a learning tool at an early concept stage of planning to determine basic impacts of a series of actions/ investments so that a decision can be made on whether to engage in more detailed analysis. No one should believe this process can by itself support making a major investment decision. These kinds of mod- els are ideally used to assess the comparative advantages among a group of alternatives of taking one action compared to some different action. Models can also offer wonderful kaleidoscopic effects allowing one to examine the same basic action/investment with different attributes. (Roberts, communication within the project panel.) The Air/Rail Diversion Model

157 Part One: The Public Policy Context—What the Model Is Intended to Do Model Overview This Chapter presents the results of the development of a new modeling tool for the examination of the competition between air services and rail services. The Air/Rail Diversion model (“the model”) is envisioned as a strategically designed forecasting tool to address specific questions about consumer preferences and trade-offs in response to conventional rail/HSR and air service modifications. The model was designed as an efficient, quick-response tool useful for realistic planning-level scenario analysis and it only includes trade-offs between air service and both conventional and HSR, omitting trade-offs with both auto travel and intercity bus. As discussed in Chapter 10, several models presently exist concerning forecast travel behavior involving mode choice decisions between rail and air. In some cases, those models are considered proprietary, such as those used by Amtrak for detailed market research in its competitive environment. In the West Coast, an elaborate model of rail demand has been developed in order the meet the very exacting requirements for the CHSRA’s project development and environmental documentation. While that model is extremely thorough in its approach, it was never designed to serve as an analysis tool for quick and cost effective analyses of public policy options. Alternative settings, contexts, and potential future scenar- ios can be entered into the program in one of two ways. First, they can be established on a global scale, applied to all ele- ments of the geographic network at once. Second, they can be entered on a geographically specific scale, with, say, major rail travel time improvements between San Jose and Bakersfield, but not in the rest of the corridor. Looking first at the pro- cess for entering global change, Figure 11-1 shows the settings established in the creation of a complex policy scenario devel- oped for and described in this part of Chapter 11. Inputs for Variation at the Global (System-wide) Level The model has been designed to allow simplified access concerning seven kinds of network assumptions, labeled as “Scenario Input Factors” in Figure 11-1. The model can sup- port the global (system-wide) alteration of data by the policy analyst concerning the following parameters: • Speed of the rail system, called “Rail In-Vehicle Travel Time” • Terminal-to-terminal time for the aircraft, called “Air In- Vehicle Travel Time” • Travel time of the auto to gain access/egress for airport or rail station, “Auto IVTT” • Rail fares • Air fares • Amount of rail service • Amount of air service Using these seven input variables, a wide variety of future conditions can be hypothesized, including variables over Figure 11-1. Scenario input screen for user interface, as used in this chapter.

158 assumptions could be manually inserted into model, by sim- ply altering the contents of the Excel spreadsheets concerning household income. The transparent presentation of input assumptions does indeed allow an exceptional level of policy exploration of the role of a wide variety of inputs assumptions; how much such exploration is justifiable is a question for further exploration. The reader is reminded that the model was not designed to predict change in rail ridership in total, only that change in rail ridership attributable to diversion to (or from) rail associated with air. Additionally, the model was not designed to predict air passenger volumes in total, only that change in air passen- gers attributable to change in competitive rail characteristics. The question of improved rail’s ability to divert from auto and bus is simply not addressed in this ACRP study. Definition of the Project Study Areas The model has been applied in two North American study areas—the Northeast Corridor (“East Coast”) and California (“West Coast”)—where there is considerable availability of both air and rail modes, meaning that many long-distance travelers have a reasonable choice between the modes. Fig- ure 11-2 and Figure 11-3 show the East and West coast study areas, respectively; they also show the county geography and the airport and rail station locations that are represented in the model. The input data summaries and model results pre- sented in the remainder of this Chapter relate to these two study areas. The geographic extent of the two study areas includes counties adjacent to the rail corridors since competition will be greatest for travelers starting and ending their trips near the rail line. Consideration was given to the definition of “adjacent” with the final study areas focused on defined corridors that follow the proposed HSR alignments without including counties that are unreasonably far away from the existing rail service. Model Exercise for the East Coast The fact that high-quality high-speed services already exist in the base case condition in the Northeast allows the ana- lyst to examine four scenarios with relative modest scale of change in the input assumptions. Four such scenarios were developed for the Northeast Corridor, with only system-wide (global) changes utilized in this set of early explorations of the model. In each case, the analyst must first create the Sce- nario for testing with the Create Scenario command, save the results, and press the Run Model command, all shown on Figure 11-4. The model creates an elaborate set of input and output file folders with the creation of each new scenario; the user is encouraged to use the Delete Scenario command when any such file can be trimmed back out of the system. which the policy maker might have direct control (e.g., setting the rail fares) and those over which the policy maker might have very little control, (e.g., congestion on the roadways lead- ing to both airports and rail terminals.) For the East Coast, this Chapter will examine the use of one base case, and four hypothetical scenarios, cumulating in the composite scenario referenced in Figure 11-1. Variables Which Can Be Altered at the Geographic Specific Level In addition to the seven factors which can be varied as input for all the geographic coverage of the system, the Air/Rail Diversion Model contains a high number of data inputs, all of which could be altered and manipulated, if the analyst were to desire to do so. For each scenario, and for the present and the future, both system descriptions and demographic data are presented that could be considered as input variables for explo- ration, if such were warranted. Alteration of these variables can be done on the spreadsheet appropriate for the content. There are five major categories of data input for each sce- nario, for the present and the future. They are 1. Socioeconomic data. This includes data about population household income, employment, all at the tract level. Also included here are total roundtrips by air and rail, orga- nized on a county to county basis. Data on party size and party demographics are included here. 2. Rail station access data. This includes data about distance to candidate rail stations, including distance, highway travel time and transit travel time. 3. Airport access data. Same as for candidate rail stations. 4. Rail service description data. Includes schedules and dis- tances for rail services. 5. Air service description data. Includes air passengers by airport pair, and various characteristics of the flight con- nections from the DB1B data base of BTS. Because the source information is transparently presented in the model, the analysis could examine the manner in which input assumptions affect the resulting predictions of rail diversion from air by scenario. From the supply side, the analyst could create a hypothetical network with significantly improved overall rail travel time between New York and Washington, with no improvement in travel time between New York and Boston, if that were a policy question being addressed at some level. Concerning the set of demographic assumptions made for the future case, the modeling process could be used to cre- ate a range of alternative futures for testing. By way of exam- ple, if the analyst was concerned that the established (e.g., Woods and Poole) forecasts might be under-predicting future income for a state (or set of states) alternative demographic

