Travel Demand Forecasting: Parameters and Techniques (2012)

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.

7 The purpose of developing travel forecasting models is to provide information that can be used to make transportation planning decisions. These decisions may require different kinds of information from the model, depending on the context. The planning context, therefore, should be used to determine the appropriate model structure, parameters, and complexity. This decision, in turn, will ensure that the travel forecasting model is appropriate for each planning context. It is useful to develop a travel forecasting model that meets most (if not all) of an agency’s current and future planning needs. This chapter discusses how the planning context affects the model’s capa- bilities and provides examples of different contexts found in U.S. urban areas. 2.1 Types of Planning Analyses The transportation planning function covers a diverse set of activities that focuses on different transportation modes and systems, timeframes, geographic scales, policy issues, and stakeholder groups. It is critical to gather input from a broad cross section of stakeholders on the types of policy consid- erations and modal analyses that need to be accounted for in the travel demand model prior to its development. Many planning requirements are directed by federal legislation, such as long-range transportation planning and air quality planning. Federal guidelines and regulations regarding transportation planning are summarized by agency in Appendix A. Planning practices for these requirements are generally consistent across areas of the same population. However, many other aspects of particular planning processes reflect state and local require- ments, and actual planning practice varies widely. Many of these transportation planning functions require forecasts of future travel or other model outputs to aid in evaluating the benefits of different plan elements and different plans. The type of analysis being performed guides the design of models and the necessary features required to produce suitable forecasts for decision making (project prioritization, for example). Typical types of transportation planning that require travel forecasts are discussed in the following sections. The planning types are adapted from “Planning and Asset Management” (FHWA, 2009b). 2.1.1 Establishing System Performance Measures The identification of individual performance measures depends on the complexity of the measures, as well as the size and characteristics of the transportation system. Standard metrics, such as vehicle-miles of travel, vehicle-hours of travel, link-based volume-to-capacity ratios, and travel speeds, can be produced by nearly all models, and some of these measures are used in model validation. (However, a model’s ability to produce an output metric does not in itself mean that the model has been validated for that metric, and due care should be taken using the results.) More advanced metrics such as travel time reliability; intersection-based, area-based, or multi- modal levels of service; hours of delay; or hours of conges- tion require both the input data and the model functions to calculate the measure for both a current base year and any horizon years. For example, a model that produces only daily traffic assignments will be unable to produce the data for calculating hours of delay without significant modifications. Transportation system performance measurement is a sig- nificant stand-alone topic related to the travel demand fore- casting process, but too great to cover in the context of this report. NCHRP Synthesis 311 (Shaw, 2003) and TCRP Report 88 (Kittelson and Associates et al., 2003) provide a starting point for understanding the development and application of per- formance measures. 2.1.2 Long-Range Transportation Planning Federal statutes require an MPO to prepare a long-range transportation plan (LRTP) and set forth many of the planning C h a p t e r 2 Planning Applications Context

8guidelines. Chief among these is a typical planning horizon of 20 to 30 years. This is not to say that other horizon years cannot be modeled, but the reliability of forecasts with a planning horizon of more than 30 years is highly questionable. Forecasts of less than 20 years may be appropriate for many of the types of planning activities listed below. In general, for long-range planning, the model must be capable of analyzing, with reasonable accuracy, the impacts of projects that are included in the LRTP. The types of projects included, of course, vary depending on the characteristics of the urban area and its transportation system. In a large urban area, the plan is likely to include both highway and transit projects; therefore, the model must be capable of analyzing the impacts of projects of all travel modes. If road pricing projects are being considered, the model should be capable of considering the effects of price on travel demand. More detail on the required model features for several project types is provided in Sections 2.1.4 through 2.1.8. If only a limited number of types of projects are included in the LRTP, which is often the case in smaller urban areas, a sim- pler modeling approach may be appropriate—unless the model is required to perform other analyses outside the long-range planning context that require additional modeling capabilities. 