159 would decrease by 10%. This scenario was named “Cheaper Rail” for short. The model is run, and automatically creates a folder containing a set of spreadsheets summarizing the results. These spreadsheets can be copied to be used by the analyst for further refinement and presentation via whatever graphic formats are desired. The format of the spreadsheets (comma delimited) allows the immediate transfer to such programs as SPSS or PowerPoint. • Scenario #3. The “Create Scenario” procedure was applied by copying it from Scenario #2, which appears on the screen. The Research Team’s scenario #3 was based on this, with the addition of the hypothetical condition that additional trains were run on the system, with a 50% increase in number of trains (e.g., two trains an hour between Boston and New York changes to three trains per hour, with no change in speed assumed). This scenario was named “Cheaper and More Rail” and the model was run. • Scenario #4 was created in the Create Scenario process by taking the settings of Scenario #3 and adding the hypo- thetical assumption that all trains in the system would run A Cumulative-Build Approach to Complex Scenario Testing In this exercise the model was used to explore a set of future policies and contexts for the East Coast model applications. Each scenario is designed to allow the direct comparison with output from the previous (simpler) model, allowing for both an incremental and cumulative analysis. East Coast Scenarios • Scenario #1 is the Base Case. The Air/Rail Diversion Model is structured to encourage the building of cumulative scenar- ios, with increasing levels of complexity. After the running of the Base Case in which all seven of the global factors remain set at 1.0 (strongly recommended), the “Create Scenario pro- cedure” in the model asks the user to define an additional scenario, and queries which scenario it could be copied from. • Scenario #2 hypothesized that fares for rail would decrease by 25% and that the number of flights within the Study Area Figure 11-2. Definition of project study area, East Coast.

160 Figure 11-3. Definition of the project study area, West Coast. Figure 11-4. The program allows new scenarios to be built from existing scenarios, within the Create Scenario command.

161 East Coast Geographic Areas Using MS Excel independently of the model, the metropoli- tan areas were restructured by geographic area for this par- ticular analysis, based on the following definitions: • All study area metro areas north of New Haven, CT, were aggregated into a category called “New England.” • All study area metro areas in New York State, and all between New Haven and north of Trenton, NJ, were aggregated into a category called “New York.” • All metro areas between Trenton and the Maryland border were aggregated into a category called “Mid-Atlantic.” • All study area metro areas in Maryland, the District of Columbia, and Virginia were aggregated into a category called “Baltimore/Washington.” As presented in the Table 11-1, volumes of rail origins by regional area were then summarized for each scenario, again outside of the model structure. From the summaries cre- ated from the “OriginCBSA” spreadsheets, as categorized by regional areas, total number of rail round trip origins was calculated; they are presented as Table 11-1. The data are expressed in terms of four regional geographic areas, and five hypothetical scenarios for service attributes and market conditions. The model output was then examined in terms of the impact of the five cumulative scenarios on the rail share of the air + rail market in the East Coast project study area, as shown in Figure 11-6. with speeds resulting in a decrease in terminal-to-terminal travel time by 25% (e.g., a 4 hour travel time from Washing- ton to Stamford would become a 3 hour travel time). This scenario was named “Cheaper, More, and Faster Rail” and is illustrated in Figure 11-4. • Scenario #5 was created from the input factors of Sce- nario #4, with the addition of the hypothetical assump- tion that air fares in the study area would increase by 25% (e.g., an airline fare of $200 is assumed to rise to a fare of $250). Illustrative Examples of the Analysis Process The output variables produced for this analysis were the total number of air trips and rail trips from all counties (and thus all SMSAs) combined. For this analysis, the model cre- ated output files for each scenario. Here the Research Team used one spreadsheet entitled “OriginCBSA” for each of the five scenarios. In this case study, the Research Team was inter- ested in the change in the absolute number of rail trips in the region, not any set trips between specific origins and destina- tions. External to the model, the spreadsheets were grouped geographically and reorganized by sub-region in the study area. An example of the output spreadsheet produced in the model is shown in Figure 11-5. The model stores each spread- sheet in a file labeled with the name of the scenario created by the user (Figure 11-5). Figure 11-5. The model creates file folders for each scenario created; these can be deleted with the Delete Scenario Command.

162 Figures 11-7 and 11-8 graphically portray these same results in terms of relative impacts of the scenarios. The results of the scenario testing are expressed from the vantage point of the rail analyst on Figure 11-7, which graphically portrays the data in Table 11-2 The rail volumes are set in the base case as 1.0, with growth in rail volume by scenario level expressed for each of the four geographic areas, and the total study area. The same data is expressed from the vantage point of the air analyst in Figure 11-8. In this format, the number of air trips in the base case is set at 1.0 with decrease by scenario level expressed for each of the four geographic areas. Both figures were developed from the rail volumes summarized in Table 11-1, and reflect two different ways to look at the policy implications of the diversion patterns. The graphic format reflects the orientation of the modal manager (air or rail), who may be interested in the relative change in markets as much as the absolute values reflected in the status quo. Table 11-1. Number of rail trips including diversion from air, by scenario and region. Scenario #1 Base condition Scenario #2 Scenario 1 plus cheaper rail Scenario #3 Scenario 2 plus more rail service Scenario #4 Scenario 3 plus faster rail service Scenario #5 Scenario 4 plus higher airline fares OriginCBSA New England 1,222,700 1,326,750 1,389,300 1,599,500 1,749,300 New York 2,465,550 2,573,900 2,640,100 2,869,750 3,006,400 Mid-Atlantic 1,580,500 1,627,000 1,673,600 1,721,000 1,743,950 Bal-Wash 1,714,500 1,850,200 1,917,250 2,162,100 2,329,800 Study Area 6,983,250 7,377,850 7,620,250 8,352,350 8,829,450 Figure 11-6. Change in rail mode share by diversion scenario and region. 1.00 1.05 1.10 1.15 1.20 1.25 1.30 1.35 1.40 1.45 1.50 base=1 cheaper rail cheaper+ more rail cheaper+ more+faster rail that+higher air fares Ra il tr ip s, re la v e to ba se Increase in Rail Trips, by Scenario New England Bal Wash New York Mid atlanc Figure 11-7. Relative increase in rail trips from diversion from air. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 base=1 cheaper rail cheaper+ more rail cheaper+ more+faster rail that+higher air fares Ai rt rip s, re la v e to ba se Diversion from Air, by Scenario Mid atlanc New York Bal Wash New England Figure 11-8. Relative change in air trips by diversion scenario. Implications for the East Coast Whether expressed as a loss to air ridership, or a gain to rail ridership, the results of this early policy exercise have interest- ing implications. First, when an analysis is undertaken on a cumulative basis, the analyst can explore both the combined impact of multiple factors (expressed as Scenario #5) and the increments of change associated with the addition of each