2.1.3 Policy Planning and Analysis Tests of different policies can range from simple to complex over several dimensions. Modeling changes in population or employment growth rates require different data than do more complex scenarios, such as congestion pricing, changes in parking costs, fuel costs, assumption of realized mode split targets, or changes in high-occupancy vehicle (HOV) policies. Forecasts for all of the above types of analyses are often con- ducted for a series of both short- and long-term horizon years. For any of these tests, a more robust model set than one used for typical LRTP preparation is required. The constituencies of many MPOs are already demanding that many of these policies be considered as part of the LRTP development, so the model functionality required to perform these types of analyses is present in many agencies, and quickly being added by others. While it may not always be possible to anticipate all of the specific policies that the model may be used to analyze, it makes sense for model developers to consult with other planners and decision makers who may request certain types of analyses. It is important for the model to include the necessary features to support the analyses required for the policies being examined. If pricing is being analyzed, vari- ables reflecting the pricing of various transportation options (tolls, parking, transit fares, etc.) must be included. If alterna- tive land use patterns are analyzed, then variables reflecting land use patterns, such as density and diversity of development, should be included. 2.1.4 Regional and Corridor Planning This type of analysis requires greater disaggregation of inputs within the study area, particularly for corridor planning. Facilities that might not be coded in a full regional travel net- work because they have a lower functional classification must be included for a corridor study, if observed data indicate the volume of traffic using the facilities is relevant to analyzing the corridor. Historically, subarea models have been devel- oped for regional and corridor planning, where the level of detail of the transportation system represented by the networks is finer in the area of interest. Many current models already have a fine level of detail throughout the model area. It may be worthwhile to consider having a fine level of resolution appro- priate for regional and corridor planning throughout the entire model, especially in smaller urban areas where the computation and model run time implications of a detailed model are not as likely to be severe. Small- and medium-sized agencies, in par- ticular, must balance this consideration against their available resources to support model development and application. 2.1.5 Project Planning and Development Forecasting the impacts of transportation projects or invest- ments (and land development projects) is even more focused than corridor planning and requires a corresponding sharper focus and disaggregation of inputs and sometimes outputs. In many project planning studies, it is now common for a refined and study area-focused travel demand forecasting model to be one step in a larger forecasting effort that may take the output model forecasts and subsequently use them as inputs to mesoscopic or microscopic dynamic traffic assign- ment (DTA) or microscopic travel simulation. In these cases, the model must be able to produce compatible outputs. Even if DTA or microsimulation is not employed for project plan- ning, it is almost inevitable that some sort of post-processing of model results must occur. It is reasonable to assume that for most projects, including studies of specific transportation improvements, either independently or as part of specific land development projects (i.e., traffic impact studies), some analysis will be conducted at the intersection level, requiring model output to be post-processed to produce reasonable intersection volumes and turning movements. This is not to say that a model is required for all such analyses; many traffic impact studies, particularly those looking at short-term fore- casts, use simpler analytical methods to produce forecasts that do not require a model. 2.1.6 Transit Planning At a minimum, forecasts for transit planning require a mode choice model and a transit network, with path building,