163 the impact of new rail travel times which were (a) 70% of cur- rent times, and (b) 50% of current times. Looking first at the East Coast, the model predicts that, under the assumption of better rail travel times 70% of the present, impacted air travel might fall from a base case of 3.8 million trips, down to about 3.3 million trips. Under the scenario in which rail improves to provide travel times as low as 50% of present times, air vol- umes would decrease to 2.9 million trips. Expressed as mode share, the airlines’ proportion of the air-plus rail would fall from 35% in the base case, to 30% and 26% under the two improved rail scenarios modeled for the East Coast. Change in Air Fares: East Coast At the same time, key policy variables are available to the airlines, particularly in the setting of fares. For the East Coast, a lowering of air fares to 70% of their present level results in an increase in predicted air travelers from 3.8 million trips in the base scenario to 4.5 million in the lowered-fare scenario. Expressed as mode share, the airlines’ proportion of the air plus rail market would increase from 35% in the base sce- nario to 41% in the lowered-fare scenario in the East Coast. Model Exercise for the West Coast Setting and Context The policy questions facing decision makers in the West Coast are fundamentally different from those presented herein, as the base case conditions for rail are so much worse than those faced in the East Coast. Rail services in the West Coast take on a different form from those in the Northeast Corridor. Between the two dominant cities of California, the present rail system could more accu- rately be described as a rail and bus system. From Los Angeles factor separately into the analysis. Of the four strategies of improvement over the base case, the largest NEC system-wide increment comes from the assumption that all travel times will improve by 25%. This is consistent with the present explora- tion by Amtrak and FRA of significantly improved travel times under various HSR futures, as is associated here with an 11% increase in the number of rail passengers solely attributable to increase in diversion from air on the basis of travel times. More relevant to policy makers is the increase attributable to many diverse factors together, as reflected in the Scenario #5 future in this analysis. Once again, the Research Team cautions that these increases in rail ridership do not include any increased diver- sion from auto and bus, which is not analyzed here. When examined from the point of view of separate regions, several patterns are revealed. Most dominant is the simple fact that service levels in the Mid-Atlantic region are already high, such that incremental increase from more air diversion is sim- ply less relevant here than it is in either New England or the southernmost region including Maryland, Virginia and the District of Columbia. Expressed differently, there are more air trips to be potentially diverted outside of the New York and Mid-Atlantic regions. Explorations of Factors Separately— East Coast The East Coast exercise described herein sought to allow the examination of several factors (both about the quality of the rail service and the competitive characteristics of the air service). The model can also be used to explore one variable at a time. Change in Travel Times–East Coast The model can be used to look at gradations in any given input variable, such as travel time. Models were run to explore Region Scenario #1 Base Condition = 1 Scenario #2 Scenario #1 plus cheaper rail Scenario #3 Scenario #2 plus more rail service Scenario #4 Scenario #3 plus faster rail service Scenario #5 Scenario #4 plus higher airline fares New England 1.00 1.09 1.14 1.31 1.43 Bal-Wash 1.00 1.08 1.12 1.26 1.36 New York 1.00 1.04 1.07 1.16 1.22 Mid-Atlantic 1.00 1.03 1.06 1.09 1.10 Study Area 1.00 1.06 1.09 1.20 1.26 Increment for each scenario Base .06 .03 .11 .06 Table 11-2. Increments of increase in rail trips due to diversion from air.

164 • Scenario #4 would simulate high-speed train service, with average speeds raised from the present 33 mph today to a very high average speed of about 100 miles per hour, for example. This is similar to average speeds between Paris and London for many years, and present HSR service between Paris and Amsterdam. The scenario is designed to test the upward limits of the model, and not to represent any assumptions about how many OD pairs could actually be provided with service this fast. • Scenario #5 takes the very high-quality train service assumed in Scenario #4, and hypothesizes that the cost of an air ticket has increased by 20% and that the number of planes oper- ated has decreased by 20%. The purpose of this scenario is to observe the model’s reaction to competing services, separate from the characteristics of the rail. West Coast Geographic Areas For this analysis, the California study area was divided into four geographic subareas. 1. “North” includes both the San Francisco and San Jose metropolitan areas and everything to the north. 2. “Central” includes areas to the South and immediately west of San Jose, including all areas to the south just short of Santa Barbara. 3. “Los Angeles” area includes everything south of San Louis Obispo, and north of San Diego. 4. “San Diego” includes San Diego and El Centro. Results of the Scenario Testing The model shows that the rail/bus system attains a share of the rail plus air market ranging from 19% in the North to 45% in the central area, which may reflect the relative pau- city of within-study area flights in the Central Area, and their relative strength in the North. Figures 11-9 and 11-10 suggest that the travelers in the Central area of California do not rely on within-state airline services to the extent that the other three subareas do. (Arguably, this seems reasonable as trip lengths would tend to be less than half the length of the state.) This seems to result in a low mode share to air, compared to the other regions, in four of the scenarios; in the fifth sce- nario, the hypothetical increase in airfare tends to bring the other three regions up in rail share, closing the gap somewhat. How much of this pattern reflects geography, and how much reflects socioeconomic characteristics remains to be explored. By contrast, residents of both San Francisco to the north, and Los Angeles are separated by about 340 miles, making the air an extremely attractive alternative to the rail-plus-bus network, reflected in the fact that the “rail” mode share for these two areas is one half of that simulated for the Central Area. Applying the Union Station one train a day goes to the Bay Area, with a stop at the waterfront (Jack London Square) in Oakland. Six more services with a single transfer from rail to bus are pro- vided by Amtrak. Accepting these definitions there are about seven bus + rail services a day. The 340 mile trip takes between 8.5 hours to 11 hours, with the slowest trip on the direct train. Thus, terminal-to-terminal speeds range from about 30 mph, to 40 mph. Policy makers in California are now facing the concept of incremental upgrades to get to the final HSR at the scale system originally envisioned, including the idea of “blended service” where early investment in electrification can improve commuter services years earlier than previously contem- plated. To better understand how the model might be applied in California, one scenario was created to reflect a major improvement in rail travel times, but one somewhat short of true high-speed rail, with an across the system assumption that travel times could be lowered by 50%. For example, to accomplish this, the average speed of (approximately) 33 mph between Los Angeles and the Bay Area would be assumed to be 66 mph in Scenario #2. Only in Scenario #4 is it assumed that the average rail speed would be 99 mph, which is associ- ated with a higher level of infrastructure investment. Defining the Scenarios and the Geographic Segments Without question, the policy questions to be addressed will be different from those in the Northeast, as the rail service is poor. The model was tested with five scenarios, including the base case using actual schedules and timetables as input. The West Coast Scenarios • Scenario #1 is the Base Case. All the service descriptions provided in the model were left intact, with all seven of the global input factors left at “1.” • Scenario #2. Double in-vehicle travel speeds (or more accurately cut terminal-to-terminal travel times by half for all areas now serviced by the rail-plus-bus network). For example, a 10-hour trip between LA and the Bay Area would now be a 5-hour trip. For model testing purposes, all trip times would be cut in half in this testing scenario. In cases where the average trip speed was 33 mph, it would now become 66 mph. • Scenario #3. Double the speed of the train, and operate three times as many. Thus, the example trip between LA and Oakland would operate 21 times a day, better than hourly service, but still less than Amtrak operates in the Northeast Corridor. This scenario explores frequency of service as a separate factor from in-vehicle travel times.

165 scenarios to the modeling process, the differences among the four sub-regions tend to dissipate with the assumption of better rail services, such that all three become more and more similar to the mode share pattern of the Central Area. All four groups are influenced by the incremental improve- ment attributable to high speed, but those living in the north- ernmost area seem to be most affected by this factor. This group experiences a 15 point increase in percentage to rail from the assumption of cutting travel times by one half in this increment (from Table 11-3). This is similar to the 12 point increase for the LA group, both of which are higher than the eight point increase for the Central area residents. Once the higher quality rail is assumed, the residents of the study area as whole are sensitive to the change in assumptions about the strength of the air competition, with a 17 point increase attributable to the increment of change included in the final scenario. As noted, once the high-quality rail service is in place the Central area residents are not influenced by the costs and frequency changes in air services that may not serve them well in the first place, as reflected in their increase of only two points for this scenario (Table 11-3). From the vantage point of the aviation manager, each of the scenario increments reflects a logically predicted decrease in air trips within the study area. Holding the base case as today’s (e.g., 2008) air volumes, the pattern of decrease from the status quo is graphed in Figure 11-10. Interestingly, the geographic group with the weakest pattern of use of air for these intra- state trips is the group that loses air volumes most dramati- cally over the first four scenarios examined. The model seems to be presenting a consistent pattern of response to the somewhat arbitrary details invented for the progression of scenarios in this exercise. In general, the results are consistent with working assumptions about the scale of impacts at major airports. If it is assumed that approximately 15% of those passengers boarding at SFO or LAX have actual 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Base Case Rail speed doubled That + rail frequency That+ rail speed tripled That + higher airfare, fewer flights Ra il Sh ar e Pe rc en ta ge Rail Share of Air+Rail by Scenario and Region Central San Diego LA North Figure 11-9. Change in rail mode share by diversion scenario and region. Figure 11-10. Relative change in air trips by diversion scenario. Scenario # 1 Scenario # 2 Scenario # 3 Scenario # 4 Scenario # 5 Base Case Rail speed doubled That + rail frequency That + rail speed tripled That + higher airfare, fewer flights Central 45% 61% 72% 80% 82% San Diego 33% 40% 47% 54% 71% LA 25% 33% 45% 57% 72% North 19% 28% 39% 54% 72% Project Study Area 24% 33% 43.5% 56% 73% Table 11-3. Rail share of air plus rail market, including project study area (West).

166 elers from 7.5 million trips in the base scenario to 7.8 million in the lowered-fare scenario. Expressed as mode share, the airlines’ proportion of the air plus rail market would increase from 80% in the base scenario to 82% in the lowered-fare scenario in the West Coast. Change in Rail Travel Times Looking at the West Coast, the model predicts that, under the assumption of better rail times of 70% of those in the base case, air volumes would fall from 7.5 million trips in the base case, to 7.3 million trips in the moderate rail improve- ment scenario. Under the scenario in which rail improves to provide travel time only 50% of present times, air volumes would decrease to 6.9 million trips. Expressed as mode share, the airlines’ proportion of the air plus rail market would fall from 80% in the base case, to 76% and 83% under the two improved rail scenarios for the West Coast. Part Two: Understanding How the Model Works Model Structure The model structure was designed to incorporate suffi- cient detail in both the models and input data. It does this by using as many of the attributes available in the trans- ferred choice model specifications as possible and by using spatially detailed input data; as a result, the model is sensi- tive enough to capture the policy changes it is intended to evaluate. The general structure of the model is shown in Figure 11-11. This structure includes a set of data inputs; a set of policy inputs (future service descriptions); two pri- mary statistical models, represented by the ovals; and two sets of outputs. The model is implemented using commonly available software and is based on travel pattern data. The model has a simple graphical user interface (GUI), implemented in Microsoft’s Excel spreadsheet application, which was selected as a familiar application that the vast majority of model users will have installed on their computers. The spreadsheet con- tains worksheets to manage scenarios, edit inputs (such as air and HSR service descriptions), and to view and compare outputs. The spreadsheet uses Visual Basic macros to pro- vide additional functions for the model user, such as allowing selection of subsets of input data for viewing and editing and automating import of model results. The model computa- tions are completed using “R,” which is an open-source statis- tical programming language that is particularly useful for the type of database computation required for these models. The spreadsheet GUI means that model users do not need to use trip destinations within the State of California (ACRP Report 31), Figure 11-10 suggest that some 60% of such intra-state travelers would not be diverted to rail under Scenario #4, which reflects a robust rail but without any assumptions about change in air services. Thus, total boardings at the two airports might be expected to drop by about six percent, under the assumptions used in this early application of the model. Limitations of the Scenario Test The purposes of the exercises included in this Chapter are to explore the propensity of the new model to produce reasonable and logical results. While the general scale of diversions implied for California’s major airports seems consistent with other pub- lished analyses, the results of the modeling process in this exer- cise are not meant be used for policy purposes. Specifically, the alteration of “global” or system-wide parameters is undertaken to better understand how the model factors interact, and work together. There is no underlying assumption in this work that every travel time in the system could be improved in this man- ner; the exercise examines what would happen if such qualita- tively superior performances could be achieved. As the process of developing and applying this model pro- ceeds, the model will be used by altering input assumptions at the geographic specific level, a process through which actual physical plans could be described, inputted and analyzed. This was not attempted in this Chapter. Results of the Scenario Testing (West Coast) The West Coast exercise described herein sought to allow the examination of several factors (both about the quality of the rail service and the competitive characteristics of the air service); the model can also be used to explore one variable at a time. Change in Rail Travel Times Looking at the West Coast, the model predicts that, under the assumption of better rail times of 70% of those in the base case, air volumes would fall from 7.5 million trips in the base case, to 7.3 million trips in the moderate rail improve- ment scenario. Under the scenario in which rail improves to provide travel time only 50% of present times, air volumes would decrease to 6.9 million trips. Expressed as mode share, the airlines’ proportion of the air plus rail market would fall from 80% in the base case, to 76% and 83% under the two improved rail scenarios for the West Coast. Change in Airline Fares Looking at the West Coast, a lowering of air fares to 70% of their present level results in an increase in predicted air trav-