9 skimming, and transit assignment capabilities. [“Skimming” sums impedances along selected paths identified as the route or path on the transit network that has the lowest cost for a traveler. Depending on the model structure, cost may be actual dollar values (fares) or monetized values of time, distance, or a combination of these and other price components.] A mode choice model, however, can have one of several differ- ent forms and specifications, ranging from a diversion table based on local survey data and a reasonable annual growth factor to a more complex nested logit structure. Regardless of the model form, the mode choice model and the entire model chain must be able to address the existing and potential new markets for transit in the study area, both regionally and for specific projects. Transit project planning, where the project may use the FTA capital funds, has its own series of guidelines and requirements, but the FTA has been careful to avoid being prescriptive about model specifications and forms when issuing guidance, focusing instead on the properties of good modeling prac- tices. Many of these properties focus on quality assurance and quality checks and rigorous model testing to ensure reliable results; these are characteristics of all good forecasts, not just those related to transit projects. The guidelines and require- ments increase based on the potential level of federal capital investment in the project: from lowest to highest, these pro- grams are currently known as Very Small Starts, Small Starts, and New Starts. Much of the current FTA guidance on model properties is included in Appendix A. As with certain types of short-term highway forecasts, forecasts for short-range transit service planning also use analytics that do not require a traditional model. 2.1.7 Road Pricing and Managed Lanes Various aspects of pricing enter into the estimation of travel demand, including tolls, transit fares, parking costs, and auto/truck operating costs, which include fuel costs. This means that, to produce accurate demand forecasts, the model must be properly sensitive to the effects of price on travel demand. This type of sensitivity might require inclusion of price in all relevant travel choice components [mode, route (i.e., assignment); destination (i.e., trip distribution); time of day, etc.], as well as precise representation of time-cost trade- offs, which requires accurate estimates of travelers’ values of time. It also may require nonconstant implied values of travel time or at least market segmentation to approximate varying values of time. Some types of projects, including congestion pricing and projects where peak spreading is likely to be an issue, may require detailed time-of-day model components. HOV lanes and carpooling incentives are analyzed in some areas using travel models. This type of analysis requires identification of roadways in the model network that require minimum occupancy levels and trip tables corresponding to each occupancy level allowed to use particular facilities. The mode choice model, therefore, must be capable of outputting these trip tables; and the highway assignment must be capable of assigning HOVs and low-occupancy vehicles to the appro- priate facilities. If facilities such as HOT lanes are to be analyzed, the model must include the capabilities of both HOV and pricing analysis. 2.1.8 Nonmotorized Transportation Planning A variety of analysis techniques is in use to forecast non- motorized travel. Several factoring methods and sketch- planning techniques, such as aggregate demand models, have been employed to address planning needs. (At the time this report was being prepared, NCHRP Project 08-78, “Estimating Bicycling and Walking for Planning and Project Develop- ment,” was under way, with a report expected by fall 2012.) The number of agencies fully integrating nonmotorized (bicycle and pedestrian) modes into travel demand fore- casting is still small; however, there is continued interest in including nonmotorized treatment as part of good planning practice. Several approaches to incorporating nonmotorized travel into regional travel demand forecasting models are in use. Many major urban areas include nonmotorized travel in their trip generation models. Some agencies then imme- diately apply factors or models to separate motorized from nonmotorized travel. Other agencies carry nonmotorized travel through trip distribution and mode choice, employing a model that includes nonmotorized modes and delivering as outputs trip tables by mode and purpose. Most such models do not include assignment procedures for nonmotorized trips. Typically, the highway network is used as the basis for both walk and bicycle trips, excluding facilities such as freeways, where pedestrians and bicycles are prohibited. Some areas, however, have opted to develop pedestrian or bicycle networks, at least for some parts of the model region. 2.1.9 Freight Planning At a minimum, an area planning to produce forecasts for freight will need truck modeling procedures incorporated within the model chain. Areas that observe significant truck traffic should model trucks separately, since passenger mod- eling procedures are not designed to accurately forecast truck movements. At least three classes of vehicles could be considered: 1. Trucks carrying freight; 2. Trucks not carrying freight (for example, service vehicles); and 3. Other modes of freight transportation (for example, trains).