167 of interest is the short-haul market whereby a traveler might bypass one airport and use HSR to access another airport and make a long-distance flight outside the corri- dor (rather than an airport transfer). This transfer market is potentially relevant in both California and the northeast, given the recent policy interest in better long-distance ser- vice to JFK Airport on Long Island. However, the Research Team considers it a difficult market to understand let alone model and elected to focus solely on the primary short- haul market, implementing a model to compare one airport alternative with one rail alternative, for each OD pair with a choice. The analysis years for the model are 2008 and 2040 (Tables 11-4–11-7). These years were chosen because they are consistent with the ana lysis years for the person-trip tables developed as part of the U.S. DOT Travel Analysis Framework project (Jenkins and Vary 2010). The 2008 trip tables by air and rail were combined to reflect overall travel demand by these two modes, and they are used as a basis for base year model calibration. The trip tables do not include any market segmentation (e.g., trip purpose or demographic data), and therefore the Research Team has used a simula- tion approach to characterize travel parties by drawing from the R language or interact with R directly to run the models, but merely have it installed on their computers by the ACRP Model installation package, working with Windows, Mac OS, or Linux computers. Travel Demand Input travel demand is provided at a county-level reso- lution, while much of the model’s simulation uses a more detailed spatial resolution of Census Tracts. This represents a trade-off between the desire for additional resolution versus the reality of data availability. The observed travel demand data is currently only available at the county level for these large corridors. For example, in the output of the United States Department of Transportation’s (U.S. DOT) Travel Analysis Framework project, modal OD tables for long-distance trips are at the county level, while certain model inputs are available at more detailed spatial resolutions. The primary purpose of the model is to understand the competition between air and HSR. The largest market of interest, and the primary focus of the model, is the short- haul market with both an origin and destination in one of the corridors that form the study areas. A smaller market * Policy inputs ** Outputs Figure 11-11. The component elements of the model.

168 considering a single HSR line, the inclusion of local transit access adds an unmanageable amount of complexity given the intent of the model. Travel time and cost data were derived from several avail- able sources. Published Amtrak schedules were used for the base condition (i.e., rail frequencies and travel times). Auto access times were derived from national roadway networks available from FHWA. Transit access times to both rail and airports were identified as a particularly difficult detail to represent, primarily because there is no corridor-wide tran- sit network to rely on for this purpose. Instead, the Research Team used data from the Census journey to work data pro- gram (now the American Community Survey) to develop a comprehensive representation of the quality of transit service at a Census Tract level of spatial detail. Airport, Station, and Mode Choice Models The model framework makes use of three choice models to select a preferred airport pair, a preferred station pair, and the main mode for the trip, comparing air versus rail. For the airport and station choice models, the Research Team used models developed for the West of Hudson Regional Tran- sit Access Study, a study sponsored by the Port Authority of New York and New Jersey and MTA’s Metro-North Railroad to evaluate improved transit connections to Stewart Airport. These models were applied to consider a set of possible airport pairs for the trip being modeled and a set of stations pairs. The model also permitted the selection of a preferred pair in each case, based on the characteristics of the access and egress trips from the airports and stations at either end of the trip and the observed distributions by trip purpose and demographics. While the trip tables are published in O-D format and do not explicitly identify the home end of the trip, the tables have been processed using assumptions described in later sections to infer the origin location and derive segmenta- tion attributes based on the origin location. Representation of Transportation Supply The Research Team considered two options for represent- ing the air and rail networks in the model. The first option is akin to a traditional travel model network depicting the physical linkages and service characteristics between sta- tions and between airports. This first approach would have required that a network be included with the model that the model user could modify for scenario testing, and this net- work would represent the relative ease of traveling from place to place by each mode. The second option, which the Research Team selected for model implementation, is a simpler approach whereby the net- work characteristics are embedded in the model in the form of predefined tabular inputs. In this approach, while it is clear what the travel time or cost is from place to place, there is no network included with the model. The predefined travel times and costs and frequencies by mode that are used as inputs to the model are viewable by the model user in the GUI; in addi- tion, the model user has the ability to efficiently modify these inputs (rather than modifying networks) in order to develop and evaluate scenarios. This second approach was more straightforward because air networks are complex to represent and visualize, and although transit networks are clear when Source: FHWA Travel Analysis Framework. State Air Rail Total CONNECTICUT DELAWARE DISTRICT OF COLUMBIA MAINE MARYLAND MASSACHUSETTS NEW HAMPSHIRE NEW JERSEY NEW YORK PENNSYLVANIA RHODE ISLAND VIRGINIA WEST VIRGINIA CALIFORNIA 186,755 12,737 154,698 96,752 596,054 1,032,723 150,059 159,937 735,790 225,271 142,715 674,518 3,123 7,062,106 406,908 104,874 853,280 55,334 606,369 700,014 56,536 522,226 1,856,841 1,375,768 211,785 254,002 1,979 2,517,740 593,663 117,611 1,007,978 152,086 1,202,423 1,732,737 206,595 682,163 2,592,631 1,601,039 354,500 928,520 5,102 9,579,846 Table 11-4. Air, rail, and total roundtrips by state (2008).

169 Eight Steps in the Model Implementation Process As noted in the previous sections, the model was developed and implemented using a combination of Microsoft Excel and custom scripts developed in the R open-source statistical analysis software. The following is a step-by-step sequence that describes the model flow as implemented in the model’s code; each step is expanded upon in the Technical Appendix with dis- cussion of data sources, model structure and specification, out- puts, and sensitivity to changes in the inputs. For each scenario tested, the model will: 1. Combine the rail and air county to county flows (2008 and 2040 trip tables are preprocessed as round trips from county to county) to create total demand for rail and air service for each county pair (Figure 11-12). 2. Enumerate a population of travelers from the total demand data, sample a selection of travel parties to simulate in the remaining models, and simulate a trip purpose and party size for each traveler in the list. 3. Allocate each travel party to specific origin and destination Census tracts using an allocation model. 4. Simulate income category and vehicle availability for the travel party based on their trip purpose for the travel party and their origin Census Tract. 5. Choose the best airport and rail station for the origin and destination ends of the trip for each party using an airport choice and a station choice model. relative levels of service on the air or rail service between the airports and stations. For the mode choice models, the team identified two existing models that have similar structures and whose coefficients have been estimated using recent data: a model developed by RSG for the most recent Toronto-Montreal- Quebec City corridor HSR study under contract to Trans- port Canada and the Ministries of Transport for Ontario and Quebec (2009/2010), and the California HSR model developed by Cambridge Systematics, modified to repre- sent the air versus HSR trade-off elements. The mode choice model that was implemented is derived from the California HSR model. Airline Response Model The airline response model predicts how airline schedules will respond to the introduction of HSR service on the North- east and California Corridors. The response may include sev- eral elements. First, to the extent that HSR captures traffic that previously traveled by air, airlines may reduce seat capac- ity on affected segments. Second, either as a result of offering reduced seat capacity or as a direct response to competition from HSR, the distribution of capacity among different types of airlines—mainline network carriers, regional affiliates, and low-cost carriers—may change. Finally, either as a direct response to HSR competition, or as a secondary response to diminished traffic, carriers may change their fleet mixes in corridors with HSR service. Source: FHWA Travel Analysis Framework. State Air Rail Total CONNECTICUT 268,048 562,465 830,513 DELAWARE 25,922 152,559 178,481 DISTRICT OF COLUMBIA 280,908 1,265,506 1,546,414 MAINE 157,657 79,027 236,684 MARYLAND 992,693 907,647 1,900,340 MASSACHUSETTS 2,046,045 968,468 3,014,513 NEW HAMPSHIRE 221,086 80,237 301,323 NEW JERSEY 311,659 737,029 1,048,688 NEW YORK 1,008,999 2,569,400 3,578,399 PENNSYLVANIA 424,709 1,933,877 2,358,586 RHODE ISLAND 192,162 291,717 483,879 VIRGINIA 1,074,605 413,035 1,487,640 WEST VIRGINIA 5,164 3,223 8,387 East Coast Total 7,009,657 9,964,190 16,973,847 CALIFORNIA 11,110,091 4,281,203 15,391,294 West Coast Total 11,110,091 4,281,203 15,391,294 Table 11-5. Air, rail, and total roundtrips by state (2040).

170 6. Apply the main mode choice model to choose between traveling by air or rail, based on the characteristics of the travel party, the accessibility of the airports and rail stations at the origin and destination end of the trip, and the level of air and rail service between them. 7. Apply the airline response model to calculate a likely response to the rail service scenario in terms of air service changes by the airlines. 8. Generate data in spreadsheet summary format. Each of the eight steps undertaken within the model is summarized here. Step 1. Combine Rail and Air County to County Flows The model’s first step is to import the preprocessed county to county OD flows for air and rail, which have been converted from one-way trips to round trips as described herein. The Figure 11-12. Air mode share by state for 2008 and 2040. State Air Rail Total CONNECTICUT 44% 38% 40% DELAWARE 104% 45% 52% DISTRICT OF COLUMBIA 82% 48% 53% MAINE 63% 43% 56% MARYLAND 67% 50% 58% MASSACHUSETTS 98% 38% 74% NEW HAMPSHIRE 47% 42% 46% NEW JERSEY 95% 41% 54% NEW YORK 37% 38% 38% PENNSYLVANIA 89% 41% 47% RHODE ISLAND 35% 38% 36% VIRGINIA 59% 63% 60% WEST VIRGINIA 65% 63% 64% East Coast Total 68% 42% 52% CALIFORNIA 57% 70% 61% West Coast Total 57% 70% 61% Table 11.6. Air, rail, and total round trips by state (growth from 2008 to 2040).

171 model then combines them to create the total demand for air and rail travel in the two study areas for both 2008 and 2040. Step 2. Enumerate a Population of Travelers, Select a Sample and Simulate Trip Purpose and Party Size This step of the model converts the summary of demand from each county to total trips and then to an enumerated list of travelers. The model then selects a sample of them for further simulation. For each selected traveler, the model simu- lates a trip purpose and party size by drawing from a joint dis- tribution of trip purpose and party size derived from airport ground access survey data and rail survey data. Step 3. Allocate to Census Tracts The simulation sample that is produced by previous steps of the model and that is input into this step is comprised of a set of travel parties with roundtrip itineraries. The origin county at the start of the travel parties’ outbound trips (and their final destination at the end of their return trip) is known, along with the destination county of their outbound trip (which is the origin county of their return trip). This model step then allocates each travel party in the simulation sample to specific origin and destination Census tracts within their origin coun- ties and destination counties using an allocation model. The population data were obtained from the 2010 U.S. Census. The study areas comprise 13,273 tracts in the East Coast study area. Step 4. Simulate Income Categories and Vehicle Availability The simulation sample that is an input to the mode choice models is comprised of a set of travel parties with round trip itineraries whose origin county and Census tract at the start of their outbound trip (and their final destination at the end of their return trip) are known. Also know are their destina- tion county, the Census tract of their outbound trip, and their trip purpose. This step in the model simulates the household income category and vehicle availability for the travel party given their trip purpose to allow the subsequent choice mod- els to be applied correctly to each travel party. The additional data used by this model step describes the distributions of household income and vehicle avail- ability among air and rail travelers as well as the household income and vehicle ownership of households in the air and rail travelers home Census tracts. The data sources are as follows: • Airport ground access surveys for the New York region, New England, the Baltimore-Washington area, Los Angeles, the Bay Area, and San Diego (FAA, MCOG, MTC) • Travel survey data collected in California by California High-Speed Rail Authority. American Community Survey data for 2006–2010 at the Census Tract level describing household income and vehicle ownership, available from the U.S. Census Bureau (ACS). Source: 2010 U.S. Census. State Populaon Total Employment Hospitality Employment CONNECTICUT 3,574,097 1,575,309 106,126 DELAWARE 897,934 392,294 31,489 DISTRICT OF COLUMBIA 601,723 621,524 54,762 MAINE 478,805 224,881 20,454 MARYLAND 5,668,368 2,383,588 184,822 MASSACHUSETTS 6,547,629 2,924,913* 232,603* NEW HAMPSHIRE 819,087 372,312 28,024 NEW JERSEY 8,791,894 3,732,237 268,508 NEW YORK 13,874,816 6,025,444 398,364 PENNSYLVANIA 7,396,902 3,196,508 233,356 RHODE ISLAND 1,052,567 435,352 39,850 VIRGINIA 5,939,131 2,595,203 216,880 WEST VIRGINIA 175,208 46,465 5,041 East Coast Total 55,818,161 24,526,030 1,820,279 CALIFORNIA 36,689,600 14,285,120 1,251,945 West Coast Total 36,689,600 14,285,120 1,251,945 Table 11.7. Population, employment, and hospitality employment by state.