10 Most urban transportation planning contexts are concerned primarily with Classes 1 and 2, although certain specialized studies, such as port or freight terminal studies, may require information on Class 3. A truck model that considers Classes 1 and 2 is, therefore, the most common type of truck/freight model found in urban travel models. The truck trip tables cre- ated by the process are assigned along with autos in the highway assignment stage. Estimates of demand for Classes 1 and 3 could be derived from a multimodal freight model, but this is difficult in urban areas since a high percentage of regional freight movements has an origin and/or destination outside the modeled area. In some states, a statewide freight model might be available to produce estimates of demand for vehicle Classes 1 and 3. However, a multimodal freight model does not consider vehicle Class 2, and so these truck trips must still be estimated. 2.1.10 Land Use Planning The “transportation-land use” connection is a complex issue that continues to be the subject of a significant amount of research. There are several land use-transportation models that are fully integrated with travel demand models. These models consider the effects of accessibility on land use and location decisions, since travel conditions ultimately impact these choices. While there is no consensus on the best type of land use-transportation model to use, most large urban areas and many smaller ones have integrated some sort of land use modeling process. Land use models have their own data requirements and must be estimated, calibrated, and validated in a process separate from the travel demand model (Parsons Brinckerhoff Quade and Douglas, Inc., 1999). 2.1.11 Environmental Planning While air quality planning has been established for some time by federal conformity requirements for MPOs, other areas, such as energy planning and carbon footprint fore- casts, are still emerging at this time. All are interrelated with the transportation system, but the needs for forecasts are still being developed (or not well understood). Air quality planning can be performed at the regional and corridor level with the use of programs, such as MOBILE, MOVES, and EMFAC [the first two programs were developed by the U.S. Environ- mental Protection Agency (EPA), and the latter was developed for use in California]. These programs, however, generally require more infor- mation than typical travel models produce. Such information includes fleet estimates by vehicle size and fuel type; traffic volume and speed information by hour of the day; the oper- ating modes of vehicles (cold start, running exhaust) at dif- ferent points in the trip; and external factors such as climatic conditions. To produce the required information, many urban areas use “post-processor” programs to convert model outputs to the required format for input into the air quality analysis program. In addition to regional air quality, global climate change and related energy issues are now considered as part of environmental planning within the transportation con- text, and an increasing number of agencies explicitly model greenhouse gas (GHG) emissions at a project level [see ICF International (2008) and John A. Volpe National Transpor- tation Systems Center (2009)]. It is likely that some of the guidance on these subjects may become formalized as part of the metropolitan planning process during the next federal reauthorization cycle. Transferable parameters are more useful for some types of transportation planning than for others. If an area is calibrating a model for long-range transportation planning, land use planning, corridor planning, project site planning, or subarea planning that does not include the evaluation of transportation demand management (TDM) or more than minimal transit service, then transferable parameters are use- ful for calibrating models that will forecast motorized vehicle use. If planning is required to determine the impact of TDM measures or the diversion of automobile trips to other modes, then transferable parameters may be of reduced value. Other approaches, such as sketch-planning methods, may be of more use for these types of planning [see TCRP Report 95 (Pratt et al., various years 2003 to 2011) and Cambridge Sys- tematics, Inc. (2000)]. 2.2 Urban Area Characteristics Affecting Planning and Modeling Independent of the type of planning analysis to be performed, many urban area characteristics (e.g., population, employment, density) greatly impact both planning and modeling. Some of these characteristics are discussed in this section, and many of TCRP Report 73: Characteristics of Urban Travel Demand (Reno et al., 2002) presents a comprehensive set of tables on various aspects of urban travel demand assembled based on data from an MPO survey, the Highway Perfor- mance Monitoring System, the National Transit Database, and the 1995 National Personal Travel Survey, including demographics, vehicle ownership, trip generation by mode and trip purpose, trip generation by characteristics or origin and destination, trip making by time of day, truck trip parameters, utilization of facilities, parking, and tele- commuting. Although the tables in TCRP Report 73 contain information largely from the 1990s, it does continue to help illustrate differences among specific metropolitan areas for many of the recorded measures.

11 them directly inform planning and modeling requirements as set forth by federal planning regulations, which are discussed in detail in Appendix A. 2.2.1 Population and Demographics Population size (greater than 50,000) is one of the urban area indicators that helps establish the formation of an MPO and the subsequent planning and modeling requirements. A separate threshold of 200,000, along with other guidelines, designates a transportation management area (TMA) and creates additional requirements. In general, the greater the population of an urban area, the more complex are the trans- portation issues, and thus the planning and modeling efforts. However, population size is not the only issue; in fact, other demographic indicators such as income, race, gender, non- native status, English as a second language, and household size all have potential impacts on aspects of travel considered in the forecasting process. Many of these characteristics are among the most common variables used in trip generation, trip distribution, and mode choice models. The average age of the population has been increasing for many years and is expected to continue to do so for the fore- seeable future. The aging of the population has significant effects on travel behavior, including the percentage of work- related travel, auto mode share, and time of day of travel. The rate of change in the age of the population differs among urban areas, and analysts should be aware of the expected trends in their regions. 2.2.2 Employment and Housing and Other Land Uses The types, location, and concentration of housing and employment are key factors in an urban area’s travel patterns. For work travel, a significant number of trips flow from home to work in the morning and the reverse in the evening. But as work hours change based on economic and travel conditions and the types of jobs in an area, and as both work and home locations become more dispersed, the travel flows become less temporally and geographically regular. This, in turn, affects nonwork travel traditionally made during off-peak periods. A travel demand model in such an area (or in a region with many such areas) would require the ability to forecast off- peak trips, and ideally would include observed off-peak and nonwork travel data for use in validation. Urban areas vary in terms of the proportion of employment located in the central business district (CBD). The amount of centralization of employment in CBDs and other major activity centers, along with the size of the region, can impact travel behavior such as trip distance, time of day, and trip chaining. 2.2.3 Geographic Size As with population size, increases in the geographic size of an urban area usually mean more complex planning and modeling issues. But it is also dependent on the land use and the density associated with the geography. All other features being equal, a large area of relatively uniform land uses and densities is more likely to produce uniform travel patterns (that is, little variability in trip purposes, time-of- day distribution, travel modes, trip distances, and other travel characteristics) than a smaller area with diverse land uses and densities. 2.2.4 Development Density, Diversity, Design, and Destinations The “four Ds” of development—density, diversity, design, and destinations—can have many different effects on planning and modeling. Population (through housing) and employ- ment density are indicators of land use intensity and, in many urban areas, are accompanied by improved pedestrian ame- nities, such as sidewalks, and transit options. Land use mix, or diversity, can affect motorized trip making; areas with greater mix often permit a wider variety of needs to be sat- isfied without needing to drive. Urban design elements, such as street pattern, block size, sidewalk coverage and continuity, and pedestrian and transit amenities, can support higher levels of walking and transit use [see TCRP Report 95, Chapter 15, “Land Use and Site Design” (Pratt et al., 2003), and Chapter 17, “Transit Oriented Development” (Pratt et al., 2007)]. Acces- sibility to a variety of destinations can affect mode shares, trip lengths, and trip chaining. Higher densities mean more people in the same unit of area, and so the number of person trips would be expected to also be greater. However, this concentration of trip ends can be more efficient to serve with good transit service and nonmotorized transportation facilities leading to differences in the type of travel mode, as compared with less dense areas. Level of density is one of the key indicators used for developing area types in travel forecasting models, and the use of such area types is discussed in Chapter 4. 2.2.5 Natural Geography Any natural feature that creates a travel barrier—from moun- tain passes to water crossings to buildable versus un buildable land (not determined solely by regulation)—affects plan- ning and modeling. Such barriers create good locations for screenlines to be used in model validation and must be key targets for practitioners to model accurately, since the facili- ties crossing them are likely to be high-profile choke points in the regional transportation system. One difference in this