172 Step 5. Airport Choice and Station Choice This step in the model applies airport and station choice models to each travel party to select likely airports and rail stations for their origins and destinations. The simulation sample that is input into this step is comprised of a set of travel parties with roundtrip itineraries where their origin county and Census Tract at the start of their outbound trip (and their final destination at the end of their return trip) are known, along with the destination county and Census Tract of their outbound trip. In addition, several characteristics of the party are known: party size, trip purpose (business versus nonbusiness), income category, and vehicle availability at the origin end of their trip. The additional data required by the airport and station choice models are as follows: • Aviation data—flight availability and fares by airport pair and airport enplanements. These data were developed using publicly sourced aviation data, including T100, on-time performance data, and DB1B. • Rail data—data items parallel to the aviation data using Amtrak OD data and Amtrak schedules (Amtrak). • Transit accessibility measures for each Census Tract—derived from CTPP journey to work data. • Highway accessibility to airports and stations—developed by applying shortest path algorithm from each Census Tract within 150 miles of each airport and station to a highway network built from a national street centerline geodatabase. Tables 11-8 and 11-9 show the airport pairs in each study area with the highest levels of air service. The source for these data is the BTS Airline on-time performance data, which shows the service scheduled and actually operated, in this case for 2008. Unlike the Official Airline Guide, this is a publicly available source of airline LOS data; therefore, this is prefer- able as the data can be included in the model for distribution without licensing issues (BTS/Transtats). Rail schedule data were developed using the published Amtrak schedule (Amtrak); these data incorporate Amtrak’s connecting buses in California (this was also incorporated on the demand side into the development of the rail county to county trip tables). For the transit accessibility measure, 2000 CTPP data describing journey to work by mode was processed for all Census Tracts and counties in the study area to calculate transit mode shares for the journey to work. These data are used in the model as a proxy for transit accessibility from that Census Tract and county rather than attempting to develop transit network data, which is beyond the scope of the type of sketch planning model developed as part of this research. While the primary choice model in the framework is the main mode choice between air and rail, the model incorpo- rates a choice model to define the particular trade-off used in the main mode choice model for each travel party. The selec- tion of particular airports and rail stations can be achieved using a simple algorithm, such as selecting the closest airport and rail stations. Instead of this approach, the model incor- porates an airport choice and a station choice model into the Source: BTS Transtats, 2008 Table 11-8. Most Flights per day–West ORIGIN DEST Flights/Day LGA BOS BOS LGA DCA LGA LGA DCA DCA BOS BOS DCA JFK BOS BOS JFK PHL BOS BOS PHL BOS BWI BWI BOS BOS IAD BWI PVD PVD BWI BOS EWR 33 33 30 30 24 24 21 20 17 17 13 13 11 11 11 11 Table 11-9. Most flights per day–East.

173 framework. As a result, travel parties first choose between the available airports and then between the available rail stations at each end of their trip to select a preferred pair of airports and pair of rail stations for use in the main mode choice model. The model that is being transferred is a joint multinomial logit airport choice and ground access mode choice model with alternatives specified for each combination of airport and mode choice. It was developed by members of the Research Team for the West of Hudson Regional Transit Access Study, WHRTAS (RSG 2010). This included airport choice between the airports in the New York region. Step 6. Main Mode Choice This step in the model applies the main mode choice model to choose between traveling by air or rail, based on the char- acteristics of the travel party, the accessibility of the airports and rail stations at origin and destination end of the trip, and the level of air and rail service between them. The output produced by the airport and station choice step is comprised of a set of travel parties with round trip itineraries that include the origin and destination Census Tracts. Several characteristics of the party are known: party size, trip purpose (business vs. nonbusiness), their income category, and their vehicle availability at the origin end of their trip. Finally, the preferred airport pair for their trip and rail station pair for their trip are known, along with variables representing the accessi- bility of the preferred airport pairs and station pairs. The data used in the airport and station choice models, describing the competing air and rail levels of service and fares, are used again in the main mode choice model. Step 7. Airline Response The airline response model predicts how airline schedules respond to the introduction of HSR service. Conceptually, that response includes several elements. First, to the extent that HSR captures traffic that previously traveled by air, air- lines may reduce seat capacity on affected segments. Second, either as a result of offering reduced seat capacity or as a direct response to competition from HSR, the distribution of capac- ity among different types of airlines—mainline network car- riers, regional affiliates, and low-cost carriers—may change. Finally, either as a direct response to HSR competition, or as a secondary response to diminished traffic, carriers may change their fleet mixes in corridors with HSR service. To model these responses, the Research Team has per- formed econometric analyses on cross-sectional airline ser- vice data in order to isolate the direct and indirect effects of competition from existing rail services. This approach assumes that the impact of introducing HSR can be captured by the change in “rail competitiveness” that results; thus, the impact can be extrapolated by observing the effect of exist- ing variation in that factor. Ways of measuring this factor are discussed further in the following section. An alternative would be to expand the geographic scope of the analysis to Europe and Japan, where HSR is already present. Limits on data availability and questions about the applicability of the results to the United States meant that the Research Team did not pursue this approach. The input data used for estimation of this model are derived from the air and rail LOS data described in the Technical Appendix to this Report. The rail stations and airports in the East Coast and West Coast study areas were mapped to the common geographic regions of Metropolitan Statistical Areas (MSA). Figure 11-13 summarizes the implications of the model research concerning the relationship between the supply of rail service and the supply of air service. Data collected in the model building show rail travel times around one hour are associated with about 40% of the flights that would be expected without rail service. This effect attenuates so that a 7-hour rail service would be associated with 80% of the flights expected under a non-rail scenario. The effect disappears entirely at a rail travel time of around 11 hours. Although the coefficient estimates are significant, there remains considerable uncertainty about the actual coefficient values, and thus about the relationship shown in this diagram. As discussed in the section of Chapter 12 con- cerning further research, additional analysis will be required to reduce this uncertainty. Step 8. Generate Data in Spreadsheet Summary Format The ACRP Air/Rail Diversion Model creates a wide vari- ety of output documentation, generally filed in the folder Figure 11-13. Relationship between rail travel time and air service without rail.