12 category is coastal versus inland urban areas. (The research team preparing this report tested a relationship between coastal and inland areas and travel characteristics using the 2001 NHTS data during initial data development for this report but found no significant relationship. Such a comparison could still be tested with local data, if available.) 2.2.6 Geographic Location within the United States Growth and population shifts in the United States since 1945 (excluding international immigration) have generally followed a north-to-south, east-to-west flow. “Newer” urban areas, such as Phoenix and Charlotte, have different travel characteristics than older areas, such as Boston and Philadelphia. Some differences may be evident on a mega-regional level as well: travelers may behave differently in the Southwest than the Northeast, or in the Midwest compared with the East Coast and West Coast. 2.2.7 Climate and Climate Change Prolonged periods of extreme temperatures, either hot or cold, can have an impact on planning and modeling, particu- larly if the climate results in degradation of or limitations to the transportation system. As noted in Section 2.1.11, global climate change and its impacts (such as rising sea levels) are now also a consideration in the planning and modeling pro- cess. However, these still-developing environmental models are considering time horizons beyond the current capabilities of travel forecasting models, so caution should be exercised when selecting analysis tools. 2.2.8 Resort/Nonresort Visitors Resort areas that experience a significant number of visitors as a percentage of their total travelers—Las Vegas and Orlando, for example—may have different travel characteristics than areas with fewer visitors. Whether the visitors to the area tend to stay for a single day or multiple days is also an issue. 2.2.9 Presence of Alternative Transportation Modes The presence of (or desire for) modes other than single- occupant vehicles (SOV) means an urban area should consider mode choice modeling. The complexity and specifications are dependent on the type of mode and type of analysis. The introduction of new fixed-guideway transit into an area has been a frequent application of transferable parameters for use in mode choice estimation, calibration, and validation. 2.2.10 Highway Network and Travel Conditions Highway mileage, both overall and by functional class, and area travel conditions may lead to different requirements for planning and modeling. Areas with significant congestion will likely need to employ travel time feedback in their models to ensure that they are accurately reflecting the effects of congestion on travel behavior. Less congested areas, where more travel is on arterials rather than freeways, will have different considerations when developing volume-delay functions for their models. One indicator of congestion that can differentiate urban areas is the Annual Urban Mobility Report (mobility.tamu.edu/ums/). 2.2.11 External and Through Travel The level of external and through travel for an urban area can affect travel conditions and may be a consideration in planning and modeling. Areas with significant through travel may be especially concerned with ways to explore diverting that through travel away from the region to help “free up” congested highways. Regions with large external travel com- ponents may need to take particular care in coordinating with neighboring jurisdictions to ensure that necessary current year data are available and that reasonable assumptions are made about future year conditions. 2.2.12 Land Use Control and Governance The ability to regulate land uses, and at what level of geography, can have an impact on planning and the type of modeling required to test future changes. An urban area with a regional government and an urban growth boundary may have different travel characteristics than an urban area with weak counties and home-rule, with local land use control in the hands of hundreds of small municipal civil divisions, such as boroughs, townships, and other municipalities. The latter case is likely to make realization of aggressive shifts in future land use difficult to achieve even if they are modeled well, so planners should consider an appropriate level of land use sensitivity/modeling as they are building their travel fore- casting model. 2.2.13 Presence of Special Generators Small- and medium-sized urbanized areas that include a major university typically have different travel patterns than similar sized cities without a large campus. Presence of a large university indicates a relatively large number of young adults in the region, likely resulting in a larger percentage of school-

13 related trips and part-time retail worker trips outside the peak period and potentially a larger share of bicycle, walking, and transit trips than other similar sized areas. The presence of a state capital can also potentially impact travel patterns when compared against a similar sized city with a higher proportion of manufacturing employment. A large state worker labor force could result in additional nonhome- based travel out to lunch and running errands; whereas, factory workers typically have minimal mobility while on a time clock. Cities with very large hub airports also have different trip characteristics reflected in a larger catchment area for their customers and a significant number of travelers spending the night at hotels in proximity to the airport property. If the airport is a freight hub, it is expected that truck traffic would potentially be higher than otherwise similar urban areas.