174 created for the specific scenario being modeled. Thus, for every model run undertaken for a specific scenario name, the model creates subfolders with spreadsheets concerning the information used and outputted in that model run. A partial list of the output for each model run includes: 1. Geographic assumptions. 2. Socioeconomic data. 3. Rail Station access data. 4. Airport access data. 5. Rail travel data. 6. Travel summaries for the present time frame. 7. Travel summaries for the future time frame. 8. For both the present time frame and future time frame output set, the model creates: 77 An airport-by-airport summary. 77 Results of the air response sub-procedure. 77 Total trips by county, by mode. 77 Total trips by SMSA, by mode. 77 Origin-destination summaries by county, by mode. 77 Origin-destination summaries by SMSA, by mode. Validation of the Model: At the City Pair Level A key element of the calibration of a model is the somewhat iterative process of “fitting” the results of the model to see how, at a given point of iteration or development, the model is replicating the conditions it was calibrated to represent. One of several examples of this validation process included in the Technical Appendix is reproduced here, in Figure 11-14. It shows a wide variety of cases in which the prediction comes close to the observed values, and suggests that some of these city pairs may have causal patterns that the general model has trouble dealing with, such as its propensity to under-predict volumes for the city pair between Los Angeles and San Jose. The interested reader is encouraged to explore the concept of validation more in the Technical Appendix. Reasonableness of the Model at the Policy Level The practitioner interested in the strengths and weaknesses of the model might also be interested in its more global appli- cation to the question of the importance of given attributes in the prediction of change in mode share for rail in the air + rail market. The Research Team undertook a series of analyses to cross check the basic level of “reasonableness” of the model for rail line haul times, the variable analyzed in Chapters 4 and 5 of this Report. Those two chapters presented a set of outcome mode shares together with total in-rail-vehicle trip times. The results of that test of reasonableness are described in the Technical Appendix. An example of the tests under- taken is here presented as Figure 11-15. The model tends to show that, when rail travel times are extremely short, the model tends to replicate the European experience of high rail share of the rail + air market. However, the European data set had fewer observations about longer train travel times. Thus, the model tends to predict a mode share of about 5% to most rail trips of over 7.5 hours. The Research Team does not find this to be unrea- sonable (e.g., Boston to Washington, DC) as there evidently 0 200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 San Francisco TO Los Angeles San Jose TO Los Angeles Los Angeles TO San Jose Los Angeles TO San Diego San Diego TO San Francisco Los Angeles TO Sacramento Riverside TO San Francisco Sacramento TO San Francisco Sacramento TO San Diego Comparison of Air Trips for West Coast CBSA pairs with >100,000 Observed Air + Rail Trips in 2008 Observed Air Trips Modeled Air Trips Figure 11-14. Results of the validation process.

175 is a loyal residual market of travelers who do not choose to fly, and they are, by definition not very time-sensitive at these travel times. The reader might well conclude that the model is more relevantly applied to situations where the rail travel times are more competitive with air. In the important area around 3.5 hours of rail travel time, there is a considerable variation in the European experience for mode share. Within that category of rail travel times, the model tends to veer towards the higher range of the observed experience. The reader is cautioned that significant variations in these mode share calculations can arise from variation in assumptions about the geographic size of the market area of origin and of destination; the reader is encouraged to review the analysis included in the Appendix to the SDG report (2006) upon which the European shares are based. That analysis suggests that the larger the market zone size, the further a candidate traveler might be from the central business district and the lower the reported rail mode share will be. In short, a given reported mode share (from either continent) will be highly affected by assumptions made about the geographic scale of market shares considered. The level of variation in the accuracy of any reported mode share could be taken into consideration when comparing results. Summary: Application of the Model The ACRP Air/Rail Diversion Model was designed to pro- vide a model that incorporates several improvements over the tools that are currently available to support integrated air and rail service planning: • The model is accessible and relatively easy for planners and analysts to use, especially when compared to the much more detailed models that have been used for new rail ser- vice planning and environmental documentation. • The model incorporates data on ground origin to ground destination air travel patterns, merging ground access pat- terns with airport-to-airport flows. • The model includes accurate data on existing intercity rail passenger flows (from Amtrak). • The model incorporates an airline response module that reflects the effects of rail/HSR services on flight volumes. This will also support HSR planners to estimate the cor- responding benefits to airports of new rail service. • The model produces output in a spreadsheet format (comma delimited from Excel) that is easily transferable to other programs for analysis (e.g., SPSS or SAS) or graphic presentation (PowerPoint or MS Publisher.) In the same process, the model produces spreadsheet summaries of the results of calculations made while running the sub- modules of the model. Uses of the Model Presented in This Chapter Chapter 11 has provided an overview of the model; it specifically illustrated how the model could be used for the examination of broad policy issues on a high level basis, and introduced the concept of tests for model validation and assessment of reasonableness of policy results. As described in much more detail in the Technical Appendix, the model can be further utilized by variation of many input parame- ters, including project specific alterations in speeds and travel times for geographically specific networks and segments of networks, down to the level of city-to-city pairs. Further Development Possibilities The model has been prepared using open-source software methods, and is specifically designed to make use of data that is provided by the agencies of the U.S. DOT. As such, practitioners can refine the model to change a wide variety of input assumptions, all of which are presented transpar- ently the program’s application. Different practitioners could take the model in different directions, based on these open- source characteristics. This further development work could be undertaken in later research projects, including possibly some within the Cooperative Research Programs. The pur- pose of Chapter 11 has been to explore the use of the model to better understand the inter-relationship between key fac- tors in both the East and West Coast project study areas. The possible directions for later research are further discussed in Chapter 12. Bibliography American Community Survey (ACS), data for 2006–2010 at the Census Tract level Census Bureau, http://www.census.gov/acs. Figure 11-15. An example of a reasonableness test for global attributes of rail.

176 Amtrak, Schedule Information, www.amtrak.com. Bureau of Transportation Statistics, Transtats Airline Data, http://www. transtats.bts.gov/DatabaseInfo.asp?DB_ID=120&Link=0. California High-Speed Rail Authority, Travel Survey Data accessible at http://www.hsr.ca.gov/Programs/Environmental_Planning/ EIR_EIS/index.html. Federal Aviation Administration. (No date) “The New England Regional Airport System Plan.” Sponsored by the New England Airport Coalition. Accessed at http://www.faa.gov/airports/new_ england/ planning_capacity/airport_system_plan/media/nerasp_ complete.pdf. Federal Aviation Administration. 2007 (May). “FAA Regional Air Ser- vice Demand Study, Summary Report.” Port Authority of New York & New Jersey, New York State Department of Transportation, and Delaware Valley Regional Planning Commission. Jenkins, D., and D. Vary. 2010 (November 5). “FHWA Traveler Analysis Framework, Part II, Status Report.” AMPO Travel Modeling Meet- ing, Washington, DC. Metropolitan Transportation Commission, Airport Survey Data, as archived, Oakland, CA http://dataportal.mtc.ca.gov/data- repository. aspx. Metropolitan Washington Council of Governments, Airport Surveys, https://www.mwcog.org/transportation/activities/airports/ documents/aps11_general_findings_report_final.pdf. Resource Systems Group. 2010. West of Hudson Regional Transit Access Study. Air Passenger Model Documentation (DRAFT), prepared for the Port Authority of New York and New Jersey. Unpublished. Steer Davies Gleave. 2006 (August). “Air and Rail Competition and Complementarity” Prepared for the European Commission DG TREN. London, UK.